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, "%#x\n", 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, "%#x\n", 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, "%#x\n", dcr->device_id);
 883 }
 884 static DEVICE_ATTR_RO(device);
 885
 886 static int num_nvdimm_formats(struct nvdimm *nvdimm)
 887 {
 888         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 889         int formats = 0;
 890
 891         if (nfit_mem->memdev_pmem)
 892                 formats++;
 893         if (nfit_mem->memdev_bdw)
 894                 formats++;
 895         return formats;
 896 }
 897
 898 static ssize_t format_show(struct device *dev,
 899                 struct device_attribute *attr, char *buf)
 900 {
 901         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 902
 903         return sprintf(buf, "%#x\n", dcr->code);
 904 }
 905 static DEVICE_ATTR_RO(format);
 906
 907 static ssize_t format1_show(struct device *dev,
 908                 struct device_attribute *attr, char *buf)
 909 {
 910         u32 handle;
 911         ssize_t rc = -ENXIO;
 912         struct nfit_mem *nfit_mem;
 913         struct nfit_memdev *nfit_memdev;
 914         struct acpi_nfit_desc *acpi_desc;
 915         struct nvdimm *nvdimm = to_nvdimm(dev);
 916         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 917
 918         nfit_mem = nvdimm_provider_data(nvdimm);
 919         acpi_desc = nfit_mem->acpi_desc;
 920         handle = to_nfit_memdev(dev)->device_handle;
 921
 922         /* assumes DIMMs have at most 2 published interface codes */
 923         mutex_lock(&acpi_desc->init_mutex);
 924         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 925                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
 926                 struct nfit_dcr *nfit_dcr;
 927
 928                 if (memdev->device_handle != handle)
 929                         continue;
 930
 931                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
 932                         if (nfit_dcr->dcr->region_index != memdev->region_index)
 933                                 continue;
 934                         if (nfit_dcr->dcr->code == dcr->code)
 935                                 continue;
 936                         rc = sprintf(buf, "%#x\n", nfit_dcr->dcr->code);
 937                         break;
 938                 }
 939                 if (rc != ENXIO)
 940                         break;
 941         }
 942         mutex_unlock(&acpi_desc->init_mutex);
 943         return rc;
 944 }
 945 static DEVICE_ATTR_RO(format1);
 946
 947 static ssize_t formats_show(struct device *dev,
 948                 struct device_attribute *attr, char *buf)
 949 {
 950         struct nvdimm *nvdimm = to_nvdimm(dev);
 951
 952         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
 953 }
 954 static DEVICE_ATTR_RO(formats);
 955
 956 static ssize_t serial_show(struct device *dev,
 957                 struct device_attribute *attr, char *buf)
 958 {
 959         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
 960
 961         return sprintf(buf, "%#x\n", dcr->serial_number);
 962 }
 963 static DEVICE_ATTR_RO(serial);
 964
 965 static ssize_t family_show(struct device *dev,
 966                 struct device_attribute *attr, char *buf)
 967 {
 968         struct nvdimm *nvdimm = to_nvdimm(dev);
 969         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 970
 971         if (nfit_mem->family < 0)
 972                 return -ENXIO;
 973         return sprintf(buf, "%d\n", nfit_mem->family);
 974 }
 975 static DEVICE_ATTR_RO(family);
 976
 977 static ssize_t dsm_mask_show(struct device *dev,
 978                 struct device_attribute *attr, char *buf)
 979 {
 980         struct nvdimm *nvdimm = to_nvdimm(dev);
 981         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 982
 983         if (nfit_mem->family < 0)
 984                 return -ENXIO;
 985         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
 986 }
 987 static DEVICE_ATTR_RO(dsm_mask);
 988
 989 static ssize_t flags_show(struct device *dev,
 990                 struct device_attribute *attr, char *buf)
 991 {
 992         u16 flags = to_nfit_memdev(dev)->flags;
 993
 994         return sprintf(buf, "%s%s%s%s%s\n",
 995                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
 996                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
 997                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
 998                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
 999                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "");
1000 }
1001 static DEVICE_ATTR_RO(flags);
1002
1003 static struct attribute *acpi_nfit_dimm_attributes[] = {
1004         &dev_attr_handle.attr,
1005         &dev_attr_phys_id.attr,
1006         &dev_attr_vendor.attr,
1007         &dev_attr_device.attr,
1008         &dev_attr_format.attr,
1009         &dev_attr_formats.attr,
1010         &dev_attr_format1.attr,
1011         &dev_attr_serial.attr,
1012         &dev_attr_rev_id.attr,
1013         &dev_attr_flags.attr,
1014         &dev_attr_family.attr,
1015         &dev_attr_dsm_mask.attr,
1016         NULL,
1017 };
1018
1019 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1020                 struct attribute *a, int n)
1021 {
1022         struct device *dev = container_of(kobj, struct device, kobj);
1023         struct nvdimm *nvdimm = to_nvdimm(dev);
1024
1025         if (!to_nfit_dcr(dev))
1026                 return 0;
1027         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1028                 return 0;
1029         return a->mode;
1030 }
1031
1032 static struct attribute_group acpi_nfit_dimm_attribute_group = {
1033         .name = "nfit",
1034         .attrs = acpi_nfit_dimm_attributes,
1035         .is_visible = acpi_nfit_dimm_attr_visible,
1036 };
1037
1038 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1039         &nvdimm_attribute_group,
1040         &nd_device_attribute_group,
1041         &acpi_nfit_dimm_attribute_group,
1042         NULL,
1043 };
1044
1045 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1046                 u32 device_handle)
1047 {
1048         struct nfit_mem *nfit_mem;
1049
1050         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1051                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1052                         return nfit_mem->nvdimm;
1053
1054         return NULL;
1055 }
1056
1057 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1058                 struct nfit_mem *nfit_mem, u32 device_handle)
1059 {
1060         struct acpi_device *adev, *adev_dimm;
1061         struct device *dev = acpi_desc->dev;
1062         unsigned long dsm_mask;
1063         const u8 *uuid;
1064         int i;
1065
1066         /* nfit test assumes 1:1 relationship between commands and dsms */
1067         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1068         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1069         adev = to_acpi_dev(acpi_desc);
1070         if (!adev)
1071                 return 0;
1072
1073         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1074         nfit_mem->adev = adev_dimm;
1075         if (!adev_dimm) {
1076                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1077                                 device_handle);
1078                 return force_enable_dimms ? 0 : -ENODEV;
1079         }
1080
1081         /*
1082          * Until standardization materializes we need to consider up to 3
1083          * different command sets.  Note, that checking for function0 (bit0)
1084          * tells us if any commands are reachable through this uuid.
1085          */
1086         for (i = NVDIMM_FAMILY_INTEL; i <= NVDIMM_FAMILY_HPE2; i++)
1087                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1088                         break;
1089
1090         /* limit the supported commands to those that are publicly documented */
1091         nfit_mem->family = i;
1092         if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1093                 dsm_mask = 0x3fe;
1094                 if (disable_vendor_specific)
1095                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1096         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1)
1097                 dsm_mask = 0x1c3c76;
1098         else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1099                 dsm_mask = 0x1fe;
1100                 if (disable_vendor_specific)
1101                         dsm_mask &= ~(1 << 8);
1102         } else {
1103                 dev_err(dev, "unknown dimm command family\n");
1104                 nfit_mem->family = -1;
1105                 return force_enable_dimms ? 0 : -ENODEV;
1106         }
1107
1108         uuid = to_nfit_uuid(nfit_mem->family);
1109         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1110                 if (acpi_check_dsm(adev_dimm->handle, uuid, 1, 1ULL << i))
1111                         set_bit(i, &nfit_mem->dsm_mask);
1112
1113         return 0;
1114 }
1115
1116 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1117 {
1118         struct nfit_mem *nfit_mem;
1119         int dimm_count = 0;
1120
1121         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1122                 unsigned long flags = 0, cmd_mask;
1123                 struct nvdimm *nvdimm;
1124                 u32 device_handle;
1125                 u16 mem_flags;
1126                 int rc;
1127
1128                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1129                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1130                 if (nvdimm) {
1131                         dimm_count++;
1132                         continue;
1133                 }
1134
1135                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1136                         flags |= NDD_ALIASING;
1137
1138                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1139                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1140                         flags |= NDD_UNARMED;
1141
1142                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1143                 if (rc)
1144                         continue;
1145
1146                 /*
1147                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1148                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1149                  * userspace interface.
1150                  */
1151                 cmd_mask = 1UL << ND_CMD_CALL;
1152                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL)
1153                         cmd_mask |= nfit_mem->dsm_mask;
1154
1155                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1156                                 acpi_nfit_dimm_attribute_groups,
1157                                 flags, cmd_mask);
1158                 if (!nvdimm)
1159                         return -ENOMEM;
1160
1161                 nfit_mem->nvdimm = nvdimm;
1162                 dimm_count++;
1163
1164                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1165                         continue;
1166
1167                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s\n",
1168                                 nvdimm_name(nvdimm),
1169                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1170                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1171                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1172                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "");
1173
1174         }
1175
1176         return nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1177 }
1178
1179 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1180 {
1181         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1182         const u8 *uuid = to_nfit_uuid(NFIT_DEV_BUS);
1183         struct acpi_device *adev;
1184         int i;
1185
1186         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1187         adev = to_acpi_dev(acpi_desc);
1188         if (!adev)
1189                 return;
1190
1191         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1192                 if (acpi_check_dsm(adev->handle, uuid, 1, 1ULL << i))
1193                         set_bit(i, &nd_desc->cmd_mask);
1194 }
1195
1196 static ssize_t range_index_show(struct device *dev,
1197                 struct device_attribute *attr, char *buf)
1198 {
1199         struct nd_region *nd_region = to_nd_region(dev);
1200         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1201
1202         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1203 }
1204 static DEVICE_ATTR_RO(range_index);
1205
1206 static struct attribute *acpi_nfit_region_attributes[] = {
1207         &dev_attr_range_index.attr,
1208         NULL,
1209 };
1210
1211 static struct attribute_group acpi_nfit_region_attribute_group = {
1212         .name = "nfit",
1213         .attrs = acpi_nfit_region_attributes,
1214 };
1215
1216 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
1217         &nd_region_attribute_group,
1218         &nd_mapping_attribute_group,
1219         &nd_device_attribute_group,
1220         &nd_numa_attribute_group,
1221         &acpi_nfit_region_attribute_group,
1222         NULL,
1223 };
1224
1225 /* enough info to uniquely specify an interleave set */
1226 struct nfit_set_info {
1227         struct nfit_set_info_map {
1228                 u64 region_offset;
1229                 u32 serial_number;
1230                 u32 pad;
1231         } mapping[0];
1232 };
1233
1234 static size_t sizeof_nfit_set_info(int num_mappings)
1235 {
1236         return sizeof(struct nfit_set_info)
1237                 + num_mappings * sizeof(struct nfit_set_info_map);
1238 }
1239
1240 static int cmp_map(const void *m0, const void *m1)
1241 {
1242         const struct nfit_set_info_map *map0 = m0;
1243         const struct nfit_set_info_map *map1 = m1;
1244
1245         return memcmp(&map0->region_offset, &map1->region_offset,
1246                         sizeof(u64));
1247 }
1248
1249 /* Retrieve the nth entry referencing this spa */
1250 static struct acpi_nfit_memory_map *memdev_from_spa(
1251                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
1252 {
1253         struct nfit_memdev *nfit_memdev;
1254
1255         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
1256                 if (nfit_memdev->memdev->range_index == range_index)
1257                         if (n-- == 0)
1258                                 return nfit_memdev->memdev;
1259         return NULL;
1260 }
1261
1262 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
1263                 struct nd_region_desc *ndr_desc,
1264                 struct acpi_nfit_system_address *spa)
1265 {
1266         int i, spa_type = nfit_spa_type(spa);
1267         struct device *dev = acpi_desc->dev;
1268         struct nd_interleave_set *nd_set;
1269         u16 nr = ndr_desc->num_mappings;
1270         struct nfit_set_info *info;
1271
1272         if (spa_type == NFIT_SPA_PM || spa_type == NFIT_SPA_VOLATILE)
1273                 /* pass */;
1274         else
1275                 return 0;
1276
1277         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
1278         if (!nd_set)
1279                 return -ENOMEM;
1280
1281         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
1282         if (!info)
1283                 return -ENOMEM;
1284         for (i = 0; i < nr; i++) {
1285                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
1286                 struct nfit_set_info_map *map = &info->mapping[i];
1287                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1288                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1289                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
1290                                 spa->range_index, i);
1291
1292                 if (!memdev || !nfit_mem->dcr) {
1293                         dev_err(dev, "%s: failed to find DCR\n", __func__);
1294                         return -ENODEV;
1295                 }
1296
1297                 map->region_offset = memdev->region_offset;
1298                 map->serial_number = nfit_mem->dcr->serial_number;
1299         }
1300
1301         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
1302                         cmp_map, NULL);
1303         nd_set->cookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
1304         ndr_desc->nd_set = nd_set;
1305         devm_kfree(dev, info);
1306
1307         return 0;
1308 }
1309
1310 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
1311 {
1312         struct acpi_nfit_interleave *idt = mmio->idt;
1313         u32 sub_line_offset, line_index, line_offset;
1314         u64 line_no, table_skip_count, table_offset;
1315
1316         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
1317         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
1318         line_offset = idt->line_offset[line_index]
1319                 * mmio->line_size;
1320         table_offset = table_skip_count * mmio->table_size;
1321
1322         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
1323 }
1324
1325 static void wmb_blk(struct nfit_blk *nfit_blk)
1326 {
1327
1328         if (nfit_blk->nvdimm_flush) {
1329                 /*
1330                  * The first wmb() is needed to 'sfence' all previous writes
1331                  * such that they are architecturally visible for the platform
1332                  * buffer flush.  Note that we've already arranged for pmem
1333                  * writes to avoid the cache via arch_memcpy_to_pmem().  The
1334                  * final wmb() ensures ordering for the NVDIMM flush write.
1335                  */
1336                 wmb();
1337                 writeq(1, nfit_blk->nvdimm_flush);
1338                 wmb();
1339         } else
1340                 wmb_pmem();
1341 }
1342
1343 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
1344 {
1345         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1346         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
1347
1348         if (mmio->num_lines)
1349                 offset = to_interleave_offset(offset, mmio);
1350
1351         return readl(mmio->addr.base + offset);
1352 }
1353
1354 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
1355                 resource_size_t dpa, unsigned int len, unsigned int write)
1356 {
1357         u64 cmd, offset;
1358         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
1359
1360         enum {
1361                 BCW_OFFSET_MASK = (1ULL << 48)-1,
1362                 BCW_LEN_SHIFT = 48,
1363                 BCW_LEN_MASK = (1ULL << 8) - 1,
1364                 BCW_CMD_SHIFT = 56,
1365         };
1366
1367         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
1368         len = len >> L1_CACHE_SHIFT;
1369         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
1370         cmd |= ((u64) write) << BCW_CMD_SHIFT;
1371
1372         offset = nfit_blk->cmd_offset + mmio->size * bw;
1373         if (mmio->num_lines)
1374                 offset = to_interleave_offset(offset, mmio);
1375
1376         writeq(cmd, mmio->addr.base + offset);
1377         wmb_blk(nfit_blk);
1378
1379         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
1380                 readq(mmio->addr.base + offset);
1381 }
1382
1383 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
1384                 resource_size_t dpa, void *iobuf, size_t len, int rw,
1385                 unsigned int lane)
1386 {
1387         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1388         unsigned int copied = 0;
1389         u64 base_offset;
1390         int rc;
1391
1392         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
1393                 + lane * mmio->size;
1394         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
1395         while (len) {
1396                 unsigned int c;
1397                 u64 offset;
1398
1399                 if (mmio->num_lines) {
1400                         u32 line_offset;
1401
1402                         offset = to_interleave_offset(base_offset + copied,
1403                                         mmio);
1404                         div_u64_rem(offset, mmio->line_size, &line_offset);
1405                         c = min_t(size_t, len, mmio->line_size - line_offset);
1406                 } else {
1407                         offset = base_offset + nfit_blk->bdw_offset;
1408                         c = len;
1409                 }
1410
1411                 if (rw)
1412                         memcpy_to_pmem(mmio->addr.aperture + offset,
1413                                         iobuf + copied, c);
1414                 else {
1415                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
1416                                 mmio_flush_range((void __force *)
1417                                         mmio->addr.aperture + offset, c);
1418
1419                         memcpy_from_pmem(iobuf + copied,
1420                                         mmio->addr.aperture + offset, c);
1421                 }
1422
1423                 copied += c;
1424                 len -= c;
1425         }
1426
1427         if (rw)
1428                 wmb_blk(nfit_blk);
1429
1430         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
1431         return rc;
1432 }
1433
1434 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
1435                 resource_size_t dpa, void *iobuf, u64 len, int rw)
1436 {
1437         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1438         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
1439         struct nd_region *nd_region = nfit_blk->nd_region;
1440         unsigned int lane, copied = 0;
1441         int rc = 0;
1442
1443         lane = nd_region_acquire_lane(nd_region);
1444         while (len) {
1445                 u64 c = min(len, mmio->size);
1446
1447                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
1448                                 iobuf + copied, c, rw, lane);
1449                 if (rc)
1450                         break;
1451
1452                 copied += c;
1453                 len -= c;
1454         }
1455         nd_region_release_lane(nd_region, lane);
1456
1457         return rc;
1458 }
1459
1460 static void nfit_spa_mapping_release(struct kref *kref)
1461 {
1462         struct nfit_spa_mapping *spa_map = to_spa_map(kref);
1463         struct acpi_nfit_system_address *spa = spa_map->spa;
1464         struct acpi_nfit_desc *acpi_desc = spa_map->acpi_desc;
1465
1466         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1467         dev_dbg(acpi_desc->dev, "%s: SPA%d\n", __func__, spa->range_index);
1468         if (spa_map->type == SPA_MAP_APERTURE)
1469                 memunmap((void __force *)spa_map->addr.aperture);
1470         else
1471                 iounmap(spa_map->addr.base);
1472         release_mem_region(spa->address, spa->length);
1473         list_del(&spa_map->list);
1474         kfree(spa_map);
1475 }
1476
1477 static struct nfit_spa_mapping *find_spa_mapping(
1478                 struct acpi_nfit_desc *acpi_desc,
1479                 struct acpi_nfit_system_address *spa)
1480 {
1481         struct nfit_spa_mapping *spa_map;
1482
1483         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1484         list_for_each_entry(spa_map, &acpi_desc->spa_maps, list)
1485                 if (spa_map->spa == spa)
1486                         return spa_map;
1487
1488         return NULL;
1489 }
1490
1491 static void nfit_spa_unmap(struct acpi_nfit_desc *acpi_desc,
1492                 struct acpi_nfit_system_address *spa)
1493 {
1494         struct nfit_spa_mapping *spa_map;
1495
1496         mutex_lock(&acpi_desc->spa_map_mutex);
1497         spa_map = find_spa_mapping(acpi_desc, spa);
1498
1499         if (spa_map)
1500                 kref_put(&spa_map->kref, nfit_spa_mapping_release);
1501         mutex_unlock(&acpi_desc->spa_map_mutex);
1502 }
1503
1504 static void __iomem *__nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1505                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1506 {
1507         resource_size_t start = spa->address;
1508         resource_size_t n = spa->length;
1509         struct nfit_spa_mapping *spa_map;
1510         struct resource *res;
1511
1512         WARN_ON(!mutex_is_locked(&acpi_desc->spa_map_mutex));
1513
1514         spa_map = find_spa_mapping(acpi_desc, spa);
1515         if (spa_map) {
1516                 kref_get(&spa_map->kref);
1517                 return spa_map->addr.base;
1518         }
1519
1520         spa_map = kzalloc(sizeof(*spa_map), GFP_KERNEL);
1521         if (!spa_map)
1522                 return NULL;
1523
1524         INIT_LIST_HEAD(&spa_map->list);
1525         spa_map->spa = spa;
1526         kref_init(&spa_map->kref);
1527         spa_map->acpi_desc = acpi_desc;
1528
1529         res = request_mem_region(start, n, dev_name(acpi_desc->dev));
1530         if (!res)
1531                 goto err_mem;
1532
1533         spa_map->type = type;
1534         if (type == SPA_MAP_APERTURE)
1535                 spa_map->addr.aperture = (void __pmem *)memremap(start, n,
1536                                                         ARCH_MEMREMAP_PMEM);
1537         else
1538                 spa_map->addr.base = ioremap_nocache(start, n);
1539
1540
1541         if (!spa_map->addr.base)
1542                 goto err_map;
1543
1544         list_add_tail(&spa_map->list, &acpi_desc->spa_maps);
1545         return spa_map->addr.base;
1546
1547  err_map:
1548         release_mem_region(start, n);
1549  err_mem:
1550         kfree(spa_map);
1551         return NULL;
1552 }
1553
1554 /**
1555  * nfit_spa_map - interleave-aware managed-mappings of acpi_nfit_system_address ranges
1556  * @nvdimm_bus: NFIT-bus that provided the spa table entry
1557  * @nfit_spa: spa table to map
1558  * @type: aperture or control region
1559  *
1560  * In the case where block-data-window apertures and
1561  * dimm-control-regions are interleaved they will end up sharing a
1562  * single request_mem_region() + ioremap() for the address range.  In
1563  * the style of devm nfit_spa_map() mappings are automatically dropped
1564  * when all region devices referencing the same mapping are disabled /
1565  * unbound.
1566  */
1567 static void __iomem *nfit_spa_map(struct acpi_nfit_desc *acpi_desc,
1568                 struct acpi_nfit_system_address *spa, enum spa_map_type type)
1569 {
1570         void __iomem *iomem;
1571
1572         mutex_lock(&acpi_desc->spa_map_mutex);
1573         iomem = __nfit_spa_map(acpi_desc, spa, type);
1574         mutex_unlock(&acpi_desc->spa_map_mutex);
1575
1576         return iomem;
1577 }
1578
1579 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
1580                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
1581 {
1582         if (idt) {
1583                 mmio->num_lines = idt->line_count;
1584                 mmio->line_size = idt->line_size;
1585                 if (interleave_ways == 0)
1586                         return -ENXIO;
1587                 mmio->table_size = mmio->num_lines * interleave_ways
1588                         * mmio->line_size;
1589         }
1590
1591         return 0;
1592 }
1593
1594 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
1595                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
1596 {
1597         struct nd_cmd_dimm_flags flags;
1598         int rc;
1599
1600         memset(&flags, 0, sizeof(flags));
1601         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
1602                         sizeof(flags), NULL);
1603
1604         if (rc >= 0 && flags.status == 0)
1605                 nfit_blk->dimm_flags = flags.flags;
1606         else if (rc == -ENOTTY) {
1607                 /* fall back to a conservative default */
1608                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
1609                 rc = 0;
1610         } else
1611                 rc = -ENXIO;
1612
1613         return rc;
1614 }
1615
1616 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
1617                 struct device *dev)
1618 {
1619         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1620         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1621         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1622         struct nfit_flush *nfit_flush;
1623         struct nfit_blk_mmio *mmio;
1624         struct nfit_blk *nfit_blk;
1625         struct nfit_mem *nfit_mem;
1626         struct nvdimm *nvdimm;
1627         int rc;
1628
1629         nvdimm = nd_blk_region_to_dimm(ndbr);
1630         nfit_mem = nvdimm_provider_data(nvdimm);
1631         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
1632                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
1633                                 nfit_mem ? "" : " nfit_mem",
1634                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
1635                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
1636                 return -ENXIO;
1637         }
1638
1639         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
1640         if (!nfit_blk)
1641                 return -ENOMEM;
1642         nd_blk_region_set_provider_data(ndbr, nfit_blk);
1643         nfit_blk->nd_region = to_nd_region(dev);
1644
1645         /* map block aperture memory */
1646         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
1647         mmio = &nfit_blk->mmio[BDW];
1648         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_bdw,
1649                         SPA_MAP_APERTURE);
1650         if (!mmio->addr.base) {
1651                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
1652                                 nvdimm_name(nvdimm));
1653                 return -ENOMEM;
1654         }
1655         mmio->size = nfit_mem->bdw->size;
1656         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
1657         mmio->idt = nfit_mem->idt_bdw;
1658         mmio->spa = nfit_mem->spa_bdw;
1659         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
1660                         nfit_mem->memdev_bdw->interleave_ways);
1661         if (rc) {
1662                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
1663                                 __func__, nvdimm_name(nvdimm));
1664                 return rc;
1665         }
1666
1667         /* map block control memory */
1668         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
1669         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
1670         mmio = &nfit_blk->mmio[DCR];
1671         mmio->addr.base = nfit_spa_map(acpi_desc, nfit_mem->spa_dcr,
1672                         SPA_MAP_CONTROL);
1673         if (!mmio->addr.base) {
1674                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
1675                                 nvdimm_name(nvdimm));
1676                 return -ENOMEM;
1677         }
1678         mmio->size = nfit_mem->dcr->window_size;
1679         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
1680         mmio->idt = nfit_mem->idt_dcr;
1681         mmio->spa = nfit_mem->spa_dcr;
1682         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
1683                         nfit_mem->memdev_dcr->interleave_ways);
1684         if (rc) {
1685                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
1686                                 __func__, nvdimm_name(nvdimm));
1687                 return rc;
1688         }
1689
1690         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
1691         if (rc < 0) {
1692                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
1693                                 __func__, nvdimm_name(nvdimm));
1694                 return rc;
1695         }
1696
1697         nfit_flush = nfit_mem->nfit_flush;
1698         if (nfit_flush && nfit_flush->flush->hint_count != 0) {
1699                 nfit_blk->nvdimm_flush = devm_ioremap_nocache(dev,
1700                                 nfit_flush->flush->hint_address[0], 8);
1701                 if (!nfit_blk->nvdimm_flush)
1702                         return -ENOMEM;
1703         }
1704
1705         if (!arch_has_wmb_pmem() && !nfit_blk->nvdimm_flush)
1706                 dev_warn(dev, "unable to guarantee persistence of writes\n");
1707
1708         if (mmio->line_size == 0)
1709                 return 0;
1710
1711         if ((u32) nfit_blk->cmd_offset % mmio->line_size
1712                         + 8 > mmio->line_size) {
1713                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
1714                 return -ENXIO;
1715         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
1716                         + 8 > mmio->line_size) {
1717                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
1718                 return -ENXIO;
1719         }
1720
1721         return 0;
1722 }
1723
1724 static void acpi_nfit_blk_region_disable(struct nvdimm_bus *nvdimm_bus,
1725                 struct device *dev)
1726 {
1727         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1728         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1729         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
1730         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
1731         int i;
1732
1733         if (!nfit_blk)
1734                 return; /* never enabled */
1735
1736         /* auto-free BLK spa mappings */
1737         for (i = 0; i < 2; i++) {
1738                 struct nfit_blk_mmio *mmio = &nfit_blk->mmio[i];
1739
1740                 if (mmio->addr.base)
1741                         nfit_spa_unmap(acpi_desc, mmio->spa);
1742         }
1743         nd_blk_region_set_provider_data(ndbr, NULL);
1744         /* devm will free nfit_blk */
1745 }
1746
1747 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
1748                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
1749 {
1750         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1751         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1752         int cmd_rc, rc;
1753
1754         cmd->address = spa->address;
1755         cmd->length = spa->length;
1756         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
1757                         sizeof(*cmd), &cmd_rc);
1758         if (rc < 0)
1759                 return rc;
1760         return cmd_rc;
1761 }
1762
1763 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
1764 {
1765         int rc;
1766         int cmd_rc;
1767         struct nd_cmd_ars_start ars_start;
1768         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1769         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1770
1771         memset(&ars_start, 0, sizeof(ars_start));
1772         ars_start.address = spa->address;
1773         ars_start.length = spa->length;
1774         if (nfit_spa_type(spa) == NFIT_SPA_PM)
1775                 ars_start.type = ND_ARS_PERSISTENT;
1776         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
1777                 ars_start.type = ND_ARS_VOLATILE;
1778         else
1779                 return -ENOTTY;
1780
1781         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1782                         sizeof(ars_start), &cmd_rc);
1783
1784         if (rc < 0)
1785                 return rc;
1786         return cmd_rc;
1787 }
1788
1789 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
1790 {
1791         int rc, cmd_rc;
1792         struct nd_cmd_ars_start ars_start;
1793         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1794         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1795
1796         memset(&ars_start, 0, sizeof(ars_start));
1797         ars_start.address = ars_status->restart_address;
1798         ars_start.length = ars_status->restart_length;
1799         ars_start.type = ars_status->type;
1800         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
1801                         sizeof(ars_start), &cmd_rc);
1802         if (rc < 0)
1803                 return rc;
1804         return cmd_rc;
1805 }
1806
1807 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
1808 {
1809         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1810         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
1811         int rc, cmd_rc;
1812
1813         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
1814                         acpi_desc->ars_status_size, &cmd_rc);
1815         if (rc < 0)
1816                 return rc;
1817         return cmd_rc;
1818 }
1819
1820 static int ars_status_process_records(struct nvdimm_bus *nvdimm_bus,
1821                 struct nd_cmd_ars_status *ars_status)
1822 {
1823         int rc;
1824         u32 i;
1825
1826         for (i = 0; i < ars_status->num_records; i++) {
1827                 rc = nvdimm_bus_add_poison(nvdimm_bus,
1828                                 ars_status->records[i].err_address,
1829                                 ars_status->records[i].length);
1830                 if (rc)
1831                         return rc;
1832         }
1833
1834         return 0;
1835 }
1836
1837 static void acpi_nfit_remove_resource(void *data)
1838 {
1839         struct resource *res = data;
1840
1841         remove_resource(res);
1842 }
1843
1844 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
1845                 struct nd_region_desc *ndr_desc)
1846 {
1847         struct resource *res, *nd_res = ndr_desc->res;
1848         int is_pmem, ret;
1849
1850         /* No operation if the region is already registered as PMEM */
1851         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
1852                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
1853         if (is_pmem == REGION_INTERSECTS)
1854                 return 0;
1855
1856         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
1857         if (!res)
1858                 return -ENOMEM;
1859
1860         res->name = "Persistent Memory";
1861         res->start = nd_res->start;
1862         res->end = nd_res->end;
1863         res->flags = IORESOURCE_MEM;
1864         res->desc = IORES_DESC_PERSISTENT_MEMORY;
1865
1866         ret = insert_resource(&iomem_resource, res);
1867         if (ret)
1868                 return ret;
1869
1870         ret = devm_add_action(acpi_desc->dev, acpi_nfit_remove_resource, res);
1871         if (ret) {
1872                 remove_resource(res);
1873                 return ret;
1874         }
1875
1876         return 0;
1877 }
1878
1879 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
1880                 struct nd_mapping *nd_mapping, struct nd_region_desc *ndr_desc,
1881                 struct acpi_nfit_memory_map *memdev,
1882                 struct nfit_spa *nfit_spa)
1883 {
1884         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
1885                         memdev->device_handle);
1886         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1887         struct nd_blk_region_desc *ndbr_desc;
1888         struct nfit_mem *nfit_mem;
1889         int blk_valid = 0;
1890
1891         if (!nvdimm) {
1892                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
1893                                 spa->range_index, memdev->device_handle);
1894                 return -ENODEV;
1895         }
1896
1897         nd_mapping->nvdimm = nvdimm;
1898         switch (nfit_spa_type(spa)) {
1899         case NFIT_SPA_PM:
1900         case NFIT_SPA_VOLATILE:
1901                 nd_mapping->start = memdev->address;
1902                 nd_mapping->size = memdev->region_size;
1903                 break;
1904         case NFIT_SPA_DCR:
1905                 nfit_mem = nvdimm_provider_data(nvdimm);
1906                 if (!nfit_mem || !nfit_mem->bdw) {
1907                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
1908                                         spa->range_index, nvdimm_name(nvdimm));
1909                 } else {
1910                         nd_mapping->size = nfit_mem->bdw->capacity;
1911                         nd_mapping->start = nfit_mem->bdw->start_address;
1912                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
1913                         blk_valid = 1;
1914                 }
1915
1916                 ndr_desc->nd_mapping = nd_mapping;
1917                 ndr_desc->num_mappings = blk_valid;
1918                 ndbr_desc = to_blk_region_desc(ndr_desc);
1919                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
1920                 ndbr_desc->disable = acpi_nfit_blk_region_disable;
1921                 ndbr_desc->do_io = acpi_desc->blk_do_io;
1922                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
1923                                 ndr_desc);
1924                 if (!nfit_spa->nd_region)
1925                         return -ENOMEM;
1926                 break;
1927         }
1928
1929         return 0;
1930 }
1931
1932 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
1933                 struct nfit_spa *nfit_spa)
1934 {
1935         static struct nd_mapping nd_mappings[ND_MAX_MAPPINGS];
1936         struct acpi_nfit_system_address *spa = nfit_spa->spa;
1937         struct nd_blk_region_desc ndbr_desc;
1938         struct nd_region_desc *ndr_desc;
1939         struct nfit_memdev *nfit_memdev;
1940         struct nvdimm_bus *nvdimm_bus;
1941         struct resource res;
1942         int count = 0, rc;
1943
1944         if (nfit_spa->nd_region)
1945                 return 0;
1946
1947         if (spa->range_index == 0) {
1948                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
1949                                 __func__);
1950                 return 0;
1951         }
1952
1953         memset(&res, 0, sizeof(res));
1954         memset(&nd_mappings, 0, sizeof(nd_mappings));
1955         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
1956         res.start = spa->address;
1957         res.end = res.start + spa->length - 1;
1958         ndr_desc = &ndbr_desc.ndr_desc;
1959         ndr_desc->res = &res;
1960         ndr_desc->provider_data = nfit_spa;
1961         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
1962         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
1963                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
1964                                                 spa->proximity_domain);
1965         else
1966                 ndr_desc->numa_node = NUMA_NO_NODE;
1967
1968         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1969                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1970                 struct nd_mapping *nd_mapping;
1971
1972                 if (memdev->range_index != spa->range_index)
1973                         continue;
1974                 if (count >= ND_MAX_MAPPINGS) {
1975                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
1976                                         spa->range_index, ND_MAX_MAPPINGS);
1977                         return -ENXIO;
1978                 }
1979                 nd_mapping = &nd_mappings[count++];
1980                 rc = acpi_nfit_init_mapping(acpi_desc, nd_mapping, ndr_desc,
1981                                 memdev, nfit_spa);
1982                 if (rc)
1983                         goto out;
1984         }
1985
1986         ndr_desc->nd_mapping = nd_mappings;
1987         ndr_desc->num_mappings = count;
1988         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
1989         if (rc)
1990                 goto out;
1991
1992         nvdimm_bus = acpi_desc->nvdimm_bus;
1993         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
1994                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
1995                 if (rc) {
1996                         dev_warn(acpi_desc->dev,
1997                                 "failed to insert pmem resource to iomem: %d\n",
1998                                 rc);
1999                         goto out;
2000                 }
2001
2002                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2003                                 ndr_desc);
2004                 if (!nfit_spa->nd_region)
2005                         rc = -ENOMEM;
2006         } else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE) {
2007                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2008                                 ndr_desc);
2009                 if (!nfit_spa->nd_region)
2010                         rc = -ENOMEM;
2011         }
2012
2013  out:
2014         if (rc)
2015                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2016                                 nfit_spa->spa->range_index);
2017         return rc;
2018 }
2019
2020 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2021                 u32 max_ars)
2022 {
2023         struct device *dev = acpi_desc->dev;
2024         struct nd_cmd_ars_status *ars_status;
2025
2026         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2027                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2028                 return 0;
2029         }
2030
2031         if (acpi_desc->ars_status)
2032                 devm_kfree(dev, acpi_desc->ars_status);
2033         acpi_desc->ars_status = NULL;
2034         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2035         if (!ars_status)
2036                 return -ENOMEM;
2037         acpi_desc->ars_status = ars_status;
2038         acpi_desc->ars_status_size = max_ars;
2039         return 0;
2040 }
2041
2042 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2043                 struct nfit_spa *nfit_spa)
2044 {
2045         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2046         int rc;
2047
2048         if (!nfit_spa->max_ars) {
2049                 struct nd_cmd_ars_cap ars_cap;
2050
2051                 memset(&ars_cap, 0, sizeof(ars_cap));
2052                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2053                 if (rc < 0)
2054                         return rc;
2055                 nfit_spa->max_ars = ars_cap.max_ars_out;
2056                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2057                 /* check that the supported scrub types match the spa type */
2058                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2059                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2060                         return -ENOTTY;
2061                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2062                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2063                         return -ENOTTY;
2064         }
2065
2066         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2067                 return -ENOMEM;
2068
2069         rc = ars_get_status(acpi_desc);
2070         if (rc < 0 && rc != -ENOSPC)
2071                 return rc;
2072
2073         if (ars_status_process_records(acpi_desc->nvdimm_bus,
2074                                 acpi_desc->ars_status))
2075                 return -ENOMEM;
2076
2077         return 0;
2078 }
2079
2080 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2081                 struct nfit_spa *nfit_spa)
2082 {
2083         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2084         unsigned int overflow_retry = scrub_overflow_abort;
2085         u64 init_ars_start = 0, init_ars_len = 0;
2086         struct device *dev = acpi_desc->dev;
2087         unsigned int tmo = scrub_timeout;
2088         int rc;
2089
2090         if (nfit_spa->ars_done || !nfit_spa->nd_region)
2091                 return;
2092
2093         rc = ars_start(acpi_desc, nfit_spa);
2094         /*
2095          * If we timed out the initial scan we'll still be busy here,
2096          * and will wait another timeout before giving up permanently.
2097          */
2098         if (rc < 0 && rc != -EBUSY)
2099                 return;
2100
2101         do {
2102                 u64 ars_start, ars_len;
2103
2104                 if (acpi_desc->cancel)
2105                         break;
2106                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2107                 if (rc == -ENOTTY)
2108                         break;
2109                 if (rc == -EBUSY && !tmo) {
2110                         dev_warn(dev, "range %d ars timeout, aborting\n",
2111                                         spa->range_index);
2112                         break;
2113                 }
2114
2115                 if (rc == -EBUSY) {
2116                         /*
2117                          * Note, entries may be appended to the list
2118                          * while the lock is dropped, but the workqueue
2119                          * being active prevents entries being deleted /
2120                          * freed.
2121                          */
2122                         mutex_unlock(&acpi_desc->init_mutex);
2123                         ssleep(1);
2124                         tmo--;
2125                         mutex_lock(&acpi_desc->init_mutex);
2126                         continue;
2127                 }
2128
2129                 /* we got some results, but there are more pending... */
2130                 if (rc == -ENOSPC && overflow_retry--) {
2131                         if (!init_ars_len) {
2132                                 init_ars_len = acpi_desc->ars_status->length;
2133                                 init_ars_start = acpi_desc->ars_status->address;
2134                         }
2135                         rc = ars_continue(acpi_desc);
2136                 }
2137
2138                 if (rc < 0) {
2139                         dev_warn(dev, "range %d ars continuation failed\n",
2140                                         spa->range_index);
2141                         break;
2142                 }
2143
2144                 if (init_ars_len) {
2145                         ars_start = init_ars_start;
2146                         ars_len = init_ars_len;
2147                 } else {
2148                         ars_start = acpi_desc->ars_status->address;
2149                         ars_len = acpi_desc->ars_status->length;
2150                 }
2151                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2152                                 spa->range_index, ars_start, ars_len);
2153                 /* notify the region about new poison entries */
2154                 nvdimm_region_notify(nfit_spa->nd_region,
2155                                 NVDIMM_REVALIDATE_POISON);
2156                 break;
2157         } while (1);
2158 }
2159
2160 static void acpi_nfit_scrub(struct work_struct *work)
2161 {
2162         struct device *dev;
2163         u64 init_scrub_length = 0;
2164         struct nfit_spa *nfit_spa;
2165         u64 init_scrub_address = 0;
2166         bool init_ars_done = false;
2167         struct acpi_nfit_desc *acpi_desc;
2168         unsigned int tmo = scrub_timeout;
2169         unsigned int overflow_retry = scrub_overflow_abort;
2170
2171         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2172         dev = acpi_desc->dev;
2173
2174         /*
2175          * We scrub in 2 phases.  The first phase waits for any platform
2176          * firmware initiated scrubs to complete and then we go search for the
2177          * affected spa regions to mark them scanned.  In the second phase we
2178          * initiate a directed scrub for every range that was not scrubbed in
2179          * phase 1.
2180          */
2181
2182         /* process platform firmware initiated scrubs */
2183  retry:
2184         mutex_lock(&acpi_desc->init_mutex);
2185         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2186                 struct nd_cmd_ars_status *ars_status;
2187                 struct acpi_nfit_system_address *spa;
2188                 u64 ars_start, ars_len;
2189                 int rc;
2190
2191                 if (acpi_desc->cancel)
2192                         break;
2193
2194                 if (nfit_spa->nd_region)
2195                         continue;
2196
2197                 if (init_ars_done) {
2198                         /*
2199                          * No need to re-query, we're now just
2200                          * reconciling all the ranges covered by the
2201                          * initial scrub
2202                          */
2203                         rc = 0;
2204                 } else
2205                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2206
2207                 if (rc == -ENOTTY) {
2208                         /* no ars capability, just register spa and move on */
2209                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2210                         continue;
2211                 }
2212
2213                 if (rc == -EBUSY && !tmo) {
2214                         /* fallthrough to directed scrub in phase 2 */
2215                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2216                         break;
2217                 } else if (rc == -EBUSY) {
2218                         mutex_unlock(&acpi_desc->init_mutex);
2219                         ssleep(1);
2220                         tmo--;
2221                         goto retry;
2222                 }
2223
2224                 /* we got some results, but there are more pending... */
2225                 if (rc == -ENOSPC && overflow_retry--) {
2226                         ars_status = acpi_desc->ars_status;
2227                         /*
2228                          * Record the original scrub range, so that we
2229                          * can recall all the ranges impacted by the
2230                          * initial scrub.
2231                          */
2232                         if (!init_scrub_length) {
2233                                 init_scrub_length = ars_status->length;
2234                                 init_scrub_address = ars_status->address;
2235                         }
2236                         rc = ars_continue(acpi_desc);
2237                         if (rc == 0) {
2238                                 mutex_unlock(&acpi_desc->init_mutex);
2239                                 goto retry;
2240                         }
2241                 }
2242
2243                 if (rc < 0) {
2244                         /*
2245                          * Initial scrub failed, we'll give it one more
2246                          * try below...
2247                          */
2248                         break;
2249                 }
2250
2251                 /* We got some final results, record completed ranges */
2252                 ars_status = acpi_desc->ars_status;
2253                 if (init_scrub_length) {
2254                         ars_start = init_scrub_address;
2255                         ars_len = ars_start + init_scrub_length;
2256                 } else {
2257                         ars_start = ars_status->address;
2258                         ars_len = ars_status->length;
2259                 }
2260                 spa = nfit_spa->spa;
2261
2262                 if (!init_ars_done) {
2263                         init_ars_done = true;
2264                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
2265                                         ars_start, ars_len);
2266                 }
2267                 if (ars_start <= spa->address && ars_start + ars_len
2268                                 >= spa->address + spa->length)
2269                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2270         }
2271
2272         /*
2273          * For all the ranges not covered by an initial scrub we still
2274          * want to see if there are errors, but it's ok to discover them
2275          * asynchronously.
2276          */
2277         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2278                 /*
2279                  * Flag all the ranges that still need scrubbing, but
2280                  * register them now to make data available.
2281                  */
2282                 if (nfit_spa->nd_region)
2283                         nfit_spa->ars_done = 1;
2284                 else
2285                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2286         }
2287
2288         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2289                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
2290         mutex_unlock(&acpi_desc->init_mutex);
2291 }
2292
2293 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
2294 {
2295         struct nfit_spa *nfit_spa;
2296         int rc;
2297
2298         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
2299                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
2300                         /* BLK regions don't need to wait for ars results */
2301                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
2302                         if (rc)
2303                                 return rc;
2304                 }
2305
2306         queue_work(nfit_wq, &acpi_desc->work);
2307         return 0;
2308 }
2309
2310 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
2311                 struct nfit_table_prev *prev)
2312 {
2313         struct device *dev = acpi_desc->dev;
2314
2315         if (!list_empty(&prev->spas) ||
2316                         !list_empty(&prev->memdevs) ||
2317                         !list_empty(&prev->dcrs) ||
2318                         !list_empty(&prev->bdws) ||
2319                         !list_empty(&prev->idts) ||
2320                         !list_empty(&prev->flushes)) {
2321                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
2322                 return -ENXIO;
2323         }
2324         return 0;
2325 }
2326
2327 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, acpi_size sz)
2328 {
2329         struct device *dev = acpi_desc->dev;
2330         struct nfit_table_prev prev;
2331         const void *end;
2332         u8 *data;
2333         int rc;
2334
2335         mutex_lock(&acpi_desc->init_mutex);
2336
2337         INIT_LIST_HEAD(&prev.spas);
2338         INIT_LIST_HEAD(&prev.memdevs);
2339         INIT_LIST_HEAD(&prev.dcrs);
2340         INIT_LIST_HEAD(&prev.bdws);
2341         INIT_LIST_HEAD(&prev.idts);
2342         INIT_LIST_HEAD(&prev.flushes);
2343
2344         list_cut_position(&prev.spas, &acpi_desc->spas,
2345                                 acpi_desc->spas.prev);
2346         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
2347                                 acpi_desc->memdevs.prev);
2348         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
2349                                 acpi_desc->dcrs.prev);
2350         list_cut_position(&prev.bdws, &acpi_desc->bdws,
2351                                 acpi_desc->bdws.prev);
2352         list_cut_position(&prev.idts, &acpi_desc->idts,
2353                                 acpi_desc->idts.prev);
2354         list_cut_position(&prev.flushes, &acpi_desc->flushes,
2355                                 acpi_desc->flushes.prev);
2356
2357         data = (u8 *) acpi_desc->nfit;
2358         end = data + sz;
2359         while (!IS_ERR_OR_NULL(data))
2360                 data = add_table(acpi_desc, &prev, data, end);
2361
2362         if (IS_ERR(data)) {
2363                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
2364                                 PTR_ERR(data));
2365                 rc = PTR_ERR(data);
2366                 goto out_unlock;
2367         }
2368
2369         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
2370         if (rc)
2371                 goto out_unlock;
2372
2373         if (nfit_mem_init(acpi_desc) != 0) {
2374                 rc = -ENOMEM;
2375                 goto out_unlock;
2376         }
2377
2378         acpi_nfit_init_dsms(acpi_desc);
2379
2380         rc = acpi_nfit_register_dimms(acpi_desc);
2381         if (rc)
2382                 goto out_unlock;
2383
2384         rc = acpi_nfit_register_regions(acpi_desc);
2385
2386  out_unlock:
2387         mutex_unlock(&acpi_desc->init_mutex);
2388         return rc;
2389 }
2390 EXPORT_SYMBOL_GPL(acpi_nfit_init);
2391
2392 struct acpi_nfit_flush_work {
2393         struct work_struct work;
2394         struct completion cmp;
2395 };
2396
2397 static void flush_probe(struct work_struct *work)
2398 {
2399         struct acpi_nfit_flush_work *flush;
2400
2401         flush = container_of(work, typeof(*flush), work);
2402         complete(&flush->cmp);
2403 }
2404
2405 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
2406 {
2407         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2408         struct device *dev = acpi_desc->dev;
2409         struct acpi_nfit_flush_work flush;
2410
2411         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
2412         device_lock(dev);
2413         device_unlock(dev);
2414
2415         /*
2416          * Scrub work could take 10s of seconds, userspace may give up so we
2417          * need to be interruptible while waiting.
2418          */
2419         INIT_WORK_ONSTACK(&flush.work, flush_probe);
2420         COMPLETION_INITIALIZER_ONSTACK(flush.cmp);
2421         queue_work(nfit_wq, &flush.work);
2422         return wait_for_completion_interruptible(&flush.cmp);
2423 }
2424
2425 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
2426                 struct nvdimm *nvdimm, unsigned int cmd)
2427 {
2428         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
2429
2430         if (nvdimm)
2431                 return 0;
2432         if (cmd != ND_CMD_ARS_START)
2433                 return 0;
2434
2435         /*
2436          * The kernel and userspace may race to initiate a scrub, but
2437          * the scrub thread is prepared to lose that initial race.  It
2438          * just needs guarantees that any ars it initiates are not
2439          * interrupted by any intervening start reqeusts from userspace.
2440          */
2441         if (work_busy(&acpi_desc->work))
2442                 return -EBUSY;
2443
2444         return 0;
2445 }
2446
2447 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
2448 {
2449         struct nvdimm_bus_descriptor *nd_desc;
2450
2451         dev_set_drvdata(dev, acpi_desc);
2452         acpi_desc->dev = dev;
2453         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
2454         nd_desc = &acpi_desc->nd_desc;
2455         nd_desc->provider_name = "ACPI.NFIT";
2456         nd_desc->ndctl = acpi_nfit_ctl;
2457         nd_desc->flush_probe = acpi_nfit_flush_probe;
2458         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
2459         nd_desc->attr_groups = acpi_nfit_attribute_groups;
2460
2461         INIT_LIST_HEAD(&acpi_desc->spa_maps);
2462         INIT_LIST_HEAD(&acpi_desc->spas);
2463         INIT_LIST_HEAD(&acpi_desc->dcrs);
2464         INIT_LIST_HEAD(&acpi_desc->bdws);
2465         INIT_LIST_HEAD(&acpi_desc->idts);
2466         INIT_LIST_HEAD(&acpi_desc->flushes);
2467         INIT_LIST_HEAD(&acpi_desc->memdevs);
2468         INIT_LIST_HEAD(&acpi_desc->dimms);
2469         mutex_init(&acpi_desc->spa_map_mutex);
2470         mutex_init(&acpi_desc->init_mutex);
2471         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
2472 }
2473 EXPORT_SYMBOL_GPL(acpi_nfit_desc_init);
2474
2475 static int acpi_nfit_add(struct acpi_device *adev)
2476 {
2477         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2478         struct acpi_nfit_desc *acpi_desc;
2479         struct device *dev = &adev->dev;
2480         struct acpi_table_header *tbl;
2481         acpi_status status = AE_OK;
2482         acpi_size sz;
2483         int rc;
2484
2485         status = acpi_get_table_with_size(ACPI_SIG_NFIT, 0, &tbl, &sz);
2486         if (ACPI_FAILURE(status)) {
2487                 /* This is ok, we could have an nvdimm hotplugged later */
2488                 dev_dbg(dev, "failed to find NFIT at startup\n");
2489                 return 0;
2490         }
2491
2492         acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2493         if (!acpi_desc)
2494                 return -ENOMEM;
2495         acpi_nfit_desc_init(acpi_desc, &adev->dev);
2496         acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2497         if (!acpi_desc->nvdimm_bus)
2498                 return -ENOMEM;
2499
2500         /*
2501          * Save the acpi header for later and then skip it,
2502          * making nfit point to the first nfit table header.
2503          */
2504         acpi_desc->acpi_header = *tbl;
2505         acpi_desc->nfit = (void *) tbl + sizeof(struct acpi_table_nfit);
2506         sz -= sizeof(struct acpi_table_nfit);
2507
2508         /* Evaluate _FIT and override with that if present */
2509         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2510         if (ACPI_SUCCESS(status) && buf.length > 0) {
2511                 union acpi_object *obj;
2512                 /*
2513                  * Adjust for the acpi_object header of the _FIT
2514                  */
2515                 obj = buf.pointer;
2516                 if (obj->type == ACPI_TYPE_BUFFER) {
2517                         acpi_desc->nfit =
2518                                 (struct acpi_nfit_header *)obj->buffer.pointer;
2519                         sz = obj->buffer.length;
2520                 } else
2521                         dev_dbg(dev, "%s invalid type %d, ignoring _FIT\n",
2522                                  __func__, (int) obj->type);
2523         }
2524
2525         rc = acpi_nfit_init(acpi_desc, sz);
2526         if (rc) {
2527                 nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2528                 return rc;
2529         }
2530         return 0;
2531 }
2532
2533 static int acpi_nfit_remove(struct acpi_device *adev)
2534 {
2535         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2536
2537         acpi_desc->cancel = 1;
2538         flush_workqueue(nfit_wq);
2539         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
2540         return 0;
2541 }
2542
2543 static void acpi_nfit_notify(struct acpi_device *adev, u32 event)
2544 {
2545         struct acpi_nfit_desc *acpi_desc = dev_get_drvdata(&adev->dev);
2546         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
2547         struct acpi_nfit_header *nfit_saved;
2548         union acpi_object *obj;
2549         struct device *dev = &adev->dev;
2550         acpi_status status;
2551         int ret;
2552
2553         dev_dbg(dev, "%s: event: %d\n", __func__, event);
2554
2555         device_lock(dev);
2556         if (!dev->driver) {
2557                 /* dev->driver may be null if we're being removed */
2558                 dev_dbg(dev, "%s: no driver found for dev\n", __func__);
2559                 goto out_unlock;
2560         }
2561
2562         if (!acpi_desc) {
2563                 acpi_desc = devm_kzalloc(dev, sizeof(*acpi_desc), GFP_KERNEL);
2564                 if (!acpi_desc)
2565                         goto out_unlock;
2566                 acpi_nfit_desc_init(acpi_desc, &adev->dev);
2567                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev, &acpi_desc->nd_desc);
2568                 if (!acpi_desc->nvdimm_bus)
2569                         goto out_unlock;
2570         } else {
2571                 /*
2572                  * Finish previous registration before considering new
2573                  * regions.
2574                  */
2575                 flush_workqueue(nfit_wq);
2576         }
2577
2578         /* Evaluate _FIT */
2579         status = acpi_evaluate_object(adev->handle, "_FIT", NULL, &buf);
2580         if (ACPI_FAILURE(status)) {
2581                 dev_err(dev, "failed to evaluate _FIT\n");
2582                 goto out_unlock;
2583         }
2584
2585         nfit_saved = acpi_desc->nfit;
2586         obj = buf.pointer;
2587         if (obj->type == ACPI_TYPE_BUFFER) {
2588                 acpi_desc->nfit =
2589                         (struct acpi_nfit_header *)obj->buffer.pointer;
2590                 ret = acpi_nfit_init(acpi_desc, obj->buffer.length);
2591                 if (ret) {
2592                         /* Merge failed, restore old nfit, and exit */
2593                         acpi_desc->nfit = nfit_saved;
2594                         dev_err(dev, "failed to merge updated NFIT\n");
2595                 }
2596         } else {
2597                 /* Bad _FIT, restore old nfit */
2598                 dev_err(dev, "Invalid _FIT\n");
2599         }
2600         kfree(buf.pointer);
2601
2602  out_unlock:
2603         device_unlock(dev);
2604 }
2605
2606 static const struct acpi_device_id acpi_nfit_ids[] = {
2607         { "ACPI0012", 0 },
2608         { "", 0 },
2609 };
2610 MODULE_DEVICE_TABLE(acpi, acpi_nfit_ids);
2611
2612 static struct acpi_driver acpi_nfit_driver = {
2613         .name = KBUILD_MODNAME,
2614         .ids = acpi_nfit_ids,
2615         .ops = {
2616                 .add = acpi_nfit_add,
2617                 .remove = acpi_nfit_remove,
2618                 .notify = acpi_nfit_notify,
2619         },
2620 };
2621
2622 static __init int nfit_init(void)
2623 {
2624         BUILD_BUG_ON(sizeof(struct acpi_table_nfit) != 40);
2625         BUILD_BUG_ON(sizeof(struct acpi_nfit_system_address) != 56);
2626         BUILD_BUG_ON(sizeof(struct acpi_nfit_memory_map) != 48);
2627         BUILD_BUG_ON(sizeof(struct acpi_nfit_interleave) != 20);
2628         BUILD_BUG_ON(sizeof(struct acpi_nfit_smbios) != 9);
2629         BUILD_BUG_ON(sizeof(struct acpi_nfit_control_region) != 80);
2630         BUILD_BUG_ON(sizeof(struct acpi_nfit_data_region) != 40);
2631
2632         acpi_str_to_uuid(UUID_VOLATILE_MEMORY, nfit_uuid[NFIT_SPA_VOLATILE]);
2633         acpi_str_to_uuid(UUID_PERSISTENT_MEMORY, nfit_uuid[NFIT_SPA_PM]);
2634         acpi_str_to_uuid(UUID_CONTROL_REGION, nfit_uuid[NFIT_SPA_DCR]);
2635         acpi_str_to_uuid(UUID_DATA_REGION, nfit_uuid[NFIT_SPA_BDW]);
2636         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_VDISK]);
2637         acpi_str_to_uuid(UUID_VOLATILE_VIRTUAL_CD, nfit_uuid[NFIT_SPA_VCD]);
2638         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_DISK, nfit_uuid[NFIT_SPA_PDISK]);
2639         acpi_str_to_uuid(UUID_PERSISTENT_VIRTUAL_CD, nfit_uuid[NFIT_SPA_PCD]);
2640         acpi_str_to_uuid(UUID_NFIT_BUS, nfit_uuid[NFIT_DEV_BUS]);
2641         acpi_str_to_uuid(UUID_NFIT_DIMM, nfit_uuid[NFIT_DEV_DIMM]);
2642         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE1, nfit_uuid[NFIT_DEV_DIMM_N_HPE1]);
2643         acpi_str_to_uuid(UUID_NFIT_DIMM_N_HPE2, nfit_uuid[NFIT_DEV_DIMM_N_HPE2]);
2644
2645         nfit_wq = create_singlethread_workqueue("nfit");
2646         if (!nfit_wq)
2647                 return -ENOMEM;
2648
2649         return acpi_bus_register_driver(&acpi_nfit_driver);
2650 }
2651
2652 static __exit void nfit_exit(void)
2653 {
2654         acpi_bus_unregister_driver(&acpi_nfit_driver);
2655         destroy_workqueue(nfit_wq);
2656 }
2657
2658 module_init(nfit_init);
2659 module_exit(nfit_exit);
2660 MODULE_LICENSE("GPL v2");
2661 MODULE_AUTHOR("Intel Corporation");