drivers/acpi/nfit.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/delay.h>
  19 #include <linux/list.h>
  20 #include <linux/acpi.h>
  21 #include <linux/sort.h>
  22 #include <linux/pmem.h>
  23 #include <linux/io.h>
  24 #include <linux/nd.h>
  25 #include <asm/cacheflush.h>
  26 #include "nfit.h"
  27
  28 /*
  29  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  30  * irrelevant.
  31  */
  32 #include <linux/io-64-nonatomic-hi-lo.h>
  33
  34 static bool force_enable_dimms;
  35 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  36 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  37
  38 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
  39 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
  40 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
  41
  42 /* after three payloads of overflow, it's dead jim */
  43 static unsigned int scrub_overflow_abort = 3;
  44 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
  45 MODULE_PARM_DESC(scrub_overflow_abort,
  46                 "Number of times we overflow ARS results before abort");
  47
  48 static struct workqueue_struct *nfit_wq;
  49
  50 struct nfit_table_prev {
  51         struct list_head spas;
  52         struct list_head memdevs;
  53         struct list_head dcrs;
  54         struct list_head bdws;
  55         struct list_head idts;
  56         struct list_head flushes;
  57 };
  58
  59 static u8 nfit_uuid[NFIT_UUID_MAX][16];
  60
  61 const u8 *to_nfit_uuid(enum nfit_uuids id)
  62 {
  63         return nfit_uuid[id];
  64 }
  65 EXPORT_SYMBOL(to_nfit_uuid);
  66
  67 static struct acpi_nfit_desc *to_acpi_nfit_desc(
  68                 struct nvdimm_bus_descriptor *nd_desc)
  69 {
  70         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
  71 }
  72
  73 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  74 {
  75         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  76
  77         /*
  78          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  79          * acpi_device.
  80          */
  81         if (!nd_desc->provider_name
  82                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  83                 return NULL;
  84
  85         return to_acpi_device(acpi_desc->dev);
  86 }
  87
  88 static int xlat_status(void *buf, unsigned int cmd)
  89 {
  90         struct nd_cmd_clear_error *clear_err;
  91         struct nd_cmd_ars_status *ars_status;
  92         struct nd_cmd_ars_start *ars_start;
  93         struct nd_cmd_ars_cap *ars_cap;
  94         u16 flags;
  95
  96         switch (cmd) {
  97         case ND_CMD_ARS_CAP:
  98                 ars_cap = buf;
  99                 if ((ars_cap->status & 0xffff) == NFIT_ARS_CAP_NONE)
 100                         return -ENOTTY;
 101
 102                 /* Command failed */
 103                 if (ars_cap->status & 0xffff)
 104                         return -EIO;
 105
 106                 /* No supported scan types for this range */
 107                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 108                 if ((ars_cap->status >> 16 & flags) == 0)
 109                         return -ENOTTY;
 110                 break;
 111         case ND_CMD_ARS_START:
 112                 ars_start = buf;
 113                 /* ARS is in progress */
 114                 if ((ars_start->status & 0xffff) == NFIT_ARS_START_BUSY)
 115                         return -EBUSY;
 116
 117                 /* Command failed */
 118                 if (ars_start->status & 0xffff)
 119                         return -EIO;
 120                 break;
 121         case ND_CMD_ARS_STATUS:
 122                 ars_status = buf;
 123                 /* Command failed */
 124                 if (ars_status->status & 0xffff)
 125                         return -EIO;
 126                 /* Check extended status (Upper two bytes) */
 127                 if (ars_status->status == NFIT_ARS_STATUS_DONE)
 128                         return 0;
 129
 130                 /* ARS is in progress */
 131                 if (ars_status->status == NFIT_ARS_STATUS_BUSY)
 132                         return -EBUSY;
 133
 134                 /* No ARS performed for the current boot */
 135                 if (ars_status->status == NFIT_ARS_STATUS_NONE)
 136                         return -EAGAIN;
 137
 138                 /*
 139                  * ARS interrupted, either we overflowed or some other
 140                  * agent wants the scan to stop.  If we didn't overflow
 141                  * then just continue with the returned results.
 142                  */
 143                 if (ars_status->status == NFIT_ARS_STATUS_INTR) {
 144                         if (ars_status->flags & NFIT_ARS_F_OVERFLOW)
 145                                 return -ENOSPC;
 146                         return 0;
 147                 }
 148
 149                 /* Unknown status */
 150                 if (ars_status->status >> 16)
 151                         return -EIO;
 152                 break;
 153         case ND_CMD_CLEAR_ERROR:
 154                 clear_err = buf;
 155                 if (clear_err->status & 0xffff)
 156                         return -EIO;
 157                 if (!clear_err->cleared)
 158                         return -EIO;
 159                 if (clear_err->length > clear_err->cleared)
 160                         return clear_err->cleared;
 161                 break;
 162         default:
 163                 break;
 164         }
 165
 166         return 0;
 167 }
 168
 169 static int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc,
 170                 struct nvdimm *nvdimm, unsigned int cmd, void *buf,
 171                 unsigned int buf_len, int *cmd_rc)
 172 {
 173         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
 174         const struct nd_cmd_desc *desc = NULL;
 175         union acpi_object in_obj, in_buf, *out_obj;
 176         struct device *dev = acpi_desc->dev;
 177         const char *cmd_name, *dimm_name;
 178         unsigned long dsm_mask;
 179         acpi_handle handle;
 180         const u8 *uuid;
 181         u32 offset;
 182         int rc, i;
 183
 184         if (nvdimm) {
 185                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 186                 struct acpi_device *adev = nfit_mem->adev;
 187
 188                 if (!adev)
 189                         return -ENOTTY;
 190                 dimm_name = nvdimm_name(nvdimm);
 191                 cmd_name = nvdimm_cmd_name(cmd);
 192                 dsm_mask = nfit_mem->dsm_mask;
 193                 desc = nd_cmd_dimm_desc(cmd);
 194                 uuid = to_nfit_uuid(NFIT_DEV_DIMM);
 195                 handle = adev->handle;
 196         } else {
 197                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 198
 199                 cmd_name = nvdimm_bus_cmd_name(cmd);
 200                 dsm_mask = nd_desc->dsm_mask;
 201                 desc = nd_cmd_bus_desc(cmd);
 202                 uuid = to_nfit_uuid(NFIT_DEV_BUS);
 203                 handle = adev->handle;
 204                 dimm_name = "bus";
 205         }
 206
 207         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 208                 return -ENOTTY;
 209
 210         if (!test_bit(cmd, &dsm_mask))
 211                 return -ENOTTY;
 212
 213         in_obj.type = ACPI_TYPE_PACKAGE;
 214         in_obj.package.count = 1;
 215         in_obj.package.elements = &in_buf;
 216         in_buf.type = ACPI_TYPE_BUFFER;
 217         in_buf.buffer.pointer = buf;
 218         in_buf.buffer.length = 0;
 219
 220         /* libnvdimm has already validated the input envelope */
 221         for (i = 0; i < desc->in_num; i++)
 222                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 223                                 i, buf);
 224
 225         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
 226                 dev_dbg(dev, "%s:%s cmd: %s input length: %d\n", __func__,
 227                                 dimm_name, cmd_name, in_buf.buffer.length);
 228                 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
 229                                 4, in_buf.buffer.pointer, min_t(u32, 128,
 230                                         in_buf.buffer.length), true);
 231         }
 232
 233         out_obj = acpi_evaluate_dsm(handle, uuid, 1, cmd, &in_obj);
 234         if (!out_obj) {
 235                 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
 236                                 cmd_name);
 237                 return -EINVAL;
 238         }
 239
 240         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 241                 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
 242                                 __func__, dimm_name, cmd_name, out_obj->type);
 243                 rc = -EINVAL;
 244                 goto out;
 245         }
 246
 247         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
 248                 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
 249                                 dimm_name, cmd_name, out_obj->buffer.length);
 250                 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
 251                                 4, out_obj->buffer.pointer, min_t(u32, 128,
 252                                         out_obj->buffer.length), true);
 253         }
 254
 255         for (i = 0, offset = 0; i < desc->out_num; i++) {
 256                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 257                                 (u32 *) out_obj->buffer.pointer);
 258
 259                 if (offset + out_size > out_obj->buffer.length) {
 260                         dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
 261                                         __func__, dimm_name, cmd_name, i);
 262                         break;
 263                 }
 264
 265                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 266                         dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
 267                                         __func__, dimm_name, cmd_name, i);
 268                         rc = -ENXIO;
 269                         goto out;
 270                 }
 271                 memcpy(buf + in_buf.buffer.length + offset,
 272                                 out_obj->buffer.pointer + offset, out_size);
 273                 offset += out_size;
 274         }
 275         if (offset + in_buf.buffer.length < buf_len) {
 276                 if (i >= 1) {
 277                         /*
 278                          * status valid, return the number of bytes left
 279                          * unfilled in the output buffer
 280                          */
 281                         rc = buf_len - offset - in_buf.buffer.length;
 282                         if (cmd_rc)
 283                                 *cmd_rc = xlat_status(buf, cmd);
 284                 } else {
 285                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 286                                         __func__, dimm_name, cmd_name, buf_len,
 287                                         offset);
 288                         rc = -ENXIO;
 289                 }
 290         } else {
 291                 rc = 0;
 292                 if (cmd_rc)
 293                         *cmd_rc = xlat_status(buf, cmd);
 294         }
 295
 296  out:
 297         ACPI_FREE(out_obj);
 298
 299         return rc;
 300 }
 301
 302 static const char *spa_type_name(u16 type)
 303 {
 304         static const char *to_name[] = {
 305                 [NFIT_SPA_VOLATILE] = "volatile",
 306                 [NFIT_SPA_PM] = "pmem",
 307                 [NFIT_SPA_DCR] = "dimm-control-region",
 308                 [NFIT_SPA_BDW] = "block-data-window",
 309                 [NFIT_SPA_VDISK] = "volatile-disk",
 310                 [NFIT_SPA_VCD] = "volatile-cd",
 311                 [NFIT_SPA_PDISK] = "persistent-disk",
 312                 [NFIT_SPA_PCD] = "persistent-cd",
 313
 314         };
 315
 316         if (type > NFIT_SPA_PCD)
 317                 return "unknown";
 318
 319         return to_name[type];
 320 }
 321
 322 static int nfit_spa_type(struct acpi_nfit_system_address *spa)
 323 {
 324         int i;
 325
 326         for (i = 0; i < NFIT_UUID_MAX; i++)
 327                 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
 328                         return i;
 329         return -1;
 330 }
 331
 332 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 333                 struct nfit_table_prev *prev,
 334                 struct acpi_nfit_system_address *spa)
 335 {
 336         size_t length = min_t(size_t, sizeof(*spa), spa->header.length);
 337         struct device *dev = acpi_desc->dev;
 338         struct nfit_spa *nfit_spa;
 339
 340         list_for_each_entry(nfit_spa, &prev->spas, list) {
 341                 if (memcmp(nfit_spa->spa, spa, length) == 0) {
 342                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 343                         return true;
 344                 }
 345         }
 346
 347         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa), GFP_KERNEL);
 348         if (!nfit_spa)
 349                 return false;
 350         INIT_LIST_HEAD(&nfit_spa->list);
 351         nfit_spa->spa = spa;
 352         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 353         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
 354                         spa->range_index,
 355                         spa_type_name(nfit_spa_type(spa)));
 356         return true;
 357 }
 358
 359 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 360                 struct nfit_table_prev *prev,
 361                 struct acpi_nfit_memory_map *memdev)
 362 {
 363         size_t length = min_t(size_t, sizeof(*memdev), memdev->header.length);
 364         struct device *dev = acpi_desc->dev;
 365         struct nfit_memdev *nfit_memdev;
 366
 367         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 368                 if (memcmp(nfit_memdev->memdev, memdev, length) == 0) {
 369                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 370                         return true;
 371                 }
 372
 373         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev), GFP_KERNEL);
 374         if (!nfit_memdev)
 375                 return false;
 376         INIT_LIST_HEAD(&nfit_memdev->list);
 377         nfit_memdev->memdev = memdev;
 378         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 379         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
 380                         __func__, memdev->device_handle, memdev->range_index,
 381                         memdev->region_index);
 382         return true;
 383 }
 384
 385 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 386                 struct nfit_table_prev *prev,
 387                 struct acpi_nfit_control_region *dcr)
 388 {
 389         size_t length = min_t(size_t, sizeof(*dcr), dcr->header.length);
 390         struct device *dev = acpi_desc->dev;
 391         struct nfit_dcr *nfit_dcr;
 392
 393         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 394                 if (memcmp(nfit_dcr->dcr, dcr, length) == 0) {
 395                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 396                         return true;
 397                 }
 398
 399         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr), GFP_KERNEL);
 400         if (!nfit_dcr)
 401                 return false;
 402         INIT_LIST_HEAD(&nfit_dcr->list);
 403         nfit_dcr->dcr = dcr;
 404         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 405         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
 406                         dcr->region_index, dcr->windows);
 407         return true;
 408 }
 409
 410 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 411                 struct nfit_table_prev *prev,
 412                 struct acpi_nfit_data_region *bdw)
 413 {
 414         size_t length = min_t(size_t, sizeof(*bdw), bdw->header.length);
 415         struct device *dev = acpi_desc->dev;
 416         struct nfit_bdw *nfit_bdw;
 417
 418         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 419                 if (memcmp(nfit_bdw->bdw, bdw, length) == 0) {
 420                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 421                         return true;
 422                 }
 423
 424         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw), GFP_KERNEL);
 425         if (!nfit_bdw)
 426                 return false;
 427         INIT_LIST_HEAD(&nfit_bdw->list);
 428         nfit_bdw->bdw = bdw;
 429         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 430         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
 431                         bdw->region_index, bdw->windows);
 432         return true;
 433 }
 434
 435 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 436                 struct nfit_table_prev *prev,
 437                 struct acpi_nfit_interleave *idt)
 438 {
 439         size_t length = min_t(size_t, sizeof(*idt), idt->header.length);
 440         struct device *dev = acpi_desc->dev;
 441         struct nfit_idt *nfit_idt;
 442
 443         list_for_each_entry(nfit_idt, &prev->idts, list)
 444                 if (memcmp(nfit_idt->idt, idt, length) == 0) {
 445                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 446                         return true;
 447                 }
 448
 449         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt), GFP_KERNEL);
 450         if (!nfit_idt)
 451                 return false;
 452         INIT_LIST_HEAD(&nfit_idt->list);
 453         nfit_idt->idt = idt;
 454         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 455         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
 456                         idt->interleave_index, idt->line_count);
 457         return true;
 458 }
 459
 460 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 461                 struct nfit_table_prev *prev,
 462                 struct acpi_nfit_flush_address *flush)
 463 {
 464         size_t length = min_t(size_t, sizeof(*flush), flush->header.length);
 465         struct device *dev = acpi_desc->dev;
 466         struct nfit_flush *nfit_flush;
 467
 468         list_for_each_entry(nfit_flush, &prev->flushes, list)
 469                 if (memcmp(nfit_flush->flush, flush, length) == 0) {
 470                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 471                         return true;
 472                 }
 473
 474         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush), GFP_KERNEL);
 475         if (!nfit_flush)
 476                 return false;
 477         INIT_LIST_HEAD(&nfit_flush->list);
 478         nfit_flush->flush = flush;
 479         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 480         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
 481                         flush->device_handle, flush->hint_count);
 482         return true;
 483 }
 484
 485 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 486                 struct nfit_table_prev *prev, void *table, const void *end)
 487 {
 488         struct device *dev = acpi_desc->dev;
 489         struct acpi_nfit_header *hdr;
 490         void *err = ERR_PTR(-ENOMEM);
 491
 492         if (table >= end)
 493                 return NULL;
 494
 495         hdr = table;
 496         if (!hdr->length) {
 497                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 498                         hdr->type);
 499                 return NULL;
 500         }
 501
 502         switch (hdr->type) {
 503         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 504                 if (!add_spa(acpi_desc, prev, table))
 505                         return err;
 506                 break;
 507         case ACPI_NFIT_TYPE_MEMORY_MAP:
 508                 if (!add_memdev(acpi_desc, prev, table))
 509                         return err;
 510                 break;
 511         case ACPI_NFIT_TYPE_CONTROL_REGION:
 512                 if (!add_dcr(acpi_desc, prev, table))
 513                         return err;
 514                 break;
 515         case ACPI_NFIT_TYPE_DATA_REGION:
 516                 if (!add_bdw(acpi_desc, prev, table))
 517                         return err;
 518                 break;
 519         case ACPI_NFIT_TYPE_INTERLEAVE:
 520                 if (!add_idt(acpi_desc, prev, table))
 521                         return err;
 522                 break;
 523         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 524                 if (!add_flush(acpi_desc, prev, table))
 525                         return err;
 526                 break;
 527         case ACPI_NFIT_TYPE_SMBIOS:
 528                 dev_dbg(dev, "%s: smbios\n", __func__);
 529                 break;
 530         default:
 531                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 532                 break;
 533         }
 534
 535         return table + hdr->length;
 536 }
 537
 538 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 539                 struct nfit_mem *nfit_mem)
 540 {
 541         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 542         u16 dcr = nfit_mem->dcr->region_index;
 543         struct nfit_spa *nfit_spa;
 544
 545         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 546                 u16 range_index = nfit_spa->spa->range_index;
 547                 int type = nfit_spa_type(nfit_spa->spa);
 548                 struct nfit_memdev *nfit_memdev;
 549
 550                 if (type != NFIT_SPA_BDW)
 551                         continue;
 552
 553                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 554                         if (nfit_memdev->memdev->range_index != range_index)
 555                                 continue;
 556                         if (nfit_memdev->memdev->device_handle != device_handle)
 557                                 continue;
 558                         if (nfit_memdev->memdev->region_index != dcr)
 559                                 continue;
 560
 561                         nfit_mem->spa_bdw = nfit_spa->spa;
 562                         return;
 563                 }
 564         }
 565
 566         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 567                         nfit_mem->spa_dcr->range_index);
 568         nfit_mem->bdw = NULL;
 569 }
 570
 571 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 572                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 573 {
 574         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 575         struct nfit_memdev *nfit_memdev;
 576         struct nfit_flush *nfit_flush;
 577         struct nfit_bdw *nfit_bdw;
 578         struct nfit_idt *nfit_idt;
 579         u16 idt_idx, range_index;
 580
 581         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 582                 if (nfit_bdw->bdw->region_index != dcr)
 583                         continue;
 584                 nfit_mem->bdw = nfit_bdw->bdw;
 585                 break;
 586         }
 587
 588         if (!nfit_mem->bdw)
 589                 return;
 590
 591         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 592
 593         if (!nfit_mem->spa_bdw)
 594                 return;
 595
 596         range_index = nfit_mem->spa_bdw->range_index;
 597         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 598                 if (nfit_memdev->memdev->range_index != range_index ||
 599                                 nfit_memdev->memdev->region_index != dcr)
 600                         continue;
 601                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
 602                 idt_idx = nfit_memdev->memdev->interleave_index;
 603                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 604                         if (nfit_idt->idt->interleave_index != idt_idx)
 605                                 continue;
 606                         nfit_mem->idt_bdw = nfit_idt->idt;
 607                         break;
 608                 }
 609
 610                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
 611                         if (nfit_flush->flush->device_handle !=
 612                                         nfit_memdev->memdev->device_handle)
 613                                 continue;
 614                         nfit_mem->nfit_flush = nfit_flush;
 615                         break;
 616                 }
 617                 break;
 618         }
 619 }
 620
 621 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
 622                 struct acpi_nfit_system_address *spa)
 623 {
 624         struct nfit_mem *nfit_mem, *found;
 625         struct nfit_memdev *nfit_memdev;
 626         int type = nfit_spa_type(spa);
 627
 628         switch (type) {
 629         case NFIT_SPA_DCR:
 630         case NFIT_SPA_PM:
 631                 break;
 632         default:
 633                 return 0;
 634         }
 635
 636         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 637                 struct nfit_dcr *nfit_dcr;
 638                 u32 device_handle;
 639                 u16 dcr;
 640
 641                 if (nfit_memdev->memdev->range_index != spa->range_index)
 642                         continue;
 643                 found = NULL;
 644                 dcr = nfit_memdev->memdev->region_index;
 645                 device_handle = nfit_memdev->memdev->device_handle;
 646                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 647                         if (__to_nfit_memdev(nfit_mem)->device_handle
 648                                         == device_handle) {
 649                                 found = nfit_mem;
 650                                 break;
 651                         }
 652
 653                 if (found)
 654                         nfit_mem = found;
 655                 else {
 656                         nfit_mem = devm_kzalloc(acpi_desc->dev,
 657                                         sizeof(*nfit_mem), GFP_KERNEL);
 658                         if (!nfit_mem)
 659                                 return -ENOMEM;
 660                         INIT_LIST_HEAD(&nfit_mem->list);
 661                         list_add(&nfit_mem->list, &acpi_desc->dimms);
 662                 }
 663
 664                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 665                         if (nfit_dcr->dcr->region_index != dcr)
 666                                 continue;
 667                         /*
 668                          * Record the control region for the dimm.  For
 669                          * the ACPI 6.1 case, where there are separate
 670                          * control regions for the pmem vs blk
 671                          * interfaces, be sure to record the extended
 672                          * blk details.
 673                          */
 674                         if (!nfit_mem->dcr)
 675                                 nfit_mem->dcr = nfit_dcr->dcr;
 676                         else if (nfit_mem->dcr->windows == 0
 677                                         && nfit_dcr->dcr->windows)
 678                                 nfit_mem->dcr = nfit_dcr->dcr;
 679                         break;
 680                 }
 681
 682                 if (dcr && !nfit_mem->dcr) {
 683                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
 684                                         spa->range_index, dcr);
 685                         return -ENODEV;
 686                 }
 687
 688                 if (type == NFIT_SPA_DCR) {
 689                         struct nfit_idt *nfit_idt;
 690                         u16 idt_idx;
 691
 692                         /* multiple dimms may share a SPA when interleaved */
 693                         nfit_mem->spa_dcr = spa;
 694                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 695                         idt_idx = nfit_memdev->memdev->interleave_index;
 696                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 697                                 if (nfit_idt->idt->interleave_index != idt_idx)
 698                                         continue;
 699                                 nfit_mem->idt_dcr = nfit_idt->idt;
 700                                 break;
 701                         }
 702                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
 703                 } else {
 704                         /*
 705                          * A single dimm may belong to multiple SPA-PM
 706                          * ranges, record at least one in addition to
 707                          * any SPA-DCR range.
 708                          */
 709                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
 710                 }
 711         }
 712
 713         return 0;
 714 }
 715
 716 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
 717 {
 718         struct nfit_mem *a = container_of(_a, typeof(*a), list);
 719         struct nfit_mem *b = container_of(_b, typeof(*b), list);
 720         u32 handleA, handleB;
 721
 722         handleA = __to_nfit_memdev(a)->device_handle;
 723         handleB = __to_nfit_memdev(b)->device_handle;
 724         if (handleA < handleB)
 725                 return -1;
 726         else if (handleA > handleB)
 727                 return 1;
 728         return 0;
 729 }
 730
 731 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
 732 {
 733         struct nfit_spa *nfit_spa;
 734
 735         /*
 736          * For each SPA-DCR or SPA-PMEM address range find its
 737          * corresponding MEMDEV(s).  From each MEMDEV find the
 738          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
 739          * try to find a SPA-BDW and a corresponding BDW that references
 740          * the DCR.  Throw it all into an nfit_mem object.  Note, that
 741          * BDWs are optional.
 742          */
 743         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 744                 int rc;
 745
 746                 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
 747                 if (rc)
 748                         return rc;
 749         }
 750
 751         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
 752
 753         return 0;
 754 }
 755
 756 static ssize_t revision_show(struct device *dev,
 757                 struct device_attribute *attr, char *buf)
 758 {
 759         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 760         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 761         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 762
 763         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
 764 }
 765 static DEVICE_ATTR_RO(revision);
 766
 767 static struct attribute *acpi_nfit_attributes[] = {
 768         &dev_attr_revision.attr,
 769         NULL,
 770 };
 771
 772 static struct attribute_group acpi_nfit_attribute_group = {
 773         .name = "nfit",
 774         .attrs = acpi_nfit_attributes,
 775 };
 776
 777 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
 778         &nvdimm_bus_attribute_group,
 779         &acpi_nfit_attribute_group,
 780         NULL,
 781 };
 782
 783 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
 784 {
 785         struct nvdimm *nvdimm = to_nvdimm(dev);
 786         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 787
 788         return __to_nfit_memdev(nfit_mem);
 789 }
 790
 791 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
 792 {
 793         struct nvdimm *nvdimm = to_nvdimm(dev);
 794         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 795
 796         return nfit_mem->dcr;
 797 }
 798
 799 static ssize_t handle_show(struct device *dev,
 800                 struct device_attribute *attr, char *buf)
 801 {
 802         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 803
 804         return sprintf(buf, "%#x\n", memdev->device_handle);
 805 }
 806 static DEVICE_ATTR_RO(handle);
 807
 808 static ssize_t phys_id_show(struct device *dev,
 809                 struct device_attribute *attr, char *buf)
 810 {
 811         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 812
 813         return sprintf(buf, "%#x\n", memdev->physical_id);
 814 }
 815 static DEVICE_ATTR_RO(phys_id);
 816
 817 static ssize_t vendor_show(struct device *dev,
 818                 struct device_attribute *attr, char *buf)
 819 {
 820         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 821
 822         return sprintf(buf, "%#x\n", dcr->vendor_id);
 823 }
 824 static DEVICE_ATTR_RO(vendor);
 825
 826 static ssize_t rev_id_show(struct device *dev,
 827                 struct device_attribute *attr, char *buf)
 828 {
 829         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 830
 831         return sprintf(buf, "%#x\n", dcr->revision_id);
 832 }
 833 static DEVICE_ATTR_RO(rev_id);
 834
 835 static ssize_t device_show(struct device *dev,
 836                 struct device_attribute *attr, char *buf)
 837 {
 838         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 839
 840         return sprintf(buf, "%#x\n", dcr->device_id);
 841 }
 842 static DEVICE_ATTR_RO(device);
 843
 844 static ssize_t format_show(struct device *dev,
 845                 struct device_attribute *attr, char *buf)
 846 {
 847         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 848
 849         return sprintf(buf, "%#x\n", dcr->code);
 850 }
 851 static DEVICE_ATTR_RO(format);
 852
 853 static ssize_t serial_show(struct device *dev,
 854                 struct device_attribute *attr, char *buf)
 855 {
 856         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 857
 858         return sprintf(buf, "%#x\n", dcr->serial_number);
 859 }
 860 static DEVICE_ATTR_RO(serial);
 861
 862 static ssize_t flags_show(struct device *dev,
 863                 struct device_attribute *attr, char *buf)
 864 {
 865         u16 flags = to_nfit_memdev(dev)->flags;
 866
 867         return sprintf(buf, "%s%s%s%s%s\n",
 868                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
 869                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
 870                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
 871                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
 872                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
 873 }
 874 static DEVICE_ATTR_RO(flags);
 875
 876 static struct attribute *acpi_nfit_dimm_attributes[] = {
 877         &dev_attr_handle.attr,
 878         &dev_attr_phys_id.attr,
 879         &dev_attr_vendor.attr,
 880         &dev_attr_device.attr,
 881         &dev_attr_format.attr,
 882         &dev_attr_serial.attr,
 883         &dev_attr_rev_id.attr,
 884         &dev_attr_flags.attr,
 885         NULL,
 886 };
 887
 888 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
 889                 struct attribute *a, int n)
 890 {
 891         struct device *dev = container_of(kobj, struct device, kobj);
 892
 893         if (to_nfit_dcr(dev))
 894                 return a->mode;
 895         else
 896                 return 0;
 897 }
 898
 899 static struct attribute_group acpi_nfit_dimm_attribute_group = {
 900         .name = "nfit",
 901         .attrs = acpi_nfit_dimm_attributes,
 902         .is_visible = acpi_nfit_dimm_attr_visible,
 903 };
 904
 905 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
 906         &nvdimm_attribute_group,
 907         &nd_device_attribute_group,
 908         &acpi_nfit_dimm_attribute_group,
 909         NULL,
 910 };
 911
 912 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
 913                 u32 device_handle)
 914 {
 915         struct nfit_mem *nfit_mem;
 916
 917         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 918                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
 919                         return nfit_mem->nvdimm;
 920
 921         return NULL;
 922 }
 923
 924 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
 925                 struct nfit_mem *nfit_mem, u32 device_handle)
 926 {
 927         struct acpi_device *adev, *adev_dimm;
 928         struct device *dev = acpi_desc->dev;
 929         const u8 *uuid = to_nfit_uuid(NFIT_DEV_DIMM);
 930         int i;
 931
 932         nfit_mem->dsm_mask = acpi_desc->dimm_dsm_force_en;
 933         adev = to_acpi_dev(acpi_desc);
 934         if (!adev)
 935                 return 0;
 936
 937         adev_dimm = acpi_find_child_device(adev, device_handle, false);
 938         nfit_mem->adev = adev_dimm;
 939         if (!adev_dimm) {
 940                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
 941                                 device_handle);
 942                 return force_enable_dimms ? 0 : -ENODEV;
 943         }
 944
 945         for (i = ND_CMD_SMART; i <= ND_CMD_VENDOR; i++)
 946                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
 947                         set_bit(i, &nfit_mem->dsm_mask);
 948
 949         return 0;
 950 }
 951
 952 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
 953 {
 954         struct nfit_mem *nfit_mem;
 955         int dimm_count = 0;
 956
 957         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 958                 struct nvdimm *nvdimm;
 959                 unsigned long flags = 0;
 960                 u32 device_handle;
 961                 u16 mem_flags;
 962                 int rc;
 963
 964                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 965                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
 966                 if (nvdimm) {
 967                         dimm_count++;
 968                         continue;
 969                 }
 970
 971                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
 972                         flags |= NDD_ALIASING;
 973
 974                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
 975                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
 976                         flags |= NDD_UNARMED;
 977
 978                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
 979                 if (rc)
 980                         continue;
 981
 982                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
 983                                 acpi_nfit_dimm_attribute_groups,
 984                                 flags, &nfit_mem->dsm_mask);
 985                 if (!nvdimm)
 986                         return -ENOMEM;
 987
 988                 nfit_mem->nvdimm = nvdimm;
 989                 dimm_count++;
 990
 991                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
 992                         continue;
 993
 994                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
 995                                 nvdimm_name(nvdimm),
 996                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
 997                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
 998                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
 999                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1000
1001         }
1002
1003         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1004 }
1005
1006 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1007 {
1008         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1009         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1010         struct acpi_device *adev;
1011         int i;
1012
1013         nd_desc->dsm_mask = acpi_desc->bus_dsm_force_en;
1014         adev = to_acpi_dev(acpi_desc);
1015         if (!adev)
1016                 return;
1017
1018         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1019                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1020                         set_bit(i, &nd_desc->dsm_mask);
1021 }
1022
1023 static ssize_t range_index_show(struct device *dev,
1024                 struct device_attribute *attr, char *buf)
1025 {
1026         struct nd_region *nd_region = to_nd_region(dev);
1027         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1028
1029         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1030 }
1031 static DEVICE_ATTR_RO(range_index);
1032
1033 static struct attribute *acpi_nfit_region_attributes[] = {
1034         &dev_attr_range_index.attr,
1035         NULL,
1036 };
1037
1038 static struct attribute_group acpi_nfit_region_attribute_group = {
1039         .name = "nfit",
1040         .attrs = acpi_nfit_region_attributes,
1041 };
1042
1043 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1044         &nd_region_attribute_group,
1045         &nd_mapping_attribute_group,
1046         &nd_device_attribute_group,
1047         &nd_numa_attribute_group,
1048         &acpi_nfit_region_attribute_group,
1049         NULL,
1050 };
1051
1052 /* enough info to uniquely specify an interleave set */
1053 struct nfit_set_info {
1054         struct nfit_set_info_map {
1055                 u64 region_offset;
1056                 u32 serial_number;
1057                 u32 pad;
1058         } mapping[0];
1059 };
1060
1061 static size_t sizeof_nfit_set_info(int num_mappings)
1062 {
1063         return sizeof(struct nfit_set_info)
1064                 + num_mappings * sizeof(struct nfit_set_info_map);
1065 }
1066
1067 static int cmp_map(const void *m0, const void *m1)
1068 {
1069         const struct nfit_set_info_map *map0 = m0;
1070         const struct nfit_set_info_map *map1 = m1;
1071
1072         return memcmp(&map0->region_offset, &map1->region_offset,
1073                         sizeof(u64));
1074 }
1075
1076 /* Retrieve the nth entry referencing this spa */
1077 static struct acpi_nfit_memory_map *memdev_from_spa(
1078                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1079 {
1080         struct nfit_memdev *nfit_memdev;
1081
1082         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1083                 if (nfit_memdev->memdev->range_index == range_index)
1084                         if (n-- == 0)
1085                                 return nfit_memdev->memdev;
1086         return NULL;
1087 }
1088
1089 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1090                 struct nd_region_desc *ndr_desc,
1091                 struct acpi_nfit_system_address *spa)
1092 {
1093         int i, spa_type = nfit_spa_type(spa);
1094         struct device *dev = acpi_desc->dev;
1095         struct nd_interleave_set *nd_set;
1096         u16 nr = ndr_desc->num_mappings;
1097         struct nfit_set_info *info;
1098
1099         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1100                 /* pass */;
1101         else
1102                 return 0;
1103
1104         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1105         if (!nd_set)
1106                 return -ENOMEM;
1107
1108         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1109         if (!info)
1110                 return -ENOMEM;
1111         for (i = 0; i < nr; i++) {
1112                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1113                 struct nfit_set_info_map *map = &info->mapping[i];
1114                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1115                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1116                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1117                                 spa->range_index, i);
1118
1119                 if (!memdev || !nfit_mem->dcr) {
1120                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1121                         return -ENODEV;
1122                 }
1123
1124                 map->region_offset = memdev->region_offset;
1125                 map->serial_number = nfit_mem->dcr->serial_number;
1126         }
1127
1128         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1129                         cmp_map, NULL);
1130         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1131         ndr_desc->nd_set = nd_set;
1132         devm_kfree(dev, info);
1133
1134         return 0;
1135 }
1136
1137 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1138 {
1139         struct acpi_nfit_interleave *idt = mmio->idt;
1140         u32 sub_line_offset, line_index, line_offset;
1141         u64 line_no, table_skip_count, table_offset;
1142
1143         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1144         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1145         line_offset = idt->line_offset[line_index]
1146                 * mmio->line_size;
1147         table_offset = table_skip_count * mmio->table_size;
1148
1149         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1150 }
1151
1152 static void wmb_blk(struct nfit_blk *nfit_blk)
1153 {
1154
1155         if (nfit_blk->nvdimm_flush) {
1156                 /*
1157                  * The first wmb() is needed to 'sfence' all previous writes
1158                  * such that they are architecturally visible for the platform
1159                  * buffer flush.  Note that we've already arranged for pmem
1160                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1161                  * final wmb() ensures ordering for the NVDIMM flush write.
1162                  */
1163                 wmb();
1164                 writeq(1, nfit_blk->nvdimm_flush);
1165                 wmb();
1166         } else
1167                 wmb_pmem();
1168 }
1169
1170 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1171 {
1172         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1173         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1174
1175         if (mmio->num_lines)
1176                 offset = to_interleave_offset(offset, mmio);
1177
1178         return readl(mmio->addr.base + offset);
1179 }
1180
1181 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1182                 resource_size_t dpa, unsigned int len, unsigned int write)
1183 {
1184         u64 cmd, offset;
1185         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1186
1187         enum {
1188                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1189                 BCW_LEN_SHIFT = 48,
1190                 BCW_LEN_MASK = (1ULL << 8) - 1,
1191                 BCW_CMD_SHIFT = 56,
1192         };
1193
1194         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1195         len = len >> L1_CACHE_SHIFT;
1196         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1197         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1198
1199         offset = nfit_blk->cmd_offset + mmio->size * bw;
1200         if (mmio->num_lines)
1201                 offset = to_interleave_offset(offset, mmio);
1202
1203         writeq(cmd, mmio->addr.base + offset);
1204         wmb_blk(nfit_blk);
1205
1206         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1207                 readq(mmio->addr.base + offset);
1208 }
1209
1210 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1211                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1212                 unsigned int lane)
1213 {
1214         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1215         unsigned int copied = 0;
1216         u64 base_offset;
1217         int rc;
1218
1219         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1220                 + lane * mmio->size;
1221         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1222         while (len) {
1223                 unsigned int c;
1224                 u64 offset;
1225
1226                 if (mmio->num_lines) {
1227                         u32 line_offset;
1228
1229                         offset = to_interleave_offset(base_offset + copied,
1230                                         mmio);
1231                         div_u64_rem(offset, mmio->line_size, &line_offset);
1232                         c = min_t(size_t, len, mmio->line_size - line_offset);
1233                 } else {
1234                         offset = base_offset + nfit_blk->bdw_offset;
1235                         c = len;
1236                 }
1237
1238                 if (rw)
1239                         memcpy_to_pmem(mmio->addr.aperture + offset,
1240                                         iobuf + copied, c);
1241                 else {
1242                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1243                                 mmio_flush_range((void __force *)
1244                                         mmio->addr.aperture + offset, c);
1245
1246                         memcpy_from_pmem(iobuf + copied,
1247                                         mmio->addr.aperture + offset, c);
1248                 }
1249
1250                 copied += c;
1251                 len -= c;
1252         }
1253
1254         if (rw)
1255                 wmb_blk(nfit_blk);
1256
1257         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1258         return rc;
1259 }
1260
1261 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1262                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1263 {
1264         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1265         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1266         struct nd_region *nd_region = nfit_blk->nd_region;
1267         unsigned int lane, copied = 0;
1268         int rc = 0;
1269
1270         lane = nd_region_acquire_lane(nd_region);
1271         while (len) {
1272                 u64 c = min(len, mmio->size);
1273
1274                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1275                                 iobuf + copied, c, rw, lane);
1276                 if (rc)
1277                         break;
1278
1279                 copied += c;
1280                 len -= c;
1281         }
1282         nd_region_release_lane(nd_region, lane);
1283
1284         return rc;
1285 }
1286
1287 static void nfit_spa_mapping_release(struct kref *kref)
1288 {
1289         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1290         struct acpi_nfit_system_address *spa = spa_map->spa;
1291         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1292
1293         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1294         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1295         if (spa_map->type == SPA_MAP_APERTURE)
1296                 memunmap((void __force *)spa_map->addr.aperture);
1297         else
1298                 iounmap(spa_map->addr.base);
1299         release_mem_region(spa->address, spa->length);
1300         list_del(&spa_map->list);
1301         kfree(spa_map);
1302 }
1303
1304 static struct nfit_spa_mapping *find_spa_mapping(
1305                 struct acpi_nfit_desc *acpi_desc,
1306                 struct acpi_nfit_system_address *spa)
1307 {
1308         struct nfit_spa_mapping *spa_map;
1309
1310         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1311         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1312                 if (spa_map->spa == spa)
1313                         return spa_map;
1314
1315         return NULL;
1316 }
1317
1318 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1319                 struct acpi_nfit_system_address *spa)
1320 {
1321         struct nfit_spa_mapping *spa_map;
1322
1323         mutex_lock(&acpi_desc->spa_map_mutex);
1324         spa_map = find_spa_mapping(acpi_desc, spa);
1325
1326         if (spa_map)
1327                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1328         mutex_unlock(&acpi_desc->spa_map_mutex);
1329 }
1330
1331 static void __iomem *__nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1332                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1333 {
1334         resource_size_t start = spa->address;
1335         resource_size_t n = spa->length;
1336         struct nfit_spa_mapping *spa_map;
1337         struct resource *res;
1338
1339         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1340
1341         spa_map = find_spa_mapping(acpi_desc, spa);
1342         if (spa_map) {
1343                 kref_get(&spa_map->kref);
1344                 return spa_map->addr.base;
1345         }
1346
1347         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1348         if (!spa_map)
1349                 return NULL;
1350
1351         INIT_LIST_HEAD(&spa_map->list);
1352         spa_map->spa = spa;
1353         kref_init(&spa_map->kref);
1354         spa_map->acpi_desc = acpi_desc;
1355
1356         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1357         if (!res)
1358                 goto err_mem;
1359
1360         spa_map->type = type;
1361         if (type == SPA_MAP_APERTURE)
1362                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1363                                                         ARCH_MEMREMAP_PMEM);
1364         else
1365                 spa_map->addr.base = ioremap_nocache(start, n);
1366
1367
1368         if (!spa_map->addr.base)
1369                 goto err_map;
1370
1371         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1372         return spa_map->addr.base;
1373
1374  err_map:
1375         release_mem_region(start, n);
1376  err_mem:
1377         kfree(spa_map);
1378         return NULL;
1379 }
1380
1381 /**
1382  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1383  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1384  * @nfit_spa: spa table to map
1385  * @type: aperture or control region
1386  *
1387  * In the case where block-data-window apertures and
1388  * dimm-control-regions are interleaved they will end up sharing a
1389  * single request_mem_region() + ioremap() for the address range.  In
1390  * the style of devm nfit_spa_map() mappings are automatically dropped
1391  * when all region devices referencing the same mapping are disabled /
1392  * unbound.
1393  */
1394 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1395                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1396 {
1397         void __iomem *iomem;
1398
1399         mutex_lock(&acpi_desc->spa_map_mutex);
1400         iomem = __nfit_spa_map(acpi_desc, spa, type);
1401         mutex_unlock(&acpi_desc->spa_map_mutex);
1402
1403         return iomem;
1404 }
1405
1406 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1407                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1408 {
1409         if (idt) {
1410                 mmio->num_lines = idt->line_count;
1411                 mmio->line_size = idt->line_size;
1412                 if (interleave_ways == 0)
1413                         return -ENXIO;
1414                 mmio->table_size = mmio->num_lines * interleave_ways
1415                         * mmio->line_size;
1416         }
1417
1418         return 0;
1419 }
1420
1421 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1422                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1423 {
1424         struct nd_cmd_dimm_flags flags;
1425         int rc;
1426
1427         memset(&flags, 0, sizeof(flags));
1428         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1429                         sizeof(flags), NULL);
1430
1431         if (rc >= 0 && flags.status == 0)
1432                 nfit_blk->dimm_flags = flags.flags;
1433         else if (rc == -ENOTTY) {
1434                 /* fall back to a conservative default */
1435                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1436                 rc = 0;
1437         } else
1438                 rc = -ENXIO;
1439
1440         return rc;
1441 }
1442
1443 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1444                 struct device *dev)
1445 {
1446         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1447         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1448         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1449         struct nfit_flush *nfit_flush;
1450         struct nfit_blk_mmio *mmio;
1451         struct nfit_blk *nfit_blk;
1452         struct nfit_mem *nfit_mem;
1453         struct nvdimm *nvdimm;
1454         int rc;
1455
1456         nvdimm = nd_blk_region_to_dimm(ndbr);
1457         nfit_mem = nvdimm_provider_data(nvdimm);
1458         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1459                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1460                                 nfit_mem ? "" : " nfit_mem",
1461                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1462                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1463                 return -ENXIO;
1464         }
1465
1466         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1467         if (!nfit_blk)
1468                 return -ENOMEM;
1469         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1470         nfit_blk->nd_region = to_nd_region(dev);
1471
1472         /* map block aperture memory */
1473         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1474         mmio = &nfit_blk->mmio[BDW];
1475         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1476                         SPA_MAP_APERTURE);
1477         if (!mmio->addr.base) {
1478                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1479                                 nvdimm_name(nvdimm));
1480                 return -ENOMEM;
1481         }
1482         mmio->size = nfit_mem->bdw->size;
1483         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1484         mmio->idt = nfit_mem->idt_bdw;
1485         mmio->spa = nfit_mem->spa_bdw;
1486         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1487                         nfit_mem->memdev_bdw->interleave_ways);
1488         if (rc) {
1489                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1490                                 __func__, nvdimm_name(nvdimm));
1491                 return rc;
1492         }
1493
1494         /* map block control memory */
1495         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1496         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1497         mmio = &nfit_blk->mmio[DCR];
1498         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1499                         SPA_MAP_CONTROL);
1500         if (!mmio->addr.base) {
1501                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1502                                 nvdimm_name(nvdimm));
1503                 return -ENOMEM;
1504         }
1505         mmio->size = nfit_mem->dcr->window_size;
1506         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1507         mmio->idt = nfit_mem->idt_dcr;
1508         mmio->spa = nfit_mem->spa_dcr;
1509         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1510                         nfit_mem->memdev_dcr->interleave_ways);
1511         if (rc) {
1512                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1513                                 __func__, nvdimm_name(nvdimm));
1514                 return rc;
1515         }
1516
1517         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1518         if (rc < 0) {
1519                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1520                                 __func__, nvdimm_name(nvdimm));
1521                 return rc;
1522         }
1523
1524         nfit_flush = nfit_mem->nfit_flush;
1525         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1526                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1527                                 nfit_flush->flush->hint_address[0], 8);
1528                 if (!nfit_blk->nvdimm_flush)
1529                         return -ENOMEM;
1530         }
1531
1532         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1533                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1534
1535         if (mmio->line_size == 0)
1536                 return 0;
1537
1538         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1539                         + 8 > mmio->line_size) {
1540                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1541                 return -ENXIO;
1542         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1543                         + 8 > mmio->line_size) {
1544                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1545                 return -ENXIO;
1546         }
1547
1548         return 0;
1549 }
1550
1551 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1552                 struct device *dev)
1553 {
1554         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1555         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1556         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1557         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1558         int i;
1559
1560         if (!nfit_blk)
1561                 return; /* never enabled */
1562
1563         /* auto-free BLK spa mappings */
1564         for (i = 0; i < 2; i++) {
1565                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1566
1567                 if (mmio->addr.base)
1568                         nfit_spa_unmap(acpi_desc, mmio->spa);
1569         }
1570         nd_blk_region_set_provider_data(ndbr, NULL);
1571         /* devm will free nfit_blk */
1572 }
1573
1574 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1575                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1576 {
1577         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1578         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1579         int cmd_rc, rc;
1580
1581         cmd->address = spa->address;
1582         cmd->length = spa->length;
1583         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1584                         sizeof(*cmd), &cmd_rc);
1585         if (rc < 0)
1586                 return rc;
1587         return cmd_rc;
1588 }
1589
1590 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1591 {
1592         int rc;
1593         int cmd_rc;
1594         struct nd_cmd_ars_start ars_start;
1595         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1596         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1597
1598         memset(&ars_start, 0, sizeof(ars_start));
1599         ars_start.address = spa->address;
1600         ars_start.length = spa->length;
1601         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1602                 ars_start.type = ND_ARS_PERSISTENT;
1603         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1604                 ars_start.type = ND_ARS_VOLATILE;
1605         else
1606                 return -ENOTTY;
1607
1608         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1609                         sizeof(ars_start), &cmd_rc);
1610
1611         if (rc < 0)
1612                 return rc;
1613         return cmd_rc;
1614 }
1615
1616 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1617 {
1618         int rc, cmd_rc;
1619         struct nd_cmd_ars_start ars_start;
1620         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1621         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1622
1623         memset(&ars_start, 0, sizeof(ars_start));
1624         ars_start.address = ars_status->restart_address;
1625         ars_start.length = ars_status->restart_length;
1626         ars_start.type = ars_status->type;
1627         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1628                         sizeof(ars_start), &cmd_rc);
1629         if (rc < 0)
1630                 return rc;
1631         return cmd_rc;
1632 }
1633
1634 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1635 {
1636         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1637         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1638         int rc, cmd_rc;
1639
1640         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1641                         acpi_desc->ars_status_size, &cmd_rc);
1642         if (rc < 0)
1643                 return rc;
1644         return cmd_rc;
1645 }
1646
1647 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1648                 struct nd_cmd_ars_status *ars_status)
1649 {
1650         int rc;
1651         u32 i;
1652
1653         for (i = 0; i < ars_status->num_records; i++) {
1654                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1655                                 ars_status->records[i].err_address,
1656                                 ars_status->records[i].length);
1657                 if (rc)
1658                         return rc;
1659         }
1660
1661         return 0;
1662 }
1663
1664 static void acpi_nfit_remove_resource(void *data)
1665 {
1666         struct resource *res = data;
1667
1668         remove_resource(res);
1669 }
1670
1671 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1672                 struct nd_region_desc *ndr_desc)
1673 {
1674         struct resource *res, *nd_res = ndr_desc->res;
1675         int is_pmem, ret;
1676
1677         /* No operation if the region is already registered as PMEM */
1678         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1679                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1680         if (is_pmem == REGION_INTERSECTS)
1681                 return 0;
1682
1683         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1684         if (!res)
1685                 return -ENOMEM;
1686
1687         res->name = "Persistent Memory";
1688         res->start = nd_res->start;
1689         res->end = nd_res->end;
1690         res->flags = IORESOURCE_MEM;
1691         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1692
1693         ret = insert_resource(&iomem_resource, res);
1694         if (ret)
1695                 return ret;
1696
1697         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1698         if (ret) {
1699                 remove_resource(res);
1700                 return ret;
1701         }
1702
1703         return 0;
1704 }
1705
1706 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1707                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1708                 struct acpi_nfit_memory_map *memdev,
1709                 struct nfit_spa *nfit_spa)
1710 {
1711         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1712                         memdev->device_handle);
1713         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1714         struct nd_blk_region_desc *ndbr_desc;
1715         struct nfit_mem *nfit_mem;
1716         int blk_valid = 0;
1717
1718         if (!nvdimm) {
1719                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1720                                 spa->range_index, memdev->device_handle);
1721                 return -ENODEV;
1722         }
1723
1724         nd_mapping->nvdimm = nvdimm;
1725         switch (nfit_spa_type(spa)) {
1726         case NFIT_SPA_PM:
1727         case NFIT_SPA_VOLATILE:
1728                 nd_mapping->start = memdev->address;
1729                 nd_mapping->size = memdev->region_size;
1730                 break;
1731         case NFIT_SPA_DCR:
1732                 nfit_mem = nvdimm_provider_data(nvdimm);
1733                 if (!nfit_mem || !nfit_mem->bdw) {
1734                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1735                                         spa->range_index, nvdimm_name(nvdimm));
1736                 } else {
1737                         nd_mapping->size = nfit_mem->bdw->capacity;
1738                         nd_mapping->start = nfit_mem->bdw->start_address;
1739                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1740                         blk_valid = 1;
1741                 }
1742
1743                 ndr_desc->nd_mapping = nd_mapping;
1744                 ndr_desc->num_mappings = blk_valid;
1745                 ndbr_desc = to_blk_region_desc(ndr_desc);
1746                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1747                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1748                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1749                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1750                                 ndr_desc);
1751                 if (!nfit_spa->nd_region)
1752                         return -ENOMEM;
1753                 break;
1754         }
1755
1756         return 0;
1757 }
1758
1759 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1760                 struct nfit_spa *nfit_spa)
1761 {
1762         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1763         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1764         struct nd_blk_region_desc ndbr_desc;
1765         struct nd_region_desc *ndr_desc;
1766         struct nfit_memdev *nfit_memdev;
1767         struct nvdimm_bus *nvdimm_bus;
1768         struct resource res;
1769         int count = 0, rc;
1770
1771         if (nfit_spa->nd_region)
1772                 return 0;
1773
1774         if (spa->range_index == 0) {
1775                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1776                                 __func__);
1777                 return 0;
1778         }
1779
1780         memset(&res, 0, sizeof(res));
1781         memset(&nd_mappings, 0, sizeof(nd_mappings));
1782         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1783         res.start = spa->address;
1784         res.end = res.start + spa->length - 1;
1785         ndr_desc = &ndbr_desc.ndr_desc;
1786         ndr_desc->res = &res;
1787         ndr_desc->provider_data = nfit_spa;
1788         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1789         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1790                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
1791                                                 spa->proximity_domain);
1792         else
1793                 ndr_desc->numa_node = NUMA_NO_NODE;
1794
1795         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1796                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1797                 struct nd_mapping *nd_mapping;
1798
1799                 if (memdev->range_index != spa->range_index)
1800                         continue;
1801                 if (count >= ND_MAX_MAPPINGS) {
1802                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
1803                                         spa->range_index, ND_MAX_MAPPINGS);
1804                         return -ENXIO;
1805                 }
1806                 nd_mapping = &nd_mappings[count++];
1807                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
1808                                 memdev, nfit_spa);
1809                 if (rc)
1810                         goto out;
1811         }
1812
1813         ndr_desc->nd_mapping = nd_mappings;
1814         ndr_desc->num_mappings = count;
1815         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
1816         if (rc)
1817                 goto out;
1818
1819         nvdimm_bus = acpi_desc->nvdimm_bus;
1820         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
1821                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
1822                 if (rc) {
1823                         dev_warn(acpi_desc->dev,
1824                                 "failed to insert pmem resource to iomem: %d\n",
1825                                 rc);
1826                         goto out;
1827                 }
1828
1829                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
1830                                 ndr_desc);
1831                 if (!nfit_spa->nd_region)
1832                         rc = -ENOMEM;
1833         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
1834                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
1835                                 ndr_desc);
1836                 if (!nfit_spa->nd_region)
1837                         rc = -ENOMEM;
1838         }
1839
1840  out:
1841         if (rc)
1842                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
1843                                 nfit_spa->spa->range_index);
1844         return rc;
1845 }
1846
1847 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
1848                 u32 max_ars)
1849 {
1850         struct device *dev = acpi_desc->dev;
1851         struct nd_cmd_ars_status *ars_status;
1852
1853         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
1854                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
1855                 return 0;
1856         }
1857
1858         if (acpi_desc->ars_status)
1859                 devm_kfree(dev, acpi_desc->ars_status);
1860         acpi_desc->ars_status = NULL;
1861         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
1862         if (!ars_status)
1863                 return -ENOMEM;
1864         acpi_desc->ars_status = ars_status;
1865         acpi_desc->ars_status_size = max_ars;
1866         return 0;
1867 }
1868
1869 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
1870                 struct nfit_spa *nfit_spa)
1871 {
1872         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1873         int rc;
1874
1875         if (!nfit_spa->max_ars) {
1876                 struct nd_cmd_ars_cap ars_cap;
1877
1878                 memset(&ars_cap, 0, sizeof(ars_cap));
1879                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
1880                 if (rc < 0)
1881                         return rc;
1882                 nfit_spa->max_ars = ars_cap.max_ars_out;
1883                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
1884                 /* check that the supported scrub types match the spa type */
1885                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
1886                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
1887                         return -ENOTTY;
1888                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
1889                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
1890                         return -ENOTTY;
1891         }
1892
1893         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
1894                 return -ENOMEM;
1895
1896         rc = ars_get_status(acpi_desc);
1897         if (rc < 0 && rc != -ENOSPC)
1898                 return rc;
1899
1900         if (ars_status_process_records(acpi_desc->nvdimm_bus,
1901                                 acpi_desc->ars_status))
1902                 return -ENOMEM;
1903
1904         return 0;
1905 }
1906
1907 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
1908                 struct nfit_spa *nfit_spa)
1909 {
1910         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1911         unsigned int overflow_retry = scrub_overflow_abort;
1912         u64 init_ars_start = 0, init_ars_len = 0;
1913         struct device *dev = acpi_desc->dev;
1914         unsigned int tmo = scrub_timeout;
1915         int rc;
1916
1917         if (nfit_spa->ars_done || !nfit_spa->nd_region)
1918                 return;
1919
1920         rc = ars_start(acpi_desc, nfit_spa);
1921         /*
1922          * If we timed out the initial scan we'll still be busy here,
1923          * and will wait another timeout before giving up permanently.
1924          */
1925         if (rc < 0 && rc != -EBUSY)
1926                 return;
1927
1928         do {
1929                 u64 ars_start, ars_len;
1930
1931                 if (acpi_desc->cancel)
1932                         break;
1933                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
1934                 if (rc == -ENOTTY)
1935                         break;
1936                 if (rc == -EBUSY && !tmo) {
1937                         dev_warn(dev, "range %d ars timeout, aborting\n",
1938                                         spa->range_index);
1939                         break;
1940                 }
1941
1942                 if (rc == -EBUSY) {
1943                         /*
1944                          * Note, entries may be appended to the list
1945                          * while the lock is dropped, but the workqueue
1946                          * being active prevents entries being deleted /
1947                          * freed.
1948                          */
1949                         mutex_unlock(&acpi_desc->init_mutex);
1950                         ssleep(1);
1951                         tmo--;
1952                         mutex_lock(&acpi_desc->init_mutex);
1953                         continue;
1954                 }
1955
1956                 /* we got some results, but there are more pending... */
1957                 if (rc == -ENOSPC && overflow_retry--) {
1958                         if (!init_ars_len) {
1959                                 init_ars_len = acpi_desc->ars_status->length;
1960                                 init_ars_start = acpi_desc->ars_status->address;
1961                         }
1962                         rc = ars_continue(acpi_desc);
1963                 }
1964
1965                 if (rc < 0) {
1966                         dev_warn(dev, "range %d ars continuation failed\n",
1967                                         spa->range_index);
1968                         break;
1969                 }
1970
1971                 if (init_ars_len) {
1972                         ars_start = init_ars_start;
1973                         ars_len = init_ars_len;
1974                 } else {
1975                         ars_start = acpi_desc->ars_status->address;
1976                         ars_len = acpi_desc->ars_status->length;
1977                 }
1978                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
1979                                 spa->range_index, ars_start, ars_len);
1980                 /* notify the region about new poison entries */
1981                 nvdimm_region_notify(nfit_spa->nd_region,
1982                                 NVDIMM_REVALIDATE_POISON);
1983                 break;
1984         } while (1);
1985 }
1986
1987 static void acpi_nfit_scrub(struct work_struct *work)
1988 {
1989         struct device *dev;
1990         u64 init_scrub_length = 0;
1991         struct nfit_spa *nfit_spa;
1992         u64 init_scrub_address = 0;
1993         bool init_ars_done = false;
1994         struct acpi_nfit_desc *acpi_desc;
1995         unsigned int tmo = scrub_timeout;
1996         unsigned int overflow_retry = scrub_overflow_abort;
1997
1998         acpi_desc = container_of(work, typeof(*acpi_desc), work);
1999         dev = acpi_desc->dev;
2000
2001         /*
2002          * We scrub in 2 phases.  The first phase waits for any platform
2003          * firmware initiated scrubs to complete and then we go search for the
2004          * affected spa regions to mark them scanned.  In the second phase we
2005          * initiate a directed scrub for every range that was not scrubbed in
2006          * phase 1.
2007          */
2008
2009         /* process platform firmware initiated scrubs */
2010  retry:
2011         mutex_lock(&acpi_desc->init_mutex);
2012         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2013                 struct nd_cmd_ars_status *ars_status;
2014                 struct acpi_nfit_system_address *spa;
2015                 u64 ars_start, ars_len;
2016                 int rc;
2017
2018                 if (acpi_desc->cancel)
2019                         break;
2020
2021                 if (nfit_spa->nd_region)
2022                         continue;
2023
2024                 if (init_ars_done) {
2025                         /*
2026                          * No need to re-query, we're now just
2027                          * reconciling all the ranges covered by the
2028                          * initial scrub
2029                          */
2030                         rc = 0;
2031                 } else
2032                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2033
2034                 if (rc == -ENOTTY) {
2035                         /* no ars capability, just register spa and move on */
2036                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2037                         continue;
2038                 }
2039
2040                 if (rc == -EBUSY && !tmo) {
2041                         /* fallthrough to directed scrub in phase 2 */
2042                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2043                         break;
2044                 } else if (rc == -EBUSY) {
2045                         mutex_unlock(&acpi_desc->init_mutex);
2046                         ssleep(1);
2047                         tmo--;
2048                         goto retry;
2049                 }
2050
2051                 /* we got some results, but there are more pending... */
2052                 if (rc == -ENOSPC && overflow_retry--) {
2053                         ars_status = acpi_desc->ars_status;
2054                         /*
2055                          * Record the original scrub range, so that we
2056                          * can recall all the ranges impacted by the
2057                          * initial scrub.
2058                          */
2059                         if (!init_scrub_length) {
2060                                 init_scrub_length = ars_status->length;
2061                                 init_scrub_address = ars_status->address;
2062                         }
2063                         rc = ars_continue(acpi_desc);
2064                         if (rc == 0) {
2065                                 mutex_unlock(&acpi_desc->init_mutex);
2066                                 goto retry;
2067                         }
2068                 }
2069
2070                 if (rc < 0) {
2071                         /*
2072                          * Initial scrub failed, we'll give it one more
2073                          * try below...
2074                          */
2075                         break;
2076                 }
2077
2078                 /* We got some final results, record completed ranges */
2079                 ars_status = acpi_desc->ars_status;
2080                 if (init_scrub_length) {
2081                         ars_start = init_scrub_address;
2082                         ars_len = ars_start + init_scrub_length;
2083                 } else {
2084                         ars_start = ars_status->address;
2085                         ars_len = ars_status->length;
2086                 }
2087                 spa = nfit_spa->spa;
2088
2089                 if (!init_ars_done) {
2090                         init_ars_done = true;
2091                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2092                                         ars_start, ars_len);
2093                 }
2094                 if (ars_start <= spa->address && ars_start + ars_len
2095                                 >= spa->address + spa->length)
2096                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2097         }
2098
2099         /*
2100          * For all the ranges not covered by an initial scrub we still
2101          * want to see if there are errors, but it's ok to discover them
2102          * asynchronously.
2103          */
2104         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2105                 /*
2106                  * Flag all the ranges that still need scrubbing, but
2107                  * register them now to make data available.
2108                  */
2109                 if (nfit_spa->nd_region)
2110                         nfit_spa->ars_done = 1;
2111                 else
2112                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2113         }
2114
2115         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2116                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2117         mutex_unlock(&acpi_desc->init_mutex);
2118 }
2119
2120 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2121 {
2122         struct nfit_spa *nfit_spa;
2123         int rc;
2124
2125         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2126                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2127                         /* BLK regions don't need to wait for ars results */
2128                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2129                         if (rc)
2130                                 return rc;
2131                 }
2132
2133         queue_work(nfit_wq, &acpi_desc->work);
2134         return 0;
2135 }
2136
2137 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2138                 struct nfit_table_prev *prev)
2139 {
2140         struct device *dev = acpi_desc->dev;
2141
2142         if (!list_empty(&prev->spas) ||
2143                         !list_empty(&prev->memdevs) ||
2144                         !list_empty(&prev->dcrs) ||
2145                         !list_empty(&prev->bdws) ||
2146                         !list_empty(&prev->idts) ||
2147                         !list_empty(&prev->flushes)) {
2148                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2149                 return -ENXIO;
2150         }
2151         return 0;
2152 }
2153
2154 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2155 {
2156         struct device *dev = acpi_desc->dev;
2157         struct nfit_table_prev prev;
2158         const void *end;
2159         u8 *data;
2160         int rc;
2161
2162         mutex_lock(&acpi_desc->init_mutex);
2163
2164         INIT_LIST_HEAD(&prev.spas);
2165         INIT_LIST_HEAD(&prev.memdevs);
2166         INIT_LIST_HEAD(&prev.dcrs);
2167         INIT_LIST_HEAD(&prev.bdws);
2168         INIT_LIST_HEAD(&prev.idts);
2169         INIT_LIST_HEAD(&prev.flushes);
2170
2171         list_cut_position(&prev.spas, &acpi_desc->spas,
2172                                 acpi_desc->spas.prev);
2173         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2174                                 acpi_desc->memdevs.prev);
2175         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2176                                 acpi_desc->dcrs.prev);
2177         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2178                                 acpi_desc->bdws.prev);
2179         list_cut_position(&prev.idts, &acpi_desc->idts,
2180                                 acpi_desc->idts.prev);
2181         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2182                                 acpi_desc->flushes.prev);
2183
2184         data = (u8 *) acpi_desc->nfit;
2185         end = data + sz;
2186         while (!IS_ERR_OR_NULL(data))
2187                 data = add_table(acpi_desc, &prev, data, end);
2188
2189         if (IS_ERR(data)) {
2190                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2191                                 PTR_ERR(data));
2192                 rc = PTR_ERR(data);
2193                 goto out_unlock;
2194         }
2195
2196         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2197         if (rc)
2198                 goto out_unlock;
2199
2200         if (nfit_mem_init(acpi_desc) != 0) {
2201                 rc = -ENOMEM;
2202                 goto out_unlock;
2203         }
2204
2205         acpi_nfit_init_dsms(acpi_desc);
2206
2207         rc = acpi_nfit_register_dimms(acpi_desc);
2208         if (rc)
2209                 goto out_unlock;
2210
2211         rc = acpi_nfit_register_regions(acpi_desc);
2212
2213  out_unlock:
2214         mutex_unlock(&acpi_desc->init_mutex);
2215         return rc;
2216 }
2217 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2218
2219 struct acpi_nfit_flush_work {
2220         struct work_struct work;
2221         struct completion cmp;
2222 };
2223
2224 static void flush_probe(struct work_struct *work)
2225 {
2226         struct acpi_nfit_flush_work *flush;
2227
2228         flush = container_of(work, typeof(*flush), work);
2229         complete(&flush->cmp);
2230 }
2231
2232 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2233 {
2234         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2235         struct device *dev = acpi_desc->dev;
2236         struct acpi_nfit_flush_work flush;
2237
2238         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2239         device_lock(dev);
2240         device_unlock(dev);
2241
2242         /*
2243          * Scrub work could take 10s of seconds, userspace may give up so we
2244          * need to be interruptible while waiting.
2245          */
2246         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2247         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2248         queue_work(nfit_wq, &flush.work);
2249         return wait_for_completion_interruptible(&flush.cmp);
2250 }
2251
2252 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2253                 struct nvdimm *nvdimm, unsigned int cmd)
2254 {
2255         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2256
2257         if (nvdimm)
2258                 return 0;
2259         if (cmd != ND_CMD_ARS_START)
2260                 return 0;
2261
2262         /*
2263          * The kernel and userspace may race to initiate a scrub, but
2264          * the scrub thread is prepared to lose that initial race.  It
2265          * just needs guarantees that any ars it initiates are not
2266          * interrupted by any intervening start reqeusts from userspace.
2267          */
2268         if (work_busy(&acpi_desc->work))
2269                 return -EBUSY;
2270
2271         return 0;
2272 }
2273
2274 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2275 {
2276         struct nvdimm_bus_descriptor *nd_desc;
2277
2278         dev_set_drvdata(dev, acpi_desc);
2279         acpi_desc->dev = dev;
2280         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2281         nd_desc = &acpi_desc->nd_desc;
2282         nd_desc->provider_name = "ACPI.NFIT";
2283         nd_desc->ndctl = acpi_nfit_ctl;
2284         nd_desc->flush_probe = acpi_nfit_flush_probe;
2285         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2286         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2287
2288         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2289         INIT_LIST_HEAD(&acpi_desc->spas);
2290         INIT_LIST_HEAD(&acpi_desc->dcrs);
2291         INIT_LIST_HEAD(&acpi_desc->bdws);
2292         INIT_LIST_HEAD(&acpi_desc->idts);
2293         INIT_LIST_HEAD(&acpi_desc->flushes);
2294         INIT_LIST_HEAD(&acpi_desc->memdevs);
2295         INIT_LIST_HEAD(&acpi_desc->dimms);
2296         mutex_init(&acpi_desc->spa_map_mutex);
2297         mutex_init(&acpi_desc->init_mutex);
2298         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2299 }
2300 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2301
2302 static int acpi_nfit_add(struct acpi_device *adev)
2303 {
2304         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2305         struct acpi_nfit_desc *acpi_desc;
2306         struct device *dev = &adev->dev;
2307         struct acpi_table_header *tbl;
2308         acpi_status status = AE_OK;
2309         acpi_size sz;
2310         int rc;
2311
2312         status = acpi_get_table_with_size("NFIT", 0, &tbl, &sz);
2313         if (ACPI_FAILURE(status)) {
2314                 /* This is ok, we could have an nvdimm hotplugged later */
2315                 dev_dbg(dev, "failed to find NFIT at startup\n");
2316                 return 0;
2317         }
2318
2319         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2320         if (!acpi_desc)
2321                 return -ENOMEM;
2322         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2323         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2324         if (!acpi_desc->nvdimm_bus)
2325                 return -ENOMEM;
2326
2327         /*
2328          * Save the acpi header for later and then skip it,
2329          * making nfit point to the first nfit table header.
2330          */
2331         acpi_desc->acpi_header = *tbl;
2332         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2333         sz -= sizeof(struct acpi_table_nfit);
2334
2335         /* Evaluate _FIT and override with that if present */
2336         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2337         if (ACPI_SUCCESS(status) && buf.length > 0) {
2338                 union acpi_object *obj;
2339                 /*
2340                  * Adjust for the acpi_object header of the _FIT
2341                  */
2342                 obj = buf.pointer;
2343                 if (obj->type == ACPI_TYPE_BUFFER) {
2344                         acpi_desc->nfit =
2345                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2346                         sz = obj->buffer.length;
2347                 } else
2348                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2349                                  __func__, (int) obj->type);
2350         }
2351
2352         rc = acpi_nfit_init(acpi_desc, sz);
2353         if (rc) {
2354                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2355                 return rc;
2356         }
2357         return 0;
2358 }
2359
2360 static int acpi_nfit_remove(struct acpi_device *adev)
2361 {
2362         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2363
2364         acpi_desc->cancel = 1;
2365         flush_workqueue(nfit_wq);
2366         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2367         return 0;
2368 }
2369
2370 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2371 {
2372         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2373         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2374         struct acpi_nfit_header *nfit_saved;
2375         union acpi_object *obj;
2376         struct device *dev = &adev->dev;
2377         acpi_status status;
2378         int ret;
2379
2380         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2381
2382         device_lock(dev);
2383         if (!dev->driver) {
2384                 /* dev->driver may be null if we're being removed */
2385                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2386                 goto out_unlock;
2387         }
2388
2389         if (!acpi_desc) {
2390                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2391                 if (!acpi_desc)
2392                         goto out_unlock;
2393                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2394                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2395                 if (!acpi_desc->nvdimm_bus)
2396                         goto out_unlock;
2397         } else {
2398                 /*
2399                  * Finish previous registration before considering new
2400                  * regions.
2401                  */
2402                 flush_workqueue(nfit_wq);
2403         }
2404
2405         /* Evaluate _FIT */
2406         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2407         if (ACPI_FAILURE(status)) {
2408                 dev_err(dev, "failed to evaluate _FIT\n");
2409                 goto out_unlock;
2410         }
2411
2412         nfit_saved = acpi_desc->nfit;
2413         obj = buf.pointer;
2414         if (obj->type == ACPI_TYPE_BUFFER) {
2415                 acpi_desc->nfit =
2416                         (struct acpi_nfit_header *)obj->buffer.pointer;
2417                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2418                 if (ret) {
2419                         /* Merge failed, restore old nfit, and exit */
2420                         acpi_desc->nfit = nfit_saved;
2421                         dev_err(dev, "failed to merge updated NFIT\n");
2422                 }
2423         } else {
2424                 /* Bad _FIT, restore old nfit */
2425                 dev_err(dev, "Invalid _FIT\n");
2426         }
2427         kfree(buf.pointer);
2428
2429  out_unlock:
2430         device_unlock(dev);
2431 }
2432
2433 static const struct acpi_device_id acpi_nfit_ids[] = {
2434         { "ACPI0012", 0 },
2435         { "", 0 },
2436 };
2437 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2438
2439 static struct acpi_driver acpi_nfit_driver = {
2440         .name = KBUILD_MODNAME,
2441         .ids = acpi_nfit_ids,
2442         .ops = {
2443                 .add = acpi_nfit_add,
2444                 .remove = acpi_nfit_remove,
2445                 .notify = acpi_nfit_notify,
2446         },
2447 };
2448
2449 static __init int nfit_init(void)
2450 {
2451         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2452         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2453         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2454         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2455         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2456         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2457         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2458
2459         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2460         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2461         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2462         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2463         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2464         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2465         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2466         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2467         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2468         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2469
2470         nfit_wq = create_singlethread_workqueue("nfit");
2471         if (!nfit_wq)
2472                 return -ENOMEM;
2473
2474         return acpi_bus_register_driver(&acpi_nfit_driver);
2475 }
2476
2477 static __exit void nfit_exit(void)
2478 {
2479         acpi_bus_unregister_driver(&acpi_nfit_driver);
2480         destroy_workqueue(nfit_wq);
2481 }
2482
2483 module_init(nfit_init);
2484 module_exit(nfit_exit);
2485 MODULE_LICENSE("GPL v2");
2486 MODULE_AUTHOR("Intel Corporation");