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