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
 293  out:
 294         ACPI_FREE(out_obj);
 295
 296         return rc;
 297 }
 298
 299 static const char *spa_type_name(u16 type)
 300 {
 301         static const char *to_name[] = {
 302                 [NFIT_SPA_VOLATILE] = "volatile",
 303                 [NFIT_SPA_PM] = "pmem",
 304                 [NFIT_SPA_DCR] = "dimm-control-region",
 305                 [NFIT_SPA_BDW] = "block-data-window",
 306                 [NFIT_SPA_VDISK] = "volatile-disk",
 307                 [NFIT_SPA_VCD] = "volatile-cd",
 308                 [NFIT_SPA_PDISK] = "persistent-disk",
 309                 [NFIT_SPA_PCD] = "persistent-cd",
 310
 311         };
 312
 313         if (type > NFIT_SPA_PCD)
 314                 return "unknown";
 315
 316         return to_name[type];
 317 }
 318
 319 static int nfit_spa_type(struct acpi_nfit_system_address *spa)
 320 {
 321         int i;
 322
 323         for (i = 0; i < NFIT_UUID_MAX; i++)
 324                 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
 325                         return i;
 326         return -1;
 327 }
 328
 329 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 330                 struct nfit_table_prev *prev,
 331                 struct acpi_nfit_system_address *spa)
 332 {
 333         size_t length = min_t(size_t, sizeof(*spa), spa->header.length);
 334         struct device *dev = acpi_desc->dev;
 335         struct nfit_spa *nfit_spa;
 336
 337         list_for_each_entry(nfit_spa, &prev->spas, list) {
 338                 if (memcmp(nfit_spa->spa, spa, length) == 0) {
 339                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 340                         return true;
 341                 }
 342         }
 343
 344         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa), GFP_KERNEL);
 345         if (!nfit_spa)
 346                 return false;
 347         INIT_LIST_HEAD(&nfit_spa->list);
 348         nfit_spa->spa = spa;
 349         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 350         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
 351                         spa->range_index,
 352                         spa_type_name(nfit_spa_type(spa)));
 353         return true;
 354 }
 355
 356 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 357                 struct nfit_table_prev *prev,
 358                 struct acpi_nfit_memory_map *memdev)
 359 {
 360         size_t length = min_t(size_t, sizeof(*memdev), memdev->header.length);
 361         struct device *dev = acpi_desc->dev;
 362         struct nfit_memdev *nfit_memdev;
 363
 364         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 365                 if (memcmp(nfit_memdev->memdev, memdev, length) == 0) {
 366                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 367                         return true;
 368                 }
 369
 370         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev), GFP_KERNEL);
 371         if (!nfit_memdev)
 372                 return false;
 373         INIT_LIST_HEAD(&nfit_memdev->list);
 374         nfit_memdev->memdev = memdev;
 375         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 376         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
 377                         __func__, memdev->device_handle, memdev->range_index,
 378                         memdev->region_index);
 379         return true;
 380 }
 381
 382 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 383                 struct nfit_table_prev *prev,
 384                 struct acpi_nfit_control_region *dcr)
 385 {
 386         size_t length = min_t(size_t, sizeof(*dcr), dcr->header.length);
 387         struct device *dev = acpi_desc->dev;
 388         struct nfit_dcr *nfit_dcr;
 389
 390         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 391                 if (memcmp(nfit_dcr->dcr, dcr, length) == 0) {
 392                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 393                         return true;
 394                 }
 395
 396         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr), GFP_KERNEL);
 397         if (!nfit_dcr)
 398                 return false;
 399         INIT_LIST_HEAD(&nfit_dcr->list);
 400         nfit_dcr->dcr = dcr;
 401         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 402         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
 403                         dcr->region_index, dcr->windows);
 404         return true;
 405 }
 406
 407 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 408                 struct nfit_table_prev *prev,
 409                 struct acpi_nfit_data_region *bdw)
 410 {
 411         size_t length = min_t(size_t, sizeof(*bdw), bdw->header.length);
 412         struct device *dev = acpi_desc->dev;
 413         struct nfit_bdw *nfit_bdw;
 414
 415         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 416                 if (memcmp(nfit_bdw->bdw, bdw, length) == 0) {
 417                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 418                         return true;
 419                 }
 420
 421         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw), GFP_KERNEL);
 422         if (!nfit_bdw)
 423                 return false;
 424         INIT_LIST_HEAD(&nfit_bdw->list);
 425         nfit_bdw->bdw = bdw;
 426         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 427         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
 428                         bdw->region_index, bdw->windows);
 429         return true;
 430 }
 431
 432 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 433                 struct nfit_table_prev *prev,
 434                 struct acpi_nfit_interleave *idt)
 435 {
 436         size_t length = min_t(size_t, sizeof(*idt), idt->header.length);
 437         struct device *dev = acpi_desc->dev;
 438         struct nfit_idt *nfit_idt;
 439
 440         list_for_each_entry(nfit_idt, &prev->idts, list)
 441                 if (memcmp(nfit_idt->idt, idt, length) == 0) {
 442                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 443                         return true;
 444                 }
 445
 446         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt), GFP_KERNEL);
 447         if (!nfit_idt)
 448                 return false;
 449         INIT_LIST_HEAD(&nfit_idt->list);
 450         nfit_idt->idt = idt;
 451         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 452         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
 453                         idt->interleave_index, idt->line_count);
 454         return true;
 455 }
 456
 457 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 458                 struct nfit_table_prev *prev,
 459                 struct acpi_nfit_flush_address *flush)
 460 {
 461         size_t length = min_t(size_t, sizeof(*flush), flush->header.length);
 462         struct device *dev = acpi_desc->dev;
 463         struct nfit_flush *nfit_flush;
 464
 465         list_for_each_entry(nfit_flush, &prev->flushes, list)
 466                 if (memcmp(nfit_flush->flush, flush, length) == 0) {
 467                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 468                         return true;
 469                 }
 470
 471         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush), GFP_KERNEL);
 472         if (!nfit_flush)
 473                 return false;
 474         INIT_LIST_HEAD(&nfit_flush->list);
 475         nfit_flush->flush = flush;
 476         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 477         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
 478                         flush->device_handle, flush->hint_count);
 479         return true;
 480 }
 481
 482 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 483                 struct nfit_table_prev *prev, void *table, const void *end)
 484 {
 485         struct device *dev = acpi_desc->dev;
 486         struct acpi_nfit_header *hdr;
 487         void *err = ERR_PTR(-ENOMEM);
 488
 489         if (table >= end)
 490                 return NULL;
 491
 492         hdr = table;
 493         if (!hdr->length) {
 494                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 495                         hdr->type);
 496                 return NULL;
 497         }
 498
 499         switch (hdr->type) {
 500         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 501                 if (!add_spa(acpi_desc, prev, table))
 502                         return err;
 503                 break;
 504         case ACPI_NFIT_TYPE_MEMORY_MAP:
 505                 if (!add_memdev(acpi_desc, prev, table))
 506                         return err;
 507                 break;
 508         case ACPI_NFIT_TYPE_CONTROL_REGION:
 509                 if (!add_dcr(acpi_desc, prev, table))
 510                         return err;
 511                 break;
 512         case ACPI_NFIT_TYPE_DATA_REGION:
 513                 if (!add_bdw(acpi_desc, prev, table))
 514                         return err;
 515                 break;
 516         case ACPI_NFIT_TYPE_INTERLEAVE:
 517                 if (!add_idt(acpi_desc, prev, table))
 518                         return err;
 519                 break;
 520         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 521                 if (!add_flush(acpi_desc, prev, table))
 522                         return err;
 523                 break;
 524         case ACPI_NFIT_TYPE_SMBIOS:
 525                 dev_dbg(dev, "%s: smbios\n", __func__);
 526                 break;
 527         default:
 528                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 529                 break;
 530         }
 531
 532         return table + hdr->length;
 533 }
 534
 535 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 536                 struct nfit_mem *nfit_mem)
 537 {
 538         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 539         u16 dcr = nfit_mem->dcr->region_index;
 540         struct nfit_spa *nfit_spa;
 541
 542         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 543                 u16 range_index = nfit_spa->spa->range_index;
 544                 int type = nfit_spa_type(nfit_spa->spa);
 545                 struct nfit_memdev *nfit_memdev;
 546
 547                 if (type != NFIT_SPA_BDW)
 548                         continue;
 549
 550                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 551                         if (nfit_memdev->memdev->range_index != range_index)
 552                                 continue;
 553                         if (nfit_memdev->memdev->device_handle != device_handle)
 554                                 continue;
 555                         if (nfit_memdev->memdev->region_index != dcr)
 556                                 continue;
 557
 558                         nfit_mem->spa_bdw = nfit_spa->spa;
 559                         return;
 560                 }
 561         }
 562
 563         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 564                         nfit_mem->spa_dcr->range_index);
 565         nfit_mem->bdw = NULL;
 566 }
 567
 568 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 569                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 570 {
 571         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 572         struct nfit_memdev *nfit_memdev;
 573         struct nfit_flush *nfit_flush;
 574         struct nfit_bdw *nfit_bdw;
 575         struct nfit_idt *nfit_idt;
 576         u16 idt_idx, range_index;
 577
 578         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 579                 if (nfit_bdw->bdw->region_index != dcr)
 580                         continue;
 581                 nfit_mem->bdw = nfit_bdw->bdw;
 582                 break;
 583         }
 584
 585         if (!nfit_mem->bdw)
 586                 return;
 587
 588         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 589
 590         if (!nfit_mem->spa_bdw)
 591                 return;
 592
 593         range_index = nfit_mem->spa_bdw->range_index;
 594         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 595                 if (nfit_memdev->memdev->range_index != range_index ||
 596                                 nfit_memdev->memdev->region_index != dcr)
 597                         continue;
 598                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
 599                 idt_idx = nfit_memdev->memdev->interleave_index;
 600                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 601                         if (nfit_idt->idt->interleave_index != idt_idx)
 602                                 continue;
 603                         nfit_mem->idt_bdw = nfit_idt->idt;
 604                         break;
 605                 }
 606
 607                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
 608                         if (nfit_flush->flush->device_handle !=
 609                                         nfit_memdev->memdev->device_handle)
 610                                 continue;
 611                         nfit_mem->nfit_flush = nfit_flush;
 612                         break;
 613                 }
 614                 break;
 615         }
 616 }
 617
 618 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
 619                 struct acpi_nfit_system_address *spa)
 620 {
 621         struct nfit_mem *nfit_mem, *found;
 622         struct nfit_memdev *nfit_memdev;
 623         int type = nfit_spa_type(spa);
 624
 625         switch (type) {
 626         case NFIT_SPA_DCR:
 627         case NFIT_SPA_PM:
 628                 break;
 629         default:
 630                 return 0;
 631         }
 632
 633         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 634                 struct nfit_dcr *nfit_dcr;
 635                 u32 device_handle;
 636                 u16 dcr;
 637
 638                 if (nfit_memdev->memdev->range_index != spa->range_index)
 639                         continue;
 640                 found = NULL;
 641                 dcr = nfit_memdev->memdev->region_index;
 642                 device_handle = nfit_memdev->memdev->device_handle;
 643                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 644                         if (__to_nfit_memdev(nfit_mem)->device_handle
 645                                         == device_handle) {
 646                                 found = nfit_mem;
 647                                 break;
 648                         }
 649
 650                 if (found)
 651                         nfit_mem = found;
 652                 else {
 653                         nfit_mem = devm_kzalloc(acpi_desc->dev,
 654                                         sizeof(*nfit_mem), GFP_KERNEL);
 655                         if (!nfit_mem)
 656                                 return -ENOMEM;
 657                         INIT_LIST_HEAD(&nfit_mem->list);
 658                         nfit_mem->acpi_desc = acpi_desc;
 659                         list_add(&nfit_mem->list, &acpi_desc->dimms);
 660                 }
 661
 662                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 663                         if (nfit_dcr->dcr->region_index != dcr)
 664                                 continue;
 665                         /*
 666                          * Record the control region for the dimm.  For
 667                          * the ACPI 6.1 case, where there are separate
 668                          * control regions for the pmem vs blk
 669                          * interfaces, be sure to record the extended
 670                          * blk details.
 671                          */
 672                         if (!nfit_mem->dcr)
 673                                 nfit_mem->dcr = nfit_dcr->dcr;
 674                         else if (nfit_mem->dcr->windows == 0
 675                                         && nfit_dcr->dcr->windows)
 676                                 nfit_mem->dcr = nfit_dcr->dcr;
 677                         break;
 678                 }
 679
 680                 if (dcr && !nfit_mem->dcr) {
 681                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
 682                                         spa->range_index, dcr);
 683                         return -ENODEV;
 684                 }
 685
 686                 if (type == NFIT_SPA_DCR) {
 687                         struct nfit_idt *nfit_idt;
 688                         u16 idt_idx;
 689
 690                         /* multiple dimms may share a SPA when interleaved */
 691                         nfit_mem->spa_dcr = spa;
 692                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 693                         idt_idx = nfit_memdev->memdev->interleave_index;
 694                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 695                                 if (nfit_idt->idt->interleave_index != idt_idx)
 696                                         continue;
 697                                 nfit_mem->idt_dcr = nfit_idt->idt;
 698                                 break;
 699                         }
 700                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
 701                 } else {
 702                         /*
 703                          * A single dimm may belong to multiple SPA-PM
 704                          * ranges, record at least one in addition to
 705                          * any SPA-DCR range.
 706                          */
 707                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
 708                 }
 709         }
 710
 711         return 0;
 712 }
 713
 714 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
 715 {
 716         struct nfit_mem *a = container_of(_a, typeof(*a), list);
 717         struct nfit_mem *b = container_of(_b, typeof(*b), list);
 718         u32 handleA, handleB;
 719
 720         handleA = __to_nfit_memdev(a)->device_handle;
 721         handleB = __to_nfit_memdev(b)->device_handle;
 722         if (handleA < handleB)
 723                 return -1;
 724         else if (handleA > handleB)
 725                 return 1;
 726         return 0;
 727 }
 728
 729 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
 730 {
 731         struct nfit_spa *nfit_spa;
 732
 733         /*
 734          * For each SPA-DCR or SPA-PMEM address range find its
 735          * corresponding MEMDEV(s).  From each MEMDEV find the
 736          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
 737          * try to find a SPA-BDW and a corresponding BDW that references
 738          * the DCR.  Throw it all into an nfit_mem object.  Note, that
 739          * BDWs are optional.
 740          */
 741         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 742                 int rc;
 743
 744                 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
 745                 if (rc)
 746                         return rc;
 747         }
 748
 749         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
 750
 751         return 0;
 752 }
 753
 754 static ssize_t revision_show(struct device *dev,
 755                 struct device_attribute *attr, char *buf)
 756 {
 757         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 758         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 759         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 760
 761         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
 762 }
 763 static DEVICE_ATTR_RO(revision);
 764
 765 static struct attribute *acpi_nfit_attributes[] = {
 766         &dev_attr_revision.attr,
 767         NULL,
 768 };
 769
 770 static struct attribute_group acpi_nfit_attribute_group = {
 771         .name = "nfit",
 772         .attrs = acpi_nfit_attributes,
 773 };
 774
 775 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
 776         &nvdimm_bus_attribute_group,
 777         &acpi_nfit_attribute_group,
 778         NULL,
 779 };
 780
 781 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
 782 {
 783         struct nvdimm *nvdimm = to_nvdimm(dev);
 784         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 785
 786         return __to_nfit_memdev(nfit_mem);
 787 }
 788
 789 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
 790 {
 791         struct nvdimm *nvdimm = to_nvdimm(dev);
 792         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 793
 794         return nfit_mem->dcr;
 795 }
 796
 797 static ssize_t handle_show(struct device *dev,
 798                 struct device_attribute *attr, char *buf)
 799 {
 800         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 801
 802         return sprintf(buf, "%#x\n", memdev->device_handle);
 803 }
 804 static DEVICE_ATTR_RO(handle);
 805
 806 static ssize_t phys_id_show(struct device *dev,
 807                 struct device_attribute *attr, char *buf)
 808 {
 809         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 810
 811         return sprintf(buf, "%#x\n", memdev->physical_id);
 812 }
 813 static DEVICE_ATTR_RO(phys_id);
 814
 815 static ssize_t vendor_show(struct device *dev,
 816                 struct device_attribute *attr, char *buf)
 817 {
 818         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 819
 820         return sprintf(buf, "%#x\n", dcr->vendor_id);
 821 }
 822 static DEVICE_ATTR_RO(vendor);
 823
 824 static ssize_t rev_id_show(struct device *dev,
 825                 struct device_attribute *attr, char *buf)
 826 {
 827         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 828
 829         return sprintf(buf, "%#x\n", dcr->revision_id);
 830 }
 831 static DEVICE_ATTR_RO(rev_id);
 832
 833 static ssize_t device_show(struct device *dev,
 834                 struct device_attribute *attr, char *buf)
 835 {
 836         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 837
 838         return sprintf(buf, "%#x\n", dcr->device_id);
 839 }
 840 static DEVICE_ATTR_RO(device);
 841
 842 static int num_nvdimm_formats(struct nvdimm *nvdimm)
 843 {
 844         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 845         int formats = 0;
 846
 847         if (nfit_mem->memdev_pmem)
 848                 formats++;
 849         if (nfit_mem->memdev_bdw)
 850                 formats++;
 851         return formats;
 852 }
 853
 854 static ssize_t format_show(struct device *dev,
 855                 struct device_attribute *attr, char *buf)
 856 {
 857         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 858
 859         return sprintf(buf, "%#x\n", dcr->code);
 860 }
 861 static DEVICE_ATTR_RO(format);
 862
 863 static ssize_t format1_show(struct device *dev,
 864                 struct device_attribute *attr, char *buf)
 865 {
 866         u32 handle;
 867         ssize_t rc = -ENXIO;
 868         struct nfit_mem *nfit_mem;
 869         struct nfit_memdev *nfit_memdev;
 870         struct acpi_nfit_desc *acpi_desc;
 871         struct nvdimm *nvdimm = to_nvdimm(dev);
 872         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 873
 874         nfit_mem = nvdimm_provider_data(nvdimm);
 875         acpi_desc = nfit_mem->acpi_desc;
 876         handle = to_nfit_memdev(dev)->device_handle;
 877
 878         /* assumes DIMMs have at most 2 published interface codes */
 879         mutex_lock(&acpi_desc->init_mutex);
 880         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 881                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
 882                 struct nfit_dcr *nfit_dcr;
 883
 884                 if (memdev->device_handle != handle)
 885                         continue;
 886
 887                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 888                         if (nfit_dcr->dcr->region_index != memdev->region_index)
 889                                 continue;
 890                         if (nfit_dcr->dcr->code == dcr->code)
 891                                 continue;
 892                         rc = sprintf(buf, "%#x\n", nfit_dcr->dcr->code);
 893                         break;
 894                 }
 895                 if (rc != ENXIO)
 896                         break;
 897         }
 898         mutex_unlock(&acpi_desc->init_mutex);
 899         return rc;
 900 }
 901 static DEVICE_ATTR_RO(format1);
 902
 903 static ssize_t formats_show(struct device *dev,
 904                 struct device_attribute *attr, char *buf)
 905 {
 906         struct nvdimm *nvdimm = to_nvdimm(dev);
 907
 908         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
 909 }
 910 static DEVICE_ATTR_RO(formats);
 911
 912 static ssize_t serial_show(struct device *dev,
 913                 struct device_attribute *attr, char *buf)
 914 {
 915         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 916
 917         return sprintf(buf, "%#x\n", dcr->serial_number);
 918 }
 919 static DEVICE_ATTR_RO(serial);
 920
 921 static ssize_t flags_show(struct device *dev,
 922                 struct device_attribute *attr, char *buf)
 923 {
 924         u16 flags = to_nfit_memdev(dev)->flags;
 925
 926         return sprintf(buf, "%s%s%s%s%s\n",
 927                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
 928                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
 929                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
 930                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
 931                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
 932 }
 933 static DEVICE_ATTR_RO(flags);
 934
 935 static struct attribute *acpi_nfit_dimm_attributes[] = {
 936         &dev_attr_handle.attr,
 937         &dev_attr_phys_id.attr,
 938         &dev_attr_vendor.attr,
 939         &dev_attr_device.attr,
 940         &dev_attr_format.attr,
 941         &dev_attr_formats.attr,
 942         &dev_attr_format1.attr,
 943         &dev_attr_serial.attr,
 944         &dev_attr_rev_id.attr,
 945         &dev_attr_flags.attr,
 946         NULL,
 947 };
 948
 949 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
 950                 struct attribute *a, int n)
 951 {
 952         struct device *dev = container_of(kobj, struct device, kobj);
 953         struct nvdimm *nvdimm = to_nvdimm(dev);
 954
 955         if (!to_nfit_dcr(dev))
 956                 return 0;
 957         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
 958                 return 0;
 959         return a->mode;
 960 }
 961
 962 static struct attribute_group acpi_nfit_dimm_attribute_group = {
 963         .name = "nfit",
 964         .attrs = acpi_nfit_dimm_attributes,
 965         .is_visible = acpi_nfit_dimm_attr_visible,
 966 };
 967
 968 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
 969         &nvdimm_attribute_group,
 970         &nd_device_attribute_group,
 971         &acpi_nfit_dimm_attribute_group,
 972         NULL,
 973 };
 974
 975 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
 976                 u32 device_handle)
 977 {
 978         struct nfit_mem *nfit_mem;
 979
 980         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 981                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
 982                         return nfit_mem->nvdimm;
 983
 984         return NULL;
 985 }
 986
 987 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
 988                 struct nfit_mem *nfit_mem, u32 device_handle)
 989 {
 990         struct acpi_device *adev, *adev_dimm;
 991         struct device *dev = acpi_desc->dev;
 992         const u8 *uuid = to_nfit_uuid(NFIT_DEV_DIMM);
 993         int i;
 994
 995         nfit_mem->dsm_mask = acpi_desc->dimm_dsm_force_en;
 996         adev = to_acpi_dev(acpi_desc);
 997         if (!adev)
 998                 return 0;
 999
1000         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1001         nfit_mem->adev = adev_dimm;
1002         if (!adev_dimm) {
1003                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1004                                 device_handle);
1005                 return force_enable_dimms ? 0 : -ENODEV;
1006         }
1007
1008         for (i = ND_CMD_SMART; i <= ND_CMD_VENDOR; i++)
1009                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1010                         set_bit(i, &nfit_mem->dsm_mask);
1011
1012         return 0;
1013 }
1014
1015 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1016 {
1017         struct nfit_mem *nfit_mem;
1018         int dimm_count = 0;
1019
1020         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1021                 struct nvdimm *nvdimm;
1022                 unsigned long flags = 0;
1023                 u32 device_handle;
1024                 u16 mem_flags;
1025                 int rc;
1026
1027                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1028                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1029                 if (nvdimm) {
1030                         dimm_count++;
1031                         continue;
1032                 }
1033
1034                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1035                         flags |= NDD_ALIASING;
1036
1037                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1038                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1039                         flags |= NDD_UNARMED;
1040
1041                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1042                 if (rc)
1043                         continue;
1044
1045                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1046                                 acpi_nfit_dimm_attribute_groups,
1047                                 flags, &nfit_mem->dsm_mask);
1048                 if (!nvdimm)
1049                         return -ENOMEM;
1050
1051                 nfit_mem->nvdimm = nvdimm;
1052                 dimm_count++;
1053
1054                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1055                         continue;
1056
1057                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1058                                 nvdimm_name(nvdimm),
1059                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1060                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1061                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1062                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1063
1064         }
1065
1066         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1067 }
1068
1069 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1070 {
1071         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1072         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1073         struct acpi_device *adev;
1074         int i;
1075
1076         nd_desc->dsm_mask = acpi_desc->bus_dsm_force_en;
1077         adev = to_acpi_dev(acpi_desc);
1078         if (!adev)
1079                 return;
1080
1081         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1082                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1083                         set_bit(i, &nd_desc->dsm_mask);
1084 }
1085
1086 static ssize_t range_index_show(struct device *dev,
1087                 struct device_attribute *attr, char *buf)
1088 {
1089         struct nd_region *nd_region = to_nd_region(dev);
1090         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1091
1092         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1093 }
1094 static DEVICE_ATTR_RO(range_index);
1095
1096 static struct attribute *acpi_nfit_region_attributes[] = {
1097         &dev_attr_range_index.attr,
1098         NULL,
1099 };
1100
1101 static struct attribute_group acpi_nfit_region_attribute_group = {
1102         .name = "nfit",
1103         .attrs = acpi_nfit_region_attributes,
1104 };
1105
1106 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1107         &nd_region_attribute_group,
1108         &nd_mapping_attribute_group,
1109         &nd_device_attribute_group,
1110         &nd_numa_attribute_group,
1111         &acpi_nfit_region_attribute_group,
1112         NULL,
1113 };
1114
1115 /* enough info to uniquely specify an interleave set */
1116 struct nfit_set_info {
1117         struct nfit_set_info_map {
1118                 u64 region_offset;
1119                 u32 serial_number;
1120                 u32 pad;
1121         } mapping[0];
1122 };
1123
1124 static size_t sizeof_nfit_set_info(int num_mappings)
1125 {
1126         return sizeof(struct nfit_set_info)
1127                 + num_mappings * sizeof(struct nfit_set_info_map);
1128 }
1129
1130 static int cmp_map(const void *m0, const void *m1)
1131 {
1132         const struct nfit_set_info_map *map0 = m0;
1133         const struct nfit_set_info_map *map1 = m1;
1134
1135         return memcmp(&map0->region_offset, &map1->region_offset,
1136                         sizeof(u64));
1137 }
1138
1139 /* Retrieve the nth entry referencing this spa */
1140 static struct acpi_nfit_memory_map *memdev_from_spa(
1141                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1142 {
1143         struct nfit_memdev *nfit_memdev;
1144
1145         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1146                 if (nfit_memdev->memdev->range_index == range_index)
1147                         if (n-- == 0)
1148                                 return nfit_memdev->memdev;
1149         return NULL;
1150 }
1151
1152 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1153                 struct nd_region_desc *ndr_desc,
1154                 struct acpi_nfit_system_address *spa)
1155 {
1156         int i, spa_type = nfit_spa_type(spa);
1157         struct device *dev = acpi_desc->dev;
1158         struct nd_interleave_set *nd_set;
1159         u16 nr = ndr_desc->num_mappings;
1160         struct nfit_set_info *info;
1161
1162         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1163                 /* pass */;
1164         else
1165                 return 0;
1166
1167         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1168         if (!nd_set)
1169                 return -ENOMEM;
1170
1171         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1172         if (!info)
1173                 return -ENOMEM;
1174         for (i = 0; i < nr; i++) {
1175                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1176                 struct nfit_set_info_map *map = &info->mapping[i];
1177                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1178                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1179                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1180                                 spa->range_index, i);
1181
1182                 if (!memdev || !nfit_mem->dcr) {
1183                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1184                         return -ENODEV;
1185                 }
1186
1187                 map->region_offset = memdev->region_offset;
1188                 map->serial_number = nfit_mem->dcr->serial_number;
1189         }
1190
1191         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1192                         cmp_map, NULL);
1193         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1194         ndr_desc->nd_set = nd_set;
1195         devm_kfree(dev, info);
1196
1197         return 0;
1198 }
1199
1200 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1201 {
1202         struct acpi_nfit_interleave *idt = mmio->idt;
1203         u32 sub_line_offset, line_index, line_offset;
1204         u64 line_no, table_skip_count, table_offset;
1205
1206         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1207         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1208         line_offset = idt->line_offset[line_index]
1209                 * mmio->line_size;
1210         table_offset = table_skip_count * mmio->table_size;
1211
1212         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1213 }
1214
1215 static void wmb_blk(struct nfit_blk *nfit_blk)
1216 {
1217
1218         if (nfit_blk->nvdimm_flush) {
1219                 /*
1220                  * The first wmb() is needed to 'sfence' all previous writes
1221                  * such that they are architecturally visible for the platform
1222                  * buffer flush.  Note that we've already arranged for pmem
1223                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1224                  * final wmb() ensures ordering for the NVDIMM flush write.
1225                  */
1226                 wmb();
1227                 writeq(1, nfit_blk->nvdimm_flush);
1228                 wmb();
1229         } else
1230                 wmb_pmem();
1231 }
1232
1233 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1234 {
1235         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1236         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1237
1238         if (mmio->num_lines)
1239                 offset = to_interleave_offset(offset, mmio);
1240
1241         return readl(mmio->addr.base + offset);
1242 }
1243
1244 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1245                 resource_size_t dpa, unsigned int len, unsigned int write)
1246 {
1247         u64 cmd, offset;
1248         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1249
1250         enum {
1251                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1252                 BCW_LEN_SHIFT = 48,
1253                 BCW_LEN_MASK = (1ULL << 8) - 1,
1254                 BCW_CMD_SHIFT = 56,
1255         };
1256
1257         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1258         len = len >> L1_CACHE_SHIFT;
1259         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1260         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1261
1262         offset = nfit_blk->cmd_offset + mmio->size * bw;
1263         if (mmio->num_lines)
1264                 offset = to_interleave_offset(offset, mmio);
1265
1266         writeq(cmd, mmio->addr.base + offset);
1267         wmb_blk(nfit_blk);
1268
1269         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1270                 readq(mmio->addr.base + offset);
1271 }
1272
1273 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1274                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1275                 unsigned int lane)
1276 {
1277         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1278         unsigned int copied = 0;
1279         u64 base_offset;
1280         int rc;
1281
1282         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1283                 + lane * mmio->size;
1284         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1285         while (len) {
1286                 unsigned int c;
1287                 u64 offset;
1288
1289                 if (mmio->num_lines) {
1290                         u32 line_offset;
1291
1292                         offset = to_interleave_offset(base_offset + copied,
1293                                         mmio);
1294                         div_u64_rem(offset, mmio->line_size, &line_offset);
1295                         c = min_t(size_t, len, mmio->line_size - line_offset);
1296                 } else {
1297                         offset = base_offset + nfit_blk->bdw_offset;
1298                         c = len;
1299                 }
1300
1301                 if (rw)
1302                         memcpy_to_pmem(mmio->addr.aperture + offset,
1303                                         iobuf + copied, c);
1304                 else {
1305                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1306                                 mmio_flush_range((void __force *)
1307                                         mmio->addr.aperture + offset, c);
1308
1309                         memcpy_from_pmem(iobuf + copied,
1310                                         mmio->addr.aperture + offset, c);
1311                 }
1312
1313                 copied += c;
1314                 len -= c;
1315         }
1316
1317         if (rw)
1318                 wmb_blk(nfit_blk);
1319
1320         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1321         return rc;
1322 }
1323
1324 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1325                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1326 {
1327         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1328         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1329         struct nd_region *nd_region = nfit_blk->nd_region;
1330         unsigned int lane, copied = 0;
1331         int rc = 0;
1332
1333         lane = nd_region_acquire_lane(nd_region);
1334         while (len) {
1335                 u64 c = min(len, mmio->size);
1336
1337                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1338                                 iobuf + copied, c, rw, lane);
1339                 if (rc)
1340                         break;
1341
1342                 copied += c;
1343                 len -= c;
1344         }
1345         nd_region_release_lane(nd_region, lane);
1346
1347         return rc;
1348 }
1349
1350 static void nfit_spa_mapping_release(struct kref *kref)
1351 {
1352         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1353         struct acpi_nfit_system_address *spa = spa_map->spa;
1354         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1355
1356         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1357         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1358         if (spa_map->type == SPA_MAP_APERTURE)
1359                 memunmap((void __force *)spa_map->addr.aperture);
1360         else
1361                 iounmap(spa_map->addr.base);
1362         release_mem_region(spa->address, spa->length);
1363         list_del(&spa_map->list);
1364         kfree(spa_map);
1365 }
1366
1367 static struct nfit_spa_mapping *find_spa_mapping(
1368                 struct acpi_nfit_desc *acpi_desc,
1369                 struct acpi_nfit_system_address *spa)
1370 {
1371         struct nfit_spa_mapping *spa_map;
1372
1373         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1374         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1375                 if (spa_map->spa == spa)
1376                         return spa_map;
1377
1378         return NULL;
1379 }
1380
1381 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1382                 struct acpi_nfit_system_address *spa)
1383 {
1384         struct nfit_spa_mapping *spa_map;
1385
1386         mutex_lock(&acpi_desc->spa_map_mutex);
1387         spa_map = find_spa_mapping(acpi_desc, spa);
1388
1389         if (spa_map)
1390                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1391         mutex_unlock(&acpi_desc->spa_map_mutex);
1392 }
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         resource_size_t start = spa->address;
1398         resource_size_t n = spa->length;
1399         struct nfit_spa_mapping *spa_map;
1400         struct resource *res;
1401
1402         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1403
1404         spa_map = find_spa_mapping(acpi_desc, spa);
1405         if (spa_map) {
1406                 kref_get(&spa_map->kref);
1407                 return spa_map->addr.base;
1408         }
1409
1410         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1411         if (!spa_map)
1412                 return NULL;
1413
1414         INIT_LIST_HEAD(&spa_map->list);
1415         spa_map->spa = spa;
1416         kref_init(&spa_map->kref);
1417         spa_map->acpi_desc = acpi_desc;
1418
1419         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1420         if (!res)
1421                 goto err_mem;
1422
1423         spa_map->type = type;
1424         if (type == SPA_MAP_APERTURE)
1425                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1426                                                         ARCH_MEMREMAP_PMEM);
1427         else
1428                 spa_map->addr.base = ioremap_nocache(start, n);
1429
1430
1431         if (!spa_map->addr.base)
1432                 goto err_map;
1433
1434         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1435         return spa_map->addr.base;
1436
1437  err_map:
1438         release_mem_region(start, n);
1439  err_mem:
1440         kfree(spa_map);
1441         return NULL;
1442 }
1443
1444 /**
1445  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1446  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1447  * @nfit_spa: spa table to map
1448  * @type: aperture or control region
1449  *
1450  * In the case where block-data-window apertures and
1451  * dimm-control-regions are interleaved they will end up sharing a
1452  * single request_mem_region() + ioremap() for the address range.  In
1453  * the style of devm nfit_spa_map() mappings are automatically dropped
1454  * when all region devices referencing the same mapping are disabled /
1455  * unbound.
1456  */
1457 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1458                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1459 {
1460         void __iomem *iomem;
1461
1462         mutex_lock(&acpi_desc->spa_map_mutex);
1463         iomem = __nfit_spa_map(acpi_desc, spa, type);
1464         mutex_unlock(&acpi_desc->spa_map_mutex);
1465
1466         return iomem;
1467 }
1468
1469 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1470                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1471 {
1472         if (idt) {
1473                 mmio->num_lines = idt->line_count;
1474                 mmio->line_size = idt->line_size;
1475                 if (interleave_ways == 0)
1476                         return -ENXIO;
1477                 mmio->table_size = mmio->num_lines * interleave_ways
1478                         * mmio->line_size;
1479         }
1480
1481         return 0;
1482 }
1483
1484 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1485                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1486 {
1487         struct nd_cmd_dimm_flags flags;
1488         int rc;
1489
1490         memset(&flags, 0, sizeof(flags));
1491         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1492                         sizeof(flags), NULL);
1493
1494         if (rc >= 0 && flags.status == 0)
1495                 nfit_blk->dimm_flags = flags.flags;
1496         else if (rc == -ENOTTY) {
1497                 /* fall back to a conservative default */
1498                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1499                 rc = 0;
1500         } else
1501                 rc = -ENXIO;
1502
1503         return rc;
1504 }
1505
1506 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1507                 struct device *dev)
1508 {
1509         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1510         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1511         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1512         struct nfit_flush *nfit_flush;
1513         struct nfit_blk_mmio *mmio;
1514         struct nfit_blk *nfit_blk;
1515         struct nfit_mem *nfit_mem;
1516         struct nvdimm *nvdimm;
1517         int rc;
1518
1519         nvdimm = nd_blk_region_to_dimm(ndbr);
1520         nfit_mem = nvdimm_provider_data(nvdimm);
1521         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1522                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1523                                 nfit_mem ? "" : " nfit_mem",
1524                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1525                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1526                 return -ENXIO;
1527         }
1528
1529         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1530         if (!nfit_blk)
1531                 return -ENOMEM;
1532         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1533         nfit_blk->nd_region = to_nd_region(dev);
1534
1535         /* map block aperture memory */
1536         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1537         mmio = &nfit_blk->mmio[BDW];
1538         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1539                         SPA_MAP_APERTURE);
1540         if (!mmio->addr.base) {
1541                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1542                                 nvdimm_name(nvdimm));
1543                 return -ENOMEM;
1544         }
1545         mmio->size = nfit_mem->bdw->size;
1546         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1547         mmio->idt = nfit_mem->idt_bdw;
1548         mmio->spa = nfit_mem->spa_bdw;
1549         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1550                         nfit_mem->memdev_bdw->interleave_ways);
1551         if (rc) {
1552                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1553                                 __func__, nvdimm_name(nvdimm));
1554                 return rc;
1555         }
1556
1557         /* map block control memory */
1558         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1559         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1560         mmio = &nfit_blk->mmio[DCR];
1561         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1562                         SPA_MAP_CONTROL);
1563         if (!mmio->addr.base) {
1564                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1565                                 nvdimm_name(nvdimm));
1566                 return -ENOMEM;
1567         }
1568         mmio->size = nfit_mem->dcr->window_size;
1569         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1570         mmio->idt = nfit_mem->idt_dcr;
1571         mmio->spa = nfit_mem->spa_dcr;
1572         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1573                         nfit_mem->memdev_dcr->interleave_ways);
1574         if (rc) {
1575                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1576                                 __func__, nvdimm_name(nvdimm));
1577                 return rc;
1578         }
1579
1580         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1581         if (rc < 0) {
1582                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1583                                 __func__, nvdimm_name(nvdimm));
1584                 return rc;
1585         }
1586
1587         nfit_flush = nfit_mem->nfit_flush;
1588         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1589                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1590                                 nfit_flush->flush->hint_address[0], 8);
1591                 if (!nfit_blk->nvdimm_flush)
1592                         return -ENOMEM;
1593         }
1594
1595         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1596                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1597
1598         if (mmio->line_size == 0)
1599                 return 0;
1600
1601         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1602                         + 8 > mmio->line_size) {
1603                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1604                 return -ENXIO;
1605         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1606                         + 8 > mmio->line_size) {
1607                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1608                 return -ENXIO;
1609         }
1610
1611         return 0;
1612 }
1613
1614 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1615                 struct device *dev)
1616 {
1617         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1618         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1619         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1620         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1621         int i;
1622
1623         if (!nfit_blk)
1624                 return; /* never enabled */
1625
1626         /* auto-free BLK spa mappings */
1627         for (i = 0; i < 2; i++) {
1628                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1629
1630                 if (mmio->addr.base)
1631                         nfit_spa_unmap(acpi_desc, mmio->spa);
1632         }
1633         nd_blk_region_set_provider_data(ndbr, NULL);
1634         /* devm will free nfit_blk */
1635 }
1636
1637 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1638                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1639 {
1640         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1641         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1642         int cmd_rc, rc;
1643
1644         cmd->address = spa->address;
1645         cmd->length = spa->length;
1646         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1647                         sizeof(*cmd), &cmd_rc);
1648         if (rc < 0)
1649                 return rc;
1650         return cmd_rc;
1651 }
1652
1653 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1654 {
1655         int rc;
1656         int cmd_rc;
1657         struct nd_cmd_ars_start ars_start;
1658         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1659         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1660
1661         memset(&ars_start, 0, sizeof(ars_start));
1662         ars_start.address = spa->address;
1663         ars_start.length = spa->length;
1664         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1665                 ars_start.type = ND_ARS_PERSISTENT;
1666         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1667                 ars_start.type = ND_ARS_VOLATILE;
1668         else
1669                 return -ENOTTY;
1670
1671         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1672                         sizeof(ars_start), &cmd_rc);
1673
1674         if (rc < 0)
1675                 return rc;
1676         return cmd_rc;
1677 }
1678
1679 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1680 {
1681         int rc, cmd_rc;
1682         struct nd_cmd_ars_start ars_start;
1683         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1684         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1685
1686         memset(&ars_start, 0, sizeof(ars_start));
1687         ars_start.address = ars_status->restart_address;
1688         ars_start.length = ars_status->restart_length;
1689         ars_start.type = ars_status->type;
1690         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1691                         sizeof(ars_start), &cmd_rc);
1692         if (rc < 0)
1693                 return rc;
1694         return cmd_rc;
1695 }
1696
1697 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1698 {
1699         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1700         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1701         int rc, cmd_rc;
1702
1703         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1704                         acpi_desc->ars_status_size, &cmd_rc);
1705         if (rc < 0)
1706                 return rc;
1707         return cmd_rc;
1708 }
1709
1710 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1711                 struct nd_cmd_ars_status *ars_status)
1712 {
1713         int rc;
1714         u32 i;
1715
1716         for (i = 0; i < ars_status->num_records; i++) {
1717                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1718                                 ars_status->records[i].err_address,
1719                                 ars_status->records[i].length);
1720                 if (rc)
1721                         return rc;
1722         }
1723
1724         return 0;
1725 }
1726
1727 static void acpi_nfit_remove_resource(void *data)
1728 {
1729         struct resource *res = data;
1730
1731         remove_resource(res);
1732 }
1733
1734 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1735                 struct nd_region_desc *ndr_desc)
1736 {
1737         struct resource *res, *nd_res = ndr_desc->res;
1738         int is_pmem, ret;
1739
1740         /* No operation if the region is already registered as PMEM */
1741         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1742                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1743         if (is_pmem == REGION_INTERSECTS)
1744                 return 0;
1745
1746         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1747         if (!res)
1748                 return -ENOMEM;
1749
1750         res->name = "Persistent Memory";
1751         res->start = nd_res->start;
1752         res->end = nd_res->end;
1753         res->flags = IORESOURCE_MEM;
1754         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1755
1756         ret = insert_resource(&iomem_resource, res);
1757         if (ret)
1758                 return ret;
1759
1760         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1761         if (ret) {
1762                 remove_resource(res);
1763                 return ret;
1764         }
1765
1766         return 0;
1767 }
1768
1769 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1770                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1771                 struct acpi_nfit_memory_map *memdev,
1772                 struct nfit_spa *nfit_spa)
1773 {
1774         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1775                         memdev->device_handle);
1776         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1777         struct nd_blk_region_desc *ndbr_desc;
1778         struct nfit_mem *nfit_mem;
1779         int blk_valid = 0;
1780
1781         if (!nvdimm) {
1782                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1783                                 spa->range_index, memdev->device_handle);
1784                 return -ENODEV;
1785         }
1786
1787         nd_mapping->nvdimm = nvdimm;
1788         switch (nfit_spa_type(spa)) {
1789         case NFIT_SPA_PM:
1790         case NFIT_SPA_VOLATILE:
1791                 nd_mapping->start = memdev->address;
1792                 nd_mapping->size = memdev->region_size;
1793                 break;
1794         case NFIT_SPA_DCR:
1795                 nfit_mem = nvdimm_provider_data(nvdimm);
1796                 if (!nfit_mem || !nfit_mem->bdw) {
1797                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1798                                         spa->range_index, nvdimm_name(nvdimm));
1799                 } else {
1800                         nd_mapping->size = nfit_mem->bdw->capacity;
1801                         nd_mapping->start = nfit_mem->bdw->start_address;
1802                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1803                         blk_valid = 1;
1804                 }
1805
1806                 ndr_desc->nd_mapping = nd_mapping;
1807                 ndr_desc->num_mappings = blk_valid;
1808                 ndbr_desc = to_blk_region_desc(ndr_desc);
1809                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1810                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1811                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1812                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1813                                 ndr_desc);
1814                 if (!nfit_spa->nd_region)
1815                         return -ENOMEM;
1816                 break;
1817         }
1818
1819         return 0;
1820 }
1821
1822 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1823                 struct nfit_spa *nfit_spa)
1824 {
1825         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1826         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1827         struct nd_blk_region_desc ndbr_desc;
1828         struct nd_region_desc *ndr_desc;
1829         struct nfit_memdev *nfit_memdev;
1830         struct nvdimm_bus *nvdimm_bus;
1831         struct resource res;
1832         int count = 0, rc;
1833
1834         if (nfit_spa->nd_region)
1835                 return 0;
1836
1837         if (spa->range_index == 0) {
1838                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1839                                 __func__);
1840                 return 0;
1841         }
1842
1843         memset(&res, 0, sizeof(res));
1844         memset(&nd_mappings, 0, sizeof(nd_mappings));
1845         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1846         res.start = spa->address;
1847         res.end = res.start + spa->length - 1;
1848         ndr_desc = &ndbr_desc.ndr_desc;
1849         ndr_desc->res = &res;
1850         ndr_desc->provider_data = nfit_spa;
1851         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1852         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1853                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
1854                                                 spa->proximity_domain);
1855         else
1856                 ndr_desc->numa_node = NUMA_NO_NODE;
1857
1858         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1859                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1860                 struct nd_mapping *nd_mapping;
1861
1862                 if (memdev->range_index != spa->range_index)
1863                         continue;
1864                 if (count >= ND_MAX_MAPPINGS) {
1865                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
1866                                         spa->range_index, ND_MAX_MAPPINGS);
1867                         return -ENXIO;
1868                 }
1869                 nd_mapping = &nd_mappings[count++];
1870                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
1871                                 memdev, nfit_spa);
1872                 if (rc)
1873                         goto out;
1874         }
1875
1876         ndr_desc->nd_mapping = nd_mappings;
1877         ndr_desc->num_mappings = count;
1878         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
1879         if (rc)
1880                 goto out;
1881
1882         nvdimm_bus = acpi_desc->nvdimm_bus;
1883         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
1884                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
1885                 if (rc) {
1886                         dev_warn(acpi_desc->dev,
1887                                 "failed to insert pmem resource to iomem: %d\n",
1888                                 rc);
1889                         goto out;
1890                 }
1891
1892                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
1893                                 ndr_desc);
1894                 if (!nfit_spa->nd_region)
1895                         rc = -ENOMEM;
1896         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
1897                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
1898                                 ndr_desc);
1899                 if (!nfit_spa->nd_region)
1900                         rc = -ENOMEM;
1901         }
1902
1903  out:
1904         if (rc)
1905                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
1906                                 nfit_spa->spa->range_index);
1907         return rc;
1908 }
1909
1910 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
1911                 u32 max_ars)
1912 {
1913         struct device *dev = acpi_desc->dev;
1914         struct nd_cmd_ars_status *ars_status;
1915
1916         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
1917                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
1918                 return 0;
1919         }
1920
1921         if (acpi_desc->ars_status)
1922                 devm_kfree(dev, acpi_desc->ars_status);
1923         acpi_desc->ars_status = NULL;
1924         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
1925         if (!ars_status)
1926                 return -ENOMEM;
1927         acpi_desc->ars_status = ars_status;
1928         acpi_desc->ars_status_size = max_ars;
1929         return 0;
1930 }
1931
1932 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
1933                 struct nfit_spa *nfit_spa)
1934 {
1935         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1936         int rc;
1937
1938         if (!nfit_spa->max_ars) {
1939                 struct nd_cmd_ars_cap ars_cap;
1940
1941                 memset(&ars_cap, 0, sizeof(ars_cap));
1942                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
1943                 if (rc < 0)
1944                         return rc;
1945                 nfit_spa->max_ars = ars_cap.max_ars_out;
1946                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
1947                 /* check that the supported scrub types match the spa type */
1948                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
1949                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
1950                         return -ENOTTY;
1951                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
1952                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
1953                         return -ENOTTY;
1954         }
1955
1956         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
1957                 return -ENOMEM;
1958
1959         rc = ars_get_status(acpi_desc);
1960         if (rc < 0 && rc != -ENOSPC)
1961                 return rc;
1962
1963         if (ars_status_process_records(acpi_desc->nvdimm_bus,
1964                                 acpi_desc->ars_status))
1965                 return -ENOMEM;
1966
1967         return 0;
1968 }
1969
1970 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
1971                 struct nfit_spa *nfit_spa)
1972 {
1973         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1974         unsigned int overflow_retry = scrub_overflow_abort;
1975         u64 init_ars_start = 0, init_ars_len = 0;
1976         struct device *dev = acpi_desc->dev;
1977         unsigned int tmo = scrub_timeout;
1978         int rc;
1979
1980         if (nfit_spa->ars_done || !nfit_spa->nd_region)
1981                 return;
1982
1983         rc = ars_start(acpi_desc, nfit_spa);
1984         /*
1985          * If we timed out the initial scan we'll still be busy here,
1986          * and will wait another timeout before giving up permanently.
1987          */
1988         if (rc < 0 && rc != -EBUSY)
1989                 return;
1990
1991         do {
1992                 u64 ars_start, ars_len;
1993
1994                 if (acpi_desc->cancel)
1995                         break;
1996                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
1997                 if (rc == -ENOTTY)
1998                         break;
1999                 if (rc == -EBUSY && !tmo) {
2000                         dev_warn(dev, "range %d ars timeout, aborting\n",
2001                                         spa->range_index);
2002                         break;
2003                 }
2004
2005                 if (rc == -EBUSY) {
2006                         /*
2007                          * Note, entries may be appended to the list
2008                          * while the lock is dropped, but the workqueue
2009                          * being active prevents entries being deleted /
2010                          * freed.
2011                          */
2012                         mutex_unlock(&acpi_desc->init_mutex);
2013                         ssleep(1);
2014                         tmo--;
2015                         mutex_lock(&acpi_desc->init_mutex);
2016                         continue;
2017                 }
2018
2019                 /* we got some results, but there are more pending... */
2020                 if (rc == -ENOSPC && overflow_retry--) {
2021                         if (!init_ars_len) {
2022                                 init_ars_len = acpi_desc->ars_status->length;
2023                                 init_ars_start = acpi_desc->ars_status->address;
2024                         }
2025                         rc = ars_continue(acpi_desc);
2026                 }
2027
2028                 if (rc < 0) {
2029                         dev_warn(dev, "range %d ars continuation failed\n",
2030                                         spa->range_index);
2031                         break;
2032                 }
2033
2034                 if (init_ars_len) {
2035                         ars_start = init_ars_start;
2036                         ars_len = init_ars_len;
2037                 } else {
2038                         ars_start = acpi_desc->ars_status->address;
2039                         ars_len = acpi_desc->ars_status->length;
2040                 }
2041                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2042                                 spa->range_index, ars_start, ars_len);
2043                 /* notify the region about new poison entries */
2044                 nvdimm_region_notify(nfit_spa->nd_region,
2045                                 NVDIMM_REVALIDATE_POISON);
2046                 break;
2047         } while (1);
2048 }
2049
2050 static void acpi_nfit_scrub(struct work_struct *work)
2051 {
2052         struct device *dev;
2053         u64 init_scrub_length = 0;
2054         struct nfit_spa *nfit_spa;
2055         u64 init_scrub_address = 0;
2056         bool init_ars_done = false;
2057         struct acpi_nfit_desc *acpi_desc;
2058         unsigned int tmo = scrub_timeout;
2059         unsigned int overflow_retry = scrub_overflow_abort;
2060
2061         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2062         dev = acpi_desc->dev;
2063
2064         /*
2065          * We scrub in 2 phases.  The first phase waits for any platform
2066          * firmware initiated scrubs to complete and then we go search for the
2067          * affected spa regions to mark them scanned.  In the second phase we
2068          * initiate a directed scrub for every range that was not scrubbed in
2069          * phase 1.
2070          */
2071
2072         /* process platform firmware initiated scrubs */
2073  retry:
2074         mutex_lock(&acpi_desc->init_mutex);
2075         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2076                 struct nd_cmd_ars_status *ars_status;
2077                 struct acpi_nfit_system_address *spa;
2078                 u64 ars_start, ars_len;
2079                 int rc;
2080
2081                 if (acpi_desc->cancel)
2082                         break;
2083
2084                 if (nfit_spa->nd_region)
2085                         continue;
2086
2087                 if (init_ars_done) {
2088                         /*
2089                          * No need to re-query, we're now just
2090                          * reconciling all the ranges covered by the
2091                          * initial scrub
2092                          */
2093                         rc = 0;
2094                 } else
2095                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2096
2097                 if (rc == -ENOTTY) {
2098                         /* no ars capability, just register spa and move on */
2099                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2100                         continue;
2101                 }
2102
2103                 if (rc == -EBUSY && !tmo) {
2104                         /* fallthrough to directed scrub in phase 2 */
2105                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2106                         break;
2107                 } else if (rc == -EBUSY) {
2108                         mutex_unlock(&acpi_desc->init_mutex);
2109                         ssleep(1);
2110                         tmo--;
2111                         goto retry;
2112                 }
2113
2114                 /* we got some results, but there are more pending... */
2115                 if (rc == -ENOSPC && overflow_retry--) {
2116                         ars_status = acpi_desc->ars_status;
2117                         /*
2118                          * Record the original scrub range, so that we
2119                          * can recall all the ranges impacted by the
2120                          * initial scrub.
2121                          */
2122                         if (!init_scrub_length) {
2123                                 init_scrub_length = ars_status->length;
2124                                 init_scrub_address = ars_status->address;
2125                         }
2126                         rc = ars_continue(acpi_desc);
2127                         if (rc == 0) {
2128                                 mutex_unlock(&acpi_desc->init_mutex);
2129                                 goto retry;
2130                         }
2131                 }
2132
2133                 if (rc < 0) {
2134                         /*
2135                          * Initial scrub failed, we'll give it one more
2136                          * try below...
2137                          */
2138                         break;
2139                 }
2140
2141                 /* We got some final results, record completed ranges */
2142                 ars_status = acpi_desc->ars_status;
2143                 if (init_scrub_length) {
2144                         ars_start = init_scrub_address;
2145                         ars_len = ars_start + init_scrub_length;
2146                 } else {
2147                         ars_start = ars_status->address;
2148                         ars_len = ars_status->length;
2149                 }
2150                 spa = nfit_spa->spa;
2151
2152                 if (!init_ars_done) {
2153                         init_ars_done = true;
2154                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2155                                         ars_start, ars_len);
2156                 }
2157                 if (ars_start <= spa->address && ars_start + ars_len
2158                                 >= spa->address + spa->length)
2159                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2160         }
2161
2162         /*
2163          * For all the ranges not covered by an initial scrub we still
2164          * want to see if there are errors, but it's ok to discover them
2165          * asynchronously.
2166          */
2167         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2168                 /*
2169                  * Flag all the ranges that still need scrubbing, but
2170                  * register them now to make data available.
2171                  */
2172                 if (nfit_spa->nd_region)
2173                         nfit_spa->ars_done = 1;
2174                 else
2175                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2176         }
2177
2178         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2179                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2180         mutex_unlock(&acpi_desc->init_mutex);
2181 }
2182
2183 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2184 {
2185         struct nfit_spa *nfit_spa;
2186         int rc;
2187
2188         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2189                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2190                         /* BLK regions don't need to wait for ars results */
2191                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2192                         if (rc)
2193                                 return rc;
2194                 }
2195
2196         queue_work(nfit_wq, &acpi_desc->work);
2197         return 0;
2198 }
2199
2200 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2201                 struct nfit_table_prev *prev)
2202 {
2203         struct device *dev = acpi_desc->dev;
2204
2205         if (!list_empty(&prev->spas) ||
2206                         !list_empty(&prev->memdevs) ||
2207                         !list_empty(&prev->dcrs) ||
2208                         !list_empty(&prev->bdws) ||
2209                         !list_empty(&prev->idts) ||
2210                         !list_empty(&prev->flushes)) {
2211                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2212                 return -ENXIO;
2213         }
2214         return 0;
2215 }
2216
2217 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2218 {
2219         struct device *dev = acpi_desc->dev;
2220         struct nfit_table_prev prev;
2221         const void *end;
2222         u8 *data;
2223         int rc;
2224
2225         mutex_lock(&acpi_desc->init_mutex);
2226
2227         INIT_LIST_HEAD(&prev.spas);
2228         INIT_LIST_HEAD(&prev.memdevs);
2229         INIT_LIST_HEAD(&prev.dcrs);
2230         INIT_LIST_HEAD(&prev.bdws);
2231         INIT_LIST_HEAD(&prev.idts);
2232         INIT_LIST_HEAD(&prev.flushes);
2233
2234         list_cut_position(&prev.spas, &acpi_desc->spas,
2235                                 acpi_desc->spas.prev);
2236         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2237                                 acpi_desc->memdevs.prev);
2238         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2239                                 acpi_desc->dcrs.prev);
2240         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2241                                 acpi_desc->bdws.prev);
2242         list_cut_position(&prev.idts, &acpi_desc->idts,
2243                                 acpi_desc->idts.prev);
2244         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2245                                 acpi_desc->flushes.prev);
2246
2247         data = (u8 *) acpi_desc->nfit;
2248         end = data + sz;
2249         while (!IS_ERR_OR_NULL(data))
2250                 data = add_table(acpi_desc, &prev, data, end);
2251
2252         if (IS_ERR(data)) {
2253                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2254                                 PTR_ERR(data));
2255                 rc = PTR_ERR(data);
2256                 goto out_unlock;
2257         }
2258
2259         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2260         if (rc)
2261                 goto out_unlock;
2262
2263         if (nfit_mem_init(acpi_desc) != 0) {
2264                 rc = -ENOMEM;
2265                 goto out_unlock;
2266         }
2267
2268         acpi_nfit_init_dsms(acpi_desc);
2269
2270         rc = acpi_nfit_register_dimms(acpi_desc);
2271         if (rc)
2272                 goto out_unlock;
2273
2274         rc = acpi_nfit_register_regions(acpi_desc);
2275
2276  out_unlock:
2277         mutex_unlock(&acpi_desc->init_mutex);
2278         return rc;
2279 }
2280 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2281
2282 struct acpi_nfit_flush_work {
2283         struct work_struct work;
2284         struct completion cmp;
2285 };
2286
2287 static void flush_probe(struct work_struct *work)
2288 {
2289         struct acpi_nfit_flush_work *flush;
2290
2291         flush = container_of(work, typeof(*flush), work);
2292         complete(&flush->cmp);
2293 }
2294
2295 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2296 {
2297         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2298         struct device *dev = acpi_desc->dev;
2299         struct acpi_nfit_flush_work flush;
2300
2301         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2302         device_lock(dev);
2303         device_unlock(dev);
2304
2305         /*
2306          * Scrub work could take 10s of seconds, userspace may give up so we
2307          * need to be interruptible while waiting.
2308          */
2309         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2310         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2311         queue_work(nfit_wq, &flush.work);
2312         return wait_for_completion_interruptible(&flush.cmp);
2313 }
2314
2315 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2316                 struct nvdimm *nvdimm, unsigned int cmd)
2317 {
2318         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2319
2320         if (nvdimm)
2321                 return 0;
2322         if (cmd != ND_CMD_ARS_START)
2323                 return 0;
2324
2325         /*
2326          * The kernel and userspace may race to initiate a scrub, but
2327          * the scrub thread is prepared to lose that initial race.  It
2328          * just needs guarantees that any ars it initiates are not
2329          * interrupted by any intervening start reqeusts from userspace.
2330          */
2331         if (work_busy(&acpi_desc->work))
2332                 return -EBUSY;
2333
2334         return 0;
2335 }
2336
2337 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2338 {
2339         struct nvdimm_bus_descriptor *nd_desc;
2340
2341         dev_set_drvdata(dev, acpi_desc);
2342         acpi_desc->dev = dev;
2343         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2344         nd_desc = &acpi_desc->nd_desc;
2345         nd_desc->provider_name = "ACPI.NFIT";
2346         nd_desc->ndctl = acpi_nfit_ctl;
2347         nd_desc->flush_probe = acpi_nfit_flush_probe;
2348         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2349         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2350
2351         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2352         INIT_LIST_HEAD(&acpi_desc->spas);
2353         INIT_LIST_HEAD(&acpi_desc->dcrs);
2354         INIT_LIST_HEAD(&acpi_desc->bdws);
2355         INIT_LIST_HEAD(&acpi_desc->idts);
2356         INIT_LIST_HEAD(&acpi_desc->flushes);
2357         INIT_LIST_HEAD(&acpi_desc->memdevs);
2358         INIT_LIST_HEAD(&acpi_desc->dimms);
2359         mutex_init(&acpi_desc->spa_map_mutex);
2360         mutex_init(&acpi_desc->init_mutex);
2361         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2362 }
2363 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2364
2365 static int acpi_nfit_add(struct acpi_device *adev)
2366 {
2367         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2368         struct acpi_nfit_desc *acpi_desc;
2369         struct device *dev = &adev->dev;
2370         struct acpi_table_header *tbl;
2371         acpi_status status = AE_OK;
2372         acpi_size sz;
2373         int rc;
2374
2375         status = acpi_get_table_with_size("NFIT", 0, &tbl, &sz);
2376         if (ACPI_FAILURE(status)) {
2377                 /* This is ok, we could have an nvdimm hotplugged later */
2378                 dev_dbg(dev, "failed to find NFIT at startup\n");
2379                 return 0;
2380         }
2381
2382         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2383         if (!acpi_desc)
2384                 return -ENOMEM;
2385         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2386         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2387         if (!acpi_desc->nvdimm_bus)
2388                 return -ENOMEM;
2389
2390         /*
2391          * Save the acpi header for later and then skip it,
2392          * making nfit point to the first nfit table header.
2393          */
2394         acpi_desc->acpi_header = *tbl;
2395         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2396         sz -= sizeof(struct acpi_table_nfit);
2397
2398         /* Evaluate _FIT and override with that if present */
2399         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2400         if (ACPI_SUCCESS(status) && buf.length > 0) {
2401                 union acpi_object *obj;
2402                 /*
2403                  * Adjust for the acpi_object header of the _FIT
2404                  */
2405                 obj = buf.pointer;
2406                 if (obj->type == ACPI_TYPE_BUFFER) {
2407                         acpi_desc->nfit =
2408                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2409                         sz = obj->buffer.length;
2410                 } else
2411                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2412                                  __func__, (int) obj->type);
2413         }
2414
2415         rc = acpi_nfit_init(acpi_desc, sz);
2416         if (rc) {
2417                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2418                 return rc;
2419         }
2420         return 0;
2421 }
2422
2423 static int acpi_nfit_remove(struct acpi_device *adev)
2424 {
2425         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2426
2427         acpi_desc->cancel = 1;
2428         flush_workqueue(nfit_wq);
2429         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2430         return 0;
2431 }
2432
2433 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2434 {
2435         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2436         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2437         struct acpi_nfit_header *nfit_saved;
2438         union acpi_object *obj;
2439         struct device *dev = &adev->dev;
2440         acpi_status status;
2441         int ret;
2442
2443         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2444
2445         device_lock(dev);
2446         if (!dev->driver) {
2447                 /* dev->driver may be null if we're being removed */
2448                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2449                 goto out_unlock;
2450         }
2451
2452         if (!acpi_desc) {
2453                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2454                 if (!acpi_desc)
2455                         goto out_unlock;
2456                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2457                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2458                 if (!acpi_desc->nvdimm_bus)
2459                         goto out_unlock;
2460         } else {
2461                 /*
2462                  * Finish previous registration before considering new
2463                  * regions.
2464                  */
2465                 flush_workqueue(nfit_wq);
2466         }
2467
2468         /* Evaluate _FIT */
2469         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2470         if (ACPI_FAILURE(status)) {
2471                 dev_err(dev, "failed to evaluate _FIT\n");
2472                 goto out_unlock;
2473         }
2474
2475         nfit_saved = acpi_desc->nfit;
2476         obj = buf.pointer;
2477         if (obj->type == ACPI_TYPE_BUFFER) {
2478                 acpi_desc->nfit =
2479                         (struct acpi_nfit_header *)obj->buffer.pointer;
2480                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2481                 if (ret) {
2482                         /* Merge failed, restore old nfit, and exit */
2483                         acpi_desc->nfit = nfit_saved;
2484                         dev_err(dev, "failed to merge updated NFIT\n");
2485                 }
2486         } else {
2487                 /* Bad _FIT, restore old nfit */
2488                 dev_err(dev, "Invalid _FIT\n");
2489         }
2490         kfree(buf.pointer);
2491
2492  out_unlock:
2493         device_unlock(dev);
2494 }
2495
2496 static const struct acpi_device_id acpi_nfit_ids[] = {
2497         { "ACPI0012", 0 },
2498         { "", 0 },
2499 };
2500 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2501
2502 static struct acpi_driver acpi_nfit_driver = {
2503         .name = KBUILD_MODNAME,
2504         .ids = acpi_nfit_ids,
2505         .ops = {
2506                 .add = acpi_nfit_add,
2507                 .remove = acpi_nfit_remove,
2508                 .notify = acpi_nfit_notify,
2509         },
2510 };
2511
2512 static __init int nfit_init(void)
2513 {
2514         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2515         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2516         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2517         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2518         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2519         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2520         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2521
2522         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2523         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2524         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2525         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2526         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2527         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2528         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2529         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2530         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2531         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2532
2533         nfit_wq = create_singlethread_workqueue("nfit");
2534         if (!nfit_wq)
2535                 return -ENOMEM;
2536
2537         return acpi_bus_register_driver(&acpi_nfit_driver);
2538 }
2539
2540 static __exit void nfit_exit(void)
2541 {
2542         acpi_bus_unregister_driver(&acpi_nfit_driver);
2543         destroy_workqueue(nfit_wq);
2544 }
2545
2546 module_init(nfit_init);
2547 module_exit(nfit_exit);
2548 MODULE_LICENSE("GPL v2");
2549 MODULE_AUTHOR("Intel Corporation");