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                         list_add(&nfit_mem->list, &acpi_desc->dimms);
 659                 }
 660
 661                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 662                         if (nfit_dcr->dcr->region_index != dcr)
 663                                 continue;
 664                         /*
 665                          * Record the control region for the dimm.  For
 666                          * the ACPI 6.1 case, where there are separate
 667                          * control regions for the pmem vs blk
 668                          * interfaces, be sure to record the extended
 669                          * blk details.
 670                          */
 671                         if (!nfit_mem->dcr)
 672                                 nfit_mem->dcr = nfit_dcr->dcr;
 673                         else if (nfit_mem->dcr->windows == 0
 674                                         && nfit_dcr->dcr->windows)
 675                                 nfit_mem->dcr = nfit_dcr->dcr;
 676                         break;
 677                 }
 678
 679                 if (dcr && !nfit_mem->dcr) {
 680                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
 681                                         spa->range_index, dcr);
 682                         return -ENODEV;
 683                 }
 684
 685                 if (type == NFIT_SPA_DCR) {
 686                         struct nfit_idt *nfit_idt;
 687                         u16 idt_idx;
 688
 689                         /* multiple dimms may share a SPA when interleaved */
 690                         nfit_mem->spa_dcr = spa;
 691                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 692                         idt_idx = nfit_memdev->memdev->interleave_index;
 693                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 694                                 if (nfit_idt->idt->interleave_index != idt_idx)
 695                                         continue;
 696                                 nfit_mem->idt_dcr = nfit_idt->idt;
 697                                 break;
 698                         }
 699                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
 700                 } else {
 701                         /*
 702                          * A single dimm may belong to multiple SPA-PM
 703                          * ranges, record at least one in addition to
 704                          * any SPA-DCR range.
 705                          */
 706                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
 707                 }
 708         }
 709
 710         return 0;
 711 }
 712
 713 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
 714 {
 715         struct nfit_mem *a = container_of(_a, typeof(*a), list);
 716         struct nfit_mem *b = container_of(_b, typeof(*b), list);
 717         u32 handleA, handleB;
 718
 719         handleA = __to_nfit_memdev(a)->device_handle;
 720         handleB = __to_nfit_memdev(b)->device_handle;
 721         if (handleA < handleB)
 722                 return -1;
 723         else if (handleA > handleB)
 724                 return 1;
 725         return 0;
 726 }
 727
 728 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
 729 {
 730         struct nfit_spa *nfit_spa;
 731
 732         /*
 733          * For each SPA-DCR or SPA-PMEM address range find its
 734          * corresponding MEMDEV(s).  From each MEMDEV find the
 735          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
 736          * try to find a SPA-BDW and a corresponding BDW that references
 737          * the DCR.  Throw it all into an nfit_mem object.  Note, that
 738          * BDWs are optional.
 739          */
 740         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 741                 int rc;
 742
 743                 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
 744                 if (rc)
 745                         return rc;
 746         }
 747
 748         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
 749
 750         return 0;
 751 }
 752
 753 static ssize_t revision_show(struct device *dev,
 754                 struct device_attribute *attr, char *buf)
 755 {
 756         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 757         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 758         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 759
 760         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
 761 }
 762 static DEVICE_ATTR_RO(revision);
 763
 764 static struct attribute *acpi_nfit_attributes[] = {
 765         &dev_attr_revision.attr,
 766         NULL,
 767 };
 768
 769 static struct attribute_group acpi_nfit_attribute_group = {
 770         .name = "nfit",
 771         .attrs = acpi_nfit_attributes,
 772 };
 773
 774 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
 775         &nvdimm_bus_attribute_group,
 776         &acpi_nfit_attribute_group,
 777         NULL,
 778 };
 779
 780 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
 781 {
 782         struct nvdimm *nvdimm = to_nvdimm(dev);
 783         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 784
 785         return __to_nfit_memdev(nfit_mem);
 786 }
 787
 788 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
 789 {
 790         struct nvdimm *nvdimm = to_nvdimm(dev);
 791         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 792
 793         return nfit_mem->dcr;
 794 }
 795
 796 static ssize_t handle_show(struct device *dev,
 797                 struct device_attribute *attr, char *buf)
 798 {
 799         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 800
 801         return sprintf(buf, "%#x\n", memdev->device_handle);
 802 }
 803 static DEVICE_ATTR_RO(handle);
 804
 805 static ssize_t phys_id_show(struct device *dev,
 806                 struct device_attribute *attr, char *buf)
 807 {
 808         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 809
 810         return sprintf(buf, "%#x\n", memdev->physical_id);
 811 }
 812 static DEVICE_ATTR_RO(phys_id);
 813
 814 static ssize_t vendor_show(struct device *dev,
 815                 struct device_attribute *attr, char *buf)
 816 {
 817         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 818
 819         return sprintf(buf, "%#x\n", dcr->vendor_id);
 820 }
 821 static DEVICE_ATTR_RO(vendor);
 822
 823 static ssize_t rev_id_show(struct device *dev,
 824                 struct device_attribute *attr, char *buf)
 825 {
 826         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 827
 828         return sprintf(buf, "%#x\n", dcr->revision_id);
 829 }
 830 static DEVICE_ATTR_RO(rev_id);
 831
 832 static ssize_t device_show(struct device *dev,
 833                 struct device_attribute *attr, char *buf)
 834 {
 835         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 836
 837         return sprintf(buf, "%#x\n", dcr->device_id);
 838 }
 839 static DEVICE_ATTR_RO(device);
 840
 841 static ssize_t format_show(struct device *dev,
 842                 struct device_attribute *attr, char *buf)
 843 {
 844         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 845
 846         return sprintf(buf, "%#x\n", dcr->code);
 847 }
 848 static DEVICE_ATTR_RO(format);
 849
 850 static ssize_t serial_show(struct device *dev,
 851                 struct device_attribute *attr, char *buf)
 852 {
 853         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 854
 855         return sprintf(buf, "%#x\n", dcr->serial_number);
 856 }
 857 static DEVICE_ATTR_RO(serial);
 858
 859 static ssize_t flags_show(struct device *dev,
 860                 struct device_attribute *attr, char *buf)
 861 {
 862         u16 flags = to_nfit_memdev(dev)->flags;
 863
 864         return sprintf(buf, "%s%s%s%s%s\n",
 865                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
 866                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
 867                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
 868                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
 869                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
 870 }
 871 static DEVICE_ATTR_RO(flags);
 872
 873 static struct attribute *acpi_nfit_dimm_attributes[] = {
 874         &dev_attr_handle.attr,
 875         &dev_attr_phys_id.attr,
 876         &dev_attr_vendor.attr,
 877         &dev_attr_device.attr,
 878         &dev_attr_format.attr,
 879         &dev_attr_serial.attr,
 880         &dev_attr_rev_id.attr,
 881         &dev_attr_flags.attr,
 882         NULL,
 883 };
 884
 885 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
 886                 struct attribute *a, int n)
 887 {
 888         struct device *dev = container_of(kobj, struct device, kobj);
 889
 890         if (to_nfit_dcr(dev))
 891                 return a->mode;
 892         else
 893                 return 0;
 894 }
 895
 896 static struct attribute_group acpi_nfit_dimm_attribute_group = {
 897         .name = "nfit",
 898         .attrs = acpi_nfit_dimm_attributes,
 899         .is_visible = acpi_nfit_dimm_attr_visible,
 900 };
 901
 902 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
 903         &nvdimm_attribute_group,
 904         &nd_device_attribute_group,
 905         &acpi_nfit_dimm_attribute_group,
 906         NULL,
 907 };
 908
 909 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
 910                 u32 device_handle)
 911 {
 912         struct nfit_mem *nfit_mem;
 913
 914         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 915                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
 916                         return nfit_mem->nvdimm;
 917
 918         return NULL;
 919 }
 920
 921 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
 922                 struct nfit_mem *nfit_mem, u32 device_handle)
 923 {
 924         struct acpi_device *adev, *adev_dimm;
 925         struct device *dev = acpi_desc->dev;
 926         const u8 *uuid = to_nfit_uuid(NFIT_DEV_DIMM);
 927         int i;
 928
 929         nfit_mem->dsm_mask = acpi_desc->dimm_dsm_force_en;
 930         adev = to_acpi_dev(acpi_desc);
 931         if (!adev)
 932                 return 0;
 933
 934         adev_dimm = acpi_find_child_device(adev, device_handle, false);
 935         nfit_mem->adev = adev_dimm;
 936         if (!adev_dimm) {
 937                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
 938                                 device_handle);
 939                 return force_enable_dimms ? 0 : -ENODEV;
 940         }
 941
 942         for (i = ND_CMD_SMART; i <= ND_CMD_VENDOR; i++)
 943                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
 944                         set_bit(i, &nfit_mem->dsm_mask);
 945
 946         return 0;
 947 }
 948
 949 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
 950 {
 951         struct nfit_mem *nfit_mem;
 952         int dimm_count = 0;
 953
 954         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 955                 struct nvdimm *nvdimm;
 956                 unsigned long flags = 0;
 957                 u32 device_handle;
 958                 u16 mem_flags;
 959                 int rc;
 960
 961                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 962                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
 963                 if (nvdimm) {
 964                         dimm_count++;
 965                         continue;
 966                 }
 967
 968                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
 969                         flags |= NDD_ALIASING;
 970
 971                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
 972                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
 973                         flags |= NDD_UNARMED;
 974
 975                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
 976                 if (rc)
 977                         continue;
 978
 979                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
 980                                 acpi_nfit_dimm_attribute_groups,
 981                                 flags, &nfit_mem->dsm_mask);
 982                 if (!nvdimm)
 983                         return -ENOMEM;
 984
 985                 nfit_mem->nvdimm = nvdimm;
 986                 dimm_count++;
 987
 988                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
 989                         continue;
 990
 991                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
 992                                 nvdimm_name(nvdimm),
 993                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
 994                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
 995                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
 996                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
 997
 998         }
 999
1000         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1001 }
1002
1003 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1004 {
1005         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1006         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1007         struct acpi_device *adev;
1008         int i;
1009
1010         nd_desc->dsm_mask = acpi_desc->bus_dsm_force_en;
1011         adev = to_acpi_dev(acpi_desc);
1012         if (!adev)
1013                 return;
1014
1015         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1016                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1017                         set_bit(i, &nd_desc->dsm_mask);
1018 }
1019
1020 static ssize_t range_index_show(struct device *dev,
1021                 struct device_attribute *attr, char *buf)
1022 {
1023         struct nd_region *nd_region = to_nd_region(dev);
1024         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1025
1026         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1027 }
1028 static DEVICE_ATTR_RO(range_index);
1029
1030 static struct attribute *acpi_nfit_region_attributes[] = {
1031         &dev_attr_range_index.attr,
1032         NULL,
1033 };
1034
1035 static struct attribute_group acpi_nfit_region_attribute_group = {
1036         .name = "nfit",
1037         .attrs = acpi_nfit_region_attributes,
1038 };
1039
1040 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1041         &nd_region_attribute_group,
1042         &nd_mapping_attribute_group,
1043         &nd_device_attribute_group,
1044         &nd_numa_attribute_group,
1045         &acpi_nfit_region_attribute_group,
1046         NULL,
1047 };
1048
1049 /* enough info to uniquely specify an interleave set */
1050 struct nfit_set_info {
1051         struct nfit_set_info_map {
1052                 u64 region_offset;
1053                 u32 serial_number;
1054                 u32 pad;
1055         } mapping[0];
1056 };
1057
1058 static size_t sizeof_nfit_set_info(int num_mappings)
1059 {
1060         return sizeof(struct nfit_set_info)
1061                 + num_mappings * sizeof(struct nfit_set_info_map);
1062 }
1063
1064 static int cmp_map(const void *m0, const void *m1)
1065 {
1066         const struct nfit_set_info_map *map0 = m0;
1067         const struct nfit_set_info_map *map1 = m1;
1068
1069         return memcmp(&map0->region_offset, &map1->region_offset,
1070                         sizeof(u64));
1071 }
1072
1073 /* Retrieve the nth entry referencing this spa */
1074 static struct acpi_nfit_memory_map *memdev_from_spa(
1075                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1076 {
1077         struct nfit_memdev *nfit_memdev;
1078
1079         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1080                 if (nfit_memdev->memdev->range_index == range_index)
1081                         if (n-- == 0)
1082                                 return nfit_memdev->memdev;
1083         return NULL;
1084 }
1085
1086 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1087                 struct nd_region_desc *ndr_desc,
1088                 struct acpi_nfit_system_address *spa)
1089 {
1090         int i, spa_type = nfit_spa_type(spa);
1091         struct device *dev = acpi_desc->dev;
1092         struct nd_interleave_set *nd_set;
1093         u16 nr = ndr_desc->num_mappings;
1094         struct nfit_set_info *info;
1095
1096         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1097                 /* pass */;
1098         else
1099                 return 0;
1100
1101         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1102         if (!nd_set)
1103                 return -ENOMEM;
1104
1105         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1106         if (!info)
1107                 return -ENOMEM;
1108         for (i = 0; i < nr; i++) {
1109                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1110                 struct nfit_set_info_map *map = &info->mapping[i];
1111                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1112                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1113                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1114                                 spa->range_index, i);
1115
1116                 if (!memdev || !nfit_mem->dcr) {
1117                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1118                         return -ENODEV;
1119                 }
1120
1121                 map->region_offset = memdev->region_offset;
1122                 map->serial_number = nfit_mem->dcr->serial_number;
1123         }
1124
1125         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1126                         cmp_map, NULL);
1127         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1128         ndr_desc->nd_set = nd_set;
1129         devm_kfree(dev, info);
1130
1131         return 0;
1132 }
1133
1134 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1135 {
1136         struct acpi_nfit_interleave *idt = mmio->idt;
1137         u32 sub_line_offset, line_index, line_offset;
1138         u64 line_no, table_skip_count, table_offset;
1139
1140         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1141         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1142         line_offset = idt->line_offset[line_index]
1143                 * mmio->line_size;
1144         table_offset = table_skip_count * mmio->table_size;
1145
1146         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1147 }
1148
1149 static void wmb_blk(struct nfit_blk *nfit_blk)
1150 {
1151
1152         if (nfit_blk->nvdimm_flush) {
1153                 /*
1154                  * The first wmb() is needed to 'sfence' all previous writes
1155                  * such that they are architecturally visible for the platform
1156                  * buffer flush.  Note that we've already arranged for pmem
1157                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1158                  * final wmb() ensures ordering for the NVDIMM flush write.
1159                  */
1160                 wmb();
1161                 writeq(1, nfit_blk->nvdimm_flush);
1162                 wmb();
1163         } else
1164                 wmb_pmem();
1165 }
1166
1167 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1168 {
1169         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1170         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1171
1172         if (mmio->num_lines)
1173                 offset = to_interleave_offset(offset, mmio);
1174
1175         return readl(mmio->addr.base + offset);
1176 }
1177
1178 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1179                 resource_size_t dpa, unsigned int len, unsigned int write)
1180 {
1181         u64 cmd, offset;
1182         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1183
1184         enum {
1185                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1186                 BCW_LEN_SHIFT = 48,
1187                 BCW_LEN_MASK = (1ULL << 8) - 1,
1188                 BCW_CMD_SHIFT = 56,
1189         };
1190
1191         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1192         len = len >> L1_CACHE_SHIFT;
1193         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1194         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1195
1196         offset = nfit_blk->cmd_offset + mmio->size * bw;
1197         if (mmio->num_lines)
1198                 offset = to_interleave_offset(offset, mmio);
1199
1200         writeq(cmd, mmio->addr.base + offset);
1201         wmb_blk(nfit_blk);
1202
1203         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1204                 readq(mmio->addr.base + offset);
1205 }
1206
1207 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1208                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1209                 unsigned int lane)
1210 {
1211         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1212         unsigned int copied = 0;
1213         u64 base_offset;
1214         int rc;
1215
1216         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1217                 + lane * mmio->size;
1218         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1219         while (len) {
1220                 unsigned int c;
1221                 u64 offset;
1222
1223                 if (mmio->num_lines) {
1224                         u32 line_offset;
1225
1226                         offset = to_interleave_offset(base_offset + copied,
1227                                         mmio);
1228                         div_u64_rem(offset, mmio->line_size, &line_offset);
1229                         c = min_t(size_t, len, mmio->line_size - line_offset);
1230                 } else {
1231                         offset = base_offset + nfit_blk->bdw_offset;
1232                         c = len;
1233                 }
1234
1235                 if (rw)
1236                         memcpy_to_pmem(mmio->addr.aperture + offset,
1237                                         iobuf + copied, c);
1238                 else {
1239                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1240                                 mmio_flush_range((void __force *)
1241                                         mmio->addr.aperture + offset, c);
1242
1243                         memcpy_from_pmem(iobuf + copied,
1244                                         mmio->addr.aperture + offset, c);
1245                 }
1246
1247                 copied += c;
1248                 len -= c;
1249         }
1250
1251         if (rw)
1252                 wmb_blk(nfit_blk);
1253
1254         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1255         return rc;
1256 }
1257
1258 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1259                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1260 {
1261         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1262         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1263         struct nd_region *nd_region = nfit_blk->nd_region;
1264         unsigned int lane, copied = 0;
1265         int rc = 0;
1266
1267         lane = nd_region_acquire_lane(nd_region);
1268         while (len) {
1269                 u64 c = min(len, mmio->size);
1270
1271                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1272                                 iobuf + copied, c, rw, lane);
1273                 if (rc)
1274                         break;
1275
1276                 copied += c;
1277                 len -= c;
1278         }
1279         nd_region_release_lane(nd_region, lane);
1280
1281         return rc;
1282 }
1283
1284 static void nfit_spa_mapping_release(struct kref *kref)
1285 {
1286         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1287         struct acpi_nfit_system_address *spa = spa_map->spa;
1288         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1289
1290         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1291         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1292         if (spa_map->type == SPA_MAP_APERTURE)
1293                 memunmap((void __force *)spa_map->addr.aperture);
1294         else
1295                 iounmap(spa_map->addr.base);
1296         release_mem_region(spa->address, spa->length);
1297         list_del(&spa_map->list);
1298         kfree(spa_map);
1299 }
1300
1301 static struct nfit_spa_mapping *find_spa_mapping(
1302                 struct acpi_nfit_desc *acpi_desc,
1303                 struct acpi_nfit_system_address *spa)
1304 {
1305         struct nfit_spa_mapping *spa_map;
1306
1307         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1308         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1309                 if (spa_map->spa == spa)
1310                         return spa_map;
1311
1312         return NULL;
1313 }
1314
1315 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1316                 struct acpi_nfit_system_address *spa)
1317 {
1318         struct nfit_spa_mapping *spa_map;
1319
1320         mutex_lock(&acpi_desc->spa_map_mutex);
1321         spa_map = find_spa_mapping(acpi_desc, spa);
1322
1323         if (spa_map)
1324                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1325         mutex_unlock(&acpi_desc->spa_map_mutex);
1326 }
1327
1328 static void __iomem *__nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1329                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1330 {
1331         resource_size_t start = spa->address;
1332         resource_size_t n = spa->length;
1333         struct nfit_spa_mapping *spa_map;
1334         struct resource *res;
1335
1336         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1337
1338         spa_map = find_spa_mapping(acpi_desc, spa);
1339         if (spa_map) {
1340                 kref_get(&spa_map->kref);
1341                 return spa_map->addr.base;
1342         }
1343
1344         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1345         if (!spa_map)
1346                 return NULL;
1347
1348         INIT_LIST_HEAD(&spa_map->list);
1349         spa_map->spa = spa;
1350         kref_init(&spa_map->kref);
1351         spa_map->acpi_desc = acpi_desc;
1352
1353         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1354         if (!res)
1355                 goto err_mem;
1356
1357         spa_map->type = type;
1358         if (type == SPA_MAP_APERTURE)
1359                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1360                                                         ARCH_MEMREMAP_PMEM);
1361         else
1362                 spa_map->addr.base = ioremap_nocache(start, n);
1363
1364
1365         if (!spa_map->addr.base)
1366                 goto err_map;
1367
1368         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1369         return spa_map->addr.base;
1370
1371  err_map:
1372         release_mem_region(start, n);
1373  err_mem:
1374         kfree(spa_map);
1375         return NULL;
1376 }
1377
1378 /**
1379  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1380  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1381  * @nfit_spa: spa table to map
1382  * @type: aperture or control region
1383  *
1384  * In the case where block-data-window apertures and
1385  * dimm-control-regions are interleaved they will end up sharing a
1386  * single request_mem_region() + ioremap() for the address range.  In
1387  * the style of devm nfit_spa_map() mappings are automatically dropped
1388  * when all region devices referencing the same mapping are disabled /
1389  * unbound.
1390  */
1391 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1392                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1393 {
1394         void __iomem *iomem;
1395
1396         mutex_lock(&acpi_desc->spa_map_mutex);
1397         iomem = __nfit_spa_map(acpi_desc, spa, type);
1398         mutex_unlock(&acpi_desc->spa_map_mutex);
1399
1400         return iomem;
1401 }
1402
1403 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1404                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1405 {
1406         if (idt) {
1407                 mmio->num_lines = idt->line_count;
1408                 mmio->line_size = idt->line_size;
1409                 if (interleave_ways == 0)
1410                         return -ENXIO;
1411                 mmio->table_size = mmio->num_lines * interleave_ways
1412                         * mmio->line_size;
1413         }
1414
1415         return 0;
1416 }
1417
1418 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1419                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1420 {
1421         struct nd_cmd_dimm_flags flags;
1422         int rc;
1423
1424         memset(&flags, 0, sizeof(flags));
1425         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1426                         sizeof(flags), NULL);
1427
1428         if (rc >= 0 && flags.status == 0)
1429                 nfit_blk->dimm_flags = flags.flags;
1430         else if (rc == -ENOTTY) {
1431                 /* fall back to a conservative default */
1432                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1433                 rc = 0;
1434         } else
1435                 rc = -ENXIO;
1436
1437         return rc;
1438 }
1439
1440 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1441                 struct device *dev)
1442 {
1443         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1444         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1445         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1446         struct nfit_flush *nfit_flush;
1447         struct nfit_blk_mmio *mmio;
1448         struct nfit_blk *nfit_blk;
1449         struct nfit_mem *nfit_mem;
1450         struct nvdimm *nvdimm;
1451         int rc;
1452
1453         nvdimm = nd_blk_region_to_dimm(ndbr);
1454         nfit_mem = nvdimm_provider_data(nvdimm);
1455         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1456                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1457                                 nfit_mem ? "" : " nfit_mem",
1458                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1459                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1460                 return -ENXIO;
1461         }
1462
1463         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1464         if (!nfit_blk)
1465                 return -ENOMEM;
1466         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1467         nfit_blk->nd_region = to_nd_region(dev);
1468
1469         /* map block aperture memory */
1470         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1471         mmio = &nfit_blk->mmio[BDW];
1472         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1473                         SPA_MAP_APERTURE);
1474         if (!mmio->addr.base) {
1475                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1476                                 nvdimm_name(nvdimm));
1477                 return -ENOMEM;
1478         }
1479         mmio->size = nfit_mem->bdw->size;
1480         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1481         mmio->idt = nfit_mem->idt_bdw;
1482         mmio->spa = nfit_mem->spa_bdw;
1483         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1484                         nfit_mem->memdev_bdw->interleave_ways);
1485         if (rc) {
1486                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1487                                 __func__, nvdimm_name(nvdimm));
1488                 return rc;
1489         }
1490
1491         /* map block control memory */
1492         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1493         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1494         mmio = &nfit_blk->mmio[DCR];
1495         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1496                         SPA_MAP_CONTROL);
1497         if (!mmio->addr.base) {
1498                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1499                                 nvdimm_name(nvdimm));
1500                 return -ENOMEM;
1501         }
1502         mmio->size = nfit_mem->dcr->window_size;
1503         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1504         mmio->idt = nfit_mem->idt_dcr;
1505         mmio->spa = nfit_mem->spa_dcr;
1506         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1507                         nfit_mem->memdev_dcr->interleave_ways);
1508         if (rc) {
1509                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1510                                 __func__, nvdimm_name(nvdimm));
1511                 return rc;
1512         }
1513
1514         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1515         if (rc < 0) {
1516                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1517                                 __func__, nvdimm_name(nvdimm));
1518                 return rc;
1519         }
1520
1521         nfit_flush = nfit_mem->nfit_flush;
1522         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1523                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1524                                 nfit_flush->flush->hint_address[0], 8);
1525                 if (!nfit_blk->nvdimm_flush)
1526                         return -ENOMEM;
1527         }
1528
1529         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1530                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1531
1532         if (mmio->line_size == 0)
1533                 return 0;
1534
1535         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1536                         + 8 > mmio->line_size) {
1537                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1538                 return -ENXIO;
1539         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1540                         + 8 > mmio->line_size) {
1541                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1542                 return -ENXIO;
1543         }
1544
1545         return 0;
1546 }
1547
1548 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1549                 struct device *dev)
1550 {
1551         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1552         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1553         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1554         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1555         int i;
1556
1557         if (!nfit_blk)
1558                 return; /* never enabled */
1559
1560         /* auto-free BLK spa mappings */
1561         for (i = 0; i < 2; i++) {
1562                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1563
1564                 if (mmio->addr.base)
1565                         nfit_spa_unmap(acpi_desc, mmio->spa);
1566         }
1567         nd_blk_region_set_provider_data(ndbr, NULL);
1568         /* devm will free nfit_blk */
1569 }
1570
1571 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1572                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1573 {
1574         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1575         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1576         int cmd_rc, rc;
1577
1578         cmd->address = spa->address;
1579         cmd->length = spa->length;
1580         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1581                         sizeof(*cmd), &cmd_rc);
1582         if (rc < 0)
1583                 return rc;
1584         return cmd_rc;
1585 }
1586
1587 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1588 {
1589         int rc;
1590         int cmd_rc;
1591         struct nd_cmd_ars_start ars_start;
1592         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1593         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1594
1595         memset(&ars_start, 0, sizeof(ars_start));
1596         ars_start.address = spa->address;
1597         ars_start.length = spa->length;
1598         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1599                 ars_start.type = ND_ARS_PERSISTENT;
1600         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1601                 ars_start.type = ND_ARS_VOLATILE;
1602         else
1603                 return -ENOTTY;
1604
1605         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1606                         sizeof(ars_start), &cmd_rc);
1607
1608         if (rc < 0)
1609                 return rc;
1610         return cmd_rc;
1611 }
1612
1613 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1614 {
1615         int rc, cmd_rc;
1616         struct nd_cmd_ars_start ars_start;
1617         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1618         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1619
1620         memset(&ars_start, 0, sizeof(ars_start));
1621         ars_start.address = ars_status->restart_address;
1622         ars_start.length = ars_status->restart_length;
1623         ars_start.type = ars_status->type;
1624         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1625                         sizeof(ars_start), &cmd_rc);
1626         if (rc < 0)
1627                 return rc;
1628         return cmd_rc;
1629 }
1630
1631 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1632 {
1633         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1634         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1635         int rc, cmd_rc;
1636
1637         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1638                         acpi_desc->ars_status_size, &cmd_rc);
1639         if (rc < 0)
1640                 return rc;
1641         return cmd_rc;
1642 }
1643
1644 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1645                 struct nd_cmd_ars_status *ars_status)
1646 {
1647         int rc;
1648         u32 i;
1649
1650         for (i = 0; i < ars_status->num_records; i++) {
1651                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1652                                 ars_status->records[i].err_address,
1653                                 ars_status->records[i].length);
1654                 if (rc)
1655                         return rc;
1656         }
1657
1658         return 0;
1659 }
1660
1661 static void acpi_nfit_remove_resource(void *data)
1662 {
1663         struct resource *res = data;
1664
1665         remove_resource(res);
1666 }
1667
1668 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1669                 struct nd_region_desc *ndr_desc)
1670 {
1671         struct resource *res, *nd_res = ndr_desc->res;
1672         int is_pmem, ret;
1673
1674         /* No operation if the region is already registered as PMEM */
1675         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1676                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1677         if (is_pmem == REGION_INTERSECTS)
1678                 return 0;
1679
1680         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1681         if (!res)
1682                 return -ENOMEM;
1683
1684         res->name = "Persistent Memory";
1685         res->start = nd_res->start;
1686         res->end = nd_res->end;
1687         res->flags = IORESOURCE_MEM;
1688         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1689
1690         ret = insert_resource(&iomem_resource, res);
1691         if (ret)
1692                 return ret;
1693
1694         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1695         if (ret) {
1696                 remove_resource(res);
1697                 return ret;
1698         }
1699
1700         return 0;
1701 }
1702
1703 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1704                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1705                 struct acpi_nfit_memory_map *memdev,
1706                 struct nfit_spa *nfit_spa)
1707 {
1708         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1709                         memdev->device_handle);
1710         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1711         struct nd_blk_region_desc *ndbr_desc;
1712         struct nfit_mem *nfit_mem;
1713         int blk_valid = 0;
1714
1715         if (!nvdimm) {
1716                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1717                                 spa->range_index, memdev->device_handle);
1718                 return -ENODEV;
1719         }
1720
1721         nd_mapping->nvdimm = nvdimm;
1722         switch (nfit_spa_type(spa)) {
1723         case NFIT_SPA_PM:
1724         case NFIT_SPA_VOLATILE:
1725                 nd_mapping->start = memdev->address;
1726                 nd_mapping->size = memdev->region_size;
1727                 break;
1728         case NFIT_SPA_DCR:
1729                 nfit_mem = nvdimm_provider_data(nvdimm);
1730                 if (!nfit_mem || !nfit_mem->bdw) {
1731                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1732                                         spa->range_index, nvdimm_name(nvdimm));
1733                 } else {
1734                         nd_mapping->size = nfit_mem->bdw->capacity;
1735                         nd_mapping->start = nfit_mem->bdw->start_address;
1736                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1737                         blk_valid = 1;
1738                 }
1739
1740                 ndr_desc->nd_mapping = nd_mapping;
1741                 ndr_desc->num_mappings = blk_valid;
1742                 ndbr_desc = to_blk_region_desc(ndr_desc);
1743                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1744                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1745                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1746                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1747                                 ndr_desc);
1748                 if (!nfit_spa->nd_region)
1749                         return -ENOMEM;
1750                 break;
1751         }
1752
1753         return 0;
1754 }
1755
1756 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1757                 struct nfit_spa *nfit_spa)
1758 {
1759         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1760         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1761         struct nd_blk_region_desc ndbr_desc;
1762         struct nd_region_desc *ndr_desc;
1763         struct nfit_memdev *nfit_memdev;
1764         struct nvdimm_bus *nvdimm_bus;
1765         struct resource res;
1766         int count = 0, rc;
1767
1768         if (nfit_spa->nd_region)
1769                 return 0;
1770
1771         if (spa->range_index == 0) {
1772                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1773                                 __func__);
1774                 return 0;
1775         }
1776
1777         memset(&res, 0, sizeof(res));
1778         memset(&nd_mappings, 0, sizeof(nd_mappings));
1779         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1780         res.start = spa->address;
1781         res.end = res.start + spa->length - 1;
1782         ndr_desc = &ndbr_desc.ndr_desc;
1783         ndr_desc->res = &res;
1784         ndr_desc->provider_data = nfit_spa;
1785         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1786         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1787                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
1788                                                 spa->proximity_domain);
1789         else
1790                 ndr_desc->numa_node = NUMA_NO_NODE;
1791
1792         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1793                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1794                 struct nd_mapping *nd_mapping;
1795
1796                 if (memdev->range_index != spa->range_index)
1797                         continue;
1798                 if (count >= ND_MAX_MAPPINGS) {
1799                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
1800                                         spa->range_index, ND_MAX_MAPPINGS);
1801                         return -ENXIO;
1802                 }
1803                 nd_mapping = &nd_mappings[count++];
1804                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
1805                                 memdev, nfit_spa);
1806                 if (rc)
1807                         goto out;
1808         }
1809
1810         ndr_desc->nd_mapping = nd_mappings;
1811         ndr_desc->num_mappings = count;
1812         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
1813         if (rc)
1814                 goto out;
1815
1816         nvdimm_bus = acpi_desc->nvdimm_bus;
1817         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
1818                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
1819                 if (rc) {
1820                         dev_warn(acpi_desc->dev,
1821                                 "failed to insert pmem resource to iomem: %d\n",
1822                                 rc);
1823                         goto out;
1824                 }
1825
1826                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
1827                                 ndr_desc);
1828                 if (!nfit_spa->nd_region)
1829                         rc = -ENOMEM;
1830         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
1831                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
1832                                 ndr_desc);
1833                 if (!nfit_spa->nd_region)
1834                         rc = -ENOMEM;
1835         }
1836
1837  out:
1838         if (rc)
1839                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
1840                                 nfit_spa->spa->range_index);
1841         return rc;
1842 }
1843
1844 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
1845                 u32 max_ars)
1846 {
1847         struct device *dev = acpi_desc->dev;
1848         struct nd_cmd_ars_status *ars_status;
1849
1850         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
1851                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
1852                 return 0;
1853         }
1854
1855         if (acpi_desc->ars_status)
1856                 devm_kfree(dev, acpi_desc->ars_status);
1857         acpi_desc->ars_status = NULL;
1858         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
1859         if (!ars_status)
1860                 return -ENOMEM;
1861         acpi_desc->ars_status = ars_status;
1862         acpi_desc->ars_status_size = max_ars;
1863         return 0;
1864 }
1865
1866 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
1867                 struct nfit_spa *nfit_spa)
1868 {
1869         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1870         int rc;
1871
1872         if (!nfit_spa->max_ars) {
1873                 struct nd_cmd_ars_cap ars_cap;
1874
1875                 memset(&ars_cap, 0, sizeof(ars_cap));
1876                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
1877                 if (rc < 0)
1878                         return rc;
1879                 nfit_spa->max_ars = ars_cap.max_ars_out;
1880                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
1881                 /* check that the supported scrub types match the spa type */
1882                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
1883                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
1884                         return -ENOTTY;
1885                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
1886                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
1887                         return -ENOTTY;
1888         }
1889
1890         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
1891                 return -ENOMEM;
1892
1893         rc = ars_get_status(acpi_desc);
1894         if (rc < 0 && rc != -ENOSPC)
1895                 return rc;
1896
1897         if (ars_status_process_records(acpi_desc->nvdimm_bus,
1898                                 acpi_desc->ars_status))
1899                 return -ENOMEM;
1900
1901         return 0;
1902 }
1903
1904 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
1905                 struct nfit_spa *nfit_spa)
1906 {
1907         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1908         unsigned int overflow_retry = scrub_overflow_abort;
1909         u64 init_ars_start = 0, init_ars_len = 0;
1910         struct device *dev = acpi_desc->dev;
1911         unsigned int tmo = scrub_timeout;
1912         int rc;
1913
1914         if (nfit_spa->ars_done || !nfit_spa->nd_region)
1915                 return;
1916
1917         rc = ars_start(acpi_desc, nfit_spa);
1918         /*
1919          * If we timed out the initial scan we'll still be busy here,
1920          * and will wait another timeout before giving up permanently.
1921          */
1922         if (rc < 0 && rc != -EBUSY)
1923                 return;
1924
1925         do {
1926                 u64 ars_start, ars_len;
1927
1928                 if (acpi_desc->cancel)
1929                         break;
1930                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
1931                 if (rc == -ENOTTY)
1932                         break;
1933                 if (rc == -EBUSY && !tmo) {
1934                         dev_warn(dev, "range %d ars timeout, aborting\n",
1935                                         spa->range_index);
1936                         break;
1937                 }
1938
1939                 if (rc == -EBUSY) {
1940                         /*
1941                          * Note, entries may be appended to the list
1942                          * while the lock is dropped, but the workqueue
1943                          * being active prevents entries being deleted /
1944                          * freed.
1945                          */
1946                         mutex_unlock(&acpi_desc->init_mutex);
1947                         ssleep(1);
1948                         tmo--;
1949                         mutex_lock(&acpi_desc->init_mutex);
1950                         continue;
1951                 }
1952
1953                 /* we got some results, but there are more pending... */
1954                 if (rc == -ENOSPC && overflow_retry--) {
1955                         if (!init_ars_len) {
1956                                 init_ars_len = acpi_desc->ars_status->length;
1957                                 init_ars_start = acpi_desc->ars_status->address;
1958                         }
1959                         rc = ars_continue(acpi_desc);
1960                 }
1961
1962                 if (rc < 0) {
1963                         dev_warn(dev, "range %d ars continuation failed\n",
1964                                         spa->range_index);
1965                         break;
1966                 }
1967
1968                 if (init_ars_len) {
1969                         ars_start = init_ars_start;
1970                         ars_len = init_ars_len;
1971                 } else {
1972                         ars_start = acpi_desc->ars_status->address;
1973                         ars_len = acpi_desc->ars_status->length;
1974                 }
1975                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
1976                                 spa->range_index, ars_start, ars_len);
1977                 /* notify the region about new poison entries */
1978                 nvdimm_region_notify(nfit_spa->nd_region,
1979                                 NVDIMM_REVALIDATE_POISON);
1980                 break;
1981         } while (1);
1982 }
1983
1984 static void acpi_nfit_scrub(struct work_struct *work)
1985 {
1986         struct device *dev;
1987         u64 init_scrub_length = 0;
1988         struct nfit_spa *nfit_spa;
1989         u64 init_scrub_address = 0;
1990         bool init_ars_done = false;
1991         struct acpi_nfit_desc *acpi_desc;
1992         unsigned int tmo = scrub_timeout;
1993         unsigned int overflow_retry = scrub_overflow_abort;
1994
1995         acpi_desc = container_of(work, typeof(*acpi_desc), work);
1996         dev = acpi_desc->dev;
1997
1998         /*
1999          * We scrub in 2 phases.  The first phase waits for any platform
2000          * firmware initiated scrubs to complete and then we go search for the
2001          * affected spa regions to mark them scanned.  In the second phase we
2002          * initiate a directed scrub for every range that was not scrubbed in
2003          * phase 1.
2004          */
2005
2006         /* process platform firmware initiated scrubs */
2007  retry:
2008         mutex_lock(&acpi_desc->init_mutex);
2009         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2010                 struct nd_cmd_ars_status *ars_status;
2011                 struct acpi_nfit_system_address *spa;
2012                 u64 ars_start, ars_len;
2013                 int rc;
2014
2015                 if (acpi_desc->cancel)
2016                         break;
2017
2018                 if (nfit_spa->nd_region)
2019                         continue;
2020
2021                 if (init_ars_done) {
2022                         /*
2023                          * No need to re-query, we're now just
2024                          * reconciling all the ranges covered by the
2025                          * initial scrub
2026                          */
2027                         rc = 0;
2028                 } else
2029                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2030
2031                 if (rc == -ENOTTY) {
2032                         /* no ars capability, just register spa and move on */
2033                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2034                         continue;
2035                 }
2036
2037                 if (rc == -EBUSY && !tmo) {
2038                         /* fallthrough to directed scrub in phase 2 */
2039                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2040                         break;
2041                 } else if (rc == -EBUSY) {
2042                         mutex_unlock(&acpi_desc->init_mutex);
2043                         ssleep(1);
2044                         tmo--;
2045                         goto retry;
2046                 }
2047
2048                 /* we got some results, but there are more pending... */
2049                 if (rc == -ENOSPC && overflow_retry--) {
2050                         ars_status = acpi_desc->ars_status;
2051                         /*
2052                          * Record the original scrub range, so that we
2053                          * can recall all the ranges impacted by the
2054                          * initial scrub.
2055                          */
2056                         if (!init_scrub_length) {
2057                                 init_scrub_length = ars_status->length;
2058                                 init_scrub_address = ars_status->address;
2059                         }
2060                         rc = ars_continue(acpi_desc);
2061                         if (rc == 0) {
2062                                 mutex_unlock(&acpi_desc->init_mutex);
2063                                 goto retry;
2064                         }
2065                 }
2066
2067                 if (rc < 0) {
2068                         /*
2069                          * Initial scrub failed, we'll give it one more
2070                          * try below...
2071                          */
2072                         break;
2073                 }
2074
2075                 /* We got some final results, record completed ranges */
2076                 ars_status = acpi_desc->ars_status;
2077                 if (init_scrub_length) {
2078                         ars_start = init_scrub_address;
2079                         ars_len = ars_start + init_scrub_length;
2080                 } else {
2081                         ars_start = ars_status->address;
2082                         ars_len = ars_status->length;
2083                 }
2084                 spa = nfit_spa->spa;
2085
2086                 if (!init_ars_done) {
2087                         init_ars_done = true;
2088                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2089                                         ars_start, ars_len);
2090                 }
2091                 if (ars_start <= spa->address && ars_start + ars_len
2092                                 >= spa->address + spa->length)
2093                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2094         }
2095
2096         /*
2097          * For all the ranges not covered by an initial scrub we still
2098          * want to see if there are errors, but it's ok to discover them
2099          * asynchronously.
2100          */
2101         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2102                 /*
2103                  * Flag all the ranges that still need scrubbing, but
2104                  * register them now to make data available.
2105                  */
2106                 if (nfit_spa->nd_region)
2107                         nfit_spa->ars_done = 1;
2108                 else
2109                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2110         }
2111
2112         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2113                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2114         mutex_unlock(&acpi_desc->init_mutex);
2115 }
2116
2117 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2118 {
2119         struct nfit_spa *nfit_spa;
2120         int rc;
2121
2122         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2123                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2124                         /* BLK regions don't need to wait for ars results */
2125                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2126                         if (rc)
2127                                 return rc;
2128                 }
2129
2130         queue_work(nfit_wq, &acpi_desc->work);
2131         return 0;
2132 }
2133
2134 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2135                 struct nfit_table_prev *prev)
2136 {
2137         struct device *dev = acpi_desc->dev;
2138
2139         if (!list_empty(&prev->spas) ||
2140                         !list_empty(&prev->memdevs) ||
2141                         !list_empty(&prev->dcrs) ||
2142                         !list_empty(&prev->bdws) ||
2143                         !list_empty(&prev->idts) ||
2144                         !list_empty(&prev->flushes)) {
2145                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2146                 return -ENXIO;
2147         }
2148         return 0;
2149 }
2150
2151 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2152 {
2153         struct device *dev = acpi_desc->dev;
2154         struct nfit_table_prev prev;
2155         const void *end;
2156         u8 *data;
2157         int rc;
2158
2159         mutex_lock(&acpi_desc->init_mutex);
2160
2161         INIT_LIST_HEAD(&prev.spas);
2162         INIT_LIST_HEAD(&prev.memdevs);
2163         INIT_LIST_HEAD(&prev.dcrs);
2164         INIT_LIST_HEAD(&prev.bdws);
2165         INIT_LIST_HEAD(&prev.idts);
2166         INIT_LIST_HEAD(&prev.flushes);
2167
2168         list_cut_position(&prev.spas, &acpi_desc->spas,
2169                                 acpi_desc->spas.prev);
2170         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2171                                 acpi_desc->memdevs.prev);
2172         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2173                                 acpi_desc->dcrs.prev);
2174         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2175                                 acpi_desc->bdws.prev);
2176         list_cut_position(&prev.idts, &acpi_desc->idts,
2177                                 acpi_desc->idts.prev);
2178         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2179                                 acpi_desc->flushes.prev);
2180
2181         data = (u8 *) acpi_desc->nfit;
2182         end = data + sz;
2183         while (!IS_ERR_OR_NULL(data))
2184                 data = add_table(acpi_desc, &prev, data, end);
2185
2186         if (IS_ERR(data)) {
2187                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2188                                 PTR_ERR(data));
2189                 rc = PTR_ERR(data);
2190                 goto out_unlock;
2191         }
2192
2193         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2194         if (rc)
2195                 goto out_unlock;
2196
2197         if (nfit_mem_init(acpi_desc) != 0) {
2198                 rc = -ENOMEM;
2199                 goto out_unlock;
2200         }
2201
2202         acpi_nfit_init_dsms(acpi_desc);
2203
2204         rc = acpi_nfit_register_dimms(acpi_desc);
2205         if (rc)
2206                 goto out_unlock;
2207
2208         rc = acpi_nfit_register_regions(acpi_desc);
2209
2210  out_unlock:
2211         mutex_unlock(&acpi_desc->init_mutex);
2212         return rc;
2213 }
2214 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2215
2216 struct acpi_nfit_flush_work {
2217         struct work_struct work;
2218         struct completion cmp;
2219 };
2220
2221 static void flush_probe(struct work_struct *work)
2222 {
2223         struct acpi_nfit_flush_work *flush;
2224
2225         flush = container_of(work, typeof(*flush), work);
2226         complete(&flush->cmp);
2227 }
2228
2229 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2230 {
2231         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2232         struct device *dev = acpi_desc->dev;
2233         struct acpi_nfit_flush_work flush;
2234
2235         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2236         device_lock(dev);
2237         device_unlock(dev);
2238
2239         /*
2240          * Scrub work could take 10s of seconds, userspace may give up so we
2241          * need to be interruptible while waiting.
2242          */
2243         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2244         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2245         queue_work(nfit_wq, &flush.work);
2246         return wait_for_completion_interruptible(&flush.cmp);
2247 }
2248
2249 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2250                 struct nvdimm *nvdimm, unsigned int cmd)
2251 {
2252         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2253
2254         if (nvdimm)
2255                 return 0;
2256         if (cmd != ND_CMD_ARS_START)
2257                 return 0;
2258
2259         /*
2260          * The kernel and userspace may race to initiate a scrub, but
2261          * the scrub thread is prepared to lose that initial race.  It
2262          * just needs guarantees that any ars it initiates are not
2263          * interrupted by any intervening start reqeusts from userspace.
2264          */
2265         if (work_busy(&acpi_desc->work))
2266                 return -EBUSY;
2267
2268         return 0;
2269 }
2270
2271 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2272 {
2273         struct nvdimm_bus_descriptor *nd_desc;
2274
2275         dev_set_drvdata(dev, acpi_desc);
2276         acpi_desc->dev = dev;
2277         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2278         nd_desc = &acpi_desc->nd_desc;
2279         nd_desc->provider_name = "ACPI.NFIT";
2280         nd_desc->ndctl = acpi_nfit_ctl;
2281         nd_desc->flush_probe = acpi_nfit_flush_probe;
2282         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2283         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2284
2285         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2286         INIT_LIST_HEAD(&acpi_desc->spas);
2287         INIT_LIST_HEAD(&acpi_desc->dcrs);
2288         INIT_LIST_HEAD(&acpi_desc->bdws);
2289         INIT_LIST_HEAD(&acpi_desc->idts);
2290         INIT_LIST_HEAD(&acpi_desc->flushes);
2291         INIT_LIST_HEAD(&acpi_desc->memdevs);
2292         INIT_LIST_HEAD(&acpi_desc->dimms);
2293         mutex_init(&acpi_desc->spa_map_mutex);
2294         mutex_init(&acpi_desc->init_mutex);
2295         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2296 }
2297 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2298
2299 static int acpi_nfit_add(struct acpi_device *adev)
2300 {
2301         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2302         struct acpi_nfit_desc *acpi_desc;
2303         struct device *dev = &adev->dev;
2304         struct acpi_table_header *tbl;
2305         acpi_status status = AE_OK;
2306         acpi_size sz;
2307         int rc;
2308
2309         status = acpi_get_table_with_size("NFIT", 0, &tbl, &sz);
2310         if (ACPI_FAILURE(status)) {
2311                 /* This is ok, we could have an nvdimm hotplugged later */
2312                 dev_dbg(dev, "failed to find NFIT at startup\n");
2313                 return 0;
2314         }
2315
2316         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2317         if (!acpi_desc)
2318                 return -ENOMEM;
2319         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2320         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2321         if (!acpi_desc->nvdimm_bus)
2322                 return -ENOMEM;
2323
2324         /*
2325          * Save the acpi header for later and then skip it,
2326          * making nfit point to the first nfit table header.
2327          */
2328         acpi_desc->acpi_header = *tbl;
2329         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2330         sz -= sizeof(struct acpi_table_nfit);
2331
2332         /* Evaluate _FIT and override with that if present */
2333         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2334         if (ACPI_SUCCESS(status) && buf.length > 0) {
2335                 union acpi_object *obj;
2336                 /*
2337                  * Adjust for the acpi_object header of the _FIT
2338                  */
2339                 obj = buf.pointer;
2340                 if (obj->type == ACPI_TYPE_BUFFER) {
2341                         acpi_desc->nfit =
2342                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2343                         sz = obj->buffer.length;
2344                 } else
2345                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2346                                  __func__, (int) obj->type);
2347         }
2348
2349         rc = acpi_nfit_init(acpi_desc, sz);
2350         if (rc) {
2351                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2352                 return rc;
2353         }
2354         return 0;
2355 }
2356
2357 static int acpi_nfit_remove(struct acpi_device *adev)
2358 {
2359         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2360
2361         acpi_desc->cancel = 1;
2362         flush_workqueue(nfit_wq);
2363         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2364         return 0;
2365 }
2366
2367 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2368 {
2369         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2370         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2371         struct acpi_nfit_header *nfit_saved;
2372         union acpi_object *obj;
2373         struct device *dev = &adev->dev;
2374         acpi_status status;
2375         int ret;
2376
2377         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2378
2379         device_lock(dev);
2380         if (!dev->driver) {
2381                 /* dev->driver may be null if we're being removed */
2382                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2383                 goto out_unlock;
2384         }
2385
2386         if (!acpi_desc) {
2387                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2388                 if (!acpi_desc)
2389                         goto out_unlock;
2390                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2391                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2392                 if (!acpi_desc->nvdimm_bus)
2393                         goto out_unlock;
2394         } else {
2395                 /*
2396                  * Finish previous registration before considering new
2397                  * regions.
2398                  */
2399                 flush_workqueue(nfit_wq);
2400         }
2401
2402         /* Evaluate _FIT */
2403         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2404         if (ACPI_FAILURE(status)) {
2405                 dev_err(dev, "failed to evaluate _FIT\n");
2406                 goto out_unlock;
2407         }
2408
2409         nfit_saved = acpi_desc->nfit;
2410         obj = buf.pointer;
2411         if (obj->type == ACPI_TYPE_BUFFER) {
2412                 acpi_desc->nfit =
2413                         (struct acpi_nfit_header *)obj->buffer.pointer;
2414                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2415                 if (ret) {
2416                         /* Merge failed, restore old nfit, and exit */
2417                         acpi_desc->nfit = nfit_saved;
2418                         dev_err(dev, "failed to merge updated NFIT\n");
2419                 }
2420         } else {
2421                 /* Bad _FIT, restore old nfit */
2422                 dev_err(dev, "Invalid _FIT\n");
2423         }
2424         kfree(buf.pointer);
2425
2426  out_unlock:
2427         device_unlock(dev);
2428 }
2429
2430 static const struct acpi_device_id acpi_nfit_ids[] = {
2431         { "ACPI0012", 0 },
2432         { "", 0 },
2433 };
2434 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2435
2436 static struct acpi_driver acpi_nfit_driver = {
2437         .name = KBUILD_MODNAME,
2438         .ids = acpi_nfit_ids,
2439         .ops = {
2440                 .add = acpi_nfit_add,
2441                 .remove = acpi_nfit_remove,
2442                 .notify = acpi_nfit_notify,
2443         },
2444 };
2445
2446 static __init int nfit_init(void)
2447 {
2448         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2449         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2450         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2451         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2452         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2453         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2454         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2455
2456         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2457         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2458         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2459         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2460         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2461         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2462         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2463         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2464         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2465         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2466
2467         nfit_wq = create_singlethread_workqueue("nfit");
2468         if (!nfit_wq)
2469                 return -ENOMEM;
2470
2471         return acpi_bus_register_driver(&acpi_nfit_driver);
2472 }
2473
2474 static __exit void nfit_exit(void)
2475 {
2476         acpi_bus_unregister_driver(&acpi_nfit_driver);
2477         destroy_workqueue(nfit_wq);
2478 }
2479
2480 module_init(nfit_init);
2481 module_exit(nfit_exit);
2482 MODULE_LICENSE("GPL v2");
2483 MODULE_AUTHOR("Intel Corporation");