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", be16_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, "%#x\n",
 965                                         be16_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_err(dev, "unknown dimm command family\n");
1155                 nfit_mem->family = -1;
1156                 return force_enable_dimms ? 0 : -ENODEV;
1157         }
1158
1159         uuid = to_nfit_uuid(nfit_mem->family);
1160         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1161                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1162                         set_bit(i, &nfit_mem->dsm_mask);
1163
1164         return 0;
1165 }
1166
1167 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1168 {
1169         struct nfit_mem *nfit_mem;
1170         int dimm_count = 0;
1171
1172         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1173                 unsigned long flags = 0, cmd_mask;
1174                 struct nvdimm *nvdimm;
1175                 u32 device_handle;
1176                 u16 mem_flags;
1177                 int rc;
1178
1179                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1180                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1181                 if (nvdimm) {
1182                         dimm_count++;
1183                         continue;
1184                 }
1185
1186                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1187                         flags |= NDD_ALIASING;
1188
1189                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1190                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1191                         flags |= NDD_UNARMED;
1192
1193                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1194                 if (rc)
1195                         continue;
1196
1197                 /*
1198                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1199                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1200                  * userspace interface.
1201                  */
1202                 cmd_mask = 1UL << ND_CMD_CALL;
1203                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1204                         cmd_mask |= nfit_mem->dsm_mask;
1205
1206                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1207                                 acpi_nfit_dimm_attribute_groups,
1208                                 flags, cmd_mask);
1209                 if (!nvdimm)
1210                         return -ENOMEM;
1211
1212                 nfit_mem->nvdimm = nvdimm;
1213                 dimm_count++;
1214
1215                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1216                         continue;
1217
1218                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1219                                 nvdimm_name(nvdimm),
1220                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1221                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1222                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1223                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1224
1225         }
1226
1227         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1228 }
1229
1230 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1231 {
1232         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1233         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1234         struct acpi_device *adev;
1235         int i;
1236
1237         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1238         adev = to_acpi_dev(acpi_desc);
1239         if (!adev)
1240                 return;
1241
1242         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1243                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1244                         set_bit(i, &nd_desc->cmd_mask);
1245 }
1246
1247 static ssize_t range_index_show(struct device *dev,
1248                 struct device_attribute *attr, char *buf)
1249 {
1250         struct nd_region *nd_region = to_nd_region(dev);
1251         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1252
1253         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1254 }
1255 static DEVICE_ATTR_RO(range_index);
1256
1257 static struct attribute *acpi_nfit_region_attributes[] = {
1258         &dev_attr_range_index.attr,
1259         NULL,
1260 };
1261
1262 static struct attribute_group acpi_nfit_region_attribute_group = {
1263         .name = "nfit",
1264         .attrs = acpi_nfit_region_attributes,
1265 };
1266
1267 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1268         &nd_region_attribute_group,
1269         &nd_mapping_attribute_group,
1270         &nd_device_attribute_group,
1271         &nd_numa_attribute_group,
1272         &acpi_nfit_region_attribute_group,
1273         NULL,
1274 };
1275
1276 /* enough info to uniquely specify an interleave set */
1277 struct nfit_set_info {
1278         struct nfit_set_info_map {
1279                 u64 region_offset;
1280                 u32 serial_number;
1281                 u32 pad;
1282         } mapping[0];
1283 };
1284
1285 static size_t sizeof_nfit_set_info(int num_mappings)
1286 {
1287         return sizeof(struct nfit_set_info)
1288                 + num_mappings * sizeof(struct nfit_set_info_map);
1289 }
1290
1291 static int cmp_map(const void *m0, const void *m1)
1292 {
1293         const struct nfit_set_info_map *map0 = m0;
1294         const struct nfit_set_info_map *map1 = m1;
1295
1296         return memcmp(&map0->region_offset, &map1->region_offset,
1297                         sizeof(u64));
1298 }
1299
1300 /* Retrieve the nth entry referencing this spa */
1301 static struct acpi_nfit_memory_map *memdev_from_spa(
1302                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1303 {
1304         struct nfit_memdev *nfit_memdev;
1305
1306         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1307                 if (nfit_memdev->memdev->range_index == range_index)
1308                         if (n-- == 0)
1309                                 return nfit_memdev->memdev;
1310         return NULL;
1311 }
1312
1313 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1314                 struct nd_region_desc *ndr_desc,
1315                 struct acpi_nfit_system_address *spa)
1316 {
1317         int i, spa_type = nfit_spa_type(spa);
1318         struct device *dev = acpi_desc->dev;
1319         struct nd_interleave_set *nd_set;
1320         u16 nr = ndr_desc->num_mappings;
1321         struct nfit_set_info *info;
1322
1323         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1324                 /* pass */;
1325         else
1326                 return 0;
1327
1328         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1329         if (!nd_set)
1330                 return -ENOMEM;
1331
1332         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1333         if (!info)
1334                 return -ENOMEM;
1335         for (i = 0; i < nr; i++) {
1336                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1337                 struct nfit_set_info_map *map = &info->mapping[i];
1338                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1339                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1340                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1341                                 spa->range_index, i);
1342
1343                 if (!memdev || !nfit_mem->dcr) {
1344                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1345                         return -ENODEV;
1346                 }
1347
1348                 map->region_offset = memdev->region_offset;
1349                 map->serial_number = nfit_mem->dcr->serial_number;
1350         }
1351
1352         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1353                         cmp_map, NULL);
1354         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1355         ndr_desc->nd_set = nd_set;
1356         devm_kfree(dev, info);
1357
1358         return 0;
1359 }
1360
1361 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1362 {
1363         struct acpi_nfit_interleave *idt = mmio->idt;
1364         u32 sub_line_offset, line_index, line_offset;
1365         u64 line_no, table_skip_count, table_offset;
1366
1367         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1368         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1369         line_offset = idt->line_offset[line_index]
1370                 * mmio->line_size;
1371         table_offset = table_skip_count * mmio->table_size;
1372
1373         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1374 }
1375
1376 static void wmb_blk(struct nfit_blk *nfit_blk)
1377 {
1378
1379         if (nfit_blk->nvdimm_flush) {
1380                 /*
1381                  * The first wmb() is needed to 'sfence' all previous writes
1382                  * such that they are architecturally visible for the platform
1383                  * buffer flush.  Note that we've already arranged for pmem
1384                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1385                  * final wmb() ensures ordering for the NVDIMM flush write.
1386                  */
1387                 wmb();
1388                 writeq(1, nfit_blk->nvdimm_flush);
1389                 wmb();
1390         } else
1391                 wmb_pmem();
1392 }
1393
1394 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1395 {
1396         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1397         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1398
1399         if (mmio->num_lines)
1400                 offset = to_interleave_offset(offset, mmio);
1401
1402         return readl(mmio->addr.base + offset);
1403 }
1404
1405 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1406                 resource_size_t dpa, unsigned int len, unsigned int write)
1407 {
1408         u64 cmd, offset;
1409         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1410
1411         enum {
1412                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1413                 BCW_LEN_SHIFT = 48,
1414                 BCW_LEN_MASK = (1ULL << 8) - 1,
1415                 BCW_CMD_SHIFT = 56,
1416         };
1417
1418         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1419         len = len >> L1_CACHE_SHIFT;
1420         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1421         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1422
1423         offset = nfit_blk->cmd_offset + mmio->size * bw;
1424         if (mmio->num_lines)
1425                 offset = to_interleave_offset(offset, mmio);
1426
1427         writeq(cmd, mmio->addr.base + offset);
1428         wmb_blk(nfit_blk);
1429
1430         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1431                 readq(mmio->addr.base + offset);
1432 }
1433
1434 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1435                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1436                 unsigned int lane)
1437 {
1438         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1439         unsigned int copied = 0;
1440         u64 base_offset;
1441         int rc;
1442
1443         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1444                 + lane * mmio->size;
1445         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1446         while (len) {
1447                 unsigned int c;
1448                 u64 offset;
1449
1450                 if (mmio->num_lines) {
1451                         u32 line_offset;
1452
1453                         offset = to_interleave_offset(base_offset + copied,
1454                                         mmio);
1455                         div_u64_rem(offset, mmio->line_size, &line_offset);
1456                         c = min_t(size_t, len, mmio->line_size - line_offset);
1457                 } else {
1458                         offset = base_offset + nfit_blk->bdw_offset;
1459                         c = len;
1460                 }
1461
1462                 if (rw)
1463                         memcpy_to_pmem(mmio->addr.aperture + offset,
1464                                         iobuf + copied, c);
1465                 else {
1466                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1467                                 mmio_flush_range((void __force *)
1468                                         mmio->addr.aperture + offset, c);
1469
1470                         memcpy_from_pmem(iobuf + copied,
1471                                         mmio->addr.aperture + offset, c);
1472                 }
1473
1474                 copied += c;
1475                 len -= c;
1476         }
1477
1478         if (rw)
1479                 wmb_blk(nfit_blk);
1480
1481         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1482         return rc;
1483 }
1484
1485 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1486                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1487 {
1488         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1489         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1490         struct nd_region *nd_region = nfit_blk->nd_region;
1491         unsigned int lane, copied = 0;
1492         int rc = 0;
1493
1494         lane = nd_region_acquire_lane(nd_region);
1495         while (len) {
1496                 u64 c = min(len, mmio->size);
1497
1498                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1499                                 iobuf + copied, c, rw, lane);
1500                 if (rc)
1501                         break;
1502
1503                 copied += c;
1504                 len -= c;
1505         }
1506         nd_region_release_lane(nd_region, lane);
1507
1508         return rc;
1509 }
1510
1511 static void nfit_spa_mapping_release(struct kref *kref)
1512 {
1513         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1514         struct acpi_nfit_system_address *spa = spa_map->spa;
1515         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1516
1517         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1518         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1519         if (spa_map->type == SPA_MAP_APERTURE)
1520                 memunmap((void __force *)spa_map->addr.aperture);
1521         else
1522                 iounmap(spa_map->addr.base);
1523         release_mem_region(spa->address, spa->length);
1524         list_del(&spa_map->list);
1525         kfree(spa_map);
1526 }
1527
1528 static struct nfit_spa_mapping *find_spa_mapping(
1529                 struct acpi_nfit_desc *acpi_desc,
1530                 struct acpi_nfit_system_address *spa)
1531 {
1532         struct nfit_spa_mapping *spa_map;
1533
1534         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1535         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1536                 if (spa_map->spa == spa)
1537                         return spa_map;
1538
1539         return NULL;
1540 }
1541
1542 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1543                 struct acpi_nfit_system_address *spa)
1544 {
1545         struct nfit_spa_mapping *spa_map;
1546
1547         mutex_lock(&acpi_desc->spa_map_mutex);
1548         spa_map = find_spa_mapping(acpi_desc, spa);
1549
1550         if (spa_map)
1551                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1552         mutex_unlock(&acpi_desc->spa_map_mutex);
1553 }
1554
1555 static void __iomem *__nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1556                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1557 {
1558         resource_size_t start = spa->address;
1559         resource_size_t n = spa->length;
1560         struct nfit_spa_mapping *spa_map;
1561         struct resource *res;
1562
1563         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1564
1565         spa_map = find_spa_mapping(acpi_desc, spa);
1566         if (spa_map) {
1567                 kref_get(&spa_map->kref);
1568                 return spa_map->addr.base;
1569         }
1570
1571         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1572         if (!spa_map)
1573                 return NULL;
1574
1575         INIT_LIST_HEAD(&spa_map->list);
1576         spa_map->spa = spa;
1577         kref_init(&spa_map->kref);
1578         spa_map->acpi_desc = acpi_desc;
1579
1580         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1581         if (!res)
1582                 goto err_mem;
1583
1584         spa_map->type = type;
1585         if (type == SPA_MAP_APERTURE)
1586                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1587                                                         ARCH_MEMREMAP_PMEM);
1588         else
1589                 spa_map->addr.base = ioremap_nocache(start, n);
1590
1591
1592         if (!spa_map->addr.base)
1593                 goto err_map;
1594
1595         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1596         return spa_map->addr.base;
1597
1598  err_map:
1599         release_mem_region(start, n);
1600  err_mem:
1601         kfree(spa_map);
1602         return NULL;
1603 }
1604
1605 /**
1606  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1607  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1608  * @nfit_spa: spa table to map
1609  * @type: aperture or control region
1610  *
1611  * In the case where block-data-window apertures and
1612  * dimm-control-regions are interleaved they will end up sharing a
1613  * single request_mem_region() + ioremap() for the address range.  In
1614  * the style of devm nfit_spa_map() mappings are automatically dropped
1615  * when all region devices referencing the same mapping are disabled /
1616  * unbound.
1617  */
1618 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1619                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1620 {
1621         void __iomem *iomem;
1622
1623         mutex_lock(&acpi_desc->spa_map_mutex);
1624         iomem = __nfit_spa_map(acpi_desc, spa, type);
1625         mutex_unlock(&acpi_desc->spa_map_mutex);
1626
1627         return iomem;
1628 }
1629
1630 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1631                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1632 {
1633         if (idt) {
1634                 mmio->num_lines = idt->line_count;
1635                 mmio->line_size = idt->line_size;
1636                 if (interleave_ways == 0)
1637                         return -ENXIO;
1638                 mmio->table_size = mmio->num_lines * interleave_ways
1639                         * mmio->line_size;
1640         }
1641
1642         return 0;
1643 }
1644
1645 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1646                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1647 {
1648         struct nd_cmd_dimm_flags flags;
1649         int rc;
1650
1651         memset(&flags, 0, sizeof(flags));
1652         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1653                         sizeof(flags), NULL);
1654
1655         if (rc >= 0 && flags.status == 0)
1656                 nfit_blk->dimm_flags = flags.flags;
1657         else if (rc == -ENOTTY) {
1658                 /* fall back to a conservative default */
1659                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1660                 rc = 0;
1661         } else
1662                 rc = -ENXIO;
1663
1664         return rc;
1665 }
1666
1667 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1668                 struct device *dev)
1669 {
1670         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1671         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1672         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1673         struct nfit_flush *nfit_flush;
1674         struct nfit_blk_mmio *mmio;
1675         struct nfit_blk *nfit_blk;
1676         struct nfit_mem *nfit_mem;
1677         struct nvdimm *nvdimm;
1678         int rc;
1679
1680         nvdimm = nd_blk_region_to_dimm(ndbr);
1681         nfit_mem = nvdimm_provider_data(nvdimm);
1682         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1683                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1684                                 nfit_mem ? "" : " nfit_mem",
1685                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1686                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1687                 return -ENXIO;
1688         }
1689
1690         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1691         if (!nfit_blk)
1692                 return -ENOMEM;
1693         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1694         nfit_blk->nd_region = to_nd_region(dev);
1695
1696         /* map block aperture memory */
1697         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1698         mmio = &nfit_blk->mmio[BDW];
1699         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1700                         SPA_MAP_APERTURE);
1701         if (!mmio->addr.base) {
1702                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1703                                 nvdimm_name(nvdimm));
1704                 return -ENOMEM;
1705         }
1706         mmio->size = nfit_mem->bdw->size;
1707         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1708         mmio->idt = nfit_mem->idt_bdw;
1709         mmio->spa = nfit_mem->spa_bdw;
1710         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1711                         nfit_mem->memdev_bdw->interleave_ways);
1712         if (rc) {
1713                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1714                                 __func__, nvdimm_name(nvdimm));
1715                 return rc;
1716         }
1717
1718         /* map block control memory */
1719         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1720         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1721         mmio = &nfit_blk->mmio[DCR];
1722         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1723                         SPA_MAP_CONTROL);
1724         if (!mmio->addr.base) {
1725                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1726                                 nvdimm_name(nvdimm));
1727                 return -ENOMEM;
1728         }
1729         mmio->size = nfit_mem->dcr->window_size;
1730         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1731         mmio->idt = nfit_mem->idt_dcr;
1732         mmio->spa = nfit_mem->spa_dcr;
1733         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1734                         nfit_mem->memdev_dcr->interleave_ways);
1735         if (rc) {
1736                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1737                                 __func__, nvdimm_name(nvdimm));
1738                 return rc;
1739         }
1740
1741         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1742         if (rc < 0) {
1743                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1744                                 __func__, nvdimm_name(nvdimm));
1745                 return rc;
1746         }
1747
1748         nfit_flush = nfit_mem->nfit_flush;
1749         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1750                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1751                                 nfit_flush->flush->hint_address[0], 8);
1752                 if (!nfit_blk->nvdimm_flush)
1753                         return -ENOMEM;
1754         }
1755
1756         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1757                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1758
1759         if (mmio->line_size == 0)
1760                 return 0;
1761
1762         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1763                         + 8 > mmio->line_size) {
1764                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1765                 return -ENXIO;
1766         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1767                         + 8 > mmio->line_size) {
1768                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1769                 return -ENXIO;
1770         }
1771
1772         return 0;
1773 }
1774
1775 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1776                 struct device *dev)
1777 {
1778         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1779         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1780         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1781         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1782         int i;
1783
1784         if (!nfit_blk)
1785                 return; /* never enabled */
1786
1787         /* auto-free BLK spa mappings */
1788         for (i = 0; i < 2; i++) {
1789                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1790
1791                 if (mmio->addr.base)
1792                         nfit_spa_unmap(acpi_desc, mmio->spa);
1793         }
1794         nd_blk_region_set_provider_data(ndbr, NULL);
1795         /* devm will free nfit_blk */
1796 }
1797
1798 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1799                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1800 {
1801         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1802         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1803         int cmd_rc, rc;
1804
1805         cmd->address = spa->address;
1806         cmd->length = spa->length;
1807         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1808                         sizeof(*cmd), &cmd_rc);
1809         if (rc < 0)
1810                 return rc;
1811         return cmd_rc;
1812 }
1813
1814 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1815 {
1816         int rc;
1817         int cmd_rc;
1818         struct nd_cmd_ars_start ars_start;
1819         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1820         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1821
1822         memset(&ars_start, 0, sizeof(ars_start));
1823         ars_start.address = spa->address;
1824         ars_start.length = spa->length;
1825         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1826                 ars_start.type = ND_ARS_PERSISTENT;
1827         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1828                 ars_start.type = ND_ARS_VOLATILE;
1829         else
1830                 return -ENOTTY;
1831
1832         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1833                         sizeof(ars_start), &cmd_rc);
1834
1835         if (rc < 0)
1836                 return rc;
1837         return cmd_rc;
1838 }
1839
1840 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1841 {
1842         int rc, cmd_rc;
1843         struct nd_cmd_ars_start ars_start;
1844         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1845         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1846
1847         memset(&ars_start, 0, sizeof(ars_start));
1848         ars_start.address = ars_status->restart_address;
1849         ars_start.length = ars_status->restart_length;
1850         ars_start.type = ars_status->type;
1851         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1852                         sizeof(ars_start), &cmd_rc);
1853         if (rc < 0)
1854                 return rc;
1855         return cmd_rc;
1856 }
1857
1858 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1859 {
1860         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1861         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1862         int rc, cmd_rc;
1863
1864         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1865                         acpi_desc->ars_status_size, &cmd_rc);
1866         if (rc < 0)
1867                 return rc;
1868         return cmd_rc;
1869 }
1870
1871 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1872                 struct nd_cmd_ars_status *ars_status)
1873 {
1874         int rc;
1875         u32 i;
1876
1877         for (i = 0; i < ars_status->num_records; i++) {
1878                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1879                                 ars_status->records[i].err_address,
1880                                 ars_status->records[i].length);
1881                 if (rc)
1882                         return rc;
1883         }
1884
1885         return 0;
1886 }
1887
1888 static void acpi_nfit_remove_resource(void *data)
1889 {
1890         struct resource *res = data;
1891
1892         remove_resource(res);
1893 }
1894
1895 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1896                 struct nd_region_desc *ndr_desc)
1897 {
1898         struct resource *res, *nd_res = ndr_desc->res;
1899         int is_pmem, ret;
1900
1901         /* No operation if the region is already registered as PMEM */
1902         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1903                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1904         if (is_pmem == REGION_INTERSECTS)
1905                 return 0;
1906
1907         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1908         if (!res)
1909                 return -ENOMEM;
1910
1911         res->name = "Persistent Memory";
1912         res->start = nd_res->start;
1913         res->end = nd_res->end;
1914         res->flags = IORESOURCE_MEM;
1915         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1916
1917         ret = insert_resource(&iomem_resource, res);
1918         if (ret)
1919                 return ret;
1920
1921         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1922         if (ret) {
1923                 remove_resource(res);
1924                 return ret;
1925         }
1926
1927         return 0;
1928 }
1929
1930 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1931                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1932                 struct acpi_nfit_memory_map *memdev,
1933                 struct nfit_spa *nfit_spa)
1934 {
1935         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1936                         memdev->device_handle);
1937         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1938         struct nd_blk_region_desc *ndbr_desc;
1939         struct nfit_mem *nfit_mem;
1940         int blk_valid = 0;
1941
1942         if (!nvdimm) {
1943                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1944                                 spa->range_index, memdev->device_handle);
1945                 return -ENODEV;
1946         }
1947
1948         nd_mapping->nvdimm = nvdimm;
1949         switch (nfit_spa_type(spa)) {
1950         case NFIT_SPA_PM:
1951         case NFIT_SPA_VOLATILE:
1952                 nd_mapping->start = memdev->address;
1953                 nd_mapping->size = memdev->region_size;
1954                 break;
1955         case NFIT_SPA_DCR:
1956                 nfit_mem = nvdimm_provider_data(nvdimm);
1957                 if (!nfit_mem || !nfit_mem->bdw) {
1958                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1959                                         spa->range_index, nvdimm_name(nvdimm));
1960                 } else {
1961                         nd_mapping->size = nfit_mem->bdw->capacity;
1962                         nd_mapping->start = nfit_mem->bdw->start_address;
1963                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1964                         blk_valid = 1;
1965                 }
1966
1967                 ndr_desc->nd_mapping = nd_mapping;
1968                 ndr_desc->num_mappings = blk_valid;
1969                 ndbr_desc = to_blk_region_desc(ndr_desc);
1970                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1971                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1972                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1973                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1974                                 ndr_desc);
1975                 if (!nfit_spa->nd_region)
1976                         return -ENOMEM;
1977                 break;
1978         }
1979
1980         return 0;
1981 }
1982
1983 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1984                 struct nfit_spa *nfit_spa)
1985 {
1986         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1987         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1988         struct nd_blk_region_desc ndbr_desc;
1989         struct nd_region_desc *ndr_desc;
1990         struct nfit_memdev *nfit_memdev;
1991         struct nvdimm_bus *nvdimm_bus;
1992         struct resource res;
1993         int count = 0, rc;
1994
1995         if (nfit_spa->nd_region)
1996                 return 0;
1997
1998         if (spa->range_index == 0) {
1999                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2000                                 __func__);
2001                 return 0;
2002         }
2003
2004         memset(&res, 0, sizeof(res));
2005         memset(&nd_mappings, 0, sizeof(nd_mappings));
2006         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2007         res.start = spa->address;
2008         res.end = res.start + spa->length - 1;
2009         ndr_desc = &ndbr_desc.ndr_desc;
2010         ndr_desc->res = &res;
2011         ndr_desc->provider_data = nfit_spa;
2012         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2013         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2014                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2015                                                 spa->proximity_domain);
2016         else
2017                 ndr_desc->numa_node = NUMA_NO_NODE;
2018
2019         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2020                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2021                 struct nd_mapping *nd_mapping;
2022
2023                 if (memdev->range_index != spa->range_index)
2024                         continue;
2025                 if (count >= ND_MAX_MAPPINGS) {
2026                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2027                                         spa->range_index, ND_MAX_MAPPINGS);
2028                         return -ENXIO;
2029                 }
2030                 nd_mapping = &nd_mappings[count++];
2031                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
2032                                 memdev, nfit_spa);
2033                 if (rc)
2034                         goto out;
2035         }
2036
2037         ndr_desc->nd_mapping = nd_mappings;
2038         ndr_desc->num_mappings = count;
2039         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2040         if (rc)
2041                 goto out;
2042
2043         nvdimm_bus = acpi_desc->nvdimm_bus;
2044         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2045                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2046                 if (rc) {
2047                         dev_warn(acpi_desc->dev,
2048                                 "failed to insert pmem resource to iomem: %d\n",
2049                                 rc);
2050                         goto out;
2051                 }
2052
2053                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2054                                 ndr_desc);
2055                 if (!nfit_spa->nd_region)
2056                         rc = -ENOMEM;
2057         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2058                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2059                                 ndr_desc);
2060                 if (!nfit_spa->nd_region)
2061                         rc = -ENOMEM;
2062         }
2063
2064  out:
2065         if (rc)
2066                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2067                                 nfit_spa->spa->range_index);
2068         return rc;
2069 }
2070
2071 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2072                 u32 max_ars)
2073 {
2074         struct device *dev = acpi_desc->dev;
2075         struct nd_cmd_ars_status *ars_status;
2076
2077         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2078                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2079                 return 0;
2080         }
2081
2082         if (acpi_desc->ars_status)
2083                 devm_kfree(dev, acpi_desc->ars_status);
2084         acpi_desc->ars_status = NULL;
2085         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2086         if (!ars_status)
2087                 return -ENOMEM;
2088         acpi_desc->ars_status = ars_status;
2089         acpi_desc->ars_status_size = max_ars;
2090         return 0;
2091 }
2092
2093 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2094                 struct nfit_spa *nfit_spa)
2095 {
2096         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2097         int rc;
2098
2099         if (!nfit_spa->max_ars) {
2100                 struct nd_cmd_ars_cap ars_cap;
2101
2102                 memset(&ars_cap, 0, sizeof(ars_cap));
2103                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2104                 if (rc < 0)
2105                         return rc;
2106                 nfit_spa->max_ars = ars_cap.max_ars_out;
2107                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2108                 /* check that the supported scrub types match the spa type */
2109                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2110                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2111                         return -ENOTTY;
2112                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2113                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2114                         return -ENOTTY;
2115         }
2116
2117         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2118                 return -ENOMEM;
2119
2120         rc = ars_get_status(acpi_desc);
2121         if (rc < 0 && rc != -ENOSPC)
2122                 return rc;
2123
2124         if (ars_status_process_records(acpi_desc->nvdimm_bus,
2125                                 acpi_desc->ars_status))
2126                 return -ENOMEM;
2127
2128         return 0;
2129 }
2130
2131 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2132                 struct nfit_spa *nfit_spa)
2133 {
2134         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2135         unsigned int overflow_retry = scrub_overflow_abort;
2136         u64 init_ars_start = 0, init_ars_len = 0;
2137         struct device *dev = acpi_desc->dev;
2138         unsigned int tmo = scrub_timeout;
2139         int rc;
2140
2141         if (nfit_spa->ars_done || !nfit_spa->nd_region)
2142                 return;
2143
2144         rc = ars_start(acpi_desc, nfit_spa);
2145         /*
2146          * If we timed out the initial scan we'll still be busy here,
2147          * and will wait another timeout before giving up permanently.
2148          */
2149         if (rc < 0 && rc != -EBUSY)
2150                 return;
2151
2152         do {
2153                 u64 ars_start, ars_len;
2154
2155                 if (acpi_desc->cancel)
2156                         break;
2157                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2158                 if (rc == -ENOTTY)
2159                         break;
2160                 if (rc == -EBUSY && !tmo) {
2161                         dev_warn(dev, "range %d ars timeout, aborting\n",
2162                                         spa->range_index);
2163                         break;
2164                 }
2165
2166                 if (rc == -EBUSY) {
2167                         /*
2168                          * Note, entries may be appended to the list
2169                          * while the lock is dropped, but the workqueue
2170                          * being active prevents entries being deleted /
2171                          * freed.
2172                          */
2173                         mutex_unlock(&acpi_desc->init_mutex);
2174                         ssleep(1);
2175                         tmo--;
2176                         mutex_lock(&acpi_desc->init_mutex);
2177                         continue;
2178                 }
2179
2180                 /* we got some results, but there are more pending... */
2181                 if (rc == -ENOSPC && overflow_retry--) {
2182                         if (!init_ars_len) {
2183                                 init_ars_len = acpi_desc->ars_status->length;
2184                                 init_ars_start = acpi_desc->ars_status->address;
2185                         }
2186                         rc = ars_continue(acpi_desc);
2187                 }
2188
2189                 if (rc < 0) {
2190                         dev_warn(dev, "range %d ars continuation failed\n",
2191                                         spa->range_index);
2192                         break;
2193                 }
2194
2195                 if (init_ars_len) {
2196                         ars_start = init_ars_start;
2197                         ars_len = init_ars_len;
2198                 } else {
2199                         ars_start = acpi_desc->ars_status->address;
2200                         ars_len = acpi_desc->ars_status->length;
2201                 }
2202                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2203                                 spa->range_index, ars_start, ars_len);
2204                 /* notify the region about new poison entries */
2205                 nvdimm_region_notify(nfit_spa->nd_region,
2206                                 NVDIMM_REVALIDATE_POISON);
2207                 break;
2208         } while (1);
2209 }
2210
2211 static void acpi_nfit_scrub(struct work_struct *work)
2212 {
2213         struct device *dev;
2214         u64 init_scrub_length = 0;
2215         struct nfit_spa *nfit_spa;
2216         u64 init_scrub_address = 0;
2217         bool init_ars_done = false;
2218         struct acpi_nfit_desc *acpi_desc;
2219         unsigned int tmo = scrub_timeout;
2220         unsigned int overflow_retry = scrub_overflow_abort;
2221
2222         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2223         dev = acpi_desc->dev;
2224
2225         /*
2226          * We scrub in 2 phases.  The first phase waits for any platform
2227          * firmware initiated scrubs to complete and then we go search for the
2228          * affected spa regions to mark them scanned.  In the second phase we
2229          * initiate a directed scrub for every range that was not scrubbed in
2230          * phase 1.
2231          */
2232
2233         /* process platform firmware initiated scrubs */
2234  retry:
2235         mutex_lock(&acpi_desc->init_mutex);
2236         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2237                 struct nd_cmd_ars_status *ars_status;
2238                 struct acpi_nfit_system_address *spa;
2239                 u64 ars_start, ars_len;
2240                 int rc;
2241
2242                 if (acpi_desc->cancel)
2243                         break;
2244
2245                 if (nfit_spa->nd_region)
2246                         continue;
2247
2248                 if (init_ars_done) {
2249                         /*
2250                          * No need to re-query, we're now just
2251                          * reconciling all the ranges covered by the
2252                          * initial scrub
2253                          */
2254                         rc = 0;
2255                 } else
2256                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2257
2258                 if (rc == -ENOTTY) {
2259                         /* no ars capability, just register spa and move on */
2260                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2261                         continue;
2262                 }
2263
2264                 if (rc == -EBUSY && !tmo) {
2265                         /* fallthrough to directed scrub in phase 2 */
2266                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2267                         break;
2268                 } else if (rc == -EBUSY) {
2269                         mutex_unlock(&acpi_desc->init_mutex);
2270                         ssleep(1);
2271                         tmo--;
2272                         goto retry;
2273                 }
2274
2275                 /* we got some results, but there are more pending... */
2276                 if (rc == -ENOSPC && overflow_retry--) {
2277                         ars_status = acpi_desc->ars_status;
2278                         /*
2279                          * Record the original scrub range, so that we
2280                          * can recall all the ranges impacted by the
2281                          * initial scrub.
2282                          */
2283                         if (!init_scrub_length) {
2284                                 init_scrub_length = ars_status->length;
2285                                 init_scrub_address = ars_status->address;
2286                         }
2287                         rc = ars_continue(acpi_desc);
2288                         if (rc == 0) {
2289                                 mutex_unlock(&acpi_desc->init_mutex);
2290                                 goto retry;
2291                         }
2292                 }
2293
2294                 if (rc < 0) {
2295                         /*
2296                          * Initial scrub failed, we'll give it one more
2297                          * try below...
2298                          */
2299                         break;
2300                 }
2301
2302                 /* We got some final results, record completed ranges */
2303                 ars_status = acpi_desc->ars_status;
2304                 if (init_scrub_length) {
2305                         ars_start = init_scrub_address;
2306                         ars_len = ars_start + init_scrub_length;
2307                 } else {
2308                         ars_start = ars_status->address;
2309                         ars_len = ars_status->length;
2310                 }
2311                 spa = nfit_spa->spa;
2312
2313                 if (!init_ars_done) {
2314                         init_ars_done = true;
2315                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2316                                         ars_start, ars_len);
2317                 }
2318                 if (ars_start <= spa->address && ars_start + ars_len
2319                                 >= spa->address + spa->length)
2320                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2321         }
2322
2323         /*
2324          * For all the ranges not covered by an initial scrub we still
2325          * want to see if there are errors, but it's ok to discover them
2326          * asynchronously.
2327          */
2328         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2329                 /*
2330                  * Flag all the ranges that still need scrubbing, but
2331                  * register them now to make data available.
2332                  */
2333                 if (nfit_spa->nd_region)
2334                         nfit_spa->ars_done = 1;
2335                 else
2336                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2337         }
2338
2339         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2340                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2341         mutex_unlock(&acpi_desc->init_mutex);
2342 }
2343
2344 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2345 {
2346         struct nfit_spa *nfit_spa;
2347         int rc;
2348
2349         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2350                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2351                         /* BLK regions don't need to wait for ars results */
2352                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2353                         if (rc)
2354                                 return rc;
2355                 }
2356
2357         queue_work(nfit_wq, &acpi_desc->work);
2358         return 0;
2359 }
2360
2361 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2362                 struct nfit_table_prev *prev)
2363 {
2364         struct device *dev = acpi_desc->dev;
2365
2366         if (!list_empty(&prev->spas) ||
2367                         !list_empty(&prev->memdevs) ||
2368                         !list_empty(&prev->dcrs) ||
2369                         !list_empty(&prev->bdws) ||
2370                         !list_empty(&prev->idts) ||
2371                         !list_empty(&prev->flushes)) {
2372                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2373                 return -ENXIO;
2374         }
2375         return 0;
2376 }
2377
2378 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2379 {
2380         struct device *dev = acpi_desc->dev;
2381         struct nfit_table_prev prev;
2382         const void *end;
2383         u8 *data;
2384         int rc;
2385
2386         mutex_lock(&acpi_desc->init_mutex);
2387
2388         INIT_LIST_HEAD(&prev.spas);
2389         INIT_LIST_HEAD(&prev.memdevs);
2390         INIT_LIST_HEAD(&prev.dcrs);
2391         INIT_LIST_HEAD(&prev.bdws);
2392         INIT_LIST_HEAD(&prev.idts);
2393         INIT_LIST_HEAD(&prev.flushes);
2394
2395         list_cut_position(&prev.spas, &acpi_desc->spas,
2396                                 acpi_desc->spas.prev);
2397         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2398                                 acpi_desc->memdevs.prev);
2399         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2400                                 acpi_desc->dcrs.prev);
2401         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2402                                 acpi_desc->bdws.prev);
2403         list_cut_position(&prev.idts, &acpi_desc->idts,
2404                                 acpi_desc->idts.prev);
2405         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2406                                 acpi_desc->flushes.prev);
2407
2408         data = (u8 *) acpi_desc->nfit;
2409         end = data + sz;
2410         while (!IS_ERR_OR_NULL(data))
2411                 data = add_table(acpi_desc, &prev, data, end);
2412
2413         if (IS_ERR(data)) {
2414                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2415                                 PTR_ERR(data));
2416                 rc = PTR_ERR(data);
2417                 goto out_unlock;
2418         }
2419
2420         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2421         if (rc)
2422                 goto out_unlock;
2423
2424         if (nfit_mem_init(acpi_desc) != 0) {
2425                 rc = -ENOMEM;
2426                 goto out_unlock;
2427         }
2428
2429         acpi_nfit_init_dsms(acpi_desc);
2430
2431         rc = acpi_nfit_register_dimms(acpi_desc);
2432         if (rc)
2433                 goto out_unlock;
2434
2435         rc = acpi_nfit_register_regions(acpi_desc);
2436
2437  out_unlock:
2438         mutex_unlock(&acpi_desc->init_mutex);
2439         return rc;
2440 }
2441 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2442
2443 struct acpi_nfit_flush_work {
2444         struct work_struct work;
2445         struct completion cmp;
2446 };
2447
2448 static void flush_probe(struct work_struct *work)
2449 {
2450         struct acpi_nfit_flush_work *flush;
2451
2452         flush = container_of(work, typeof(*flush), work);
2453         complete(&flush->cmp);
2454 }
2455
2456 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2457 {
2458         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2459         struct device *dev = acpi_desc->dev;
2460         struct acpi_nfit_flush_work flush;
2461
2462         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2463         device_lock(dev);
2464         device_unlock(dev);
2465
2466         /*
2467          * Scrub work could take 10s of seconds, userspace may give up so we
2468          * need to be interruptible while waiting.
2469          */
2470         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2471         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2472         queue_work(nfit_wq, &flush.work);
2473         return wait_for_completion_interruptible(&flush.cmp);
2474 }
2475
2476 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2477                 struct nvdimm *nvdimm, unsigned int cmd)
2478 {
2479         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2480
2481         if (nvdimm)
2482                 return 0;
2483         if (cmd != ND_CMD_ARS_START)
2484                 return 0;
2485
2486         /*
2487          * The kernel and userspace may race to initiate a scrub, but
2488          * the scrub thread is prepared to lose that initial race.  It
2489          * just needs guarantees that any ars it initiates are not
2490          * interrupted by any intervening start reqeusts from userspace.
2491          */
2492         if (work_busy(&acpi_desc->work))
2493                 return -EBUSY;
2494
2495         return 0;
2496 }
2497
2498 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2499 {
2500         struct nvdimm_bus_descriptor *nd_desc;
2501
2502         dev_set_drvdata(dev, acpi_desc);
2503         acpi_desc->dev = dev;
2504         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2505         nd_desc = &acpi_desc->nd_desc;
2506         nd_desc->provider_name = "ACPI.NFIT";
2507         nd_desc->ndctl = acpi_nfit_ctl;
2508         nd_desc->flush_probe = acpi_nfit_flush_probe;
2509         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2510         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2511
2512         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2513         INIT_LIST_HEAD(&acpi_desc->spas);
2514         INIT_LIST_HEAD(&acpi_desc->dcrs);
2515         INIT_LIST_HEAD(&acpi_desc->bdws);
2516         INIT_LIST_HEAD(&acpi_desc->idts);
2517         INIT_LIST_HEAD(&acpi_desc->flushes);
2518         INIT_LIST_HEAD(&acpi_desc->memdevs);
2519         INIT_LIST_HEAD(&acpi_desc->dimms);
2520         mutex_init(&acpi_desc->spa_map_mutex);
2521         mutex_init(&acpi_desc->init_mutex);
2522         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2523 }
2524 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2525
2526 static int acpi_nfit_add(struct acpi_device *adev)
2527 {
2528         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2529         struct acpi_nfit_desc *acpi_desc;
2530         struct device *dev = &adev->dev;
2531         struct acpi_table_header *tbl;
2532         acpi_status status = AE_OK;
2533         acpi_size sz;
2534         int rc;
2535
2536         status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2537         if (ACPI_FAILURE(status)) {
2538                 /* This is ok, we could have an nvdimm hotplugged later */
2539                 dev_dbg(dev, "failed to find NFIT at startup\n");
2540                 return 0;
2541         }
2542
2543         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2544         if (!acpi_desc)
2545                 return -ENOMEM;
2546         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2547         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2548         if (!acpi_desc->nvdimm_bus)
2549                 return -ENOMEM;
2550
2551         /*
2552          * Save the acpi header for later and then skip it,
2553          * making nfit point to the first nfit table header.
2554          */
2555         acpi_desc->acpi_header = *tbl;
2556         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2557         sz -= sizeof(struct acpi_table_nfit);
2558
2559         /* Evaluate _FIT and override with that if present */
2560         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2561         if (ACPI_SUCCESS(status) && buf.length > 0) {
2562                 union acpi_object *obj;
2563                 /*
2564                  * Adjust for the acpi_object header of the _FIT
2565                  */
2566                 obj = buf.pointer;
2567                 if (obj->type == ACPI_TYPE_BUFFER) {
2568                         acpi_desc->nfit =
2569                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2570                         sz = obj->buffer.length;
2571                 } else
2572                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2573                                  __func__, (int) obj->type);
2574         }
2575
2576         rc = acpi_nfit_init(acpi_desc, sz);
2577         if (rc) {
2578                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2579                 return rc;
2580         }
2581         return 0;
2582 }
2583
2584 static int acpi_nfit_remove(struct acpi_device *adev)
2585 {
2586         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2587
2588         acpi_desc->cancel = 1;
2589         flush_workqueue(nfit_wq);
2590         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2591         return 0;
2592 }
2593
2594 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2595 {
2596         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2597         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2598         struct acpi_nfit_header *nfit_saved;
2599         union acpi_object *obj;
2600         struct device *dev = &adev->dev;
2601         acpi_status status;
2602         int ret;
2603
2604         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2605
2606         device_lock(dev);
2607         if (!dev->driver) {
2608                 /* dev->driver may be null if we're being removed */
2609                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2610                 goto out_unlock;
2611         }
2612
2613         if (!acpi_desc) {
2614                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2615                 if (!acpi_desc)
2616                         goto out_unlock;
2617                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2618                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2619                 if (!acpi_desc->nvdimm_bus)
2620                         goto out_unlock;
2621         } else {
2622                 /*
2623                  * Finish previous registration before considering new
2624                  * regions.
2625                  */
2626                 flush_workqueue(nfit_wq);
2627         }
2628
2629         /* Evaluate _FIT */
2630         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2631         if (ACPI_FAILURE(status)) {
2632                 dev_err(dev, "failed to evaluate _FIT\n");
2633                 goto out_unlock;
2634         }
2635
2636         nfit_saved = acpi_desc->nfit;
2637         obj = buf.pointer;
2638         if (obj->type == ACPI_TYPE_BUFFER) {
2639                 acpi_desc->nfit =
2640                         (struct acpi_nfit_header *)obj->buffer.pointer;
2641                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2642                 if (ret) {
2643                         /* Merge failed, restore old nfit, and exit */
2644                         acpi_desc->nfit = nfit_saved;
2645                         dev_err(dev, "failed to merge updated NFIT\n");
2646                 }
2647         } else {
2648                 /* Bad _FIT, restore old nfit */
2649                 dev_err(dev, "Invalid _FIT\n");
2650         }
2651         kfree(buf.pointer);
2652
2653  out_unlock:
2654         device_unlock(dev);
2655 }
2656
2657 static const struct acpi_device_id acpi_nfit_ids[] = {
2658         { "ACPI0012", 0 },
2659         { "", 0 },
2660 };
2661 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2662
2663 static struct acpi_driver acpi_nfit_driver = {
2664         .name = KBUILD_MODNAME,
2665         .ids = acpi_nfit_ids,
2666         .ops = {
2667                 .add = acpi_nfit_add,
2668                 .remove = acpi_nfit_remove,
2669                 .notify = acpi_nfit_notify,
2670         },
2671 };
2672
2673 static __init int nfit_init(void)
2674 {
2675         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2676         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2677         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2678         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2679         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2680         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2681         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2682
2683         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2684         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2685         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2686         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2687         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2688         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2689         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2690         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2691         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2692         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2693         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2694         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2695
2696         nfit_wq = create_singlethread_workqueue("nfit");
2697         if (!nfit_wq)
2698                 return -ENOMEM;
2699
2700         return acpi_bus_register_driver(&acpi_nfit_driver);
2701 }
2702
2703 static __exit void nfit_exit(void)
2704 {
2705         acpi_bus_unregister_driver(&acpi_nfit_driver);
2706         destroy_workqueue(nfit_wq);
2707 }
2708
2709 module_init(nfit_init);
2710 module_exit(nfit_exit);
2711 MODULE_LICENSE("GPL v2");
2712 MODULE_AUTHOR("Intel Corporation");