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 "nfit.h"
  27
  28 /*
  29  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  30  * irrelevant.
  31  */
  32 #include <linux/io-64-nonatomic-hi-lo.h>
  33
  34 static bool force_enable_dimms;
  35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  37
  38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
  39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
  40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
  41
  42 /* after three payloads of overflow, it's dead jim */
  43 static unsigned int scrub_overflow_abort = 3;
  44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
  45 MODULE_PARM_DESC(scrub_overflow_abort,
  46                 "Number of times we overflow ARS results before abort");
  47
  48 static bool disable_vendor_specific;
  49 module_param(disable_vendor_specific, bool, S_IRUGO);
  50 MODULE_PARM_DESC(disable_vendor_specific,
  51                 "Limit commands to the publicly specified set");
  52
  53 static unsigned long override_dsm_mask;
  54 module_param(override_dsm_mask, ulong, S_IRUGO);
  55 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  56
  57 static int default_dsm_family = -1;
  58 module_param(default_dsm_family, int, S_IRUGO);
  59 MODULE_PARM_DESC(default_dsm_family,
  60                 "Try this DSM type first when identifying NVDIMM family");
  61
  62 LIST_HEAD(acpi_descs);
  63 DEFINE_MUTEX(acpi_desc_lock);
  64
  65 static struct workqueue_struct *nfit_wq;
  66
  67 struct nfit_table_prev {
  68         struct list_head spas;
  69         struct list_head memdevs;
  70         struct list_head dcrs;
  71         struct list_head bdws;
  72         struct list_head idts;
  73         struct list_head flushes;
  74 };
  75
  76 static guid_t nfit_uuid[NFIT_UUID_MAX];
  77
  78 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  79 {
  80         return &nfit_uuid[id];
  81 }
  82 EXPORT_SYMBOL(to_nfit_uuid);
  83
  84 static struct acpi_nfit_desc *to_acpi_nfit_desc(
  85                 struct nvdimm_bus_descriptor *nd_desc)
  86 {
  87         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
  88 }
  89
  90 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  91 {
  92         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  93
  94         /*
  95          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  96          * acpi_device.
  97          */
  98         if (!nd_desc->provider_name
  99                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
 100                 return NULL;
 101
 102         return to_acpi_device(acpi_desc->dev);
 103 }
 104
 105 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
 106 {
 107         struct nd_cmd_clear_error *clear_err;
 108         struct nd_cmd_ars_status *ars_status;
 109         u16 flags;
 110
 111         switch (cmd) {
 112         case ND_CMD_ARS_CAP:
 113                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 114                         return -ENOTTY;
 115
 116                 /* Command failed */
 117                 if (status & 0xffff)
 118                         return -EIO;
 119
 120                 /* No supported scan types for this range */
 121                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 122                 if ((status >> 16 & flags) == 0)
 123                         return -ENOTTY;
 124                 return 0;
 125         case ND_CMD_ARS_START:
 126                 /* ARS is in progress */
 127                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 128                         return -EBUSY;
 129
 130                 /* Command failed */
 131                 if (status & 0xffff)
 132                         return -EIO;
 133                 return 0;
 134         case ND_CMD_ARS_STATUS:
 135                 ars_status = buf;
 136                 /* Command failed */
 137                 if (status & 0xffff)
 138                         return -EIO;
 139                 /* Check extended status (Upper two bytes) */
 140                 if (status == NFIT_ARS_STATUS_DONE)
 141                         return 0;
 142
 143                 /* ARS is in progress */
 144                 if (status == NFIT_ARS_STATUS_BUSY)
 145                         return -EBUSY;
 146
 147                 /* No ARS performed for the current boot */
 148                 if (status == NFIT_ARS_STATUS_NONE)
 149                         return -EAGAIN;
 150
 151                 /*
 152                  * ARS interrupted, either we overflowed or some other
 153                  * agent wants the scan to stop.  If we didn't overflow
 154                  * then just continue with the returned results.
 155                  */
 156                 if (status == NFIT_ARS_STATUS_INTR) {
 157                         if (ars_status->out_length >= 40 && (ars_status->flags
 158                                                 & NFIT_ARS_F_OVERFLOW))
 159                                 return -ENOSPC;
 160                         return 0;
 161                 }
 162
 163                 /* Unknown status */
 164                 if (status >> 16)
 165                         return -EIO;
 166                 return 0;
 167         case ND_CMD_CLEAR_ERROR:
 168                 clear_err = buf;
 169                 if (status & 0xffff)
 170                         return -EIO;
 171                 if (!clear_err->cleared)
 172                         return -EIO;
 173                 if (clear_err->length > clear_err->cleared)
 174                         return clear_err->cleared;
 175                 return 0;
 176         default:
 177                 break;
 178         }
 179
 180         /* all other non-zero status results in an error */
 181         if (status)
 182                 return -EIO;
 183         return 0;
 184 }
 185
 186 static int xlat_nvdimm_status(void *buf, unsigned int cmd, u32 status)
 187 {
 188         switch (cmd) {
 189         case ND_CMD_GET_CONFIG_SIZE:
 190                 if (status >> 16 & ND_CONFIG_LOCKED)
 191                         return -EACCES;
 192                 break;
 193         default:
 194                 break;
 195         }
 196
 197         /* all other non-zero status results in an error */
 198         if (status)
 199                 return -EIO;
 200         return 0;
 201 }
 202
 203 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 204                 u32 status)
 205 {
 206         if (!nvdimm)
 207                 return xlat_bus_status(buf, cmd, status);
 208         return xlat_nvdimm_status(buf, cmd, status);
 209 }
 210
 211 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 212                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 213 {
 214         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
 215         union acpi_object in_obj, in_buf, *out_obj;
 216         const struct nd_cmd_desc *desc = NULL;
 217         struct device *dev = acpi_desc->dev;
 218         struct nd_cmd_pkg *call_pkg = NULL;
 219         const char *cmd_name, *dimm_name;
 220         unsigned long cmd_mask, dsm_mask;
 221         u32 offset, fw_status = 0;
 222         acpi_handle handle;
 223         unsigned int func;
 224         const guid_t *guid;
 225         int rc, i;
 226
 227         func = cmd;
 228         if (cmd == ND_CMD_CALL) {
 229                 call_pkg = buf;
 230                 func = call_pkg->nd_command;
 231         }
 232
 233         if (nvdimm) {
 234                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 235                 struct acpi_device *adev = nfit_mem->adev;
 236
 237                 if (!adev)
 238                         return -ENOTTY;
 239                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
 240                         return -ENOTTY;
 241
 242                 dimm_name = nvdimm_name(nvdimm);
 243                 cmd_name = nvdimm_cmd_name(cmd);
 244                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 245                 dsm_mask = nfit_mem->dsm_mask;
 246                 desc = nd_cmd_dimm_desc(cmd);
 247                 guid = to_nfit_uuid(nfit_mem->family);
 248                 handle = adev->handle;
 249         } else {
 250                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 251
 252                 cmd_name = nvdimm_bus_cmd_name(cmd);
 253                 cmd_mask = nd_desc->cmd_mask;
 254                 dsm_mask = cmd_mask;
 255                 if (cmd == ND_CMD_CALL)
 256                         dsm_mask = nd_desc->bus_dsm_mask;
 257                 desc = nd_cmd_bus_desc(cmd);
 258                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 259                 handle = adev->handle;
 260                 dimm_name = "bus";
 261         }
 262
 263         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 264                 return -ENOTTY;
 265
 266         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
 267                 return -ENOTTY;
 268
 269         in_obj.type = ACPI_TYPE_PACKAGE;
 270         in_obj.package.count = 1;
 271         in_obj.package.elements = &in_buf;
 272         in_buf.type = ACPI_TYPE_BUFFER;
 273         in_buf.buffer.pointer = buf;
 274         in_buf.buffer.length = 0;
 275
 276         /* libnvdimm has already validated the input envelope */
 277         for (i = 0; i < desc->in_num; i++)
 278                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 279                                 i, buf);
 280
 281         if (call_pkg) {
 282                 /* skip over package wrapper */
 283                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 284                 in_buf.buffer.length = call_pkg->nd_size_in;
 285         }
 286
 287         dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
 288                         __func__, dimm_name, cmd, func, in_buf.buffer.length);
 289         print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 290                         in_buf.buffer.pointer,
 291                         min_t(u32, 256, in_buf.buffer.length), true);
 292
 293         out_obj = acpi_evaluate_dsm(handle, guid, 1, func, &in_obj);
 294         if (!out_obj) {
 295                 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
 296                                 cmd_name);
 297                 return -EINVAL;
 298         }
 299
 300         if (call_pkg) {
 301                 call_pkg->nd_fw_size = out_obj->buffer.length;
 302                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 303                         out_obj->buffer.pointer,
 304                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 305
 306                 ACPI_FREE(out_obj);
 307                 /*
 308                  * Need to support FW function w/o known size in advance.
 309                  * Caller can determine required size based upon nd_fw_size.
 310                  * If we return an error (like elsewhere) then caller wouldn't
 311                  * be able to rely upon data returned to make calculation.
 312                  */
 313                 return 0;
 314         }
 315
 316         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 317                 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
 318                                 __func__, dimm_name, cmd_name, out_obj->type);
 319                 rc = -EINVAL;
 320                 goto out;
 321         }
 322
 323         dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__, dimm_name,
 324                         cmd_name, out_obj->buffer.length);
 325         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 326                         out_obj->buffer.pointer,
 327                         min_t(u32, 128, out_obj->buffer.length), true);
 328
 329         for (i = 0, offset = 0; i < desc->out_num; i++) {
 330                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 331                                 (u32 *) out_obj->buffer.pointer,
 332                                 out_obj->buffer.length - offset);
 333
 334                 if (offset + out_size > out_obj->buffer.length) {
 335                         dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
 336                                         __func__, dimm_name, cmd_name, i);
 337                         break;
 338                 }
 339
 340                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 341                         dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
 342                                         __func__, dimm_name, cmd_name, i);
 343                         rc = -ENXIO;
 344                         goto out;
 345                 }
 346                 memcpy(buf + in_buf.buffer.length + offset,
 347                                 out_obj->buffer.pointer + offset, out_size);
 348                 offset += out_size;
 349         }
 350
 351         /*
 352          * Set fw_status for all the commands with a known format to be
 353          * later interpreted by xlat_status().
 354          */
 355         if (i >= 1 && ((cmd >= ND_CMD_ARS_CAP && cmd <= ND_CMD_CLEAR_ERROR)
 356                         || (cmd >= ND_CMD_SMART && cmd <= ND_CMD_VENDOR)))
 357                 fw_status = *(u32 *) out_obj->buffer.pointer;
 358
 359         if (offset + in_buf.buffer.length < buf_len) {
 360                 if (i >= 1) {
 361                         /*
 362                          * status valid, return the number of bytes left
 363                          * unfilled in the output buffer
 364                          */
 365                         rc = buf_len - offset - in_buf.buffer.length;
 366                         if (cmd_rc)
 367                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 368                                                 fw_status);
 369                 } else {
 370                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 371                                         __func__, dimm_name, cmd_name, buf_len,
 372                                         offset);
 373                         rc = -ENXIO;
 374                 }
 375         } else {
 376                 rc = 0;
 377                 if (cmd_rc)
 378                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 379         }
 380
 381  out:
 382         ACPI_FREE(out_obj);
 383
 384         return rc;
 385 }
 386 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 387
 388 static const char *spa_type_name(u16 type)
 389 {
 390         static const char *to_name[] = {
 391                 [NFIT_SPA_VOLATILE] = "volatile",
 392                 [NFIT_SPA_PM] = "pmem",
 393                 [NFIT_SPA_DCR] = "dimm-control-region",
 394                 [NFIT_SPA_BDW] = "block-data-window",
 395                 [NFIT_SPA_VDISK] = "volatile-disk",
 396                 [NFIT_SPA_VCD] = "volatile-cd",
 397                 [NFIT_SPA_PDISK] = "persistent-disk",
 398                 [NFIT_SPA_PCD] = "persistent-cd",
 399
 400         };
 401
 402         if (type > NFIT_SPA_PCD)
 403                 return "unknown";
 404
 405         return to_name[type];
 406 }
 407
 408 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 409 {
 410         int i;
 411
 412         for (i = 0; i < NFIT_UUID_MAX; i++)
 413                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 414                         return i;
 415         return -1;
 416 }
 417
 418 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 419                 struct nfit_table_prev *prev,
 420                 struct acpi_nfit_system_address *spa)
 421 {
 422         struct device *dev = acpi_desc->dev;
 423         struct nfit_spa *nfit_spa;
 424
 425         if (spa->header.length != sizeof(*spa))
 426                 return false;
 427
 428         list_for_each_entry(nfit_spa, &prev->spas, list) {
 429                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 430                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 431                         return true;
 432                 }
 433         }
 434
 435         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 436                         GFP_KERNEL);
 437         if (!nfit_spa)
 438                 return false;
 439         INIT_LIST_HEAD(&nfit_spa->list);
 440         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 441         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 442         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
 443                         spa->range_index,
 444                         spa_type_name(nfit_spa_type(spa)));
 445         return true;
 446 }
 447
 448 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 449                 struct nfit_table_prev *prev,
 450                 struct acpi_nfit_memory_map *memdev)
 451 {
 452         struct device *dev = acpi_desc->dev;
 453         struct nfit_memdev *nfit_memdev;
 454
 455         if (memdev->header.length != sizeof(*memdev))
 456                 return false;
 457
 458         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 459                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 460                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 461                         return true;
 462                 }
 463
 464         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 465                         GFP_KERNEL);
 466         if (!nfit_memdev)
 467                 return false;
 468         INIT_LIST_HEAD(&nfit_memdev->list);
 469         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 470         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 471         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 472                         __func__, memdev->device_handle, memdev->range_index,
 473                         memdev->region_index, memdev->flags);
 474         return true;
 475 }
 476
 477 /*
 478  * An implementation may provide a truncated control region if no block windows
 479  * are defined.
 480  */
 481 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 482 {
 483         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 484                                 window_size))
 485                 return 0;
 486         if (dcr->windows)
 487                 return sizeof(*dcr);
 488         return offsetof(struct acpi_nfit_control_region, window_size);
 489 }
 490
 491 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 492                 struct nfit_table_prev *prev,
 493                 struct acpi_nfit_control_region *dcr)
 494 {
 495         struct device *dev = acpi_desc->dev;
 496         struct nfit_dcr *nfit_dcr;
 497
 498         if (!sizeof_dcr(dcr))
 499                 return false;
 500
 501         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 502                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 503                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 504                         return true;
 505                 }
 506
 507         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 508                         GFP_KERNEL);
 509         if (!nfit_dcr)
 510                 return false;
 511         INIT_LIST_HEAD(&nfit_dcr->list);
 512         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 513         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 514         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
 515                         dcr->region_index, dcr->windows);
 516         return true;
 517 }
 518
 519 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 520                 struct nfit_table_prev *prev,
 521                 struct acpi_nfit_data_region *bdw)
 522 {
 523         struct device *dev = acpi_desc->dev;
 524         struct nfit_bdw *nfit_bdw;
 525
 526         if (bdw->header.length != sizeof(*bdw))
 527                 return false;
 528         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 529                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 530                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 531                         return true;
 532                 }
 533
 534         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 535                         GFP_KERNEL);
 536         if (!nfit_bdw)
 537                 return false;
 538         INIT_LIST_HEAD(&nfit_bdw->list);
 539         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 540         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 541         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
 542                         bdw->region_index, bdw->windows);
 543         return true;
 544 }
 545
 546 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 547 {
 548         if (idt->header.length < sizeof(*idt))
 549                 return 0;
 550         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 551 }
 552
 553 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 554                 struct nfit_table_prev *prev,
 555                 struct acpi_nfit_interleave *idt)
 556 {
 557         struct device *dev = acpi_desc->dev;
 558         struct nfit_idt *nfit_idt;
 559
 560         if (!sizeof_idt(idt))
 561                 return false;
 562
 563         list_for_each_entry(nfit_idt, &prev->idts, list) {
 564                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 565                         continue;
 566
 567                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 568                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 569                         return true;
 570                 }
 571         }
 572
 573         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 574                         GFP_KERNEL);
 575         if (!nfit_idt)
 576                 return false;
 577         INIT_LIST_HEAD(&nfit_idt->list);
 578         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 579         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 580         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
 581                         idt->interleave_index, idt->line_count);
 582         return true;
 583 }
 584
 585 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 586 {
 587         if (flush->header.length < sizeof(*flush))
 588                 return 0;
 589         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 590 }
 591
 592 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 593                 struct nfit_table_prev *prev,
 594                 struct acpi_nfit_flush_address *flush)
 595 {
 596         struct device *dev = acpi_desc->dev;
 597         struct nfit_flush *nfit_flush;
 598
 599         if (!sizeof_flush(flush))
 600                 return false;
 601
 602         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 603                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 604                         continue;
 605
 606                 if (memcmp(nfit_flush->flush, flush,
 607                                         sizeof_flush(flush)) == 0) {
 608                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 609                         return true;
 610                 }
 611         }
 612
 613         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 614                         + sizeof_flush(flush), GFP_KERNEL);
 615         if (!nfit_flush)
 616                 return false;
 617         INIT_LIST_HEAD(&nfit_flush->list);
 618         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 619         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 620         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
 621                         flush->device_handle, flush->hint_count);
 622         return true;
 623 }
 624
 625 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 626                 struct nfit_table_prev *prev, void *table, const void *end)
 627 {
 628         struct device *dev = acpi_desc->dev;
 629         struct acpi_nfit_header *hdr;
 630         void *err = ERR_PTR(-ENOMEM);
 631
 632         if (table >= end)
 633                 return NULL;
 634
 635         hdr = table;
 636         if (!hdr->length) {
 637                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 638                         hdr->type);
 639                 return NULL;
 640         }
 641
 642         switch (hdr->type) {
 643         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 644                 if (!add_spa(acpi_desc, prev, table))
 645                         return err;
 646                 break;
 647         case ACPI_NFIT_TYPE_MEMORY_MAP:
 648                 if (!add_memdev(acpi_desc, prev, table))
 649                         return err;
 650                 break;
 651         case ACPI_NFIT_TYPE_CONTROL_REGION:
 652                 if (!add_dcr(acpi_desc, prev, table))
 653                         return err;
 654                 break;
 655         case ACPI_NFIT_TYPE_DATA_REGION:
 656                 if (!add_bdw(acpi_desc, prev, table))
 657                         return err;
 658                 break;
 659         case ACPI_NFIT_TYPE_INTERLEAVE:
 660                 if (!add_idt(acpi_desc, prev, table))
 661                         return err;
 662                 break;
 663         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 664                 if (!add_flush(acpi_desc, prev, table))
 665                         return err;
 666                 break;
 667         case ACPI_NFIT_TYPE_SMBIOS:
 668                 dev_dbg(dev, "%s: smbios\n", __func__);
 669                 break;
 670         default:
 671                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 672                 break;
 673         }
 674
 675         return table + hdr->length;
 676 }
 677
 678 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 679                 struct nfit_mem *nfit_mem)
 680 {
 681         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 682         u16 dcr = nfit_mem->dcr->region_index;
 683         struct nfit_spa *nfit_spa;
 684
 685         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 686                 u16 range_index = nfit_spa->spa->range_index;
 687                 int type = nfit_spa_type(nfit_spa->spa);
 688                 struct nfit_memdev *nfit_memdev;
 689
 690                 if (type != NFIT_SPA_BDW)
 691                         continue;
 692
 693                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 694                         if (nfit_memdev->memdev->range_index != range_index)
 695                                 continue;
 696                         if (nfit_memdev->memdev->device_handle != device_handle)
 697                                 continue;
 698                         if (nfit_memdev->memdev->region_index != dcr)
 699                                 continue;
 700
 701                         nfit_mem->spa_bdw = nfit_spa->spa;
 702                         return;
 703                 }
 704         }
 705
 706         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 707                         nfit_mem->spa_dcr->range_index);
 708         nfit_mem->bdw = NULL;
 709 }
 710
 711 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 712                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 713 {
 714         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 715         struct nfit_memdev *nfit_memdev;
 716         struct nfit_bdw *nfit_bdw;
 717         struct nfit_idt *nfit_idt;
 718         u16 idt_idx, range_index;
 719
 720         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 721                 if (nfit_bdw->bdw->region_index != dcr)
 722                         continue;
 723                 nfit_mem->bdw = nfit_bdw->bdw;
 724                 break;
 725         }
 726
 727         if (!nfit_mem->bdw)
 728                 return;
 729
 730         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 731
 732         if (!nfit_mem->spa_bdw)
 733                 return;
 734
 735         range_index = nfit_mem->spa_bdw->range_index;
 736         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 737                 if (nfit_memdev->memdev->range_index != range_index ||
 738                                 nfit_memdev->memdev->region_index != dcr)
 739                         continue;
 740                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
 741                 idt_idx = nfit_memdev->memdev->interleave_index;
 742                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 743                         if (nfit_idt->idt->interleave_index != idt_idx)
 744                                 continue;
 745                         nfit_mem->idt_bdw = nfit_idt->idt;
 746                         break;
 747                 }
 748                 break;
 749         }
 750 }
 751
 752 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
 753                 struct acpi_nfit_system_address *spa)
 754 {
 755         struct nfit_mem *nfit_mem, *found;
 756         struct nfit_memdev *nfit_memdev;
 757         int type = spa ? nfit_spa_type(spa) : 0;
 758
 759         switch (type) {
 760         case NFIT_SPA_DCR:
 761         case NFIT_SPA_PM:
 762                 break;
 763         default:
 764                 if (spa)
 765                         return 0;
 766         }
 767
 768         /*
 769          * This loop runs in two modes, when a dimm is mapped the loop
 770          * adds memdev associations to an existing dimm, or creates a
 771          * dimm. In the unmapped dimm case this loop sweeps for memdev
 772          * instances with an invalid / zero range_index and adds those
 773          * dimms without spa associations.
 774          */
 775         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 776                 struct nfit_flush *nfit_flush;
 777                 struct nfit_dcr *nfit_dcr;
 778                 u32 device_handle;
 779                 u16 dcr;
 780
 781                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
 782                         continue;
 783                 if (!spa && nfit_memdev->memdev->range_index)
 784                         continue;
 785                 found = NULL;
 786                 dcr = nfit_memdev->memdev->region_index;
 787                 device_handle = nfit_memdev->memdev->device_handle;
 788                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 789                         if (__to_nfit_memdev(nfit_mem)->device_handle
 790                                         == device_handle) {
 791                                 found = nfit_mem;
 792                                 break;
 793                         }
 794
 795                 if (found)
 796                         nfit_mem = found;
 797                 else {
 798                         nfit_mem = devm_kzalloc(acpi_desc->dev,
 799                                         sizeof(*nfit_mem), GFP_KERNEL);
 800                         if (!nfit_mem)
 801                                 return -ENOMEM;
 802                         INIT_LIST_HEAD(&nfit_mem->list);
 803                         nfit_mem->acpi_desc = acpi_desc;
 804                         list_add(&nfit_mem->list, &acpi_desc->dimms);
 805                 }
 806
 807                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 808                         if (nfit_dcr->dcr->region_index != dcr)
 809                                 continue;
 810                         /*
 811                          * Record the control region for the dimm.  For
 812                          * the ACPI 6.1 case, where there are separate
 813                          * control regions for the pmem vs blk
 814                          * interfaces, be sure to record the extended
 815                          * blk details.
 816                          */
 817                         if (!nfit_mem->dcr)
 818                                 nfit_mem->dcr = nfit_dcr->dcr;
 819                         else if (nfit_mem->dcr->windows == 0
 820                                         && nfit_dcr->dcr->windows)
 821                                 nfit_mem->dcr = nfit_dcr->dcr;
 822                         break;
 823                 }
 824
 825                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
 826                         struct acpi_nfit_flush_address *flush;
 827                         u16 i;
 828
 829                         if (nfit_flush->flush->device_handle != device_handle)
 830                                 continue;
 831                         nfit_mem->nfit_flush = nfit_flush;
 832                         flush = nfit_flush->flush;
 833                         nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
 834                                         flush->hint_count
 835                                         * sizeof(struct resource), GFP_KERNEL);
 836                         if (!nfit_mem->flush_wpq)
 837                                 return -ENOMEM;
 838                         for (i = 0; i < flush->hint_count; i++) {
 839                                 struct resource *res = &nfit_mem->flush_wpq[i];
 840
 841                                 res->start = flush->hint_address[i];
 842                                 res->end = res->start + 8 - 1;
 843                         }
 844                         break;
 845                 }
 846
 847                 if (dcr && !nfit_mem->dcr) {
 848                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
 849                                         spa->range_index, dcr);
 850                         return -ENODEV;
 851                 }
 852
 853                 if (type == NFIT_SPA_DCR) {
 854                         struct nfit_idt *nfit_idt;
 855                         u16 idt_idx;
 856
 857                         /* multiple dimms may share a SPA when interleaved */
 858                         nfit_mem->spa_dcr = spa;
 859                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 860                         idt_idx = nfit_memdev->memdev->interleave_index;
 861                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 862                                 if (nfit_idt->idt->interleave_index != idt_idx)
 863                                         continue;
 864                                 nfit_mem->idt_dcr = nfit_idt->idt;
 865                                 break;
 866                         }
 867                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
 868                 } else if (type == NFIT_SPA_PM) {
 869                         /*
 870                          * A single dimm may belong to multiple SPA-PM
 871                          * ranges, record at least one in addition to
 872                          * any SPA-DCR range.
 873                          */
 874                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
 875                 } else
 876                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 877         }
 878
 879         return 0;
 880 }
 881
 882 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
 883 {
 884         struct nfit_mem *a = container_of(_a, typeof(*a), list);
 885         struct nfit_mem *b = container_of(_b, typeof(*b), list);
 886         u32 handleA, handleB;
 887
 888         handleA = __to_nfit_memdev(a)->device_handle;
 889         handleB = __to_nfit_memdev(b)->device_handle;
 890         if (handleA < handleB)
 891                 return -1;
 892         else if (handleA > handleB)
 893                 return 1;
 894         return 0;
 895 }
 896
 897 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
 898 {
 899         struct nfit_spa *nfit_spa;
 900         int rc;
 901
 902
 903         /*
 904          * For each SPA-DCR or SPA-PMEM address range find its
 905          * corresponding MEMDEV(s).  From each MEMDEV find the
 906          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
 907          * try to find a SPA-BDW and a corresponding BDW that references
 908          * the DCR.  Throw it all into an nfit_mem object.  Note, that
 909          * BDWs are optional.
 910          */
 911         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 912                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
 913                 if (rc)
 914                         return rc;
 915         }
 916
 917         /*
 918          * If a DIMM has failed to be mapped into SPA there will be no
 919          * SPA entries above. Find and register all the unmapped DIMMs
 920          * for reporting and recovery purposes.
 921          */
 922         rc = __nfit_mem_init(acpi_desc, NULL);
 923         if (rc)
 924                 return rc;
 925
 926         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
 927
 928         return 0;
 929 }
 930
 931 static ssize_t bus_dsm_mask_show(struct device *dev,
 932                 struct device_attribute *attr, char *buf)
 933 {
 934         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 935         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 936
 937         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
 938 }
 939 static struct device_attribute dev_attr_bus_dsm_mask =
 940                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
 941
 942 static ssize_t revision_show(struct device *dev,
 943                 struct device_attribute *attr, char *buf)
 944 {
 945         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 946         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 947         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 948
 949         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
 950 }
 951 static DEVICE_ATTR_RO(revision);
 952
 953 static ssize_t hw_error_scrub_show(struct device *dev,
 954                 struct device_attribute *attr, char *buf)
 955 {
 956         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 957         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 958         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 959
 960         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
 961 }
 962
 963 /*
 964  * The 'hw_error_scrub' attribute can have the following values written to it:
 965  * '0': Switch to the default mode where an exception will only insert
 966  *      the address of the memory error into the poison and badblocks lists.
 967  * '1': Enable a full scrub to happen if an exception for a memory error is
 968  *      received.
 969  */
 970 static ssize_t hw_error_scrub_store(struct device *dev,
 971                 struct device_attribute *attr, const char *buf, size_t size)
 972 {
 973         struct nvdimm_bus_descriptor *nd_desc;
 974         ssize_t rc;
 975         long val;
 976
 977         rc = kstrtol(buf, 0, &val);
 978         if (rc)
 979                 return rc;
 980
 981         device_lock(dev);
 982         nd_desc = dev_get_drvdata(dev);
 983         if (nd_desc) {
 984                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 985
 986                 switch (val) {
 987                 case HW_ERROR_SCRUB_ON:
 988                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
 989                         break;
 990                 case HW_ERROR_SCRUB_OFF:
 991                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
 992                         break;
 993                 default:
 994                         rc = -EINVAL;
 995                         break;
 996                 }
 997         }
 998         device_unlock(dev);
 999         if (rc)
1000                 return rc;
1001         return size;
1002 }
1003 static DEVICE_ATTR_RW(hw_error_scrub);
1004
1005 /*
1006  * This shows the number of full Address Range Scrubs that have been
1007  * completed since driver load time. Userspace can wait on this using
1008  * select/poll etc. A '+' at the end indicates an ARS is in progress
1009  */
1010 static ssize_t scrub_show(struct device *dev,
1011                 struct device_attribute *attr, char *buf)
1012 {
1013         struct nvdimm_bus_descriptor *nd_desc;
1014         ssize_t rc = -ENXIO;
1015
1016         device_lock(dev);
1017         nd_desc = dev_get_drvdata(dev);
1018         if (nd_desc) {
1019                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1020
1021                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1022                                 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1023         }
1024         device_unlock(dev);
1025         return rc;
1026 }
1027
1028 static ssize_t scrub_store(struct device *dev,
1029                 struct device_attribute *attr, const char *buf, size_t size)
1030 {
1031         struct nvdimm_bus_descriptor *nd_desc;
1032         ssize_t rc;
1033         long val;
1034
1035         rc = kstrtol(buf, 0, &val);
1036         if (rc)
1037                 return rc;
1038         if (val != 1)
1039                 return -EINVAL;
1040
1041         device_lock(dev);
1042         nd_desc = dev_get_drvdata(dev);
1043         if (nd_desc) {
1044                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1045
1046                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1047         }
1048         device_unlock(dev);
1049         if (rc)
1050                 return rc;
1051         return size;
1052 }
1053 static DEVICE_ATTR_RW(scrub);
1054
1055 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1056 {
1057         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1058         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1059                 | 1 << ND_CMD_ARS_STATUS;
1060
1061         return (nd_desc->cmd_mask & mask) == mask;
1062 }
1063
1064 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1065 {
1066         struct device *dev = container_of(kobj, struct device, kobj);
1067         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1068
1069         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1070                 return 0;
1071         return a->mode;
1072 }
1073
1074 static struct attribute *acpi_nfit_attributes[] = {
1075         &dev_attr_revision.attr,
1076         &dev_attr_scrub.attr,
1077         &dev_attr_hw_error_scrub.attr,
1078         &dev_attr_bus_dsm_mask.attr,
1079         NULL,
1080 };
1081
1082 static const struct attribute_group acpi_nfit_attribute_group = {
1083         .name = "nfit",
1084         .attrs = acpi_nfit_attributes,
1085         .is_visible = nfit_visible,
1086 };
1087
1088 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1089         &nvdimm_bus_attribute_group,
1090         &acpi_nfit_attribute_group,
1091         NULL,
1092 };
1093
1094 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1095 {
1096         struct nvdimm *nvdimm = to_nvdimm(dev);
1097         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1098
1099         return __to_nfit_memdev(nfit_mem);
1100 }
1101
1102 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1103 {
1104         struct nvdimm *nvdimm = to_nvdimm(dev);
1105         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1106
1107         return nfit_mem->dcr;
1108 }
1109
1110 static ssize_t handle_show(struct device *dev,
1111                 struct device_attribute *attr, char *buf)
1112 {
1113         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1114
1115         return sprintf(buf, "%#x\n", memdev->device_handle);
1116 }
1117 static DEVICE_ATTR_RO(handle);
1118
1119 static ssize_t phys_id_show(struct device *dev,
1120                 struct device_attribute *attr, char *buf)
1121 {
1122         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1123
1124         return sprintf(buf, "%#x\n", memdev->physical_id);
1125 }
1126 static DEVICE_ATTR_RO(phys_id);
1127
1128 static ssize_t vendor_show(struct device *dev,
1129                 struct device_attribute *attr, char *buf)
1130 {
1131         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1132
1133         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1134 }
1135 static DEVICE_ATTR_RO(vendor);
1136
1137 static ssize_t rev_id_show(struct device *dev,
1138                 struct device_attribute *attr, char *buf)
1139 {
1140         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1141
1142         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1143 }
1144 static DEVICE_ATTR_RO(rev_id);
1145
1146 static ssize_t device_show(struct device *dev,
1147                 struct device_attribute *attr, char *buf)
1148 {
1149         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1150
1151         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1152 }
1153 static DEVICE_ATTR_RO(device);
1154
1155 static ssize_t subsystem_vendor_show(struct device *dev,
1156                 struct device_attribute *attr, char *buf)
1157 {
1158         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1159
1160         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1161 }
1162 static DEVICE_ATTR_RO(subsystem_vendor);
1163
1164 static ssize_t subsystem_rev_id_show(struct device *dev,
1165                 struct device_attribute *attr, char *buf)
1166 {
1167         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1168
1169         return sprintf(buf, "0x%04x\n",
1170                         be16_to_cpu(dcr->subsystem_revision_id));
1171 }
1172 static DEVICE_ATTR_RO(subsystem_rev_id);
1173
1174 static ssize_t subsystem_device_show(struct device *dev,
1175                 struct device_attribute *attr, char *buf)
1176 {
1177         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1178
1179         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1180 }
1181 static DEVICE_ATTR_RO(subsystem_device);
1182
1183 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1184 {
1185         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1186         int formats = 0;
1187
1188         if (nfit_mem->memdev_pmem)
1189                 formats++;
1190         if (nfit_mem->memdev_bdw)
1191                 formats++;
1192         return formats;
1193 }
1194
1195 static ssize_t format_show(struct device *dev,
1196                 struct device_attribute *attr, char *buf)
1197 {
1198         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1199
1200         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1201 }
1202 static DEVICE_ATTR_RO(format);
1203
1204 static ssize_t format1_show(struct device *dev,
1205                 struct device_attribute *attr, char *buf)
1206 {
1207         u32 handle;
1208         ssize_t rc = -ENXIO;
1209         struct nfit_mem *nfit_mem;
1210         struct nfit_memdev *nfit_memdev;
1211         struct acpi_nfit_desc *acpi_desc;
1212         struct nvdimm *nvdimm = to_nvdimm(dev);
1213         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1214
1215         nfit_mem = nvdimm_provider_data(nvdimm);
1216         acpi_desc = nfit_mem->acpi_desc;
1217         handle = to_nfit_memdev(dev)->device_handle;
1218
1219         /* assumes DIMMs have at most 2 published interface codes */
1220         mutex_lock(&acpi_desc->init_mutex);
1221         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1222                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1223                 struct nfit_dcr *nfit_dcr;
1224
1225                 if (memdev->device_handle != handle)
1226                         continue;
1227
1228                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1229                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1230                                 continue;
1231                         if (nfit_dcr->dcr->code == dcr->code)
1232                                 continue;
1233                         rc = sprintf(buf, "0x%04x\n",
1234                                         le16_to_cpu(nfit_dcr->dcr->code));
1235                         break;
1236                 }
1237                 if (rc != ENXIO)
1238                         break;
1239         }
1240         mutex_unlock(&acpi_desc->init_mutex);
1241         return rc;
1242 }
1243 static DEVICE_ATTR_RO(format1);
1244
1245 static ssize_t formats_show(struct device *dev,
1246                 struct device_attribute *attr, char *buf)
1247 {
1248         struct nvdimm *nvdimm = to_nvdimm(dev);
1249
1250         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1251 }
1252 static DEVICE_ATTR_RO(formats);
1253
1254 static ssize_t serial_show(struct device *dev,
1255                 struct device_attribute *attr, char *buf)
1256 {
1257         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1258
1259         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1260 }
1261 static DEVICE_ATTR_RO(serial);
1262
1263 static ssize_t family_show(struct device *dev,
1264                 struct device_attribute *attr, char *buf)
1265 {
1266         struct nvdimm *nvdimm = to_nvdimm(dev);
1267         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1268
1269         if (nfit_mem->family < 0)
1270                 return -ENXIO;
1271         return sprintf(buf, "%d\n", nfit_mem->family);
1272 }
1273 static DEVICE_ATTR_RO(family);
1274
1275 static ssize_t dsm_mask_show(struct device *dev,
1276                 struct device_attribute *attr, char *buf)
1277 {
1278         struct nvdimm *nvdimm = to_nvdimm(dev);
1279         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1280
1281         if (nfit_mem->family < 0)
1282                 return -ENXIO;
1283         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1284 }
1285 static DEVICE_ATTR_RO(dsm_mask);
1286
1287 static ssize_t flags_show(struct device *dev,
1288                 struct device_attribute *attr, char *buf)
1289 {
1290         u16 flags = to_nfit_memdev(dev)->flags;
1291
1292         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1293                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1294                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1295                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1296                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1297                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1298                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1299                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1300 }
1301 static DEVICE_ATTR_RO(flags);
1302
1303 static ssize_t id_show(struct device *dev,
1304                 struct device_attribute *attr, char *buf)
1305 {
1306         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1307
1308         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1309                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1310                                 be16_to_cpu(dcr->vendor_id),
1311                                 dcr->manufacturing_location,
1312                                 be16_to_cpu(dcr->manufacturing_date),
1313                                 be32_to_cpu(dcr->serial_number));
1314         else
1315                 return sprintf(buf, "%04x-%08x\n",
1316                                 be16_to_cpu(dcr->vendor_id),
1317                                 be32_to_cpu(dcr->serial_number));
1318 }
1319 static DEVICE_ATTR_RO(id);
1320
1321 static struct attribute *acpi_nfit_dimm_attributes[] = {
1322         &dev_attr_handle.attr,
1323         &dev_attr_phys_id.attr,
1324         &dev_attr_vendor.attr,
1325         &dev_attr_device.attr,
1326         &dev_attr_rev_id.attr,
1327         &dev_attr_subsystem_vendor.attr,
1328         &dev_attr_subsystem_device.attr,
1329         &dev_attr_subsystem_rev_id.attr,
1330         &dev_attr_format.attr,
1331         &dev_attr_formats.attr,
1332         &dev_attr_format1.attr,
1333         &dev_attr_serial.attr,
1334         &dev_attr_flags.attr,
1335         &dev_attr_id.attr,
1336         &dev_attr_family.attr,
1337         &dev_attr_dsm_mask.attr,
1338         NULL,
1339 };
1340
1341 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1342                 struct attribute *a, int n)
1343 {
1344         struct device *dev = container_of(kobj, struct device, kobj);
1345         struct nvdimm *nvdimm = to_nvdimm(dev);
1346
1347         if (!to_nfit_dcr(dev)) {
1348                 /* Without a dcr only the memdev attributes can be surfaced */
1349                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1350                                 || a == &dev_attr_flags.attr
1351                                 || a == &dev_attr_family.attr
1352                                 || a == &dev_attr_dsm_mask.attr)
1353                         return a->mode;
1354                 return 0;
1355         }
1356
1357         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1358                 return 0;
1359         return a->mode;
1360 }
1361
1362 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1363         .name = "nfit",
1364         .attrs = acpi_nfit_dimm_attributes,
1365         .is_visible = acpi_nfit_dimm_attr_visible,
1366 };
1367
1368 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1369         &nvdimm_attribute_group,
1370         &nd_device_attribute_group,
1371         &acpi_nfit_dimm_attribute_group,
1372         NULL,
1373 };
1374
1375 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1376                 u32 device_handle)
1377 {
1378         struct nfit_mem *nfit_mem;
1379
1380         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1381                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1382                         return nfit_mem->nvdimm;
1383
1384         return NULL;
1385 }
1386
1387 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1388 {
1389         struct nfit_mem *nfit_mem;
1390         struct acpi_nfit_desc *acpi_desc;
1391
1392         dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1393                         event);
1394
1395         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1396                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1397                                 event);
1398                 return;
1399         }
1400
1401         acpi_desc = dev_get_drvdata(dev->parent);
1402         if (!acpi_desc)
1403                 return;
1404
1405         /*
1406          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1407          * is still valid.
1408          */
1409         nfit_mem = dev_get_drvdata(dev);
1410         if (nfit_mem && nfit_mem->flags_attr)
1411                 sysfs_notify_dirent(nfit_mem->flags_attr);
1412 }
1413 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1414
1415 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1416 {
1417         struct acpi_device *adev = data;
1418         struct device *dev = &adev->dev;
1419
1420         device_lock(dev->parent);
1421         __acpi_nvdimm_notify(dev, event);
1422         device_unlock(dev->parent);
1423 }
1424
1425 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1426                 struct nfit_mem *nfit_mem, u32 device_handle)
1427 {
1428         struct acpi_device *adev, *adev_dimm;
1429         struct device *dev = acpi_desc->dev;
1430         unsigned long dsm_mask;
1431         const guid_t *guid;
1432         int i;
1433         int family = -1;
1434
1435         /* nfit test assumes 1:1 relationship between commands and dsms */
1436         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1437         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1438         adev = to_acpi_dev(acpi_desc);
1439         if (!adev)
1440                 return 0;
1441
1442         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1443         nfit_mem->adev = adev_dimm;
1444         if (!adev_dimm) {
1445                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1446                                 device_handle);
1447                 return force_enable_dimms ? 0 : -ENODEV;
1448         }
1449
1450         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1451                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1452                 dev_err(dev, "%s: notification registration failed\n",
1453                                 dev_name(&adev_dimm->dev));
1454                 return -ENXIO;
1455         }
1456
1457         /*
1458          * Until standardization materializes we need to consider 4
1459          * different command sets.  Note, that checking for function0 (bit0)
1460          * tells us if any commands are reachable through this GUID.
1461          */
1462         for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_MSFT; i++)
1463                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1464                         if (family < 0 || i == default_dsm_family)
1465                                 family = i;
1466
1467         /* limit the supported commands to those that are publicly documented */
1468         nfit_mem->family = family;
1469         if (override_dsm_mask && !disable_vendor_specific)
1470                 dsm_mask = override_dsm_mask;
1471         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1472                 dsm_mask = 0x3fe;
1473                 if (disable_vendor_specific)
1474                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1475         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1476                 dsm_mask = 0x1c3c76;
1477         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1478                 dsm_mask = 0x1fe;
1479                 if (disable_vendor_specific)
1480                         dsm_mask &= ~(1 << 8);
1481         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1482                 dsm_mask = 0xffffffff;
1483         } else {
1484                 dev_dbg(dev, "unknown dimm command family\n");
1485                 nfit_mem->family = -1;
1486                 /* DSMs are optional, continue loading the driver... */
1487                 return 0;
1488         }
1489
1490         guid = to_nfit_uuid(nfit_mem->family);
1491         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1492                 if (acpi_check_dsm(adev_dimm->handle, guid, 1, 1ULL << i))
1493                         set_bit(i, &nfit_mem->dsm_mask);
1494
1495         return 0;
1496 }
1497
1498 static void shutdown_dimm_notify(void *data)
1499 {
1500         struct acpi_nfit_desc *acpi_desc = data;
1501         struct nfit_mem *nfit_mem;
1502
1503         mutex_lock(&acpi_desc->init_mutex);
1504         /*
1505          * Clear out the nfit_mem->flags_attr and shut down dimm event
1506          * notifications.
1507          */
1508         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1509                 struct acpi_device *adev_dimm = nfit_mem->adev;
1510
1511                 if (nfit_mem->flags_attr) {
1512                         sysfs_put(nfit_mem->flags_attr);
1513                         nfit_mem->flags_attr = NULL;
1514                 }
1515                 if (adev_dimm)
1516                         acpi_remove_notify_handler(adev_dimm->handle,
1517                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1518         }
1519         mutex_unlock(&acpi_desc->init_mutex);
1520 }
1521
1522 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1523 {
1524         struct nfit_mem *nfit_mem;
1525         int dimm_count = 0, rc;
1526         struct nvdimm *nvdimm;
1527
1528         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1529                 struct acpi_nfit_flush_address *flush;
1530                 unsigned long flags = 0, cmd_mask;
1531                 struct nfit_memdev *nfit_memdev;
1532                 u32 device_handle;
1533                 u16 mem_flags;
1534
1535                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1536                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1537                 if (nvdimm) {
1538                         dimm_count++;
1539                         continue;
1540                 }
1541
1542                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1543                         set_bit(NDD_ALIASING, &flags);
1544
1545                 /* collate flags across all memdevs for this dimm */
1546                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1547                         struct acpi_nfit_memory_map *dimm_memdev;
1548
1549                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1550                         if (dimm_memdev->device_handle
1551                                         != nfit_memdev->memdev->device_handle)
1552                                 continue;
1553                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1554                 }
1555
1556                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1557                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1558                         set_bit(NDD_UNARMED, &flags);
1559
1560                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1561                 if (rc)
1562                         continue;
1563
1564                 /*
1565                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1566                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1567                  * userspace interface.
1568                  */
1569                 cmd_mask = 1UL << ND_CMD_CALL;
1570                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1571                         cmd_mask |= nfit_mem->dsm_mask;
1572
1573                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1574                         : NULL;
1575                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1576                                 acpi_nfit_dimm_attribute_groups,
1577                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1578                                 nfit_mem->flush_wpq);
1579                 if (!nvdimm)
1580                         return -ENOMEM;
1581
1582                 nfit_mem->nvdimm = nvdimm;
1583                 dimm_count++;
1584
1585                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1586                         continue;
1587
1588                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1589                                 nvdimm_name(nvdimm),
1590                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1591                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1592                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1593                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1594                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1595
1596         }
1597
1598         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1599         if (rc)
1600                 return rc;
1601
1602         /*
1603          * Now that dimms are successfully registered, and async registration
1604          * is flushed, attempt to enable event notification.
1605          */
1606         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1607                 struct kernfs_node *nfit_kernfs;
1608
1609                 nvdimm = nfit_mem->nvdimm;
1610                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1611                 if (nfit_kernfs)
1612                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1613                                         "flags");
1614                 sysfs_put(nfit_kernfs);
1615                 if (!nfit_mem->flags_attr)
1616                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1617                                         nvdimm_name(nvdimm));
1618         }
1619
1620         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1621                         acpi_desc);
1622 }
1623
1624 /*
1625  * These constants are private because there are no kernel consumers of
1626  * these commands.
1627  */
1628 enum nfit_aux_cmds {
1629         NFIT_CMD_TRANSLATE_SPA = 5,
1630         NFIT_CMD_ARS_INJECT_SET = 7,
1631         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1632         NFIT_CMD_ARS_INJECT_GET = 9,
1633 };
1634
1635 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1636 {
1637         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1638         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1639         struct acpi_device *adev;
1640         unsigned long dsm_mask;
1641         int i;
1642
1643         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1644         adev = to_acpi_dev(acpi_desc);
1645         if (!adev)
1646                 return;
1647
1648         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1649                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1650                         set_bit(i, &nd_desc->cmd_mask);
1651         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1652
1653         dsm_mask =
1654                 (1 << ND_CMD_ARS_CAP) |
1655                 (1 << ND_CMD_ARS_START) |
1656                 (1 << ND_CMD_ARS_STATUS) |
1657                 (1 << ND_CMD_CLEAR_ERROR) |
1658                 (1 << NFIT_CMD_TRANSLATE_SPA) |
1659                 (1 << NFIT_CMD_ARS_INJECT_SET) |
1660                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1661                 (1 << NFIT_CMD_ARS_INJECT_GET);
1662         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1663                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1664                         set_bit(i, &nd_desc->bus_dsm_mask);
1665 }
1666
1667 static ssize_t range_index_show(struct device *dev,
1668                 struct device_attribute *attr, char *buf)
1669 {
1670         struct nd_region *nd_region = to_nd_region(dev);
1671         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1672
1673         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1674 }
1675 static DEVICE_ATTR_RO(range_index);
1676
1677 static struct attribute *acpi_nfit_region_attributes[] = {
1678         &dev_attr_range_index.attr,
1679         NULL,
1680 };
1681
1682 static const struct attribute_group acpi_nfit_region_attribute_group = {
1683         .name = "nfit",
1684         .attrs = acpi_nfit_region_attributes,
1685 };
1686
1687 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1688         &nd_region_attribute_group,
1689         &nd_mapping_attribute_group,
1690         &nd_device_attribute_group,
1691         &nd_numa_attribute_group,
1692         &acpi_nfit_region_attribute_group,
1693         NULL,
1694 };
1695
1696 /* enough info to uniquely specify an interleave set */
1697 struct nfit_set_info {
1698         struct nfit_set_info_map {
1699                 u64 region_offset;
1700                 u32 serial_number;
1701                 u32 pad;
1702         } mapping[0];
1703 };
1704
1705 struct nfit_set_info2 {
1706         struct nfit_set_info_map2 {
1707                 u64 region_offset;
1708                 u32 serial_number;
1709                 u16 vendor_id;
1710                 u16 manufacturing_date;
1711                 u8  manufacturing_location;
1712                 u8  reserved[31];
1713         } mapping[0];
1714 };
1715
1716 static size_t sizeof_nfit_set_info(int num_mappings)
1717 {
1718         return sizeof(struct nfit_set_info)
1719                 + num_mappings * sizeof(struct nfit_set_info_map);
1720 }
1721
1722 static size_t sizeof_nfit_set_info2(int num_mappings)
1723 {
1724         return sizeof(struct nfit_set_info2)
1725                 + num_mappings * sizeof(struct nfit_set_info_map2);
1726 }
1727
1728 static int cmp_map_compat(const void *m0, const void *m1)
1729 {
1730         const struct nfit_set_info_map *map0 = m0;
1731         const struct nfit_set_info_map *map1 = m1;
1732
1733         return memcmp(&map0->region_offset, &map1->region_offset,
1734                         sizeof(u64));
1735 }
1736
1737 static int cmp_map(const void *m0, const void *m1)
1738 {
1739         const struct nfit_set_info_map *map0 = m0;
1740         const struct nfit_set_info_map *map1 = m1;
1741
1742         if (map0->region_offset < map1->region_offset)
1743                 return -1;
1744         else if (map0->region_offset > map1->region_offset)
1745                 return 1;
1746         return 0;
1747 }
1748
1749 static int cmp_map2(const void *m0, const void *m1)
1750 {
1751         const struct nfit_set_info_map2 *map0 = m0;
1752         const struct nfit_set_info_map2 *map1 = m1;
1753
1754         if (map0->region_offset < map1->region_offset)
1755                 return -1;
1756         else if (map0->region_offset > map1->region_offset)
1757                 return 1;
1758         return 0;
1759 }
1760
1761 /* Retrieve the nth entry referencing this spa */
1762 static struct acpi_nfit_memory_map *memdev_from_spa(
1763                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1764 {
1765         struct nfit_memdev *nfit_memdev;
1766
1767         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1768                 if (nfit_memdev->memdev->range_index == range_index)
1769                         if (n-- == 0)
1770                                 return nfit_memdev->memdev;
1771         return NULL;
1772 }
1773
1774 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1775                 struct nd_region_desc *ndr_desc,
1776                 struct acpi_nfit_system_address *spa)
1777 {
1778         struct device *dev = acpi_desc->dev;
1779         struct nd_interleave_set *nd_set;
1780         u16 nr = ndr_desc->num_mappings;
1781         struct nfit_set_info2 *info2;
1782         struct nfit_set_info *info;
1783         int i;
1784
1785         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1786         if (!nd_set)
1787                 return -ENOMEM;
1788         ndr_desc->nd_set = nd_set;
1789         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
1790
1791         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1792         if (!info)
1793                 return -ENOMEM;
1794
1795         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
1796         if (!info2)
1797                 return -ENOMEM;
1798
1799         for (i = 0; i < nr; i++) {
1800                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1801                 struct nfit_set_info_map *map = &info->mapping[i];
1802                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
1803                 struct nvdimm *nvdimm = mapping->nvdimm;
1804                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1805                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1806                                 spa->range_index, i);
1807
1808                 if (!memdev || !nfit_mem->dcr) {
1809                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1810                         return -ENODEV;
1811                 }
1812
1813                 map->region_offset = memdev->region_offset;
1814                 map->serial_number = nfit_mem->dcr->serial_number;
1815
1816                 map2->region_offset = memdev->region_offset;
1817                 map2->serial_number = nfit_mem->dcr->serial_number;
1818                 map2->vendor_id = nfit_mem->dcr->vendor_id;
1819                 map2->manufacturing_date = nfit_mem->dcr->manufacturing_date;
1820                 map2->manufacturing_location = nfit_mem->dcr->manufacturing_location;
1821         }
1822
1823         /* v1.1 namespaces */
1824         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1825                         cmp_map, NULL);
1826         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1827
1828         /* v1.2 namespaces */
1829         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
1830                         cmp_map2, NULL);
1831         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
1832
1833         /* support v1.1 namespaces created with the wrong sort order */
1834         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1835                         cmp_map_compat, NULL);
1836         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1837
1838         ndr_desc->nd_set = nd_set;
1839         devm_kfree(dev, info);
1840         devm_kfree(dev, info2);
1841
1842         return 0;
1843 }
1844
1845 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1846 {
1847         struct acpi_nfit_interleave *idt = mmio->idt;
1848         u32 sub_line_offset, line_index, line_offset;
1849         u64 line_no, table_skip_count, table_offset;
1850
1851         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1852         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1853         line_offset = idt->line_offset[line_index]
1854                 * mmio->line_size;
1855         table_offset = table_skip_count * mmio->table_size;
1856
1857         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1858 }
1859
1860 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1861 {
1862         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1863         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1864         const u32 STATUS_MASK = 0x80000037;
1865
1866         if (mmio->num_lines)
1867                 offset = to_interleave_offset(offset, mmio);
1868
1869         return readl(mmio->addr.base + offset) & STATUS_MASK;
1870 }
1871
1872 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1873                 resource_size_t dpa, unsigned int len, unsigned int write)
1874 {
1875         u64 cmd, offset;
1876         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1877
1878         enum {
1879                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1880                 BCW_LEN_SHIFT = 48,
1881                 BCW_LEN_MASK = (1ULL << 8) - 1,
1882                 BCW_CMD_SHIFT = 56,
1883         };
1884
1885         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1886         len = len >> L1_CACHE_SHIFT;
1887         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1888         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1889
1890         offset = nfit_blk->cmd_offset + mmio->size * bw;
1891         if (mmio->num_lines)
1892                 offset = to_interleave_offset(offset, mmio);
1893
1894         writeq(cmd, mmio->addr.base + offset);
1895         nvdimm_flush(nfit_blk->nd_region);
1896
1897         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1898                 readq(mmio->addr.base + offset);
1899 }
1900
1901 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1902                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1903                 unsigned int lane)
1904 {
1905         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1906         unsigned int copied = 0;
1907         u64 base_offset;
1908         int rc;
1909
1910         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1911                 + lane * mmio->size;
1912         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1913         while (len) {
1914                 unsigned int c;
1915                 u64 offset;
1916
1917                 if (mmio->num_lines) {
1918                         u32 line_offset;
1919
1920                         offset = to_interleave_offset(base_offset + copied,
1921                                         mmio);
1922                         div_u64_rem(offset, mmio->line_size, &line_offset);
1923                         c = min_t(size_t, len, mmio->line_size - line_offset);
1924                 } else {
1925                         offset = base_offset + nfit_blk->bdw_offset;
1926                         c = len;
1927                 }
1928
1929                 if (rw)
1930                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
1931                 else {
1932                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1933                                 mmio_flush_range((void __force *)
1934                                         mmio->addr.aperture + offset, c);
1935
1936                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
1937                 }
1938
1939                 copied += c;
1940                 len -= c;
1941         }
1942
1943         if (rw)
1944                 nvdimm_flush(nfit_blk->nd_region);
1945
1946         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1947         return rc;
1948 }
1949
1950 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1951                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1952 {
1953         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1954         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1955         struct nd_region *nd_region = nfit_blk->nd_region;
1956         unsigned int lane, copied = 0;
1957         int rc = 0;
1958
1959         lane = nd_region_acquire_lane(nd_region);
1960         while (len) {
1961                 u64 c = min(len, mmio->size);
1962
1963                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1964                                 iobuf + copied, c, rw, lane);
1965                 if (rc)
1966                         break;
1967
1968                 copied += c;
1969                 len -= c;
1970         }
1971         nd_region_release_lane(nd_region, lane);
1972
1973         return rc;
1974 }
1975
1976 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1977                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1978 {
1979         if (idt) {
1980                 mmio->num_lines = idt->line_count;
1981                 mmio->line_size = idt->line_size;
1982                 if (interleave_ways == 0)
1983                         return -ENXIO;
1984                 mmio->table_size = mmio->num_lines * interleave_ways
1985                         * mmio->line_size;
1986         }
1987
1988         return 0;
1989 }
1990
1991 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1992                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1993 {
1994         struct nd_cmd_dimm_flags flags;
1995         int rc;
1996
1997         memset(&flags, 0, sizeof(flags));
1998         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1999                         sizeof(flags), NULL);
2000
2001         if (rc >= 0 && flags.status == 0)
2002                 nfit_blk->dimm_flags = flags.flags;
2003         else if (rc == -ENOTTY) {
2004                 /* fall back to a conservative default */
2005                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2006                 rc = 0;
2007         } else
2008                 rc = -ENXIO;
2009
2010         return rc;
2011 }
2012
2013 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2014                 struct device *dev)
2015 {
2016         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2017         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2018         struct nfit_blk_mmio *mmio;
2019         struct nfit_blk *nfit_blk;
2020         struct nfit_mem *nfit_mem;
2021         struct nvdimm *nvdimm;
2022         int rc;
2023
2024         nvdimm = nd_blk_region_to_dimm(ndbr);
2025         nfit_mem = nvdimm_provider_data(nvdimm);
2026         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2027                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2028                                 nfit_mem ? "" : " nfit_mem",
2029                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2030                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2031                 return -ENXIO;
2032         }
2033
2034         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2035         if (!nfit_blk)
2036                 return -ENOMEM;
2037         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2038         nfit_blk->nd_region = to_nd_region(dev);
2039
2040         /* map block aperture memory */
2041         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2042         mmio = &nfit_blk->mmio[BDW];
2043         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2044                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2045         if (!mmio->addr.base) {
2046                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2047                                 nvdimm_name(nvdimm));
2048                 return -ENOMEM;
2049         }
2050         mmio->size = nfit_mem->bdw->size;
2051         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2052         mmio->idt = nfit_mem->idt_bdw;
2053         mmio->spa = nfit_mem->spa_bdw;
2054         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2055                         nfit_mem->memdev_bdw->interleave_ways);
2056         if (rc) {
2057                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2058                                 __func__, nvdimm_name(nvdimm));
2059                 return rc;
2060         }
2061
2062         /* map block control memory */
2063         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2064         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2065         mmio = &nfit_blk->mmio[DCR];
2066         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2067                         nfit_mem->spa_dcr->length);
2068         if (!mmio->addr.base) {
2069                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2070                                 nvdimm_name(nvdimm));
2071                 return -ENOMEM;
2072         }
2073         mmio->size = nfit_mem->dcr->window_size;
2074         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2075         mmio->idt = nfit_mem->idt_dcr;
2076         mmio->spa = nfit_mem->spa_dcr;
2077         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2078                         nfit_mem->memdev_dcr->interleave_ways);
2079         if (rc) {
2080                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2081                                 __func__, nvdimm_name(nvdimm));
2082                 return rc;
2083         }
2084
2085         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2086         if (rc < 0) {
2087                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2088                                 __func__, nvdimm_name(nvdimm));
2089                 return rc;
2090         }
2091
2092         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2093                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2094
2095         if (mmio->line_size == 0)
2096                 return 0;
2097
2098         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2099                         + 8 > mmio->line_size) {
2100                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2101                 return -ENXIO;
2102         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2103                         + 8 > mmio->line_size) {
2104                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2105                 return -ENXIO;
2106         }
2107
2108         return 0;
2109 }
2110
2111 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2112                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2113 {
2114         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2115         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2116         int cmd_rc, rc;
2117
2118         cmd->address = spa->address;
2119         cmd->length = spa->length;
2120         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2121                         sizeof(*cmd), &cmd_rc);
2122         if (rc < 0)
2123                 return rc;
2124         return cmd_rc;
2125 }
2126
2127 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2128 {
2129         int rc;
2130         int cmd_rc;
2131         struct nd_cmd_ars_start ars_start;
2132         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2133         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2134
2135         memset(&ars_start, 0, sizeof(ars_start));
2136         ars_start.address = spa->address;
2137         ars_start.length = spa->length;
2138         ars_start.flags = acpi_desc->ars_start_flags;
2139         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2140                 ars_start.type = ND_ARS_PERSISTENT;
2141         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2142                 ars_start.type = ND_ARS_VOLATILE;
2143         else
2144                 return -ENOTTY;
2145
2146         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2147                         sizeof(ars_start), &cmd_rc);
2148
2149         if (rc < 0)
2150                 return rc;
2151         return cmd_rc;
2152 }
2153
2154 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2155 {
2156         int rc, cmd_rc;
2157         struct nd_cmd_ars_start ars_start;
2158         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2159         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2160
2161         memset(&ars_start, 0, sizeof(ars_start));
2162         ars_start.address = ars_status->restart_address;
2163         ars_start.length = ars_status->restart_length;
2164         ars_start.type = ars_status->type;
2165         ars_start.flags = acpi_desc->ars_start_flags;
2166         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2167                         sizeof(ars_start), &cmd_rc);
2168         if (rc < 0)
2169                 return rc;
2170         return cmd_rc;
2171 }
2172
2173 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2174 {
2175         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2176         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2177         int rc, cmd_rc;
2178
2179         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2180                         acpi_desc->ars_status_size, &cmd_rc);
2181         if (rc < 0)
2182                 return rc;
2183         return cmd_rc;
2184 }
2185
2186 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2187                 struct nd_cmd_ars_status *ars_status)
2188 {
2189         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2190         int rc;
2191         u32 i;
2192
2193         /*
2194          * First record starts at 44 byte offset from the start of the
2195          * payload.
2196          */
2197         if (ars_status->out_length < 44)
2198                 return 0;
2199         for (i = 0; i < ars_status->num_records; i++) {
2200                 /* only process full records */
2201                 if (ars_status->out_length
2202                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2203                         break;
2204                 rc = nvdimm_bus_add_poison(nvdimm_bus,
2205                                 ars_status->records[i].err_address,
2206                                 ars_status->records[i].length);
2207                 if (rc)
2208                         return rc;
2209         }
2210         if (i < ars_status->num_records)
2211                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2212
2213         return 0;
2214 }
2215
2216 static void acpi_nfit_remove_resource(void *data)
2217 {
2218         struct resource *res = data;
2219
2220         remove_resource(res);
2221 }
2222
2223 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2224                 struct nd_region_desc *ndr_desc)
2225 {
2226         struct resource *res, *nd_res = ndr_desc->res;
2227         int is_pmem, ret;
2228
2229         /* No operation if the region is already registered as PMEM */
2230         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2231                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2232         if (is_pmem == REGION_INTERSECTS)
2233                 return 0;
2234
2235         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2236         if (!res)
2237                 return -ENOMEM;
2238
2239         res->name = "Persistent Memory";
2240         res->start = nd_res->start;
2241         res->end = nd_res->end;
2242         res->flags = IORESOURCE_MEM;
2243         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2244
2245         ret = insert_resource(&iomem_resource, res);
2246         if (ret)
2247                 return ret;
2248
2249         ret = devm_add_action_or_reset(acpi_desc->dev,
2250                                         acpi_nfit_remove_resource,
2251                                         res);
2252         if (ret)
2253                 return ret;
2254
2255         return 0;
2256 }
2257
2258 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2259                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2260                 struct acpi_nfit_memory_map *memdev,
2261                 struct nfit_spa *nfit_spa)
2262 {
2263         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2264                         memdev->device_handle);
2265         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2266         struct nd_blk_region_desc *ndbr_desc;
2267         struct nfit_mem *nfit_mem;
2268         int blk_valid = 0, rc;
2269
2270         if (!nvdimm) {
2271                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2272                                 spa->range_index, memdev->device_handle);
2273                 return -ENODEV;
2274         }
2275
2276         mapping->nvdimm = nvdimm;
2277         switch (nfit_spa_type(spa)) {
2278         case NFIT_SPA_PM:
2279         case NFIT_SPA_VOLATILE:
2280                 mapping->start = memdev->address;
2281                 mapping->size = memdev->region_size;
2282                 break;
2283         case NFIT_SPA_DCR:
2284                 nfit_mem = nvdimm_provider_data(nvdimm);
2285                 if (!nfit_mem || !nfit_mem->bdw) {
2286                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2287                                         spa->range_index, nvdimm_name(nvdimm));
2288                 } else {
2289                         mapping->size = nfit_mem->bdw->capacity;
2290                         mapping->start = nfit_mem->bdw->start_address;
2291                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
2292                         blk_valid = 1;
2293                 }
2294
2295                 ndr_desc->mapping = mapping;
2296                 ndr_desc->num_mappings = blk_valid;
2297                 ndbr_desc = to_blk_region_desc(ndr_desc);
2298                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2299                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2300                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2301                 if (rc)
2302                         return rc;
2303                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2304                                 ndr_desc);
2305                 if (!nfit_spa->nd_region)
2306                         return -ENOMEM;
2307                 break;
2308         }
2309
2310         return 0;
2311 }
2312
2313 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2314 {
2315         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2316                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2317                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2318                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2319 }
2320
2321 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2322 {
2323         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2324                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2325                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2326 }
2327
2328 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2329                 struct nfit_spa *nfit_spa)
2330 {
2331         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2332         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2333         struct nd_blk_region_desc ndbr_desc;
2334         struct nd_region_desc *ndr_desc;
2335         struct nfit_memdev *nfit_memdev;
2336         struct nvdimm_bus *nvdimm_bus;
2337         struct resource res;
2338         int count = 0, rc;
2339
2340         if (nfit_spa->nd_region)
2341                 return 0;
2342
2343         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2344                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2345                                 __func__);
2346                 return 0;
2347         }
2348
2349         memset(&res, 0, sizeof(res));
2350         memset(&mappings, 0, sizeof(mappings));
2351         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2352         res.start = spa->address;
2353         res.end = res.start + spa->length - 1;
2354         ndr_desc = &ndbr_desc.ndr_desc;
2355         ndr_desc->res = &res;
2356         ndr_desc->provider_data = nfit_spa;
2357         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2358         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2359                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2360                                                 spa->proximity_domain);
2361         else
2362                 ndr_desc->numa_node = NUMA_NO_NODE;
2363
2364         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2365                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2366                 struct nd_mapping_desc *mapping;
2367
2368                 if (memdev->range_index != spa->range_index)
2369                         continue;
2370                 if (count >= ND_MAX_MAPPINGS) {
2371                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2372                                         spa->range_index, ND_MAX_MAPPINGS);
2373                         return -ENXIO;
2374                 }
2375                 mapping = &mappings[count++];
2376                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2377                                 memdev, nfit_spa);
2378                 if (rc)
2379                         goto out;
2380         }
2381
2382         ndr_desc->mapping = mappings;
2383         ndr_desc->num_mappings = count;
2384         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2385         if (rc)
2386                 goto out;
2387
2388         nvdimm_bus = acpi_desc->nvdimm_bus;
2389         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2390                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2391                 if (rc) {
2392                         dev_warn(acpi_desc->dev,
2393                                 "failed to insert pmem resource to iomem: %d\n",
2394                                 rc);
2395                         goto out;
2396                 }
2397
2398                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2399                                 ndr_desc);
2400                 if (!nfit_spa->nd_region)
2401                         rc = -ENOMEM;
2402         } else if (nfit_spa_is_volatile(spa)) {
2403                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2404                                 ndr_desc);
2405                 if (!nfit_spa->nd_region)
2406                         rc = -ENOMEM;
2407         } else if (nfit_spa_is_virtual(spa)) {
2408                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2409                                 ndr_desc);
2410                 if (!nfit_spa->nd_region)
2411                         rc = -ENOMEM;
2412         }
2413
2414  out:
2415         if (rc)
2416                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2417                                 nfit_spa->spa->range_index);
2418         return rc;
2419 }
2420
2421 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2422                 u32 max_ars)
2423 {
2424         struct device *dev = acpi_desc->dev;
2425         struct nd_cmd_ars_status *ars_status;
2426
2427         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2428                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2429                 return 0;
2430         }
2431
2432         if (acpi_desc->ars_status)
2433                 devm_kfree(dev, acpi_desc->ars_status);
2434         acpi_desc->ars_status = NULL;
2435         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2436         if (!ars_status)
2437                 return -ENOMEM;
2438         acpi_desc->ars_status = ars_status;
2439         acpi_desc->ars_status_size = max_ars;
2440         return 0;
2441 }
2442
2443 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2444                 struct nfit_spa *nfit_spa)
2445 {
2446         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2447         int rc;
2448
2449         if (!nfit_spa->max_ars) {
2450                 struct nd_cmd_ars_cap ars_cap;
2451
2452                 memset(&ars_cap, 0, sizeof(ars_cap));
2453                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2454                 if (rc < 0)
2455                         return rc;
2456                 nfit_spa->max_ars = ars_cap.max_ars_out;
2457                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2458                 /* check that the supported scrub types match the spa type */
2459                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2460                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2461                         return -ENOTTY;
2462                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2463                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2464                         return -ENOTTY;
2465         }
2466
2467         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2468                 return -ENOMEM;
2469
2470         rc = ars_get_status(acpi_desc);
2471         if (rc < 0 && rc != -ENOSPC)
2472                 return rc;
2473
2474         if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2475                 return -ENOMEM;
2476
2477         return 0;
2478 }
2479
2480 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2481                 struct nfit_spa *nfit_spa)
2482 {
2483         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2484         unsigned int overflow_retry = scrub_overflow_abort;
2485         u64 init_ars_start = 0, init_ars_len = 0;
2486         struct device *dev = acpi_desc->dev;
2487         unsigned int tmo = scrub_timeout;
2488         int rc;
2489
2490         if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2491                 return;
2492
2493         rc = ars_start(acpi_desc, nfit_spa);
2494         /*
2495          * If we timed out the initial scan we'll still be busy here,
2496          * and will wait another timeout before giving up permanently.
2497          */
2498         if (rc < 0 && rc != -EBUSY)
2499                 return;
2500
2501         do {
2502                 u64 ars_start, ars_len;
2503
2504                 if (acpi_desc->cancel)
2505                         break;
2506                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2507                 if (rc == -ENOTTY)
2508                         break;
2509                 if (rc == -EBUSY && !tmo) {
2510                         dev_warn(dev, "range %d ars timeout, aborting\n",
2511                                         spa->range_index);
2512                         break;
2513                 }
2514
2515                 if (rc == -EBUSY) {
2516                         /*
2517                          * Note, entries may be appended to the list
2518                          * while the lock is dropped, but the workqueue
2519                          * being active prevents entries being deleted /
2520                          * freed.
2521                          */
2522                         mutex_unlock(&acpi_desc->init_mutex);
2523                         ssleep(1);
2524                         tmo--;
2525                         mutex_lock(&acpi_desc->init_mutex);
2526                         continue;
2527                 }
2528
2529                 /* we got some results, but there are more pending... */
2530                 if (rc == -ENOSPC && overflow_retry--) {
2531                         if (!init_ars_len) {
2532                                 init_ars_len = acpi_desc->ars_status->length;
2533                                 init_ars_start = acpi_desc->ars_status->address;
2534                         }
2535                         rc = ars_continue(acpi_desc);
2536                 }
2537
2538                 if (rc < 0) {
2539                         dev_warn(dev, "range %d ars continuation failed\n",
2540                                         spa->range_index);
2541                         break;
2542                 }
2543
2544                 if (init_ars_len) {
2545                         ars_start = init_ars_start;
2546                         ars_len = init_ars_len;
2547                 } else {
2548                         ars_start = acpi_desc->ars_status->address;
2549                         ars_len = acpi_desc->ars_status->length;
2550                 }
2551                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2552                                 spa->range_index, ars_start, ars_len);
2553                 /* notify the region about new poison entries */
2554                 nvdimm_region_notify(nfit_spa->nd_region,
2555                                 NVDIMM_REVALIDATE_POISON);
2556                 break;
2557         } while (1);
2558 }
2559
2560 static void acpi_nfit_scrub(struct work_struct *work)
2561 {
2562         struct device *dev;
2563         u64 init_scrub_length = 0;
2564         struct nfit_spa *nfit_spa;
2565         u64 init_scrub_address = 0;
2566         bool init_ars_done = false;
2567         struct acpi_nfit_desc *acpi_desc;
2568         unsigned int tmo = scrub_timeout;
2569         unsigned int overflow_retry = scrub_overflow_abort;
2570
2571         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2572         dev = acpi_desc->dev;
2573
2574         /*
2575          * We scrub in 2 phases.  The first phase waits for any platform
2576          * firmware initiated scrubs to complete and then we go search for the
2577          * affected spa regions to mark them scanned.  In the second phase we
2578          * initiate a directed scrub for every range that was not scrubbed in
2579          * phase 1. If we're called for a 'rescan', we harmlessly pass through
2580          * the first phase, but really only care about running phase 2, where
2581          * regions can be notified of new poison.
2582          */
2583
2584         /* process platform firmware initiated scrubs */
2585  retry:
2586         mutex_lock(&acpi_desc->init_mutex);
2587         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2588                 struct nd_cmd_ars_status *ars_status;
2589                 struct acpi_nfit_system_address *spa;
2590                 u64 ars_start, ars_len;
2591                 int rc;
2592
2593                 if (acpi_desc->cancel)
2594                         break;
2595
2596                 if (nfit_spa->nd_region)
2597                         continue;
2598
2599                 if (init_ars_done) {
2600                         /*
2601                          * No need to re-query, we're now just
2602                          * reconciling all the ranges covered by the
2603                          * initial scrub
2604                          */
2605                         rc = 0;
2606                 } else
2607                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2608
2609                 if (rc == -ENOTTY) {
2610                         /* no ars capability, just register spa and move on */
2611                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2612                         continue;
2613                 }
2614
2615                 if (rc == -EBUSY && !tmo) {
2616                         /* fallthrough to directed scrub in phase 2 */
2617                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2618                         break;
2619                 } else if (rc == -EBUSY) {
2620                         mutex_unlock(&acpi_desc->init_mutex);
2621                         ssleep(1);
2622                         tmo--;
2623                         goto retry;
2624                 }
2625
2626                 /* we got some results, but there are more pending... */
2627                 if (rc == -ENOSPC && overflow_retry--) {
2628                         ars_status = acpi_desc->ars_status;
2629                         /*
2630                          * Record the original scrub range, so that we
2631                          * can recall all the ranges impacted by the
2632                          * initial scrub.
2633                          */
2634                         if (!init_scrub_length) {
2635                                 init_scrub_length = ars_status->length;
2636                                 init_scrub_address = ars_status->address;
2637                         }
2638                         rc = ars_continue(acpi_desc);
2639                         if (rc == 0) {
2640                                 mutex_unlock(&acpi_desc->init_mutex);
2641                                 goto retry;
2642                         }
2643                 }
2644
2645                 if (rc < 0) {
2646                         /*
2647                          * Initial scrub failed, we'll give it one more
2648                          * try below...
2649                          */
2650                         break;
2651                 }
2652
2653                 /* We got some final results, record completed ranges */
2654                 ars_status = acpi_desc->ars_status;
2655                 if (init_scrub_length) {
2656                         ars_start = init_scrub_address;
2657                         ars_len = ars_start + init_scrub_length;
2658                 } else {
2659                         ars_start = ars_status->address;
2660                         ars_len = ars_status->length;
2661                 }
2662                 spa = nfit_spa->spa;
2663
2664                 if (!init_ars_done) {
2665                         init_ars_done = true;
2666                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2667                                         ars_start, ars_len);
2668                 }
2669                 if (ars_start <= spa->address && ars_start + ars_len
2670                                 >= spa->address + spa->length)
2671                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2672         }
2673
2674         /*
2675          * For all the ranges not covered by an initial scrub we still
2676          * want to see if there are errors, but it's ok to discover them
2677          * asynchronously.
2678          */
2679         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2680                 /*
2681                  * Flag all the ranges that still need scrubbing, but
2682                  * register them now to make data available.
2683                  */
2684                 if (!nfit_spa->nd_region) {
2685                         nfit_spa->ars_required = 1;
2686                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2687                 }
2688         }
2689         acpi_desc->init_complete = 1;
2690
2691         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2692                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2693         acpi_desc->scrub_count++;
2694         acpi_desc->ars_start_flags = 0;
2695         if (acpi_desc->scrub_count_state)
2696                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
2697         mutex_unlock(&acpi_desc->init_mutex);
2698 }
2699
2700 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2701 {
2702         struct nfit_spa *nfit_spa;
2703         int rc;
2704
2705         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2706                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2707                         /* BLK regions don't need to wait for ars results */
2708                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2709                         if (rc)
2710                                 return rc;
2711                 }
2712
2713         acpi_desc->ars_start_flags = 0;
2714         if (!acpi_desc->cancel)
2715                 queue_work(nfit_wq, &acpi_desc->work);
2716         return 0;
2717 }
2718
2719 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2720                 struct nfit_table_prev *prev)
2721 {
2722         struct device *dev = acpi_desc->dev;
2723
2724         if (!list_empty(&prev->spas) ||
2725                         !list_empty(&prev->memdevs) ||
2726                         !list_empty(&prev->dcrs) ||
2727                         !list_empty(&prev->bdws) ||
2728                         !list_empty(&prev->idts) ||
2729                         !list_empty(&prev->flushes)) {
2730                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2731                 return -ENXIO;
2732         }
2733         return 0;
2734 }
2735
2736 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
2737 {
2738         struct device *dev = acpi_desc->dev;
2739         struct kernfs_node *nfit;
2740         struct device *bus_dev;
2741
2742         if (!ars_supported(acpi_desc->nvdimm_bus))
2743                 return 0;
2744
2745         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2746         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
2747         if (!nfit) {
2748                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
2749                 return -ENODEV;
2750         }
2751         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
2752         sysfs_put(nfit);
2753         if (!acpi_desc->scrub_count_state) {
2754                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
2755                 return -ENODEV;
2756         }
2757
2758         return 0;
2759 }
2760
2761 static void acpi_nfit_unregister(void *data)
2762 {
2763         struct acpi_nfit_desc *acpi_desc = data;
2764
2765         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2766 }
2767
2768 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
2769 {
2770         struct device *dev = acpi_desc->dev;
2771         struct nfit_table_prev prev;
2772         const void *end;
2773         int rc;
2774
2775         if (!acpi_desc->nvdimm_bus) {
2776                 acpi_nfit_init_dsms(acpi_desc);
2777
2778                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
2779                                 &acpi_desc->nd_desc);
2780                 if (!acpi_desc->nvdimm_bus)
2781                         return -ENOMEM;
2782
2783                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
2784                                 acpi_desc);
2785                 if (rc)
2786                         return rc;
2787
2788                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
2789                 if (rc)
2790                         return rc;
2791
2792                 /* register this acpi_desc for mce notifications */
2793                 mutex_lock(&acpi_desc_lock);
2794                 list_add_tail(&acpi_desc->list, &acpi_descs);
2795                 mutex_unlock(&acpi_desc_lock);
2796         }
2797
2798         mutex_lock(&acpi_desc->init_mutex);
2799
2800         INIT_LIST_HEAD(&prev.spas);
2801         INIT_LIST_HEAD(&prev.memdevs);
2802         INIT_LIST_HEAD(&prev.dcrs);
2803         INIT_LIST_HEAD(&prev.bdws);
2804         INIT_LIST_HEAD(&prev.idts);
2805         INIT_LIST_HEAD(&prev.flushes);
2806
2807         list_cut_position(&prev.spas, &acpi_desc->spas,
2808                                 acpi_desc->spas.prev);
2809         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2810                                 acpi_desc->memdevs.prev);
2811         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2812                                 acpi_desc->dcrs.prev);
2813         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2814                                 acpi_desc->bdws.prev);
2815         list_cut_position(&prev.idts, &acpi_desc->idts,
2816                                 acpi_desc->idts.prev);
2817         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2818                                 acpi_desc->flushes.prev);
2819
2820         end = data + sz;
2821         while (!IS_ERR_OR_NULL(data))
2822                 data = add_table(acpi_desc, &prev, data, end);
2823
2824         if (IS_ERR(data)) {
2825                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2826                                 PTR_ERR(data));
2827                 rc = PTR_ERR(data);
2828                 goto out_unlock;
2829         }
2830
2831         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2832         if (rc)
2833                 goto out_unlock;
2834
2835         rc = nfit_mem_init(acpi_desc);
2836         if (rc)
2837                 goto out_unlock;
2838
2839         rc = acpi_nfit_register_dimms(acpi_desc);
2840         if (rc)
2841                 goto out_unlock;
2842
2843         rc = acpi_nfit_register_regions(acpi_desc);
2844
2845  out_unlock:
2846         mutex_unlock(&acpi_desc->init_mutex);
2847         return rc;
2848 }
2849 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2850
2851 struct acpi_nfit_flush_work {
2852         struct work_struct work;
2853         struct completion cmp;
2854 };
2855
2856 static void flush_probe(struct work_struct *work)
2857 {
2858         struct acpi_nfit_flush_work *flush;
2859
2860         flush = container_of(work, typeof(*flush), work);
2861         complete(&flush->cmp);
2862 }
2863
2864 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2865 {
2866         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2867         struct device *dev = acpi_desc->dev;
2868         struct acpi_nfit_flush_work flush;
2869         int rc;
2870
2871         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2872         device_lock(dev);
2873         device_unlock(dev);
2874
2875         /* bounce the init_mutex to make init_complete valid */
2876         mutex_lock(&acpi_desc->init_mutex);
2877         if (acpi_desc->cancel || acpi_desc->init_complete) {
2878                 mutex_unlock(&acpi_desc->init_mutex);
2879                 return 0;
2880         }
2881
2882         /*
2883          * Scrub work could take 10s of seconds, userspace may give up so we
2884          * need to be interruptible while waiting.
2885          */
2886         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2887         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2888         queue_work(nfit_wq, &flush.work);
2889         mutex_unlock(&acpi_desc->init_mutex);
2890
2891         rc = wait_for_completion_interruptible(&flush.cmp);
2892         cancel_work_sync(&flush.work);
2893         return rc;
2894 }
2895
2896 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2897                 struct nvdimm *nvdimm, unsigned int cmd)
2898 {
2899         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2900
2901         if (nvdimm)
2902                 return 0;
2903         if (cmd != ND_CMD_ARS_START)
2904                 return 0;
2905
2906         /*
2907          * The kernel and userspace may race to initiate a scrub, but
2908          * the scrub thread is prepared to lose that initial race.  It
2909          * just needs guarantees that any ars it initiates are not
2910          * interrupted by any intervening start reqeusts from userspace.
2911          */
2912         if (work_busy(&acpi_desc->work))
2913                 return -EBUSY;
2914
2915         return 0;
2916 }
2917
2918 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
2919 {
2920         struct device *dev = acpi_desc->dev;
2921         struct nfit_spa *nfit_spa;
2922
2923         if (work_busy(&acpi_desc->work))
2924                 return -EBUSY;
2925
2926         mutex_lock(&acpi_desc->init_mutex);
2927         if (acpi_desc->cancel) {
2928                 mutex_unlock(&acpi_desc->init_mutex);
2929                 return 0;
2930         }
2931
2932         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2933                 struct acpi_nfit_system_address *spa = nfit_spa->spa;
2934
2935                 if (nfit_spa_type(spa) != NFIT_SPA_PM)
2936                         continue;
2937
2938                 nfit_spa->ars_required = 1;
2939         }
2940         acpi_desc->ars_start_flags = flags;
2941         queue_work(nfit_wq, &acpi_desc->work);
2942         dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
2943         mutex_unlock(&acpi_desc->init_mutex);
2944
2945         return 0;
2946 }
2947
2948 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2949 {
2950         struct nvdimm_bus_descriptor *nd_desc;
2951
2952         dev_set_drvdata(dev, acpi_desc);
2953         acpi_desc->dev = dev;
2954         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2955         nd_desc = &acpi_desc->nd_desc;
2956         nd_desc->provider_name = "ACPI.NFIT";
2957         nd_desc->module = THIS_MODULE;
2958         nd_desc->ndctl = acpi_nfit_ctl;
2959         nd_desc->flush_probe = acpi_nfit_flush_probe;
2960         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2961         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2962
2963         INIT_LIST_HEAD(&acpi_desc->spas);
2964         INIT_LIST_HEAD(&acpi_desc->dcrs);
2965         INIT_LIST_HEAD(&acpi_desc->bdws);
2966         INIT_LIST_HEAD(&acpi_desc->idts);
2967         INIT_LIST_HEAD(&acpi_desc->flushes);
2968         INIT_LIST_HEAD(&acpi_desc->memdevs);
2969         INIT_LIST_HEAD(&acpi_desc->dimms);
2970         INIT_LIST_HEAD(&acpi_desc->list);
2971         mutex_init(&acpi_desc->init_mutex);
2972         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2973 }
2974 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2975
2976 static void acpi_nfit_put_table(void *table)
2977 {
2978         acpi_put_table(table);
2979 }
2980
2981 void acpi_nfit_shutdown(void *data)
2982 {
2983         struct acpi_nfit_desc *acpi_desc = data;
2984         struct device *bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
2985
2986         /*
2987          * Destruct under acpi_desc_lock so that nfit_handle_mce does not
2988          * race teardown
2989          */
2990         mutex_lock(&acpi_desc_lock);
2991         list_del(&acpi_desc->list);
2992         mutex_unlock(&acpi_desc_lock);
2993
2994         mutex_lock(&acpi_desc->init_mutex);
2995         acpi_desc->cancel = 1;
2996         mutex_unlock(&acpi_desc->init_mutex);
2997
2998         /*
2999          * Bounce the nvdimm bus lock to make sure any in-flight
3000          * acpi_nfit_ars_rescan() submissions have had a chance to
3001          * either submit or see ->cancel set.
3002          */
3003         device_lock(bus_dev);
3004         device_unlock(bus_dev);
3005
3006         flush_workqueue(nfit_wq);
3007 }
3008 EXPORT_SYMBOL_GPL(acpi_nfit_shutdown);
3009
3010 static int acpi_nfit_add(struct acpi_device *adev)
3011 {
3012         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3013         struct acpi_nfit_desc *acpi_desc;
3014         struct device *dev = &adev->dev;
3015         struct acpi_table_header *tbl;
3016         acpi_status status = AE_OK;
3017         acpi_size sz;
3018         int rc = 0;
3019
3020         status = acpi_get_table(ACPI_SIG_NFIT, 0, &tbl);
3021         if (ACPI_FAILURE(status)) {
3022                 /* This is ok, we could have an nvdimm hotplugged later */
3023                 dev_dbg(dev, "failed to find NFIT at startup\n");
3024                 return 0;
3025         }
3026
3027         rc = devm_add_action_or_reset(dev, acpi_nfit_put_table, tbl);
3028         if (rc)
3029                 return rc;
3030         sz = tbl->length;
3031
3032         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3033         if (!acpi_desc)
3034                 return -ENOMEM;
3035         acpi_nfit_desc_init(acpi_desc, &adev->dev);
3036
3037         /* Save the acpi header for exporting the revision via sysfs */
3038         acpi_desc->acpi_header = *tbl;
3039
3040         /* Evaluate _FIT and override with that if present */
3041         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
3042         if (ACPI_SUCCESS(status) && buf.length > 0) {
3043                 union acpi_object *obj = buf.pointer;
3044
3045                 if (obj->type == ACPI_TYPE_BUFFER)
3046                         rc = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3047                                         obj->buffer.length);
3048                 else
3049                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
3050                                  __func__, (int) obj->type);
3051                 kfree(buf.pointer);
3052         } else
3053                 /* skip over the lead-in header table */
3054                 rc = acpi_nfit_init(acpi_desc, (void *) tbl
3055                                 + sizeof(struct acpi_table_nfit),
3056                                 sz - sizeof(struct acpi_table_nfit));
3057
3058         if (rc)
3059                 return rc;
3060         return devm_add_action_or_reset(dev, acpi_nfit_shutdown, acpi_desc);
3061 }
3062
3063 static int acpi_nfit_remove(struct acpi_device *adev)
3064 {
3065         /* see acpi_nfit_unregister */
3066         return 0;
3067 }
3068
3069 static void acpi_nfit_update_notify(struct device *dev, acpi_handle handle)
3070 {
3071         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3072         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
3073         union acpi_object *obj;
3074         acpi_status status;
3075         int ret;
3076
3077         if (!dev->driver) {
3078                 /* dev->driver may be null if we're being removed */
3079                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
3080                 return;
3081         }
3082
3083         if (!acpi_desc) {
3084                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
3085                 if (!acpi_desc)
3086                         return;
3087                 acpi_nfit_desc_init(acpi_desc, dev);
3088         } else {
3089                 /*
3090                  * Finish previous registration before considering new
3091                  * regions.
3092                  */
3093                 flush_workqueue(nfit_wq);
3094         }
3095
3096         /* Evaluate _FIT */
3097         status = acpi_evaluate_object(handle, "_FIT", NULL, &buf);
3098         if (ACPI_FAILURE(status)) {
3099                 dev_err(dev, "failed to evaluate _FIT\n");
3100                 return;
3101         }
3102
3103         obj = buf.pointer;
3104         if (obj->type == ACPI_TYPE_BUFFER) {
3105                 ret = acpi_nfit_init(acpi_desc, obj->buffer.pointer,
3106                                 obj->buffer.length);
3107                 if (ret)
3108                         dev_err(dev, "failed to merge updated NFIT\n");
3109         } else
3110                 dev_err(dev, "Invalid _FIT\n");
3111         kfree(buf.pointer);
3112 }
3113
3114 static void acpi_nfit_uc_error_notify(struct device *dev, acpi_handle handle)
3115 {
3116         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(dev);
3117         u8 flags = (acpi_desc->scrub_mode == HW_ERROR_SCRUB_ON) ?
3118                         0 : ND_ARS_RETURN_PREV_DATA;
3119
3120         acpi_nfit_ars_rescan(acpi_desc, flags);
3121 }
3122
3123 void __acpi_nfit_notify(struct device *dev, acpi_handle handle, u32 event)
3124 {
3125         dev_dbg(dev, "%s: event: 0x%x\n", __func__, event);
3126
3127         switch (event) {
3128         case NFIT_NOTIFY_UPDATE:
3129                 return acpi_nfit_update_notify(dev, handle);
3130         case NFIT_NOTIFY_UC_MEMORY_ERROR:
3131                 return acpi_nfit_uc_error_notify(dev, handle);
3132         default:
3133                 return;
3134         }
3135 }
3136 EXPORT_SYMBOL_GPL(__acpi_nfit_notify);
3137
3138 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
3139 {
3140         device_lock(&adev->dev);
3141         __acpi_nfit_notify(&adev->dev, adev->handle, event);
3142         device_unlock(&adev->dev);
3143 }
3144
3145 static const struct acpi_device_id acpi_nfit_ids[] = {
3146         { "ACPI0012", 0 },
3147         { "", 0 },
3148 };
3149 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
3150
3151 static struct acpi_driver acpi_nfit_driver = {
3152         .name = KBUILD_MODNAME,
3153         .ids = acpi_nfit_ids,
3154         .ops = {
3155                 .add = acpi_nfit_add,
3156                 .remove = acpi_nfit_remove,
3157                 .notify = acpi_nfit_notify,
3158         },
3159 };
3160
3161 static __init int nfit_init(void)
3162 {
3163         int ret;
3164
3165         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
3166         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
3167         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
3168         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
3169         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
3170         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
3171         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
3172
3173         guid_parse(UUID_VOLATILE_MEMORY, &nfit_uuid[NFIT_SPA_VOLATILE]);
3174         guid_parse(UUID_PERSISTENT_MEMORY, &nfit_uuid[NFIT_SPA_PM]);
3175         guid_parse(UUID_CONTROL_REGION, &nfit_uuid[NFIT_SPA_DCR]);
3176         guid_parse(UUID_DATA_REGION, &nfit_uuid[NFIT_SPA_BDW]);
3177         guid_parse(UUID_VOLATILE_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_VDISK]);
3178         guid_parse(UUID_VOLATILE_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_VCD]);
3179         guid_parse(UUID_PERSISTENT_VIRTUAL_DISK, &nfit_uuid[NFIT_SPA_PDISK]);
3180         guid_parse(UUID_PERSISTENT_VIRTUAL_CD, &nfit_uuid[NFIT_SPA_PCD]);
3181         guid_parse(UUID_NFIT_BUS, &nfit_uuid[NFIT_DEV_BUS]);
3182         guid_parse(UUID_NFIT_DIMM, &nfit_uuid[NFIT_DEV_DIMM]);
3183         guid_parse(UUID_NFIT_DIMM_N_HPE1, &nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
3184         guid_parse(UUID_NFIT_DIMM_N_HPE2, &nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
3185         guid_parse(UUID_NFIT_DIMM_N_MSFT, &nfit_uuid[NFIT_DEV_DIMM_N_MSFT]);
3186
3187         nfit_wq = create_singlethread_workqueue("nfit");
3188         if (!nfit_wq)
3189                 return -ENOMEM;
3190
3191         nfit_mce_register();
3192         ret = acpi_bus_register_driver(&acpi_nfit_driver);
3193         if (ret) {
3194                 nfit_mce_unregister();
3195                 destroy_workqueue(nfit_wq);
3196         }
3197
3198         return ret;
3199
3200 }
3201
3202 static __exit void nfit_exit(void)
3203 {
3204         nfit_mce_unregister();
3205         acpi_bus_unregister_driver(&acpi_nfit_driver);
3206         destroy_workqueue(nfit_wq);
3207         WARN_ON(!list_empty(&acpi_descs));
3208 }
3209
3210 module_init(nfit_init);
3211 module_exit(nfit_exit);
3212 MODULE_LICENSE("GPL v2");
3213 MODULE_AUTHOR("Intel Corporation");