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 bool disable_vendor_specific;
  49 module_param(disable_vendor_specific, bool, S_IRUGO);
  50 MODULE_PARM_DESC(disable_vendor_specific,
  51                 "Limit commands to the publicly specified set\n");
  52
  53 static struct workqueue_struct *nfit_wq;
  54
  55 struct nfit_table_prev {
  56         struct list_head spas;
  57         struct list_head memdevs;
  58         struct list_head dcrs;
  59         struct list_head bdws;
  60         struct list_head idts;
  61         struct list_head flushes;
  62 };
  63
  64 static u8 nfit_uuid[NFIT_UUID_MAX][16];
  65
  66 const u8 *to_nfit_uuid(enum nfit_uuids id)
  67 {
  68         return nfit_uuid[id];
  69 }
  70 EXPORT_SYMBOL(to_nfit_uuid);
  71
  72 static struct acpi_nfit_desc *to_acpi_nfit_desc(
  73                 struct nvdimm_bus_descriptor *nd_desc)
  74 {
  75         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
  76 }
  77
  78 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  79 {
  80         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  81
  82         /*
  83          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  84          * acpi_device.
  85          */
  86         if (!nd_desc->provider_name
  87                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
  88                 return NULL;
  89
  90         return to_acpi_device(acpi_desc->dev);
  91 }
  92
  93 static int xlat_status(void *buf, unsigned int cmd)
  94 {
  95         struct nd_cmd_clear_error *clear_err;
  96         struct nd_cmd_ars_status *ars_status;
  97         struct nd_cmd_ars_start *ars_start;
  98         struct nd_cmd_ars_cap *ars_cap;
  99         u16 flags;
 100
 101         switch (cmd) {
 102         case ND_CMD_ARS_CAP:
 103                 ars_cap = buf;
 104                 if ((ars_cap->status & 0xffff) == NFIT_ARS_CAP_NONE)
 105                         return -ENOTTY;
 106
 107                 /* Command failed */
 108                 if (ars_cap->status & 0xffff)
 109                         return -EIO;
 110
 111                 /* No supported scan types for this range */
 112                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 113                 if ((ars_cap->status >> 16 & flags) == 0)
 114                         return -ENOTTY;
 115                 break;
 116         case ND_CMD_ARS_START:
 117                 ars_start = buf;
 118                 /* ARS is in progress */
 119                 if ((ars_start->status & 0xffff) == NFIT_ARS_START_BUSY)
 120                         return -EBUSY;
 121
 122                 /* Command failed */
 123                 if (ars_start->status & 0xffff)
 124                         return -EIO;
 125                 break;
 126         case ND_CMD_ARS_STATUS:
 127                 ars_status = buf;
 128                 /* Command failed */
 129                 if (ars_status->status & 0xffff)
 130                         return -EIO;
 131                 /* Check extended status (Upper two bytes) */
 132                 if (ars_status->status == NFIT_ARS_STATUS_DONE)
 133                         return 0;
 134
 135                 /* ARS is in progress */
 136                 if (ars_status->status == NFIT_ARS_STATUS_BUSY)
 137                         return -EBUSY;
 138
 139                 /* No ARS performed for the current boot */
 140                 if (ars_status->status == NFIT_ARS_STATUS_NONE)
 141                         return -EAGAIN;
 142
 143                 /*
 144                  * ARS interrupted, either we overflowed or some other
 145                  * agent wants the scan to stop.  If we didn't overflow
 146                  * then just continue with the returned results.
 147                  */
 148                 if (ars_status->status == NFIT_ARS_STATUS_INTR) {
 149                         if (ars_status->flags & NFIT_ARS_F_OVERFLOW)
 150                                 return -ENOSPC;
 151                         return 0;
 152                 }
 153
 154                 /* Unknown status */
 155                 if (ars_status->status >> 16)
 156                         return -EIO;
 157                 break;
 158         case ND_CMD_CLEAR_ERROR:
 159                 clear_err = buf;
 160                 if (clear_err->status & 0xffff)
 161                         return -EIO;
 162                 if (!clear_err->cleared)
 163                         return -EIO;
 164                 if (clear_err->length > clear_err->cleared)
 165                         return clear_err->cleared;
 166                 break;
 167         default:
 168                 break;
 169         }
 170
 171         return 0;
 172 }
 173
 174 static int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc,
 175                 struct nvdimm *nvdimm, unsigned int cmd, void *buf,
 176                 unsigned int buf_len, int *cmd_rc)
 177 {
 178         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
 179         union acpi_object in_obj, in_buf, *out_obj;
 180         const struct nd_cmd_desc *desc = NULL;
 181         struct device *dev = acpi_desc->dev;
 182         struct nd_cmd_pkg *call_pkg = NULL;
 183         const char *cmd_name, *dimm_name;
 184         unsigned long cmd_mask, dsm_mask;
 185         acpi_handle handle;
 186         unsigned int func;
 187         const u8 *uuid;
 188         u32 offset;
 189         int rc, i;
 190
 191         func = cmd;
 192         if (cmd == ND_CMD_CALL) {
 193                 call_pkg = buf;
 194                 func = call_pkg->nd_command;
 195         }
 196
 197         if (nvdimm) {
 198                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 199                 struct acpi_device *adev = nfit_mem->adev;
 200
 201                 if (!adev)
 202                         return -ENOTTY;
 203                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
 204                         return -ENOTTY;
 205
 206                 dimm_name = nvdimm_name(nvdimm);
 207                 cmd_name = nvdimm_cmd_name(cmd);
 208                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 209                 dsm_mask = nfit_mem->dsm_mask;
 210                 desc = nd_cmd_dimm_desc(cmd);
 211                 uuid = to_nfit_uuid(nfit_mem->family);
 212                 handle = adev->handle;
 213         } else {
 214                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 215
 216                 cmd_name = nvdimm_bus_cmd_name(cmd);
 217                 cmd_mask = nd_desc->cmd_mask;
 218                 dsm_mask = cmd_mask;
 219                 desc = nd_cmd_bus_desc(cmd);
 220                 uuid = to_nfit_uuid(NFIT_DEV_BUS);
 221                 handle = adev->handle;
 222                 dimm_name = "bus";
 223         }
 224
 225         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 226                 return -ENOTTY;
 227
 228         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
 229                 return -ENOTTY;
 230
 231         in_obj.type = ACPI_TYPE_PACKAGE;
 232         in_obj.package.count = 1;
 233         in_obj.package.elements = &in_buf;
 234         in_buf.type = ACPI_TYPE_BUFFER;
 235         in_buf.buffer.pointer = buf;
 236         in_buf.buffer.length = 0;
 237
 238         /* libnvdimm has already validated the input envelope */
 239         for (i = 0; i < desc->in_num; i++)
 240                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 241                                 i, buf);
 242
 243         if (call_pkg) {
 244                 /* skip over package wrapper */
 245                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 246                 in_buf.buffer.length = call_pkg->nd_size_in;
 247         }
 248
 249         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
 250                 dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
 251                                 __func__, dimm_name, cmd, func,
 252                                 in_buf.buffer.length);
 253                 print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 254                         in_buf.buffer.pointer,
 255                         min_t(u32, 256, in_buf.buffer.length), true);
 256         }
 257
 258         out_obj = acpi_evaluate_dsm(handle, uuid, 1, func, &in_obj);
 259         if (!out_obj) {
 260                 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
 261                                 cmd_name);
 262                 return -EINVAL;
 263         }
 264
 265         if (call_pkg) {
 266                 call_pkg->nd_fw_size = out_obj->buffer.length;
 267                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 268                         out_obj->buffer.pointer,
 269                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 270
 271                 ACPI_FREE(out_obj);
 272                 /*
 273                  * Need to support FW function w/o known size in advance.
 274                  * Caller can determine required size based upon nd_fw_size.
 275                  * If we return an error (like elsewhere) then caller wouldn't
 276                  * be able to rely upon data returned to make calculation.
 277                  */
 278                 return 0;
 279         }
 280
 281         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 282                 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
 283                                 __func__, dimm_name, cmd_name, out_obj->type);
 284                 rc = -EINVAL;
 285                 goto out;
 286         }
 287
 288         if (IS_ENABLED(CONFIG_ACPI_NFIT_DEBUG)) {
 289                 dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__,
 290                                 dimm_name, cmd_name, out_obj->buffer.length);
 291                 print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4,
 292                                 4, out_obj->buffer.pointer, min_t(u32, 128,
 293                                         out_obj->buffer.length), true);
 294         }
 295
 296         for (i = 0, offset = 0; i < desc->out_num; i++) {
 297                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 298                                 (u32 *) out_obj->buffer.pointer);
 299
 300                 if (offset + out_size > out_obj->buffer.length) {
 301                         dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
 302                                         __func__, dimm_name, cmd_name, i);
 303                         break;
 304                 }
 305
 306                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 307                         dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
 308                                         __func__, dimm_name, cmd_name, i);
 309                         rc = -ENXIO;
 310                         goto out;
 311                 }
 312                 memcpy(buf + in_buf.buffer.length + offset,
 313                                 out_obj->buffer.pointer + offset, out_size);
 314                 offset += out_size;
 315         }
 316         if (offset + in_buf.buffer.length < buf_len) {
 317                 if (i >= 1) {
 318                         /*
 319                          * status valid, return the number of bytes left
 320                          * unfilled in the output buffer
 321                          */
 322                         rc = buf_len - offset - in_buf.buffer.length;
 323                         if (cmd_rc)
 324                                 *cmd_rc = xlat_status(buf, cmd);
 325                 } else {
 326                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 327                                         __func__, dimm_name, cmd_name, buf_len,
 328                                         offset);
 329                         rc = -ENXIO;
 330                 }
 331         } else {
 332                 rc = 0;
 333                 if (cmd_rc)
 334                         *cmd_rc = xlat_status(buf, cmd);
 335         }
 336
 337  out:
 338         ACPI_FREE(out_obj);
 339
 340         return rc;
 341 }
 342
 343 static const char *spa_type_name(u16 type)
 344 {
 345         static const char *to_name[] = {
 346                 [NFIT_SPA_VOLATILE] = "volatile",
 347                 [NFIT_SPA_PM] = "pmem",
 348                 [NFIT_SPA_DCR] = "dimm-control-region",
 349                 [NFIT_SPA_BDW] = "block-data-window",
 350                 [NFIT_SPA_VDISK] = "volatile-disk",
 351                 [NFIT_SPA_VCD] = "volatile-cd",
 352                 [NFIT_SPA_PDISK] = "persistent-disk",
 353                 [NFIT_SPA_PCD] = "persistent-cd",
 354
 355         };
 356
 357         if (type > NFIT_SPA_PCD)
 358                 return "unknown";
 359
 360         return to_name[type];
 361 }
 362
 363 static int nfit_spa_type(struct acpi_nfit_system_address *spa)
 364 {
 365         int i;
 366
 367         for (i = 0; i < NFIT_UUID_MAX; i++)
 368                 if (memcmp(to_nfit_uuid(i), spa->range_guid, 16) == 0)
 369                         return i;
 370         return -1;
 371 }
 372
 373 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 374                 struct nfit_table_prev *prev,
 375                 struct acpi_nfit_system_address *spa)
 376 {
 377         size_t length = min_t(size_t, sizeof(*spa), spa->header.length);
 378         struct device *dev = acpi_desc->dev;
 379         struct nfit_spa *nfit_spa;
 380
 381         list_for_each_entry(nfit_spa, &prev->spas, list) {
 382                 if (memcmp(nfit_spa->spa, spa, length) == 0) {
 383                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 384                         return true;
 385                 }
 386         }
 387
 388         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa), GFP_KERNEL);
 389         if (!nfit_spa)
 390                 return false;
 391         INIT_LIST_HEAD(&nfit_spa->list);
 392         nfit_spa->spa = spa;
 393         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 394         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
 395                         spa->range_index,
 396                         spa_type_name(nfit_spa_type(spa)));
 397         return true;
 398 }
 399
 400 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 401                 struct nfit_table_prev *prev,
 402                 struct acpi_nfit_memory_map *memdev)
 403 {
 404         size_t length = min_t(size_t, sizeof(*memdev), memdev->header.length);
 405         struct device *dev = acpi_desc->dev;
 406         struct nfit_memdev *nfit_memdev;
 407
 408         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 409                 if (memcmp(nfit_memdev->memdev, memdev, length) == 0) {
 410                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 411                         return true;
 412                 }
 413
 414         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev), GFP_KERNEL);
 415         if (!nfit_memdev)
 416                 return false;
 417         INIT_LIST_HEAD(&nfit_memdev->list);
 418         nfit_memdev->memdev = memdev;
 419         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 420         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d\n",
 421                         __func__, memdev->device_handle, memdev->range_index,
 422                         memdev->region_index);
 423         return true;
 424 }
 425
 426 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 427                 struct nfit_table_prev *prev,
 428                 struct acpi_nfit_control_region *dcr)
 429 {
 430         size_t length = min_t(size_t, sizeof(*dcr), dcr->header.length);
 431         struct device *dev = acpi_desc->dev;
 432         struct nfit_dcr *nfit_dcr;
 433
 434         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 435                 if (memcmp(nfit_dcr->dcr, dcr, length) == 0) {
 436                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 437                         return true;
 438                 }
 439
 440         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr), GFP_KERNEL);
 441         if (!nfit_dcr)
 442                 return false;
 443         INIT_LIST_HEAD(&nfit_dcr->list);
 444         nfit_dcr->dcr = dcr;
 445         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 446         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
 447                         dcr->region_index, dcr->windows);
 448         return true;
 449 }
 450
 451 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 452                 struct nfit_table_prev *prev,
 453                 struct acpi_nfit_data_region *bdw)
 454 {
 455         size_t length = min_t(size_t, sizeof(*bdw), bdw->header.length);
 456         struct device *dev = acpi_desc->dev;
 457         struct nfit_bdw *nfit_bdw;
 458
 459         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 460                 if (memcmp(nfit_bdw->bdw, bdw, length) == 0) {
 461                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 462                         return true;
 463                 }
 464
 465         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw), GFP_KERNEL);
 466         if (!nfit_bdw)
 467                 return false;
 468         INIT_LIST_HEAD(&nfit_bdw->list);
 469         nfit_bdw->bdw = bdw;
 470         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 471         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
 472                         bdw->region_index, bdw->windows);
 473         return true;
 474 }
 475
 476 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 477                 struct nfit_table_prev *prev,
 478                 struct acpi_nfit_interleave *idt)
 479 {
 480         size_t length = min_t(size_t, sizeof(*idt), idt->header.length);
 481         struct device *dev = acpi_desc->dev;
 482         struct nfit_idt *nfit_idt;
 483
 484         list_for_each_entry(nfit_idt, &prev->idts, list)
 485                 if (memcmp(nfit_idt->idt, idt, length) == 0) {
 486                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 487                         return true;
 488                 }
 489
 490         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt), GFP_KERNEL);
 491         if (!nfit_idt)
 492                 return false;
 493         INIT_LIST_HEAD(&nfit_idt->list);
 494         nfit_idt->idt = idt;
 495         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 496         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
 497                         idt->interleave_index, idt->line_count);
 498         return true;
 499 }
 500
 501 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 502                 struct nfit_table_prev *prev,
 503                 struct acpi_nfit_flush_address *flush)
 504 {
 505         size_t length = min_t(size_t, sizeof(*flush), flush->header.length);
 506         struct device *dev = acpi_desc->dev;
 507         struct nfit_flush *nfit_flush;
 508
 509         list_for_each_entry(nfit_flush, &prev->flushes, list)
 510                 if (memcmp(nfit_flush->flush, flush, length) == 0) {
 511                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 512                         return true;
 513                 }
 514
 515         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush), GFP_KERNEL);
 516         if (!nfit_flush)
 517                 return false;
 518         INIT_LIST_HEAD(&nfit_flush->list);
 519         nfit_flush->flush = flush;
 520         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 521         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
 522                         flush->device_handle, flush->hint_count);
 523         return true;
 524 }
 525
 526 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 527                 struct nfit_table_prev *prev, void *table, const void *end)
 528 {
 529         struct device *dev = acpi_desc->dev;
 530         struct acpi_nfit_header *hdr;
 531         void *err = ERR_PTR(-ENOMEM);
 532
 533         if (table >= end)
 534                 return NULL;
 535
 536         hdr = table;
 537         if (!hdr->length) {
 538                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 539                         hdr->type);
 540                 return NULL;
 541         }
 542
 543         switch (hdr->type) {
 544         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 545                 if (!add_spa(acpi_desc, prev, table))
 546                         return err;
 547                 break;
 548         case ACPI_NFIT_TYPE_MEMORY_MAP:
 549                 if (!add_memdev(acpi_desc, prev, table))
 550                         return err;
 551                 break;
 552         case ACPI_NFIT_TYPE_CONTROL_REGION:
 553                 if (!add_dcr(acpi_desc, prev, table))
 554                         return err;
 555                 break;
 556         case ACPI_NFIT_TYPE_DATA_REGION:
 557                 if (!add_bdw(acpi_desc, prev, table))
 558                         return err;
 559                 break;
 560         case ACPI_NFIT_TYPE_INTERLEAVE:
 561                 if (!add_idt(acpi_desc, prev, table))
 562                         return err;
 563                 break;
 564         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 565                 if (!add_flush(acpi_desc, prev, table))
 566                         return err;
 567                 break;
 568         case ACPI_NFIT_TYPE_SMBIOS:
 569                 dev_dbg(dev, "%s: smbios\n", __func__);
 570                 break;
 571         default:
 572                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 573                 break;
 574         }
 575
 576         return table + hdr->length;
 577 }
 578
 579 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 580                 struct nfit_mem *nfit_mem)
 581 {
 582         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 583         u16 dcr = nfit_mem->dcr->region_index;
 584         struct nfit_spa *nfit_spa;
 585
 586         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 587                 u16 range_index = nfit_spa->spa->range_index;
 588                 int type = nfit_spa_type(nfit_spa->spa);
 589                 struct nfit_memdev *nfit_memdev;
 590
 591                 if (type != NFIT_SPA_BDW)
 592                         continue;
 593
 594                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 595                         if (nfit_memdev->memdev->range_index != range_index)
 596                                 continue;
 597                         if (nfit_memdev->memdev->device_handle != device_handle)
 598                                 continue;
 599                         if (nfit_memdev->memdev->region_index != dcr)
 600                                 continue;
 601
 602                         nfit_mem->spa_bdw = nfit_spa->spa;
 603                         return;
 604                 }
 605         }
 606
 607         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 608                         nfit_mem->spa_dcr->range_index);
 609         nfit_mem->bdw = NULL;
 610 }
 611
 612 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 613                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 614 {
 615         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 616         struct nfit_memdev *nfit_memdev;
 617         struct nfit_flush *nfit_flush;
 618         struct nfit_bdw *nfit_bdw;
 619         struct nfit_idt *nfit_idt;
 620         u16 idt_idx, range_index;
 621
 622         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 623                 if (nfit_bdw->bdw->region_index != dcr)
 624                         continue;
 625                 nfit_mem->bdw = nfit_bdw->bdw;
 626                 break;
 627         }
 628
 629         if (!nfit_mem->bdw)
 630                 return;
 631
 632         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 633
 634         if (!nfit_mem->spa_bdw)
 635                 return;
 636
 637         range_index = nfit_mem->spa_bdw->range_index;
 638         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 639                 if (nfit_memdev->memdev->range_index != range_index ||
 640                                 nfit_memdev->memdev->region_index != dcr)
 641                         continue;
 642                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
 643                 idt_idx = nfit_memdev->memdev->interleave_index;
 644                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 645                         if (nfit_idt->idt->interleave_index != idt_idx)
 646                                 continue;
 647                         nfit_mem->idt_bdw = nfit_idt->idt;
 648                         break;
 649                 }
 650
 651                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
 652                         if (nfit_flush->flush->device_handle !=
 653                                         nfit_memdev->memdev->device_handle)
 654                                 continue;
 655                         nfit_mem->nfit_flush = nfit_flush;
 656                         break;
 657                 }
 658                 break;
 659         }
 660 }
 661
 662 static int nfit_mem_dcr_init(struct acpi_nfit_desc *acpi_desc,
 663                 struct acpi_nfit_system_address *spa)
 664 {
 665         struct nfit_mem *nfit_mem, *found;
 666         struct nfit_memdev *nfit_memdev;
 667         int type = nfit_spa_type(spa);
 668
 669         switch (type) {
 670         case NFIT_SPA_DCR:
 671         case NFIT_SPA_PM:
 672                 break;
 673         default:
 674                 return 0;
 675         }
 676
 677         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 678                 struct nfit_dcr *nfit_dcr;
 679                 u32 device_handle;
 680                 u16 dcr;
 681
 682                 if (nfit_memdev->memdev->range_index != spa->range_index)
 683                         continue;
 684                 found = NULL;
 685                 dcr = nfit_memdev->memdev->region_index;
 686                 device_handle = nfit_memdev->memdev->device_handle;
 687                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
 688                         if (__to_nfit_memdev(nfit_mem)->device_handle
 689                                         == device_handle) {
 690                                 found = nfit_mem;
 691                                 break;
 692                         }
 693
 694                 if (found)
 695                         nfit_mem = found;
 696                 else {
 697                         nfit_mem = devm_kzalloc(acpi_desc->dev,
 698                                         sizeof(*nfit_mem), GFP_KERNEL);
 699                         if (!nfit_mem)
 700                                 return -ENOMEM;
 701                         INIT_LIST_HEAD(&nfit_mem->list);
 702                         nfit_mem->acpi_desc = acpi_desc;
 703                         list_add(&nfit_mem->list, &acpi_desc->dimms);
 704                 }
 705
 706                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 707                         if (nfit_dcr->dcr->region_index != dcr)
 708                                 continue;
 709                         /*
 710                          * Record the control region for the dimm.  For
 711                          * the ACPI 6.1 case, where there are separate
 712                          * control regions for the pmem vs blk
 713                          * interfaces, be sure to record the extended
 714                          * blk details.
 715                          */
 716                         if (!nfit_mem->dcr)
 717                                 nfit_mem->dcr = nfit_dcr->dcr;
 718                         else if (nfit_mem->dcr->windows == 0
 719                                         && nfit_dcr->dcr->windows)
 720                                 nfit_mem->dcr = nfit_dcr->dcr;
 721                         break;
 722                 }
 723
 724                 if (dcr && !nfit_mem->dcr) {
 725                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
 726                                         spa->range_index, dcr);
 727                         return -ENODEV;
 728                 }
 729
 730                 if (type == NFIT_SPA_DCR) {
 731                         struct nfit_idt *nfit_idt;
 732                         u16 idt_idx;
 733
 734                         /* multiple dimms may share a SPA when interleaved */
 735                         nfit_mem->spa_dcr = spa;
 736                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
 737                         idt_idx = nfit_memdev->memdev->interleave_index;
 738                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
 739                                 if (nfit_idt->idt->interleave_index != idt_idx)
 740                                         continue;
 741                                 nfit_mem->idt_dcr = nfit_idt->idt;
 742                                 break;
 743                         }
 744                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
 745                 } else {
 746                         /*
 747                          * A single dimm may belong to multiple SPA-PM
 748                          * ranges, record at least one in addition to
 749                          * any SPA-DCR range.
 750                          */
 751                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
 752                 }
 753         }
 754
 755         return 0;
 756 }
 757
 758 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
 759 {
 760         struct nfit_mem *a = container_of(_a, typeof(*a), list);
 761         struct nfit_mem *b = container_of(_b, typeof(*b), list);
 762         u32 handleA, handleB;
 763
 764         handleA = __to_nfit_memdev(a)->device_handle;
 765         handleB = __to_nfit_memdev(b)->device_handle;
 766         if (handleA < handleB)
 767                 return -1;
 768         else if (handleA > handleB)
 769                 return 1;
 770         return 0;
 771 }
 772
 773 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
 774 {
 775         struct nfit_spa *nfit_spa;
 776
 777         /*
 778          * For each SPA-DCR or SPA-PMEM address range find its
 779          * corresponding MEMDEV(s).  From each MEMDEV find the
 780          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
 781          * try to find a SPA-BDW and a corresponding BDW that references
 782          * the DCR.  Throw it all into an nfit_mem object.  Note, that
 783          * BDWs are optional.
 784          */
 785         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 786                 int rc;
 787
 788                 rc = nfit_mem_dcr_init(acpi_desc, nfit_spa->spa);
 789                 if (rc)
 790                         return rc;
 791         }
 792
 793         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
 794
 795         return 0;
 796 }
 797
 798 static ssize_t revision_show(struct device *dev,
 799                 struct device_attribute *attr, char *buf)
 800 {
 801         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 802         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
 803         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
 804
 805         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
 806 }
 807 static DEVICE_ATTR_RO(revision);
 808
 809 static struct attribute *acpi_nfit_attributes[] = {
 810         &dev_attr_revision.attr,
 811         NULL,
 812 };
 813
 814 static struct attribute_group acpi_nfit_attribute_group = {
 815         .name = "nfit",
 816         .attrs = acpi_nfit_attributes,
 817 };
 818
 819 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
 820         &nvdimm_bus_attribute_group,
 821         &acpi_nfit_attribute_group,
 822         NULL,
 823 };
 824
 825 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
 826 {
 827         struct nvdimm *nvdimm = to_nvdimm(dev);
 828         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 829
 830         return __to_nfit_memdev(nfit_mem);
 831 }
 832
 833 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
 834 {
 835         struct nvdimm *nvdimm = to_nvdimm(dev);
 836         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 837
 838         return nfit_mem->dcr;
 839 }
 840
 841 static ssize_t handle_show(struct device *dev,
 842                 struct device_attribute *attr, char *buf)
 843 {
 844         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 845
 846         return sprintf(buf, "%#x\n", memdev->device_handle);
 847 }
 848 static DEVICE_ATTR_RO(handle);
 849
 850 static ssize_t phys_id_show(struct device *dev,
 851                 struct device_attribute *attr, char *buf)
 852 {
 853         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
 854
 855         return sprintf(buf, "%#x\n", memdev->physical_id);
 856 }
 857 static DEVICE_ATTR_RO(phys_id);
 858
 859 static ssize_t vendor_show(struct device *dev,
 860                 struct device_attribute *attr, char *buf)
 861 {
 862         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 863
 864         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
 865 }
 866 static DEVICE_ATTR_RO(vendor);
 867
 868 static ssize_t rev_id_show(struct device *dev,
 869                 struct device_attribute *attr, char *buf)
 870 {
 871         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 872
 873         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
 874 }
 875 static DEVICE_ATTR_RO(rev_id);
 876
 877 static ssize_t device_show(struct device *dev,
 878                 struct device_attribute *attr, char *buf)
 879 {
 880         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 881
 882         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
 883 }
 884 static DEVICE_ATTR_RO(device);
 885
 886 static ssize_t subsystem_vendor_show(struct device *dev,
 887                 struct device_attribute *attr, char *buf)
 888 {
 889         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 890
 891         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
 892 }
 893 static DEVICE_ATTR_RO(subsystem_vendor);
 894
 895 static ssize_t subsystem_rev_id_show(struct device *dev,
 896                 struct device_attribute *attr, char *buf)
 897 {
 898         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 899
 900         return sprintf(buf, "0x%04x\n",
 901                         be16_to_cpu(dcr->subsystem_revision_id));
 902 }
 903 static DEVICE_ATTR_RO(subsystem_rev_id);
 904
 905 static ssize_t subsystem_device_show(struct device *dev,
 906                 struct device_attribute *attr, char *buf)
 907 {
 908         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 909
 910         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
 911 }
 912 static DEVICE_ATTR_RO(subsystem_device);
 913
 914 static int num_nvdimm_formats(struct nvdimm *nvdimm)
 915 {
 916         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 917         int formats = 0;
 918
 919         if (nfit_mem->memdev_pmem)
 920                 formats++;
 921         if (nfit_mem->memdev_bdw)
 922                 formats++;
 923         return formats;
 924 }
 925
 926 static ssize_t format_show(struct device *dev,
 927                 struct device_attribute *attr, char *buf)
 928 {
 929         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 930
 931         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
 932 }
 933 static DEVICE_ATTR_RO(format);
 934
 935 static ssize_t format1_show(struct device *dev,
 936                 struct device_attribute *attr, char *buf)
 937 {
 938         u32 handle;
 939         ssize_t rc = -ENXIO;
 940         struct nfit_mem *nfit_mem;
 941         struct nfit_memdev *nfit_memdev;
 942         struct acpi_nfit_desc *acpi_desc;
 943         struct nvdimm *nvdimm = to_nvdimm(dev);
 944         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 945
 946         nfit_mem = nvdimm_provider_data(nvdimm);
 947         acpi_desc = nfit_mem->acpi_desc;
 948         handle = to_nfit_memdev(dev)->device_handle;
 949
 950         /* assumes DIMMs have at most 2 published interface codes */
 951         mutex_lock(&acpi_desc->init_mutex);
 952         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 953                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
 954                 struct nfit_dcr *nfit_dcr;
 955
 956                 if (memdev->device_handle != handle)
 957                         continue;
 958
 959                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 960                         if (nfit_dcr->dcr->region_index != memdev->region_index)
 961                                 continue;
 962                         if (nfit_dcr->dcr->code == dcr->code)
 963                                 continue;
 964                         rc = sprintf(buf, "0x%04x\n",
 965                                         le16_to_cpu(nfit_dcr->dcr->code));
 966                         break;
 967                 }
 968                 if (rc != ENXIO)
 969                         break;
 970         }
 971         mutex_unlock(&acpi_desc->init_mutex);
 972         return rc;
 973 }
 974 static DEVICE_ATTR_RO(format1);
 975
 976 static ssize_t formats_show(struct device *dev,
 977                 struct device_attribute *attr, char *buf)
 978 {
 979         struct nvdimm *nvdimm = to_nvdimm(dev);
 980
 981         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
 982 }
 983 static DEVICE_ATTR_RO(formats);
 984
 985 static ssize_t serial_show(struct device *dev,
 986                 struct device_attribute *attr, char *buf)
 987 {
 988         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 989
 990         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
 991 }
 992 static DEVICE_ATTR_RO(serial);
 993
 994 static ssize_t family_show(struct device *dev,
 995                 struct device_attribute *attr, char *buf)
 996 {
 997         struct nvdimm *nvdimm = to_nvdimm(dev);
 998         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 999
1000         if (nfit_mem->family < 0)
1001                 return -ENXIO;
1002         return sprintf(buf, "%d\n", nfit_mem->family);
1003 }
1004 static DEVICE_ATTR_RO(family);
1005
1006 static ssize_t dsm_mask_show(struct device *dev,
1007                 struct device_attribute *attr, char *buf)
1008 {
1009         struct nvdimm *nvdimm = to_nvdimm(dev);
1010         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1011
1012         if (nfit_mem->family < 0)
1013                 return -ENXIO;
1014         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1015 }
1016 static DEVICE_ATTR_RO(dsm_mask);
1017
1018 static ssize_t flags_show(struct device *dev,
1019                 struct device_attribute *attr, char *buf)
1020 {
1021         u16 flags = to_nfit_memdev(dev)->flags;
1022
1023         return sprintf(buf, "%s%s%s%s%s\n",
1024                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1025                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1026                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1027                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1028                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1029 }
1030 static DEVICE_ATTR_RO(flags);
1031
1032 static ssize_t id_show(struct device *dev,
1033                 struct device_attribute *attr, char *buf)
1034 {
1035         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1036
1037         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1038                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1039                                 be16_to_cpu(dcr->vendor_id),
1040                                 dcr->manufacturing_location,
1041                                 be16_to_cpu(dcr->manufacturing_date),
1042                                 be32_to_cpu(dcr->serial_number));
1043         else
1044                 return sprintf(buf, "%04x-%08x\n",
1045                                 be16_to_cpu(dcr->vendor_id),
1046                                 be32_to_cpu(dcr->serial_number));
1047 }
1048 static DEVICE_ATTR_RO(id);
1049
1050 static struct attribute *acpi_nfit_dimm_attributes[] = {
1051         &dev_attr_handle.attr,
1052         &dev_attr_phys_id.attr,
1053         &dev_attr_vendor.attr,
1054         &dev_attr_device.attr,
1055         &dev_attr_rev_id.attr,
1056         &dev_attr_subsystem_vendor.attr,
1057         &dev_attr_subsystem_device.attr,
1058         &dev_attr_subsystem_rev_id.attr,
1059         &dev_attr_format.attr,
1060         &dev_attr_formats.attr,
1061         &dev_attr_format1.attr,
1062         &dev_attr_serial.attr,
1063         &dev_attr_flags.attr,
1064         &dev_attr_id.attr,
1065         &dev_attr_family.attr,
1066         &dev_attr_dsm_mask.attr,
1067         NULL,
1068 };
1069
1070 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1071                 struct attribute *a, int n)
1072 {
1073         struct device *dev = container_of(kobj, struct device, kobj);
1074         struct nvdimm *nvdimm = to_nvdimm(dev);
1075
1076         if (!to_nfit_dcr(dev))
1077                 return 0;
1078         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1079                 return 0;
1080         return a->mode;
1081 }
1082
1083 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1084         .name = "nfit",
1085         .attrs = acpi_nfit_dimm_attributes,
1086         .is_visible = acpi_nfit_dimm_attr_visible,
1087 };
1088
1089 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1090         &nvdimm_attribute_group,
1091         &nd_device_attribute_group,
1092         &acpi_nfit_dimm_attribute_group,
1093         NULL,
1094 };
1095
1096 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1097                 u32 device_handle)
1098 {
1099         struct nfit_mem *nfit_mem;
1100
1101         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1102                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1103                         return nfit_mem->nvdimm;
1104
1105         return NULL;
1106 }
1107
1108 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1109                 struct nfit_mem *nfit_mem, u32 device_handle)
1110 {
1111         struct acpi_device *adev, *adev_dimm;
1112         struct device *dev = acpi_desc->dev;
1113         unsigned long dsm_mask;
1114         const u8 *uuid;
1115         int i;
1116
1117         /* nfit test assumes 1:1 relationship between commands and dsms */
1118         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1119         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1120         adev = to_acpi_dev(acpi_desc);
1121         if (!adev)
1122                 return 0;
1123
1124         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1125         nfit_mem->adev = adev_dimm;
1126         if (!adev_dimm) {
1127                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1128                                 device_handle);
1129                 return force_enable_dimms ? 0 : -ENODEV;
1130         }
1131
1132         /*
1133          * Until standardization materializes we need to consider up to 3
1134          * different command sets.  Note, that checking for function0 (bit0)
1135          * tells us if any commands are reachable through this uuid.
1136          */
1137         for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_HPE2; i++)
1138                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1139                         break;
1140
1141         /* limit the supported commands to those that are publicly documented */
1142         nfit_mem->family = i;
1143         if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1144                 dsm_mask = 0x3fe;
1145                 if (disable_vendor_specific)
1146                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1147         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1)
1148                 dsm_mask = 0x1c3c76;
1149         else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1150                 dsm_mask = 0x1fe;
1151                 if (disable_vendor_specific)
1152                         dsm_mask &= ~(1 << 8);
1153         } else {
1154                 dev_dbg(dev, "unknown dimm command family\n");
1155                 nfit_mem->family = -1;
1156                 /* DSMs are optional, continue loading the driver... */
1157                 return 0;
1158         }
1159
1160         uuid = to_nfit_uuid(nfit_mem->family);
1161         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1162                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1163                         set_bit(i, &nfit_mem->dsm_mask);
1164
1165         return 0;
1166 }
1167
1168 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1169 {
1170         struct nfit_mem *nfit_mem;
1171         int dimm_count = 0;
1172
1173         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1174                 unsigned long flags = 0, cmd_mask;
1175                 struct nvdimm *nvdimm;
1176                 u32 device_handle;
1177                 u16 mem_flags;
1178                 int rc;
1179
1180                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1181                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1182                 if (nvdimm) {
1183                         dimm_count++;
1184                         continue;
1185                 }
1186
1187                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1188                         flags |= NDD_ALIASING;
1189
1190                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1191                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1192                         flags |= NDD_UNARMED;
1193
1194                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1195                 if (rc)
1196                         continue;
1197
1198                 /*
1199                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1200                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1201                  * userspace interface.
1202                  */
1203                 cmd_mask = 1UL << ND_CMD_CALL;
1204                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1205                         cmd_mask |= nfit_mem->dsm_mask;
1206
1207                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1208                                 acpi_nfit_dimm_attribute_groups,
1209                                 flags, cmd_mask);
1210                 if (!nvdimm)
1211                         return -ENOMEM;
1212
1213                 nfit_mem->nvdimm = nvdimm;
1214                 dimm_count++;
1215
1216                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1217                         continue;
1218
1219                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1220                                 nvdimm_name(nvdimm),
1221                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1222                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1223                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1224                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1225
1226         }
1227
1228         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1229 }
1230
1231 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1232 {
1233         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1234         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1235         struct acpi_device *adev;
1236         int i;
1237
1238         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1239         adev = to_acpi_dev(acpi_desc);
1240         if (!adev)
1241                 return;
1242
1243         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1244                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1245                         set_bit(i, &nd_desc->cmd_mask);
1246 }
1247
1248 static ssize_t range_index_show(struct device *dev,
1249                 struct device_attribute *attr, char *buf)
1250 {
1251         struct nd_region *nd_region = to_nd_region(dev);
1252         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1253
1254         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1255 }
1256 static DEVICE_ATTR_RO(range_index);
1257
1258 static struct attribute *acpi_nfit_region_attributes[] = {
1259         &dev_attr_range_index.attr,
1260         NULL,
1261 };
1262
1263 static struct attribute_group acpi_nfit_region_attribute_group = {
1264         .name = "nfit",
1265         .attrs = acpi_nfit_region_attributes,
1266 };
1267
1268 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1269         &nd_region_attribute_group,
1270         &nd_mapping_attribute_group,
1271         &nd_device_attribute_group,
1272         &nd_numa_attribute_group,
1273         &acpi_nfit_region_attribute_group,
1274         NULL,
1275 };
1276
1277 /* enough info to uniquely specify an interleave set */
1278 struct nfit_set_info {
1279         struct nfit_set_info_map {
1280                 u64 region_offset;
1281                 u32 serial_number;
1282                 u32 pad;
1283         } mapping[0];
1284 };
1285
1286 static size_t sizeof_nfit_set_info(int num_mappings)
1287 {
1288         return sizeof(struct nfit_set_info)
1289                 + num_mappings * sizeof(struct nfit_set_info_map);
1290 }
1291
1292 static int cmp_map(const void *m0, const void *m1)
1293 {
1294         const struct nfit_set_info_map *map0 = m0;
1295         const struct nfit_set_info_map *map1 = m1;
1296
1297         return memcmp(&map0->region_offset, &map1->region_offset,
1298                         sizeof(u64));
1299 }
1300
1301 /* Retrieve the nth entry referencing this spa */
1302 static struct acpi_nfit_memory_map *memdev_from_spa(
1303                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1304 {
1305         struct nfit_memdev *nfit_memdev;
1306
1307         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1308                 if (nfit_memdev->memdev->range_index == range_index)
1309                         if (n-- == 0)
1310                                 return nfit_memdev->memdev;
1311         return NULL;
1312 }
1313
1314 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1315                 struct nd_region_desc *ndr_desc,
1316                 struct acpi_nfit_system_address *spa)
1317 {
1318         int i, spa_type = nfit_spa_type(spa);
1319         struct device *dev = acpi_desc->dev;
1320         struct nd_interleave_set *nd_set;
1321         u16 nr = ndr_desc->num_mappings;
1322         struct nfit_set_info *info;
1323
1324         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1325                 /* pass */;
1326         else
1327                 return 0;
1328
1329         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1330         if (!nd_set)
1331                 return -ENOMEM;
1332
1333         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1334         if (!info)
1335                 return -ENOMEM;
1336         for (i = 0; i < nr; i++) {
1337                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1338                 struct nfit_set_info_map *map = &info->mapping[i];
1339                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1340                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1341                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1342                                 spa->range_index, i);
1343
1344                 if (!memdev || !nfit_mem->dcr) {
1345                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1346                         return -ENODEV;
1347                 }
1348
1349                 map->region_offset = memdev->region_offset;
1350                 map->serial_number = nfit_mem->dcr->serial_number;
1351         }
1352
1353         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1354                         cmp_map, NULL);
1355         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1356         ndr_desc->nd_set = nd_set;
1357         devm_kfree(dev, info);
1358
1359         return 0;
1360 }
1361
1362 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1363 {
1364         struct acpi_nfit_interleave *idt = mmio->idt;
1365         u32 sub_line_offset, line_index, line_offset;
1366         u64 line_no, table_skip_count, table_offset;
1367
1368         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1369         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1370         line_offset = idt->line_offset[line_index]
1371                 * mmio->line_size;
1372         table_offset = table_skip_count * mmio->table_size;
1373
1374         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1375 }
1376
1377 static void wmb_blk(struct nfit_blk *nfit_blk)
1378 {
1379
1380         if (nfit_blk->nvdimm_flush) {
1381                 /*
1382                  * The first wmb() is needed to 'sfence' all previous writes
1383                  * such that they are architecturally visible for the platform
1384                  * buffer flush.  Note that we've already arranged for pmem
1385                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1386                  * final wmb() ensures ordering for the NVDIMM flush write.
1387                  */
1388                 wmb();
1389                 writeq(1, nfit_blk->nvdimm_flush);
1390                 wmb();
1391         } else
1392                 wmb_pmem();
1393 }
1394
1395 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1396 {
1397         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1398         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1399
1400         if (mmio->num_lines)
1401                 offset = to_interleave_offset(offset, mmio);
1402
1403         return readl(mmio->addr.base + offset);
1404 }
1405
1406 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1407                 resource_size_t dpa, unsigned int len, unsigned int write)
1408 {
1409         u64 cmd, offset;
1410         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1411
1412         enum {
1413                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1414                 BCW_LEN_SHIFT = 48,
1415                 BCW_LEN_MASK = (1ULL << 8) - 1,
1416                 BCW_CMD_SHIFT = 56,
1417         };
1418
1419         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1420         len = len >> L1_CACHE_SHIFT;
1421         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1422         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1423
1424         offset = nfit_blk->cmd_offset + mmio->size * bw;
1425         if (mmio->num_lines)
1426                 offset = to_interleave_offset(offset, mmio);
1427
1428         writeq(cmd, mmio->addr.base + offset);
1429         wmb_blk(nfit_blk);
1430
1431         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1432                 readq(mmio->addr.base + offset);
1433 }
1434
1435 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1436                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1437                 unsigned int lane)
1438 {
1439         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1440         unsigned int copied = 0;
1441         u64 base_offset;
1442         int rc;
1443
1444         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1445                 + lane * mmio->size;
1446         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1447         while (len) {
1448                 unsigned int c;
1449                 u64 offset;
1450
1451                 if (mmio->num_lines) {
1452                         u32 line_offset;
1453
1454                         offset = to_interleave_offset(base_offset + copied,
1455                                         mmio);
1456                         div_u64_rem(offset, mmio->line_size, &line_offset);
1457                         c = min_t(size_t, len, mmio->line_size - line_offset);
1458                 } else {
1459                         offset = base_offset + nfit_blk->bdw_offset;
1460                         c = len;
1461                 }
1462
1463                 if (rw)
1464                         memcpy_to_pmem(mmio->addr.aperture + offset,
1465                                         iobuf + copied, c);
1466                 else {
1467                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1468                                 mmio_flush_range((void __force *)
1469                                         mmio->addr.aperture + offset, c);
1470
1471                         memcpy_from_pmem(iobuf + copied,
1472                                         mmio->addr.aperture + offset, c);
1473                 }
1474
1475                 copied += c;
1476                 len -= c;
1477         }
1478
1479         if (rw)
1480                 wmb_blk(nfit_blk);
1481
1482         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1483         return rc;
1484 }
1485
1486 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1487                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1488 {
1489         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1490         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1491         struct nd_region *nd_region = nfit_blk->nd_region;
1492         unsigned int lane, copied = 0;
1493         int rc = 0;
1494
1495         lane = nd_region_acquire_lane(nd_region);
1496         while (len) {
1497                 u64 c = min(len, mmio->size);
1498
1499                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1500                                 iobuf + copied, c, rw, lane);
1501                 if (rc)
1502                         break;
1503
1504                 copied += c;
1505                 len -= c;
1506         }
1507         nd_region_release_lane(nd_region, lane);
1508
1509         return rc;
1510 }
1511
1512 static void nfit_spa_mapping_release(struct kref *kref)
1513 {
1514         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1515         struct acpi_nfit_system_address *spa = spa_map->spa;
1516         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1517
1518         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1519         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1520         if (spa_map->type == SPA_MAP_APERTURE)
1521                 memunmap((void __force *)spa_map->addr.aperture);
1522         else
1523                 iounmap(spa_map->addr.base);
1524         release_mem_region(spa->address, spa->length);
1525         list_del(&spa_map->list);
1526         kfree(spa_map);
1527 }
1528
1529 static struct nfit_spa_mapping *find_spa_mapping(
1530                 struct acpi_nfit_desc *acpi_desc,
1531                 struct acpi_nfit_system_address *spa)
1532 {
1533         struct nfit_spa_mapping *spa_map;
1534
1535         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1536         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1537                 if (spa_map->spa == spa)
1538                         return spa_map;
1539
1540         return NULL;
1541 }
1542
1543 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1544                 struct acpi_nfit_system_address *spa)
1545 {
1546         struct nfit_spa_mapping *spa_map;
1547
1548         mutex_lock(&acpi_desc->spa_map_mutex);
1549         spa_map = find_spa_mapping(acpi_desc, spa);
1550
1551         if (spa_map)
1552                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1553         mutex_unlock(&acpi_desc->spa_map_mutex);
1554 }
1555
1556 static void __iomem *__nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1557                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1558 {
1559         resource_size_t start = spa->address;
1560         resource_size_t n = spa->length;
1561         struct nfit_spa_mapping *spa_map;
1562         struct resource *res;
1563
1564         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1565
1566         spa_map = find_spa_mapping(acpi_desc, spa);
1567         if (spa_map) {
1568                 kref_get(&spa_map->kref);
1569                 return spa_map->addr.base;
1570         }
1571
1572         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1573         if (!spa_map)
1574                 return NULL;
1575
1576         INIT_LIST_HEAD(&spa_map->list);
1577         spa_map->spa = spa;
1578         kref_init(&spa_map->kref);
1579         spa_map->acpi_desc = acpi_desc;
1580
1581         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1582         if (!res)
1583                 goto err_mem;
1584
1585         spa_map->type = type;
1586         if (type == SPA_MAP_APERTURE)
1587                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1588                                                         ARCH_MEMREMAP_PMEM);
1589         else
1590                 spa_map->addr.base = ioremap_nocache(start, n);
1591
1592
1593         if (!spa_map->addr.base)
1594                 goto err_map;
1595
1596         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1597         return spa_map->addr.base;
1598
1599  err_map:
1600         release_mem_region(start, n);
1601  err_mem:
1602         kfree(spa_map);
1603         return NULL;
1604 }
1605
1606 /**
1607  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1608  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1609  * @nfit_spa: spa table to map
1610  * @type: aperture or control region
1611  *
1612  * In the case where block-data-window apertures and
1613  * dimm-control-regions are interleaved they will end up sharing a
1614  * single request_mem_region() + ioremap() for the address range.  In
1615  * the style of devm nfit_spa_map() mappings are automatically dropped
1616  * when all region devices referencing the same mapping are disabled /
1617  * unbound.
1618  */
1619 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1620                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1621 {
1622         void __iomem *iomem;
1623
1624         mutex_lock(&acpi_desc->spa_map_mutex);
1625         iomem = __nfit_spa_map(acpi_desc, spa, type);
1626         mutex_unlock(&acpi_desc->spa_map_mutex);
1627
1628         return iomem;
1629 }
1630
1631 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1632                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1633 {
1634         if (idt) {
1635                 mmio->num_lines = idt->line_count;
1636                 mmio->line_size = idt->line_size;
1637                 if (interleave_ways == 0)
1638                         return -ENXIO;
1639                 mmio->table_size = mmio->num_lines * interleave_ways
1640                         * mmio->line_size;
1641         }
1642
1643         return 0;
1644 }
1645
1646 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1647                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1648 {
1649         struct nd_cmd_dimm_flags flags;
1650         int rc;
1651
1652         memset(&flags, 0, sizeof(flags));
1653         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1654                         sizeof(flags), NULL);
1655
1656         if (rc >= 0 && flags.status == 0)
1657                 nfit_blk->dimm_flags = flags.flags;
1658         else if (rc == -ENOTTY) {
1659                 /* fall back to a conservative default */
1660                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1661                 rc = 0;
1662         } else
1663                 rc = -ENXIO;
1664
1665         return rc;
1666 }
1667
1668 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1669                 struct device *dev)
1670 {
1671         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1672         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1673         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1674         struct nfit_flush *nfit_flush;
1675         struct nfit_blk_mmio *mmio;
1676         struct nfit_blk *nfit_blk;
1677         struct nfit_mem *nfit_mem;
1678         struct nvdimm *nvdimm;
1679         int rc;
1680
1681         nvdimm = nd_blk_region_to_dimm(ndbr);
1682         nfit_mem = nvdimm_provider_data(nvdimm);
1683         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1684                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1685                                 nfit_mem ? "" : " nfit_mem",
1686                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1687                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1688                 return -ENXIO;
1689         }
1690
1691         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1692         if (!nfit_blk)
1693                 return -ENOMEM;
1694         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1695         nfit_blk->nd_region = to_nd_region(dev);
1696
1697         /* map block aperture memory */
1698         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1699         mmio = &nfit_blk->mmio[BDW];
1700         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1701                         SPA_MAP_APERTURE);
1702         if (!mmio->addr.base) {
1703                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1704                                 nvdimm_name(nvdimm));
1705                 return -ENOMEM;
1706         }
1707         mmio->size = nfit_mem->bdw->size;
1708         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1709         mmio->idt = nfit_mem->idt_bdw;
1710         mmio->spa = nfit_mem->spa_bdw;
1711         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1712                         nfit_mem->memdev_bdw->interleave_ways);
1713         if (rc) {
1714                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1715                                 __func__, nvdimm_name(nvdimm));
1716                 return rc;
1717         }
1718
1719         /* map block control memory */
1720         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1721         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1722         mmio = &nfit_blk->mmio[DCR];
1723         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1724                         SPA_MAP_CONTROL);
1725         if (!mmio->addr.base) {
1726                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1727                                 nvdimm_name(nvdimm));
1728                 return -ENOMEM;
1729         }
1730         mmio->size = nfit_mem->dcr->window_size;
1731         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1732         mmio->idt = nfit_mem->idt_dcr;
1733         mmio->spa = nfit_mem->spa_dcr;
1734         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1735                         nfit_mem->memdev_dcr->interleave_ways);
1736         if (rc) {
1737                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1738                                 __func__, nvdimm_name(nvdimm));
1739                 return rc;
1740         }
1741
1742         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1743         if (rc < 0) {
1744                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1745                                 __func__, nvdimm_name(nvdimm));
1746                 return rc;
1747         }
1748
1749         nfit_flush = nfit_mem->nfit_flush;
1750         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1751                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1752                                 nfit_flush->flush->hint_address[0], 8);
1753                 if (!nfit_blk->nvdimm_flush)
1754                         return -ENOMEM;
1755         }
1756
1757         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1758                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1759
1760         if (mmio->line_size == 0)
1761                 return 0;
1762
1763         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1764                         + 8 > mmio->line_size) {
1765                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1766                 return -ENXIO;
1767         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1768                         + 8 > mmio->line_size) {
1769                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1770                 return -ENXIO;
1771         }
1772
1773         return 0;
1774 }
1775
1776 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1777                 struct device *dev)
1778 {
1779         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1780         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1781         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1782         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1783         int i;
1784
1785         if (!nfit_blk)
1786                 return; /* never enabled */
1787
1788         /* auto-free BLK spa mappings */
1789         for (i = 0; i < 2; i++) {
1790                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1791
1792                 if (mmio->addr.base)
1793                         nfit_spa_unmap(acpi_desc, mmio->spa);
1794         }
1795         nd_blk_region_set_provider_data(ndbr, NULL);
1796         /* devm will free nfit_blk */
1797 }
1798
1799 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1800                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1801 {
1802         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1803         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1804         int cmd_rc, rc;
1805
1806         cmd->address = spa->address;
1807         cmd->length = spa->length;
1808         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1809                         sizeof(*cmd), &cmd_rc);
1810         if (rc < 0)
1811                 return rc;
1812         return cmd_rc;
1813 }
1814
1815 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1816 {
1817         int rc;
1818         int cmd_rc;
1819         struct nd_cmd_ars_start ars_start;
1820         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1821         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1822
1823         memset(&ars_start, 0, sizeof(ars_start));
1824         ars_start.address = spa->address;
1825         ars_start.length = spa->length;
1826         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1827                 ars_start.type = ND_ARS_PERSISTENT;
1828         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1829                 ars_start.type = ND_ARS_VOLATILE;
1830         else
1831                 return -ENOTTY;
1832
1833         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1834                         sizeof(ars_start), &cmd_rc);
1835
1836         if (rc < 0)
1837                 return rc;
1838         return cmd_rc;
1839 }
1840
1841 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1842 {
1843         int rc, cmd_rc;
1844         struct nd_cmd_ars_start ars_start;
1845         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1846         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1847
1848         memset(&ars_start, 0, sizeof(ars_start));
1849         ars_start.address = ars_status->restart_address;
1850         ars_start.length = ars_status->restart_length;
1851         ars_start.type = ars_status->type;
1852         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1853                         sizeof(ars_start), &cmd_rc);
1854         if (rc < 0)
1855                 return rc;
1856         return cmd_rc;
1857 }
1858
1859 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1860 {
1861         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1862         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1863         int rc, cmd_rc;
1864
1865         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1866                         acpi_desc->ars_status_size, &cmd_rc);
1867         if (rc < 0)
1868                 return rc;
1869         return cmd_rc;
1870 }
1871
1872 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1873                 struct nd_cmd_ars_status *ars_status)
1874 {
1875         int rc;
1876         u32 i;
1877
1878         for (i = 0; i < ars_status->num_records; i++) {
1879                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1880                                 ars_status->records[i].err_address,
1881                                 ars_status->records[i].length);
1882                 if (rc)
1883                         return rc;
1884         }
1885
1886         return 0;
1887 }
1888
1889 static void acpi_nfit_remove_resource(void *data)
1890 {
1891         struct resource *res = data;
1892
1893         remove_resource(res);
1894 }
1895
1896 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1897                 struct nd_region_desc *ndr_desc)
1898 {
1899         struct resource *res, *nd_res = ndr_desc->res;
1900         int is_pmem, ret;
1901
1902         /* No operation if the region is already registered as PMEM */
1903         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1904                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1905         if (is_pmem == REGION_INTERSECTS)
1906                 return 0;
1907
1908         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1909         if (!res)
1910                 return -ENOMEM;
1911
1912         res->name = "Persistent Memory";
1913         res->start = nd_res->start;
1914         res->end = nd_res->end;
1915         res->flags = IORESOURCE_MEM;
1916         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1917
1918         ret = insert_resource(&iomem_resource, res);
1919         if (ret)
1920                 return ret;
1921
1922         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1923         if (ret) {
1924                 remove_resource(res);
1925                 return ret;
1926         }
1927
1928         return 0;
1929 }
1930
1931 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1932                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1933                 struct acpi_nfit_memory_map *memdev,
1934                 struct nfit_spa *nfit_spa)
1935 {
1936         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1937                         memdev->device_handle);
1938         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1939         struct nd_blk_region_desc *ndbr_desc;
1940         struct nfit_mem *nfit_mem;
1941         int blk_valid = 0;
1942
1943         if (!nvdimm) {
1944                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1945                                 spa->range_index, memdev->device_handle);
1946                 return -ENODEV;
1947         }
1948
1949         nd_mapping->nvdimm = nvdimm;
1950         switch (nfit_spa_type(spa)) {
1951         case NFIT_SPA_PM:
1952         case NFIT_SPA_VOLATILE:
1953                 nd_mapping->start = memdev->address;
1954                 nd_mapping->size = memdev->region_size;
1955                 break;
1956         case NFIT_SPA_DCR:
1957                 nfit_mem = nvdimm_provider_data(nvdimm);
1958                 if (!nfit_mem || !nfit_mem->bdw) {
1959                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1960                                         spa->range_index, nvdimm_name(nvdimm));
1961                 } else {
1962                         nd_mapping->size = nfit_mem->bdw->capacity;
1963                         nd_mapping->start = nfit_mem->bdw->start_address;
1964                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1965                         blk_valid = 1;
1966                 }
1967
1968                 ndr_desc->nd_mapping = nd_mapping;
1969                 ndr_desc->num_mappings = blk_valid;
1970                 ndbr_desc = to_blk_region_desc(ndr_desc);
1971                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1972                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1973                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1974                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1975                                 ndr_desc);
1976                 if (!nfit_spa->nd_region)
1977                         return -ENOMEM;
1978                 break;
1979         }
1980
1981         return 0;
1982 }
1983
1984 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1985                 struct nfit_spa *nfit_spa)
1986 {
1987         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1988         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1989         struct nd_blk_region_desc ndbr_desc;
1990         struct nd_region_desc *ndr_desc;
1991         struct nfit_memdev *nfit_memdev;
1992         struct nvdimm_bus *nvdimm_bus;
1993         struct resource res;
1994         int count = 0, rc;
1995
1996         if (nfit_spa->nd_region)
1997                 return 0;
1998
1999         if (spa->range_index == 0) {
2000                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2001                                 __func__);
2002                 return 0;
2003         }
2004
2005         memset(&res, 0, sizeof(res));
2006         memset(&nd_mappings, 0, sizeof(nd_mappings));
2007         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2008         res.start = spa->address;
2009         res.end = res.start + spa->length - 1;
2010         ndr_desc = &ndbr_desc.ndr_desc;
2011         ndr_desc->res = &res;
2012         ndr_desc->provider_data = nfit_spa;
2013         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2014         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2015                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2016                                                 spa->proximity_domain);
2017         else
2018                 ndr_desc->numa_node = NUMA_NO_NODE;
2019
2020         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2021                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2022                 struct nd_mapping *nd_mapping;
2023
2024                 if (memdev->range_index != spa->range_index)
2025                         continue;
2026                 if (count >= ND_MAX_MAPPINGS) {
2027                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2028                                         spa->range_index, ND_MAX_MAPPINGS);
2029                         return -ENXIO;
2030                 }
2031                 nd_mapping = &nd_mappings[count++];
2032                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
2033                                 memdev, nfit_spa);
2034                 if (rc)
2035                         goto out;
2036         }
2037
2038         ndr_desc->nd_mapping = nd_mappings;
2039         ndr_desc->num_mappings = count;
2040         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2041         if (rc)
2042                 goto out;
2043
2044         nvdimm_bus = acpi_desc->nvdimm_bus;
2045         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2046                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2047                 if (rc) {
2048                         dev_warn(acpi_desc->dev,
2049                                 "failed to insert pmem resource to iomem: %d\n",
2050                                 rc);
2051                         goto out;
2052                 }
2053
2054                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2055                                 ndr_desc);
2056                 if (!nfit_spa->nd_region)
2057                         rc = -ENOMEM;
2058         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2059                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2060                                 ndr_desc);
2061                 if (!nfit_spa->nd_region)
2062                         rc = -ENOMEM;
2063         }
2064
2065  out:
2066         if (rc)
2067                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2068                                 nfit_spa->spa->range_index);
2069         return rc;
2070 }
2071
2072 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2073                 u32 max_ars)
2074 {
2075         struct device *dev = acpi_desc->dev;
2076         struct nd_cmd_ars_status *ars_status;
2077
2078         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2079                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2080                 return 0;
2081         }
2082
2083         if (acpi_desc->ars_status)
2084                 devm_kfree(dev, acpi_desc->ars_status);
2085         acpi_desc->ars_status = NULL;
2086         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2087         if (!ars_status)
2088                 return -ENOMEM;
2089         acpi_desc->ars_status = ars_status;
2090         acpi_desc->ars_status_size = max_ars;
2091         return 0;
2092 }
2093
2094 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2095                 struct nfit_spa *nfit_spa)
2096 {
2097         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2098         int rc;
2099
2100         if (!nfit_spa->max_ars) {
2101                 struct nd_cmd_ars_cap ars_cap;
2102
2103                 memset(&ars_cap, 0, sizeof(ars_cap));
2104                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2105                 if (rc < 0)
2106                         return rc;
2107                 nfit_spa->max_ars = ars_cap.max_ars_out;
2108                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2109                 /* check that the supported scrub types match the spa type */
2110                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2111                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2112                         return -ENOTTY;
2113                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2114                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2115                         return -ENOTTY;
2116         }
2117
2118         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2119                 return -ENOMEM;
2120
2121         rc = ars_get_status(acpi_desc);
2122         if (rc < 0 && rc != -ENOSPC)
2123                 return rc;
2124
2125         if (ars_status_process_records(acpi_desc->nvdimm_bus,
2126                                 acpi_desc->ars_status))
2127                 return -ENOMEM;
2128
2129         return 0;
2130 }
2131
2132 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2133                 struct nfit_spa *nfit_spa)
2134 {
2135         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2136         unsigned int overflow_retry = scrub_overflow_abort;
2137         u64 init_ars_start = 0, init_ars_len = 0;
2138         struct device *dev = acpi_desc->dev;
2139         unsigned int tmo = scrub_timeout;
2140         int rc;
2141
2142         if (nfit_spa->ars_done || !nfit_spa->nd_region)
2143                 return;
2144
2145         rc = ars_start(acpi_desc, nfit_spa);
2146         /*
2147          * If we timed out the initial scan we'll still be busy here,
2148          * and will wait another timeout before giving up permanently.
2149          */
2150         if (rc < 0 && rc != -EBUSY)
2151                 return;
2152
2153         do {
2154                 u64 ars_start, ars_len;
2155
2156                 if (acpi_desc->cancel)
2157                         break;
2158                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2159                 if (rc == -ENOTTY)
2160                         break;
2161                 if (rc == -EBUSY && !tmo) {
2162                         dev_warn(dev, "range %d ars timeout, aborting\n",
2163                                         spa->range_index);
2164                         break;
2165                 }
2166
2167                 if (rc == -EBUSY) {
2168                         /*
2169                          * Note, entries may be appended to the list
2170                          * while the lock is dropped, but the workqueue
2171                          * being active prevents entries being deleted /
2172                          * freed.
2173                          */
2174                         mutex_unlock(&acpi_desc->init_mutex);
2175                         ssleep(1);
2176                         tmo--;
2177                         mutex_lock(&acpi_desc->init_mutex);
2178                         continue;
2179                 }
2180
2181                 /* we got some results, but there are more pending... */
2182                 if (rc == -ENOSPC && overflow_retry--) {
2183                         if (!init_ars_len) {
2184                                 init_ars_len = acpi_desc->ars_status->length;
2185                                 init_ars_start = acpi_desc->ars_status->address;
2186                         }
2187                         rc = ars_continue(acpi_desc);
2188                 }
2189
2190                 if (rc < 0) {
2191                         dev_warn(dev, "range %d ars continuation failed\n",
2192                                         spa->range_index);
2193                         break;
2194                 }
2195
2196                 if (init_ars_len) {
2197                         ars_start = init_ars_start;
2198                         ars_len = init_ars_len;
2199                 } else {
2200                         ars_start = acpi_desc->ars_status->address;
2201                         ars_len = acpi_desc->ars_status->length;
2202                 }
2203                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2204                                 spa->range_index, ars_start, ars_len);
2205                 /* notify the region about new poison entries */
2206                 nvdimm_region_notify(nfit_spa->nd_region,
2207                                 NVDIMM_REVALIDATE_POISON);
2208                 break;
2209         } while (1);
2210 }
2211
2212 static void acpi_nfit_scrub(struct work_struct *work)
2213 {
2214         struct device *dev;
2215         u64 init_scrub_length = 0;
2216         struct nfit_spa *nfit_spa;
2217         u64 init_scrub_address = 0;
2218         bool init_ars_done = false;
2219         struct acpi_nfit_desc *acpi_desc;
2220         unsigned int tmo = scrub_timeout;
2221         unsigned int overflow_retry = scrub_overflow_abort;
2222
2223         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2224         dev = acpi_desc->dev;
2225
2226         /*
2227          * We scrub in 2 phases.  The first phase waits for any platform
2228          * firmware initiated scrubs to complete and then we go search for the
2229          * affected spa regions to mark them scanned.  In the second phase we
2230          * initiate a directed scrub for every range that was not scrubbed in
2231          * phase 1.
2232          */
2233
2234         /* process platform firmware initiated scrubs */
2235  retry:
2236         mutex_lock(&acpi_desc->init_mutex);
2237         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2238                 struct nd_cmd_ars_status *ars_status;
2239                 struct acpi_nfit_system_address *spa;
2240                 u64 ars_start, ars_len;
2241                 int rc;
2242
2243                 if (acpi_desc->cancel)
2244                         break;
2245
2246                 if (nfit_spa->nd_region)
2247                         continue;
2248
2249                 if (init_ars_done) {
2250                         /*
2251                          * No need to re-query, we're now just
2252                          * reconciling all the ranges covered by the
2253                          * initial scrub
2254                          */
2255                         rc = 0;
2256                 } else
2257                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2258
2259                 if (rc == -ENOTTY) {
2260                         /* no ars capability, just register spa and move on */
2261                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2262                         continue;
2263                 }
2264
2265                 if (rc == -EBUSY && !tmo) {
2266                         /* fallthrough to directed scrub in phase 2 */
2267                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2268                         break;
2269                 } else if (rc == -EBUSY) {
2270                         mutex_unlock(&acpi_desc->init_mutex);
2271                         ssleep(1);
2272                         tmo--;
2273                         goto retry;
2274                 }
2275
2276                 /* we got some results, but there are more pending... */
2277                 if (rc == -ENOSPC && overflow_retry--) {
2278                         ars_status = acpi_desc->ars_status;
2279                         /*
2280                          * Record the original scrub range, so that we
2281                          * can recall all the ranges impacted by the
2282                          * initial scrub.
2283                          */
2284                         if (!init_scrub_length) {
2285                                 init_scrub_length = ars_status->length;
2286                                 init_scrub_address = ars_status->address;
2287                         }
2288                         rc = ars_continue(acpi_desc);
2289                         if (rc == 0) {
2290                                 mutex_unlock(&acpi_desc->init_mutex);
2291                                 goto retry;
2292                         }
2293                 }
2294
2295                 if (rc < 0) {
2296                         /*
2297                          * Initial scrub failed, we'll give it one more
2298                          * try below...
2299                          */
2300                         break;
2301                 }
2302
2303                 /* We got some final results, record completed ranges */
2304                 ars_status = acpi_desc->ars_status;
2305                 if (init_scrub_length) {
2306                         ars_start = init_scrub_address;
2307                         ars_len = ars_start + init_scrub_length;
2308                 } else {
2309                         ars_start = ars_status->address;
2310                         ars_len = ars_status->length;
2311                 }
2312                 spa = nfit_spa->spa;
2313
2314                 if (!init_ars_done) {
2315                         init_ars_done = true;
2316                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2317                                         ars_start, ars_len);
2318                 }
2319                 if (ars_start <= spa->address && ars_start + ars_len
2320                                 >= spa->address + spa->length)
2321                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2322         }
2323
2324         /*
2325          * For all the ranges not covered by an initial scrub we still
2326          * want to see if there are errors, but it's ok to discover them
2327          * asynchronously.
2328          */
2329         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2330                 /*
2331                  * Flag all the ranges that still need scrubbing, but
2332                  * register them now to make data available.
2333                  */
2334                 if (nfit_spa->nd_region)
2335                         nfit_spa->ars_done = 1;
2336                 else
2337                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2338         }
2339
2340         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2341                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2342         mutex_unlock(&acpi_desc->init_mutex);
2343 }
2344
2345 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2346 {
2347         struct nfit_spa *nfit_spa;
2348         int rc;
2349
2350         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2351                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2352                         /* BLK regions don't need to wait for ars results */
2353                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2354                         if (rc)
2355                                 return rc;
2356                 }
2357
2358         queue_work(nfit_wq, &acpi_desc->work);
2359         return 0;
2360 }
2361
2362 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2363                 struct nfit_table_prev *prev)
2364 {
2365         struct device *dev = acpi_desc->dev;
2366
2367         if (!list_empty(&prev->spas) ||
2368                         !list_empty(&prev->memdevs) ||
2369                         !list_empty(&prev->dcrs) ||
2370                         !list_empty(&prev->bdws) ||
2371                         !list_empty(&prev->idts) ||
2372                         !list_empty(&prev->flushes)) {
2373                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2374                 return -ENXIO;
2375         }
2376         return 0;
2377 }
2378
2379 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2380 {
2381         struct device *dev = acpi_desc->dev;
2382         struct nfit_table_prev prev;
2383         const void *end;
2384         u8 *data;
2385         int rc;
2386
2387         mutex_lock(&acpi_desc->init_mutex);
2388
2389         INIT_LIST_HEAD(&prev.spas);
2390         INIT_LIST_HEAD(&prev.memdevs);
2391         INIT_LIST_HEAD(&prev.dcrs);
2392         INIT_LIST_HEAD(&prev.bdws);
2393         INIT_LIST_HEAD(&prev.idts);
2394         INIT_LIST_HEAD(&prev.flushes);
2395
2396         list_cut_position(&prev.spas, &acpi_desc->spas,
2397                                 acpi_desc->spas.prev);
2398         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2399                                 acpi_desc->memdevs.prev);
2400         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2401                                 acpi_desc->dcrs.prev);
2402         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2403                                 acpi_desc->bdws.prev);
2404         list_cut_position(&prev.idts, &acpi_desc->idts,
2405                                 acpi_desc->idts.prev);
2406         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2407                                 acpi_desc->flushes.prev);
2408
2409         data = (u8 *) acpi_desc->nfit;
2410         end = data + sz;
2411         while (!IS_ERR_OR_NULL(data))
2412                 data = add_table(acpi_desc, &prev, data, end);
2413
2414         if (IS_ERR(data)) {
2415                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2416                                 PTR_ERR(data));
2417                 rc = PTR_ERR(data);
2418                 goto out_unlock;
2419         }
2420
2421         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2422         if (rc)
2423                 goto out_unlock;
2424
2425         if (nfit_mem_init(acpi_desc) != 0) {
2426                 rc = -ENOMEM;
2427                 goto out_unlock;
2428         }
2429
2430         acpi_nfit_init_dsms(acpi_desc);
2431
2432         rc = acpi_nfit_register_dimms(acpi_desc);
2433         if (rc)
2434                 goto out_unlock;
2435
2436         rc = acpi_nfit_register_regions(acpi_desc);
2437
2438  out_unlock:
2439         mutex_unlock(&acpi_desc->init_mutex);
2440         return rc;
2441 }
2442 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2443
2444 struct acpi_nfit_flush_work {
2445         struct work_struct work;
2446         struct completion cmp;
2447 };
2448
2449 static void flush_probe(struct work_struct *work)
2450 {
2451         struct acpi_nfit_flush_work *flush;
2452
2453         flush = container_of(work, typeof(*flush), work);
2454         complete(&flush->cmp);
2455 }
2456
2457 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2458 {
2459         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2460         struct device *dev = acpi_desc->dev;
2461         struct acpi_nfit_flush_work flush;
2462
2463         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2464         device_lock(dev);
2465         device_unlock(dev);
2466
2467         /*
2468          * Scrub work could take 10s of seconds, userspace may give up so we
2469          * need to be interruptible while waiting.
2470          */
2471         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2472         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2473         queue_work(nfit_wq, &flush.work);
2474         return wait_for_completion_interruptible(&flush.cmp);
2475 }
2476
2477 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2478                 struct nvdimm *nvdimm, unsigned int cmd)
2479 {
2480         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2481
2482         if (nvdimm)
2483                 return 0;
2484         if (cmd != ND_CMD_ARS_START)
2485                 return 0;
2486
2487         /*
2488          * The kernel and userspace may race to initiate a scrub, but
2489          * the scrub thread is prepared to lose that initial race.  It
2490          * just needs guarantees that any ars it initiates are not
2491          * interrupted by any intervening start reqeusts from userspace.
2492          */
2493         if (work_busy(&acpi_desc->work))
2494                 return -EBUSY;
2495
2496         return 0;
2497 }
2498
2499 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2500 {
2501         struct nvdimm_bus_descriptor *nd_desc;
2502
2503         dev_set_drvdata(dev, acpi_desc);
2504         acpi_desc->dev = dev;
2505         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2506         nd_desc = &acpi_desc->nd_desc;
2507         nd_desc->provider_name = "ACPI.NFIT";
2508         nd_desc->ndctl = acpi_nfit_ctl;
2509         nd_desc->flush_probe = acpi_nfit_flush_probe;
2510         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2511         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2512
2513         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2514         INIT_LIST_HEAD(&acpi_desc->spas);
2515         INIT_LIST_HEAD(&acpi_desc->dcrs);
2516         INIT_LIST_HEAD(&acpi_desc->bdws);
2517         INIT_LIST_HEAD(&acpi_desc->idts);
2518         INIT_LIST_HEAD(&acpi_desc->flushes);
2519         INIT_LIST_HEAD(&acpi_desc->memdevs);
2520         INIT_LIST_HEAD(&acpi_desc->dimms);
2521         mutex_init(&acpi_desc->spa_map_mutex);
2522         mutex_init(&acpi_desc->init_mutex);
2523         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2524 }
2525 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2526
2527 static int acpi_nfit_add(struct acpi_device *adev)
2528 {
2529         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2530         struct acpi_nfit_desc *acpi_desc;
2531         struct device *dev = &adev->dev;
2532         struct acpi_table_header *tbl;
2533         acpi_status status = AE_OK;
2534         acpi_size sz;
2535         int rc;
2536
2537         status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2538         if (ACPI_FAILURE(status)) {
2539                 /* This is ok, we could have an nvdimm hotplugged later */
2540                 dev_dbg(dev, "failed to find NFIT at startup\n");
2541                 return 0;
2542         }
2543
2544         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2545         if (!acpi_desc)
2546                 return -ENOMEM;
2547         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2548         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2549         if (!acpi_desc->nvdimm_bus)
2550                 return -ENOMEM;
2551
2552         /*
2553          * Save the acpi header for later and then skip it,
2554          * making nfit point to the first nfit table header.
2555          */
2556         acpi_desc->acpi_header = *tbl;
2557         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2558         sz -= sizeof(struct acpi_table_nfit);
2559
2560         /* Evaluate _FIT and override with that if present */
2561         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2562         if (ACPI_SUCCESS(status) && buf.length > 0) {
2563                 union acpi_object *obj;
2564                 /*
2565                  * Adjust for the acpi_object header of the _FIT
2566                  */
2567                 obj = buf.pointer;
2568                 if (obj->type == ACPI_TYPE_BUFFER) {
2569                         acpi_desc->nfit =
2570                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2571                         sz = obj->buffer.length;
2572                 } else
2573                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2574                                  __func__, (int) obj->type);
2575         }
2576
2577         rc = acpi_nfit_init(acpi_desc, sz);
2578         if (rc) {
2579                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2580                 return rc;
2581         }
2582         return 0;
2583 }
2584
2585 static int acpi_nfit_remove(struct acpi_device *adev)
2586 {
2587         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2588
2589         acpi_desc->cancel = 1;
2590         flush_workqueue(nfit_wq);
2591         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2592         return 0;
2593 }
2594
2595 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2596 {
2597         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2598         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2599         struct acpi_nfit_header *nfit_saved;
2600         union acpi_object *obj;
2601         struct device *dev = &adev->dev;
2602         acpi_status status;
2603         int ret;
2604
2605         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2606
2607         device_lock(dev);
2608         if (!dev->driver) {
2609                 /* dev->driver may be null if we're being removed */
2610                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2611                 goto out_unlock;
2612         }
2613
2614         if (!acpi_desc) {
2615                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2616                 if (!acpi_desc)
2617                         goto out_unlock;
2618                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2619                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2620                 if (!acpi_desc->nvdimm_bus)
2621                         goto out_unlock;
2622         } else {
2623                 /*
2624                  * Finish previous registration before considering new
2625                  * regions.
2626                  */
2627                 flush_workqueue(nfit_wq);
2628         }
2629
2630         /* Evaluate _FIT */
2631         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2632         if (ACPI_FAILURE(status)) {
2633                 dev_err(dev, "failed to evaluate _FIT\n");
2634                 goto out_unlock;
2635         }
2636
2637         nfit_saved = acpi_desc->nfit;
2638         obj = buf.pointer;
2639         if (obj->type == ACPI_TYPE_BUFFER) {
2640                 acpi_desc->nfit =
2641                         (struct acpi_nfit_header *)obj->buffer.pointer;
2642                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2643                 if (ret) {
2644                         /* Merge failed, restore old nfit, and exit */
2645                         acpi_desc->nfit = nfit_saved;
2646                         dev_err(dev, "failed to merge updated NFIT\n");
2647                 }
2648         } else {
2649                 /* Bad _FIT, restore old nfit */
2650                 dev_err(dev, "Invalid _FIT\n");
2651         }
2652         kfree(buf.pointer);
2653
2654  out_unlock:
2655         device_unlock(dev);
2656 }
2657
2658 static const struct acpi_device_id acpi_nfit_ids[] = {
2659         { "ACPI0012", 0 },
2660         { "", 0 },
2661 };
2662 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2663
2664 static struct acpi_driver acpi_nfit_driver = {
2665         .name = KBUILD_MODNAME,
2666         .ids = acpi_nfit_ids,
2667         .ops = {
2668                 .add = acpi_nfit_add,
2669                 .remove = acpi_nfit_remove,
2670                 .notify = acpi_nfit_notify,
2671         },
2672 };
2673
2674 static __init int nfit_init(void)
2675 {
2676         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2677         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2678         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2679         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2680         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2681         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2682         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2683
2684         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2685         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2686         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2687         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2688         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2689         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2690         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2691         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2692         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2693         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2694         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2695         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2696
2697         nfit_wq = create_singlethread_workqueue("nfit");
2698         if (!nfit_wq)
2699                 return -ENOMEM;
2700
2701         return acpi_bus_register_driver(&acpi_nfit_driver);
2702 }
2703
2704 static __exit void nfit_exit(void)
2705 {
2706         acpi_bus_unregister_driver(&acpi_nfit_driver);
2707         destroy_workqueue(nfit_wq);
2708 }
2709
2710 module_init(nfit_init);
2711 module_exit(nfit_exit);
2712 MODULE_LICENSE("GPL v2");
2713 MODULE_AUTHOR("Intel Corporation");