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