drivers/acpi/nfit/core.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/sysfs.h>
  19 #include <linux/delay.h>
  20 #include <linux/list.h>
  21 #include <linux/acpi.h>
  22 #include <linux/sort.h>
  23 #include <linux/io.h>
  24 #include <linux/nd.h>
  25 #include <asm/cacheflush.h>
  26 #include <acpi/nfit.h>
  27 #include "nfit.h"
  28
  29 /*
  30  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  31  * irrelevant.
  32  */
  33 #include <linux/io-64-nonatomic-hi-lo.h>
  34
  35 static bool force_enable_dimms;
  36 module_param(force_enable_dimms, bool, S_IRUGO|S_IWUSR);
  37 MODULE_PARM_DESC(force_enable_dimms, "Ignore _STA (ACPI DIMM device) status");
  38
  39 static unsigned int scrub_timeout = NFIT_ARS_TIMEOUT;
  40 module_param(scrub_timeout, uint, S_IRUGO|S_IWUSR);
  41 MODULE_PARM_DESC(scrub_timeout, "Initial scrub timeout in seconds");
  42
  43 /* after three payloads of overflow, it's dead jim */
  44 static unsigned int scrub_overflow_abort = 3;
  45 module_param(scrub_overflow_abort, uint, S_IRUGO|S_IWUSR);
  46 MODULE_PARM_DESC(scrub_overflow_abort,
  47                 "Number of times we overflow ARS results before abort");
  48
  49 static bool disable_vendor_specific;
  50 module_param(disable_vendor_specific, bool, S_IRUGO);
  51 MODULE_PARM_DESC(disable_vendor_specific,
  52                 "Limit commands to the publicly specified set");
  53
  54 static unsigned long override_dsm_mask;
  55 module_param(override_dsm_mask, ulong, S_IRUGO);
  56 MODULE_PARM_DESC(override_dsm_mask, "Bitmask of allowed NVDIMM DSM functions");
  57
  58 static int default_dsm_family = -1;
  59 module_param(default_dsm_family, int, S_IRUGO);
  60 MODULE_PARM_DESC(default_dsm_family,
  61                 "Try this DSM type first when identifying NVDIMM family");
  62
  63 LIST_HEAD(acpi_descs);
  64 DEFINE_MUTEX(acpi_desc_lock);
  65
  66 static struct workqueue_struct *nfit_wq;
  67
  68 struct nfit_table_prev {
  69         struct list_head spas;
  70         struct list_head memdevs;
  71         struct list_head dcrs;
  72         struct list_head bdws;
  73         struct list_head idts;
  74         struct list_head flushes;
  75 };
  76
  77 static guid_t nfit_uuid[NFIT_UUID_MAX];
  78
  79 const guid_t *to_nfit_uuid(enum nfit_uuids id)
  80 {
  81         return &nfit_uuid[id];
  82 }
  83 EXPORT_SYMBOL(to_nfit_uuid);
  84
  85 static struct acpi_nfit_desc *to_acpi_nfit_desc(
  86                 struct nvdimm_bus_descriptor *nd_desc)
  87 {
  88         return container_of(nd_desc, struct acpi_nfit_desc, nd_desc);
  89 }
  90
  91 static struct acpi_device *to_acpi_dev(struct acpi_nfit_desc *acpi_desc)
  92 {
  93         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
  94
  95         /*
  96          * If provider == 'ACPI.NFIT' we can assume 'dev' is a struct
  97          * acpi_device.
  98          */
  99         if (!nd_desc->provider_name
 100                         || strcmp(nd_desc->provider_name, "ACPI.NFIT") != 0)
 101                 return NULL;
 102
 103         return to_acpi_device(acpi_desc->dev);
 104 }
 105
 106 static int xlat_bus_status(void *buf, unsigned int cmd, u32 status)
 107 {
 108         struct nd_cmd_clear_error *clear_err;
 109         struct nd_cmd_ars_status *ars_status;
 110         u16 flags;
 111
 112         switch (cmd) {
 113         case ND_CMD_ARS_CAP:
 114                 if ((status & 0xffff) == NFIT_ARS_CAP_NONE)
 115                         return -ENOTTY;
 116
 117                 /* Command failed */
 118                 if (status & 0xffff)
 119                         return -EIO;
 120
 121                 /* No supported scan types for this range */
 122                 flags = ND_ARS_PERSISTENT | ND_ARS_VOLATILE;
 123                 if ((status >> 16 & flags) == 0)
 124                         return -ENOTTY;
 125                 return 0;
 126         case ND_CMD_ARS_START:
 127                 /* ARS is in progress */
 128                 if ((status & 0xffff) == NFIT_ARS_START_BUSY)
 129                         return -EBUSY;
 130
 131                 /* Command failed */
 132                 if (status & 0xffff)
 133                         return -EIO;
 134                 return 0;
 135         case ND_CMD_ARS_STATUS:
 136                 ars_status = buf;
 137                 /* Command failed */
 138                 if (status & 0xffff)
 139                         return -EIO;
 140                 /* Check extended status (Upper two bytes) */
 141                 if (status == NFIT_ARS_STATUS_DONE)
 142                         return 0;
 143
 144                 /* ARS is in progress */
 145                 if (status == NFIT_ARS_STATUS_BUSY)
 146                         return -EBUSY;
 147
 148                 /* No ARS performed for the current boot */
 149                 if (status == NFIT_ARS_STATUS_NONE)
 150                         return -EAGAIN;
 151
 152                 /*
 153                  * ARS interrupted, either we overflowed or some other
 154                  * agent wants the scan to stop.  If we didn't overflow
 155                  * then just continue with the returned results.
 156                  */
 157                 if (status == NFIT_ARS_STATUS_INTR) {
 158                         if (ars_status->out_length >= 40 && (ars_status->flags
 159                                                 & NFIT_ARS_F_OVERFLOW))
 160                                 return -ENOSPC;
 161                         return 0;
 162                 }
 163
 164                 /* Unknown status */
 165                 if (status >> 16)
 166                         return -EIO;
 167                 return 0;
 168         case ND_CMD_CLEAR_ERROR:
 169                 clear_err = buf;
 170                 if (status & 0xffff)
 171                         return -EIO;
 172                 if (!clear_err->cleared)
 173                         return -EIO;
 174                 if (clear_err->length > clear_err->cleared)
 175                         return clear_err->cleared;
 176                 return 0;
 177         default:
 178                 break;
 179         }
 180
 181         /* all other non-zero status results in an error */
 182         if (status)
 183                 return -EIO;
 184         return 0;
 185 }
 186
 187 #define ACPI_LABELS_LOCKED 3
 188
 189 static int xlat_nvdimm_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 190                 u32 status)
 191 {
 192         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 193
 194         switch (cmd) {
 195         case ND_CMD_GET_CONFIG_SIZE:
 196                 /*
 197                  * In the _LSI, _LSR, _LSW case the locked status is
 198                  * communicated via the read/write commands
 199                  */
 200                 if (nfit_mem->has_lsi)
 201                         break;
 202
 203                 if (status >> 16 & ND_CONFIG_LOCKED)
 204                         return -EACCES;
 205                 break;
 206         case ND_CMD_GET_CONFIG_DATA:
 207                 if (nfit_mem->has_lsr && status == ACPI_LABELS_LOCKED)
 208                         return -EACCES;
 209                 break;
 210         case ND_CMD_SET_CONFIG_DATA:
 211                 if (nfit_mem->has_lsw && status == ACPI_LABELS_LOCKED)
 212                         return -EACCES;
 213                 break;
 214         default:
 215                 break;
 216         }
 217
 218         /* all other non-zero status results in an error */
 219         if (status)
 220                 return -EIO;
 221         return 0;
 222 }
 223
 224 static int xlat_status(struct nvdimm *nvdimm, void *buf, unsigned int cmd,
 225                 u32 status)
 226 {
 227         if (!nvdimm)
 228                 return xlat_bus_status(buf, cmd, status);
 229         return xlat_nvdimm_status(nvdimm, buf, cmd, status);
 230 }
 231
 232 /* convert _LS{I,R} packages to the buffer object acpi_nfit_ctl expects */
 233 static union acpi_object *pkg_to_buf(union acpi_object *pkg)
 234 {
 235         int i;
 236         void *dst;
 237         size_t size = 0;
 238         union acpi_object *buf = NULL;
 239
 240         if (pkg->type != ACPI_TYPE_PACKAGE) {
 241                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 242                                 pkg->type);
 243                 goto err;
 244         }
 245
 246         for (i = 0; i < pkg->package.count; i++) {
 247                 union acpi_object *obj = &pkg->package.elements[i];
 248
 249                 if (obj->type == ACPI_TYPE_INTEGER)
 250                         size += 4;
 251                 else if (obj->type == ACPI_TYPE_BUFFER)
 252                         size += obj->buffer.length;
 253                 else {
 254                         WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 255                                         obj->type);
 256                         goto err;
 257                 }
 258         }
 259
 260         buf = ACPI_ALLOCATE(sizeof(*buf) + size);
 261         if (!buf)
 262                 goto err;
 263
 264         dst = buf + 1;
 265         buf->type = ACPI_TYPE_BUFFER;
 266         buf->buffer.length = size;
 267         buf->buffer.pointer = dst;
 268         for (i = 0; i < pkg->package.count; i++) {
 269                 union acpi_object *obj = &pkg->package.elements[i];
 270
 271                 if (obj->type == ACPI_TYPE_INTEGER) {
 272                         memcpy(dst, &obj->integer.value, 4);
 273                         dst += 4;
 274                 } else if (obj->type == ACPI_TYPE_BUFFER) {
 275                         memcpy(dst, obj->buffer.pointer, obj->buffer.length);
 276                         dst += obj->buffer.length;
 277                 }
 278         }
 279 err:
 280         ACPI_FREE(pkg);
 281         return buf;
 282 }
 283
 284 static union acpi_object *int_to_buf(union acpi_object *integer)
 285 {
 286         union acpi_object *buf = ACPI_ALLOCATE(sizeof(*buf) + 4);
 287         void *dst = NULL;
 288
 289         if (!buf)
 290                 goto err;
 291
 292         if (integer->type != ACPI_TYPE_INTEGER) {
 293                 WARN_ONCE(1, "BIOS bug, unexpected element type: %d\n",
 294                                 integer->type);
 295                 goto err;
 296         }
 297
 298         dst = buf + 1;
 299         buf->type = ACPI_TYPE_BUFFER;
 300         buf->buffer.length = 4;
 301         buf->buffer.pointer = dst;
 302         memcpy(dst, &integer->integer.value, 4);
 303 err:
 304         ACPI_FREE(integer);
 305         return buf;
 306 }
 307
 308 static union acpi_object *acpi_label_write(acpi_handle handle, u32 offset,
 309                 u32 len, void *data)
 310 {
 311         acpi_status rc;
 312         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 313         struct acpi_object_list input = {
 314                 .count = 3,
 315                 .pointer = (union acpi_object []) {
 316                         [0] = {
 317                                 .integer.type = ACPI_TYPE_INTEGER,
 318                                 .integer.value = offset,
 319                         },
 320                         [1] = {
 321                                 .integer.type = ACPI_TYPE_INTEGER,
 322                                 .integer.value = len,
 323                         },
 324                         [2] = {
 325                                 .buffer.type = ACPI_TYPE_BUFFER,
 326                                 .buffer.pointer = data,
 327                                 .buffer.length = len,
 328                         },
 329                 },
 330         };
 331
 332         rc = acpi_evaluate_object(handle, "_LSW", &input, &buf);
 333         if (ACPI_FAILURE(rc))
 334                 return NULL;
 335         return int_to_buf(buf.pointer);
 336 }
 337
 338 static union acpi_object *acpi_label_read(acpi_handle handle, u32 offset,
 339                 u32 len)
 340 {
 341         acpi_status rc;
 342         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 343         struct acpi_object_list input = {
 344                 .count = 2,
 345                 .pointer = (union acpi_object []) {
 346                         [0] = {
 347                                 .integer.type = ACPI_TYPE_INTEGER,
 348                                 .integer.value = offset,
 349                         },
 350                         [1] = {
 351                                 .integer.type = ACPI_TYPE_INTEGER,
 352                                 .integer.value = len,
 353                         },
 354                 },
 355         };
 356
 357         rc = acpi_evaluate_object(handle, "_LSR", &input, &buf);
 358         if (ACPI_FAILURE(rc))
 359                 return NULL;
 360         return pkg_to_buf(buf.pointer);
 361 }
 362
 363 static union acpi_object *acpi_label_info(acpi_handle handle)
 364 {
 365         acpi_status rc;
 366         struct acpi_buffer buf = { ACPI_ALLOCATE_BUFFER, NULL };
 367
 368         rc = acpi_evaluate_object(handle, "_LSI", NULL, &buf);
 369         if (ACPI_FAILURE(rc))
 370                 return NULL;
 371         return pkg_to_buf(buf.pointer);
 372 }
 373
 374 static u8 nfit_dsm_revid(unsigned family, unsigned func)
 375 {
 376         static const u8 revid_table[NVDIMM_FAMILY_MAX+1][32] = {
 377                 [NVDIMM_FAMILY_INTEL] = {
 378                         [NVDIMM_INTEL_GET_MODES] = 2,
 379                         [NVDIMM_INTEL_GET_FWINFO] = 2,
 380                         [NVDIMM_INTEL_START_FWUPDATE] = 2,
 381                         [NVDIMM_INTEL_SEND_FWUPDATE] = 2,
 382                         [NVDIMM_INTEL_FINISH_FWUPDATE] = 2,
 383                         [NVDIMM_INTEL_QUERY_FWUPDATE] = 2,
 384                         [NVDIMM_INTEL_SET_THRESHOLD] = 2,
 385                         [NVDIMM_INTEL_INJECT_ERROR] = 2,
 386                 },
 387         };
 388         u8 id;
 389
 390         if (family > NVDIMM_FAMILY_MAX)
 391                 return 0;
 392         if (func > 31)
 393                 return 0;
 394         id = revid_table[family][func];
 395         if (id == 0)
 396                 return 1; /* default */
 397         return id;
 398 }
 399
 400 int acpi_nfit_ctl(struct nvdimm_bus_descriptor *nd_desc, struct nvdimm *nvdimm,
 401                 unsigned int cmd, void *buf, unsigned int buf_len, int *cmd_rc)
 402 {
 403         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
 404         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
 405         union acpi_object in_obj, in_buf, *out_obj;
 406         const struct nd_cmd_desc *desc = NULL;
 407         struct device *dev = acpi_desc->dev;
 408         struct nd_cmd_pkg *call_pkg = NULL;
 409         const char *cmd_name, *dimm_name;
 410         unsigned long cmd_mask, dsm_mask;
 411         u32 offset, fw_status = 0;
 412         acpi_handle handle;
 413         unsigned int func;
 414         const guid_t *guid;
 415         int rc, i;
 416
 417         func = cmd;
 418         if (cmd == ND_CMD_CALL) {
 419                 call_pkg = buf;
 420                 func = call_pkg->nd_command;
 421
 422                 for (i = 0; i < ARRAY_SIZE(call_pkg->nd_reserved2); i++)
 423                         if (call_pkg->nd_reserved2[i])
 424                                 return -EINVAL;
 425         }
 426
 427         if (nvdimm) {
 428                 struct acpi_device *adev = nfit_mem->adev;
 429
 430                 if (!adev)
 431                         return -ENOTTY;
 432                 if (call_pkg && nfit_mem->family != call_pkg->nd_family)
 433                         return -ENOTTY;
 434
 435                 dimm_name = nvdimm_name(nvdimm);
 436                 cmd_name = nvdimm_cmd_name(cmd);
 437                 cmd_mask = nvdimm_cmd_mask(nvdimm);
 438                 dsm_mask = nfit_mem->dsm_mask;
 439                 desc = nd_cmd_dimm_desc(cmd);
 440                 guid = to_nfit_uuid(nfit_mem->family);
 441                 handle = adev->handle;
 442         } else {
 443                 struct acpi_device *adev = to_acpi_dev(acpi_desc);
 444
 445                 cmd_name = nvdimm_bus_cmd_name(cmd);
 446                 cmd_mask = nd_desc->cmd_mask;
 447                 dsm_mask = cmd_mask;
 448                 if (cmd == ND_CMD_CALL)
 449                         dsm_mask = nd_desc->bus_dsm_mask;
 450                 desc = nd_cmd_bus_desc(cmd);
 451                 guid = to_nfit_uuid(NFIT_DEV_BUS);
 452                 handle = adev->handle;
 453                 dimm_name = "bus";
 454         }
 455
 456         if (!desc || (cmd && (desc->out_num + desc->in_num == 0)))
 457                 return -ENOTTY;
 458
 459         if (!test_bit(cmd, &cmd_mask) || !test_bit(func, &dsm_mask))
 460                 return -ENOTTY;
 461
 462         in_obj.type = ACPI_TYPE_PACKAGE;
 463         in_obj.package.count = 1;
 464         in_obj.package.elements = &in_buf;
 465         in_buf.type = ACPI_TYPE_BUFFER;
 466         in_buf.buffer.pointer = buf;
 467         in_buf.buffer.length = 0;
 468
 469         /* libnvdimm has already validated the input envelope */
 470         for (i = 0; i < desc->in_num; i++)
 471                 in_buf.buffer.length += nd_cmd_in_size(nvdimm, cmd, desc,
 472                                 i, buf);
 473
 474         if (call_pkg) {
 475                 /* skip over package wrapper */
 476                 in_buf.buffer.pointer = (void *) &call_pkg->nd_payload;
 477                 in_buf.buffer.length = call_pkg->nd_size_in;
 478         }
 479
 480         dev_dbg(dev, "%s:%s cmd: %d: func: %d input length: %d\n",
 481                         __func__, dimm_name, cmd, func, in_buf.buffer.length);
 482         print_hex_dump_debug("nvdimm in  ", DUMP_PREFIX_OFFSET, 4, 4,
 483                         in_buf.buffer.pointer,
 484                         min_t(u32, 256, in_buf.buffer.length), true);
 485
 486         /* call the BIOS, prefer the named methods over _DSM if available */
 487         if (nvdimm && cmd == ND_CMD_GET_CONFIG_SIZE && nfit_mem->has_lsi)
 488                 out_obj = acpi_label_info(handle);
 489         else if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && nfit_mem->has_lsr) {
 490                 struct nd_cmd_get_config_data_hdr *p = buf;
 491
 492                 out_obj = acpi_label_read(handle, p->in_offset, p->in_length);
 493         } else if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA
 494                         && nfit_mem->has_lsw) {
 495                 struct nd_cmd_set_config_hdr *p = buf;
 496
 497                 out_obj = acpi_label_write(handle, p->in_offset, p->in_length,
 498                                 p->in_buf);
 499         } else {
 500                 u8 revid;
 501
 502                 if (nvdimm)
 503                         revid = nfit_dsm_revid(nfit_mem->family, func);
 504                 else
 505                         revid = 1;
 506                 out_obj = acpi_evaluate_dsm(handle, guid, revid, func, &in_obj);
 507         }
 508
 509         if (!out_obj) {
 510                 dev_dbg(dev, "%s:%s _DSM failed cmd: %s\n", __func__, dimm_name,
 511                                 cmd_name);
 512                 return -EINVAL;
 513         }
 514
 515         if (call_pkg) {
 516                 call_pkg->nd_fw_size = out_obj->buffer.length;
 517                 memcpy(call_pkg->nd_payload + call_pkg->nd_size_in,
 518                         out_obj->buffer.pointer,
 519                         min(call_pkg->nd_fw_size, call_pkg->nd_size_out));
 520
 521                 ACPI_FREE(out_obj);
 522                 /*
 523                  * Need to support FW function w/o known size in advance.
 524                  * Caller can determine required size based upon nd_fw_size.
 525                  * If we return an error (like elsewhere) then caller wouldn't
 526                  * be able to rely upon data returned to make calculation.
 527                  */
 528                 return 0;
 529         }
 530
 531         if (out_obj->package.type != ACPI_TYPE_BUFFER) {
 532                 dev_dbg(dev, "%s:%s unexpected output object type cmd: %s type: %d\n",
 533                                 __func__, dimm_name, cmd_name, out_obj->type);
 534                 rc = -EINVAL;
 535                 goto out;
 536         }
 537
 538         dev_dbg(dev, "%s:%s cmd: %s output length: %d\n", __func__, dimm_name,
 539                         cmd_name, out_obj->buffer.length);
 540         print_hex_dump_debug(cmd_name, DUMP_PREFIX_OFFSET, 4, 4,
 541                         out_obj->buffer.pointer,
 542                         min_t(u32, 128, out_obj->buffer.length), true);
 543
 544         for (i = 0, offset = 0; i < desc->out_num; i++) {
 545                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i, buf,
 546                                 (u32 *) out_obj->buffer.pointer,
 547                                 out_obj->buffer.length - offset);
 548
 549                 if (offset + out_size > out_obj->buffer.length) {
 550                         dev_dbg(dev, "%s:%s output object underflow cmd: %s field: %d\n",
 551                                         __func__, dimm_name, cmd_name, i);
 552                         break;
 553                 }
 554
 555                 if (in_buf.buffer.length + offset + out_size > buf_len) {
 556                         dev_dbg(dev, "%s:%s output overrun cmd: %s field: %d\n",
 557                                         __func__, dimm_name, cmd_name, i);
 558                         rc = -ENXIO;
 559                         goto out;
 560                 }
 561                 memcpy(buf + in_buf.buffer.length + offset,
 562                                 out_obj->buffer.pointer + offset, out_size);
 563                 offset += out_size;
 564         }
 565
 566         /*
 567          * Set fw_status for all the commands with a known format to be
 568          * later interpreted by xlat_status().
 569          */
 570         if (i >= 1 && ((!nvdimm && cmd >= ND_CMD_ARS_CAP
 571                                         && cmd <= ND_CMD_CLEAR_ERROR)
 572                                 || (nvdimm && cmd >= ND_CMD_SMART
 573                                         && cmd <= ND_CMD_VENDOR)))
 574                 fw_status = *(u32 *) out_obj->buffer.pointer;
 575
 576         if (offset + in_buf.buffer.length < buf_len) {
 577                 if (i >= 1) {
 578                         /*
 579                          * status valid, return the number of bytes left
 580                          * unfilled in the output buffer
 581                          */
 582                         rc = buf_len - offset - in_buf.buffer.length;
 583                         if (cmd_rc)
 584                                 *cmd_rc = xlat_status(nvdimm, buf, cmd,
 585                                                 fw_status);
 586                 } else {
 587                         dev_err(dev, "%s:%s underrun cmd: %s buf_len: %d out_len: %d\n",
 588                                         __func__, dimm_name, cmd_name, buf_len,
 589                                         offset);
 590                         rc = -ENXIO;
 591                 }
 592         } else {
 593                 rc = 0;
 594                 if (cmd_rc)
 595                         *cmd_rc = xlat_status(nvdimm, buf, cmd, fw_status);
 596         }
 597
 598  out:
 599         ACPI_FREE(out_obj);
 600
 601         return rc;
 602 }
 603 EXPORT_SYMBOL_GPL(acpi_nfit_ctl);
 604
 605 static const char *spa_type_name(u16 type)
 606 {
 607         static const char *to_name[] = {
 608                 [NFIT_SPA_VOLATILE] = "volatile",
 609                 [NFIT_SPA_PM] = "pmem",
 610                 [NFIT_SPA_DCR] = "dimm-control-region",
 611                 [NFIT_SPA_BDW] = "block-data-window",
 612                 [NFIT_SPA_VDISK] = "volatile-disk",
 613                 [NFIT_SPA_VCD] = "volatile-cd",
 614                 [NFIT_SPA_PDISK] = "persistent-disk",
 615                 [NFIT_SPA_PCD] = "persistent-cd",
 616
 617         };
 618
 619         if (type > NFIT_SPA_PCD)
 620                 return "unknown";
 621
 622         return to_name[type];
 623 }
 624
 625 int nfit_spa_type(struct acpi_nfit_system_address *spa)
 626 {
 627         int i;
 628
 629         for (i = 0; i < NFIT_UUID_MAX; i++)
 630                 if (guid_equal(to_nfit_uuid(i), (guid_t *)&spa->range_guid))
 631                         return i;
 632         return -1;
 633 }
 634
 635 static bool add_spa(struct acpi_nfit_desc *acpi_desc,
 636                 struct nfit_table_prev *prev,
 637                 struct acpi_nfit_system_address *spa)
 638 {
 639         struct device *dev = acpi_desc->dev;
 640         struct nfit_spa *nfit_spa;
 641
 642         if (spa->header.length != sizeof(*spa))
 643                 return false;
 644
 645         list_for_each_entry(nfit_spa, &prev->spas, list) {
 646                 if (memcmp(nfit_spa->spa, spa, sizeof(*spa)) == 0) {
 647                         list_move_tail(&nfit_spa->list, &acpi_desc->spas);
 648                         return true;
 649                 }
 650         }
 651
 652         nfit_spa = devm_kzalloc(dev, sizeof(*nfit_spa) + sizeof(*spa),
 653                         GFP_KERNEL);
 654         if (!nfit_spa)
 655                 return false;
 656         INIT_LIST_HEAD(&nfit_spa->list);
 657         memcpy(nfit_spa->spa, spa, sizeof(*spa));
 658         list_add_tail(&nfit_spa->list, &acpi_desc->spas);
 659         dev_dbg(dev, "%s: spa index: %d type: %s\n", __func__,
 660                         spa->range_index,
 661                         spa_type_name(nfit_spa_type(spa)));
 662         return true;
 663 }
 664
 665 static bool add_memdev(struct acpi_nfit_desc *acpi_desc,
 666                 struct nfit_table_prev *prev,
 667                 struct acpi_nfit_memory_map *memdev)
 668 {
 669         struct device *dev = acpi_desc->dev;
 670         struct nfit_memdev *nfit_memdev;
 671
 672         if (memdev->header.length != sizeof(*memdev))
 673                 return false;
 674
 675         list_for_each_entry(nfit_memdev, &prev->memdevs, list)
 676                 if (memcmp(nfit_memdev->memdev, memdev, sizeof(*memdev)) == 0) {
 677                         list_move_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 678                         return true;
 679                 }
 680
 681         nfit_memdev = devm_kzalloc(dev, sizeof(*nfit_memdev) + sizeof(*memdev),
 682                         GFP_KERNEL);
 683         if (!nfit_memdev)
 684                 return false;
 685         INIT_LIST_HEAD(&nfit_memdev->list);
 686         memcpy(nfit_memdev->memdev, memdev, sizeof(*memdev));
 687         list_add_tail(&nfit_memdev->list, &acpi_desc->memdevs);
 688         dev_dbg(dev, "%s: memdev handle: %#x spa: %d dcr: %d flags: %#x\n",
 689                         __func__, memdev->device_handle, memdev->range_index,
 690                         memdev->region_index, memdev->flags);
 691         return true;
 692 }
 693
 694 int nfit_get_smbios_id(u32 device_handle, u16 *flags)
 695 {
 696         struct acpi_nfit_memory_map *memdev;
 697         struct acpi_nfit_desc *acpi_desc;
 698         struct nfit_mem *nfit_mem;
 699
 700         mutex_lock(&acpi_desc_lock);
 701         list_for_each_entry(acpi_desc, &acpi_descs, list) {
 702                 mutex_lock(&acpi_desc->init_mutex);
 703                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
 704                         memdev = __to_nfit_memdev(nfit_mem);
 705                         if (memdev->device_handle == device_handle) {
 706                                 mutex_unlock(&acpi_desc->init_mutex);
 707                                 mutex_unlock(&acpi_desc_lock);
 708                                 *flags = memdev->flags;
 709                                 return memdev->physical_id;
 710                         }
 711                 }
 712                 mutex_unlock(&acpi_desc->init_mutex);
 713         }
 714         mutex_unlock(&acpi_desc_lock);
 715
 716         return -ENODEV;
 717 }
 718 EXPORT_SYMBOL_GPL(nfit_get_smbios_id);
 719
 720 /*
 721  * An implementation may provide a truncated control region if no block windows
 722  * are defined.
 723  */
 724 static size_t sizeof_dcr(struct acpi_nfit_control_region *dcr)
 725 {
 726         if (dcr->header.length < offsetof(struct acpi_nfit_control_region,
 727                                 window_size))
 728                 return 0;
 729         if (dcr->windows)
 730                 return sizeof(*dcr);
 731         return offsetof(struct acpi_nfit_control_region, window_size);
 732 }
 733
 734 static bool add_dcr(struct acpi_nfit_desc *acpi_desc,
 735                 struct nfit_table_prev *prev,
 736                 struct acpi_nfit_control_region *dcr)
 737 {
 738         struct device *dev = acpi_desc->dev;
 739         struct nfit_dcr *nfit_dcr;
 740
 741         if (!sizeof_dcr(dcr))
 742                 return false;
 743
 744         list_for_each_entry(nfit_dcr, &prev->dcrs, list)
 745                 if (memcmp(nfit_dcr->dcr, dcr, sizeof_dcr(dcr)) == 0) {
 746                         list_move_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 747                         return true;
 748                 }
 749
 750         nfit_dcr = devm_kzalloc(dev, sizeof(*nfit_dcr) + sizeof(*dcr),
 751                         GFP_KERNEL);
 752         if (!nfit_dcr)
 753                 return false;
 754         INIT_LIST_HEAD(&nfit_dcr->list);
 755         memcpy(nfit_dcr->dcr, dcr, sizeof_dcr(dcr));
 756         list_add_tail(&nfit_dcr->list, &acpi_desc->dcrs);
 757         dev_dbg(dev, "%s: dcr index: %d windows: %d\n", __func__,
 758                         dcr->region_index, dcr->windows);
 759         return true;
 760 }
 761
 762 static bool add_bdw(struct acpi_nfit_desc *acpi_desc,
 763                 struct nfit_table_prev *prev,
 764                 struct acpi_nfit_data_region *bdw)
 765 {
 766         struct device *dev = acpi_desc->dev;
 767         struct nfit_bdw *nfit_bdw;
 768
 769         if (bdw->header.length != sizeof(*bdw))
 770                 return false;
 771         list_for_each_entry(nfit_bdw, &prev->bdws, list)
 772                 if (memcmp(nfit_bdw->bdw, bdw, sizeof(*bdw)) == 0) {
 773                         list_move_tail(&nfit_bdw->list, &acpi_desc->bdws);
 774                         return true;
 775                 }
 776
 777         nfit_bdw = devm_kzalloc(dev, sizeof(*nfit_bdw) + sizeof(*bdw),
 778                         GFP_KERNEL);
 779         if (!nfit_bdw)
 780                 return false;
 781         INIT_LIST_HEAD(&nfit_bdw->list);
 782         memcpy(nfit_bdw->bdw, bdw, sizeof(*bdw));
 783         list_add_tail(&nfit_bdw->list, &acpi_desc->bdws);
 784         dev_dbg(dev, "%s: bdw dcr: %d windows: %d\n", __func__,
 785                         bdw->region_index, bdw->windows);
 786         return true;
 787 }
 788
 789 static size_t sizeof_idt(struct acpi_nfit_interleave *idt)
 790 {
 791         if (idt->header.length < sizeof(*idt))
 792                 return 0;
 793         return sizeof(*idt) + sizeof(u32) * (idt->line_count - 1);
 794 }
 795
 796 static bool add_idt(struct acpi_nfit_desc *acpi_desc,
 797                 struct nfit_table_prev *prev,
 798                 struct acpi_nfit_interleave *idt)
 799 {
 800         struct device *dev = acpi_desc->dev;
 801         struct nfit_idt *nfit_idt;
 802
 803         if (!sizeof_idt(idt))
 804                 return false;
 805
 806         list_for_each_entry(nfit_idt, &prev->idts, list) {
 807                 if (sizeof_idt(nfit_idt->idt) != sizeof_idt(idt))
 808                         continue;
 809
 810                 if (memcmp(nfit_idt->idt, idt, sizeof_idt(idt)) == 0) {
 811                         list_move_tail(&nfit_idt->list, &acpi_desc->idts);
 812                         return true;
 813                 }
 814         }
 815
 816         nfit_idt = devm_kzalloc(dev, sizeof(*nfit_idt) + sizeof_idt(idt),
 817                         GFP_KERNEL);
 818         if (!nfit_idt)
 819                 return false;
 820         INIT_LIST_HEAD(&nfit_idt->list);
 821         memcpy(nfit_idt->idt, idt, sizeof_idt(idt));
 822         list_add_tail(&nfit_idt->list, &acpi_desc->idts);
 823         dev_dbg(dev, "%s: idt index: %d num_lines: %d\n", __func__,
 824                         idt->interleave_index, idt->line_count);
 825         return true;
 826 }
 827
 828 static size_t sizeof_flush(struct acpi_nfit_flush_address *flush)
 829 {
 830         if (flush->header.length < sizeof(*flush))
 831                 return 0;
 832         return sizeof(*flush) + sizeof(u64) * (flush->hint_count - 1);
 833 }
 834
 835 static bool add_flush(struct acpi_nfit_desc *acpi_desc,
 836                 struct nfit_table_prev *prev,
 837                 struct acpi_nfit_flush_address *flush)
 838 {
 839         struct device *dev = acpi_desc->dev;
 840         struct nfit_flush *nfit_flush;
 841
 842         if (!sizeof_flush(flush))
 843                 return false;
 844
 845         list_for_each_entry(nfit_flush, &prev->flushes, list) {
 846                 if (sizeof_flush(nfit_flush->flush) != sizeof_flush(flush))
 847                         continue;
 848
 849                 if (memcmp(nfit_flush->flush, flush,
 850                                         sizeof_flush(flush)) == 0) {
 851                         list_move_tail(&nfit_flush->list, &acpi_desc->flushes);
 852                         return true;
 853                 }
 854         }
 855
 856         nfit_flush = devm_kzalloc(dev, sizeof(*nfit_flush)
 857                         + sizeof_flush(flush), GFP_KERNEL);
 858         if (!nfit_flush)
 859                 return false;
 860         INIT_LIST_HEAD(&nfit_flush->list);
 861         memcpy(nfit_flush->flush, flush, sizeof_flush(flush));
 862         list_add_tail(&nfit_flush->list, &acpi_desc->flushes);
 863         dev_dbg(dev, "%s: nfit_flush handle: %d hint_count: %d\n", __func__,
 864                         flush->device_handle, flush->hint_count);
 865         return true;
 866 }
 867
 868 static bool add_platform_cap(struct acpi_nfit_desc *acpi_desc,
 869                 struct acpi_nfit_capabilities *pcap)
 870 {
 871         struct device *dev = acpi_desc->dev;
 872         u32 mask;
 873
 874         mask = (1 << (pcap->highest_capability + 1)) - 1;
 875         acpi_desc->platform_cap = pcap->capabilities & mask;
 876         dev_dbg(dev, "%s: cap: %#x\n", __func__, acpi_desc->platform_cap);
 877         return true;
 878 }
 879
 880 static void *add_table(struct acpi_nfit_desc *acpi_desc,
 881                 struct nfit_table_prev *prev, void *table, const void *end)
 882 {
 883         struct device *dev = acpi_desc->dev;
 884         struct acpi_nfit_header *hdr;
 885         void *err = ERR_PTR(-ENOMEM);
 886
 887         if (table >= end)
 888                 return NULL;
 889
 890         hdr = table;
 891         if (!hdr->length) {
 892                 dev_warn(dev, "found a zero length table '%d' parsing nfit\n",
 893                         hdr->type);
 894                 return NULL;
 895         }
 896
 897         switch (hdr->type) {
 898         case ACPI_NFIT_TYPE_SYSTEM_ADDRESS:
 899                 if (!add_spa(acpi_desc, prev, table))
 900                         return err;
 901                 break;
 902         case ACPI_NFIT_TYPE_MEMORY_MAP:
 903                 if (!add_memdev(acpi_desc, prev, table))
 904                         return err;
 905                 break;
 906         case ACPI_NFIT_TYPE_CONTROL_REGION:
 907                 if (!add_dcr(acpi_desc, prev, table))
 908                         return err;
 909                 break;
 910         case ACPI_NFIT_TYPE_DATA_REGION:
 911                 if (!add_bdw(acpi_desc, prev, table))
 912                         return err;
 913                 break;
 914         case ACPI_NFIT_TYPE_INTERLEAVE:
 915                 if (!add_idt(acpi_desc, prev, table))
 916                         return err;
 917                 break;
 918         case ACPI_NFIT_TYPE_FLUSH_ADDRESS:
 919                 if (!add_flush(acpi_desc, prev, table))
 920                         return err;
 921                 break;
 922         case ACPI_NFIT_TYPE_SMBIOS:
 923                 dev_dbg(dev, "%s: smbios\n", __func__);
 924                 break;
 925         case ACPI_NFIT_TYPE_CAPABILITIES:
 926                 if (!add_platform_cap(acpi_desc, table))
 927                         return err;
 928                 break;
 929         default:
 930                 dev_err(dev, "unknown table '%d' parsing nfit\n", hdr->type);
 931                 break;
 932         }
 933
 934         return table + hdr->length;
 935 }
 936
 937 static void nfit_mem_find_spa_bdw(struct acpi_nfit_desc *acpi_desc,
 938                 struct nfit_mem *nfit_mem)
 939 {
 940         u32 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
 941         u16 dcr = nfit_mem->dcr->region_index;
 942         struct nfit_spa *nfit_spa;
 943
 944         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
 945                 u16 range_index = nfit_spa->spa->range_index;
 946                 int type = nfit_spa_type(nfit_spa->spa);
 947                 struct nfit_memdev *nfit_memdev;
 948
 949                 if (type != NFIT_SPA_BDW)
 950                         continue;
 951
 952                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 953                         if (nfit_memdev->memdev->range_index != range_index)
 954                                 continue;
 955                         if (nfit_memdev->memdev->device_handle != device_handle)
 956                                 continue;
 957                         if (nfit_memdev->memdev->region_index != dcr)
 958                                 continue;
 959
 960                         nfit_mem->spa_bdw = nfit_spa->spa;
 961                         return;
 962                 }
 963         }
 964
 965         dev_dbg(acpi_desc->dev, "SPA-BDW not found for SPA-DCR %d\n",
 966                         nfit_mem->spa_dcr->range_index);
 967         nfit_mem->bdw = NULL;
 968 }
 969
 970 static void nfit_mem_init_bdw(struct acpi_nfit_desc *acpi_desc,
 971                 struct nfit_mem *nfit_mem, struct acpi_nfit_system_address *spa)
 972 {
 973         u16 dcr = __to_nfit_memdev(nfit_mem)->region_index;
 974         struct nfit_memdev *nfit_memdev;
 975         struct nfit_bdw *nfit_bdw;
 976         struct nfit_idt *nfit_idt;
 977         u16 idt_idx, range_index;
 978
 979         list_for_each_entry(nfit_bdw, &acpi_desc->bdws, list) {
 980                 if (nfit_bdw->bdw->region_index != dcr)
 981                         continue;
 982                 nfit_mem->bdw = nfit_bdw->bdw;
 983                 break;
 984         }
 985
 986         if (!nfit_mem->bdw)
 987                 return;
 988
 989         nfit_mem_find_spa_bdw(acpi_desc, nfit_mem);
 990
 991         if (!nfit_mem->spa_bdw)
 992                 return;
 993
 994         range_index = nfit_mem->spa_bdw->range_index;
 995         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
 996                 if (nfit_memdev->memdev->range_index != range_index ||
 997                                 nfit_memdev->memdev->region_index != dcr)
 998                         continue;
 999                 nfit_mem->memdev_bdw = nfit_memdev->memdev;
1000                 idt_idx = nfit_memdev->memdev->interleave_index;
1001                 list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1002                         if (nfit_idt->idt->interleave_index != idt_idx)
1003                                 continue;
1004                         nfit_mem->idt_bdw = nfit_idt->idt;
1005                         break;
1006                 }
1007                 break;
1008         }
1009 }
1010
1011 static int __nfit_mem_init(struct acpi_nfit_desc *acpi_desc,
1012                 struct acpi_nfit_system_address *spa)
1013 {
1014         struct nfit_mem *nfit_mem, *found;
1015         struct nfit_memdev *nfit_memdev;
1016         int type = spa ? nfit_spa_type(spa) : 0;
1017
1018         switch (type) {
1019         case NFIT_SPA_DCR:
1020         case NFIT_SPA_PM:
1021                 break;
1022         default:
1023                 if (spa)
1024                         return 0;
1025         }
1026
1027         /*
1028          * This loop runs in two modes, when a dimm is mapped the loop
1029          * adds memdev associations to an existing dimm, or creates a
1030          * dimm. In the unmapped dimm case this loop sweeps for memdev
1031          * instances with an invalid / zero range_index and adds those
1032          * dimms without spa associations.
1033          */
1034         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1035                 struct nfit_flush *nfit_flush;
1036                 struct nfit_dcr *nfit_dcr;
1037                 u32 device_handle;
1038                 u16 dcr;
1039
1040                 if (spa && nfit_memdev->memdev->range_index != spa->range_index)
1041                         continue;
1042                 if (!spa && nfit_memdev->memdev->range_index)
1043                         continue;
1044                 found = NULL;
1045                 dcr = nfit_memdev->memdev->region_index;
1046                 device_handle = nfit_memdev->memdev->device_handle;
1047                 list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1048                         if (__to_nfit_memdev(nfit_mem)->device_handle
1049                                         == device_handle) {
1050                                 found = nfit_mem;
1051                                 break;
1052                         }
1053
1054                 if (found)
1055                         nfit_mem = found;
1056                 else {
1057                         nfit_mem = devm_kzalloc(acpi_desc->dev,
1058                                         sizeof(*nfit_mem), GFP_KERNEL);
1059                         if (!nfit_mem)
1060                                 return -ENOMEM;
1061                         INIT_LIST_HEAD(&nfit_mem->list);
1062                         nfit_mem->acpi_desc = acpi_desc;
1063                         list_add(&nfit_mem->list, &acpi_desc->dimms);
1064                 }
1065
1066                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1067                         if (nfit_dcr->dcr->region_index != dcr)
1068                                 continue;
1069                         /*
1070                          * Record the control region for the dimm.  For
1071                          * the ACPI 6.1 case, where there are separate
1072                          * control regions for the pmem vs blk
1073                          * interfaces, be sure to record the extended
1074                          * blk details.
1075                          */
1076                         if (!nfit_mem->dcr)
1077                                 nfit_mem->dcr = nfit_dcr->dcr;
1078                         else if (nfit_mem->dcr->windows == 0
1079                                         && nfit_dcr->dcr->windows)
1080                                 nfit_mem->dcr = nfit_dcr->dcr;
1081                         break;
1082                 }
1083
1084                 list_for_each_entry(nfit_flush, &acpi_desc->flushes, list) {
1085                         struct acpi_nfit_flush_address *flush;
1086                         u16 i;
1087
1088                         if (nfit_flush->flush->device_handle != device_handle)
1089                                 continue;
1090                         nfit_mem->nfit_flush = nfit_flush;
1091                         flush = nfit_flush->flush;
1092                         nfit_mem->flush_wpq = devm_kzalloc(acpi_desc->dev,
1093                                         flush->hint_count
1094                                         * sizeof(struct resource), GFP_KERNEL);
1095                         if (!nfit_mem->flush_wpq)
1096                                 return -ENOMEM;
1097                         for (i = 0; i < flush->hint_count; i++) {
1098                                 struct resource *res = &nfit_mem->flush_wpq[i];
1099
1100                                 res->start = flush->hint_address[i];
1101                                 res->end = res->start + 8 - 1;
1102                         }
1103                         break;
1104                 }
1105
1106                 if (dcr && !nfit_mem->dcr) {
1107                         dev_err(acpi_desc->dev, "SPA %d missing DCR %d\n",
1108                                         spa->range_index, dcr);
1109                         return -ENODEV;
1110                 }
1111
1112                 if (type == NFIT_SPA_DCR) {
1113                         struct nfit_idt *nfit_idt;
1114                         u16 idt_idx;
1115
1116                         /* multiple dimms may share a SPA when interleaved */
1117                         nfit_mem->spa_dcr = spa;
1118                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1119                         idt_idx = nfit_memdev->memdev->interleave_index;
1120                         list_for_each_entry(nfit_idt, &acpi_desc->idts, list) {
1121                                 if (nfit_idt->idt->interleave_index != idt_idx)
1122                                         continue;
1123                                 nfit_mem->idt_dcr = nfit_idt->idt;
1124                                 break;
1125                         }
1126                         nfit_mem_init_bdw(acpi_desc, nfit_mem, spa);
1127                 } else if (type == NFIT_SPA_PM) {
1128                         /*
1129                          * A single dimm may belong to multiple SPA-PM
1130                          * ranges, record at least one in addition to
1131                          * any SPA-DCR range.
1132                          */
1133                         nfit_mem->memdev_pmem = nfit_memdev->memdev;
1134                 } else
1135                         nfit_mem->memdev_dcr = nfit_memdev->memdev;
1136         }
1137
1138         return 0;
1139 }
1140
1141 static int nfit_mem_cmp(void *priv, struct list_head *_a, struct list_head *_b)
1142 {
1143         struct nfit_mem *a = container_of(_a, typeof(*a), list);
1144         struct nfit_mem *b = container_of(_b, typeof(*b), list);
1145         u32 handleA, handleB;
1146
1147         handleA = __to_nfit_memdev(a)->device_handle;
1148         handleB = __to_nfit_memdev(b)->device_handle;
1149         if (handleA < handleB)
1150                 return -1;
1151         else if (handleA > handleB)
1152                 return 1;
1153         return 0;
1154 }
1155
1156 static int nfit_mem_init(struct acpi_nfit_desc *acpi_desc)
1157 {
1158         struct nfit_spa *nfit_spa;
1159         int rc;
1160
1161
1162         /*
1163          * For each SPA-DCR or SPA-PMEM address range find its
1164          * corresponding MEMDEV(s).  From each MEMDEV find the
1165          * corresponding DCR.  Then, if we're operating on a SPA-DCR,
1166          * try to find a SPA-BDW and a corresponding BDW that references
1167          * the DCR.  Throw it all into an nfit_mem object.  Note, that
1168          * BDWs are optional.
1169          */
1170         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
1171                 rc = __nfit_mem_init(acpi_desc, nfit_spa->spa);
1172                 if (rc)
1173                         return rc;
1174         }
1175
1176         /*
1177          * If a DIMM has failed to be mapped into SPA there will be no
1178          * SPA entries above. Find and register all the unmapped DIMMs
1179          * for reporting and recovery purposes.
1180          */
1181         rc = __nfit_mem_init(acpi_desc, NULL);
1182         if (rc)
1183                 return rc;
1184
1185         list_sort(NULL, &acpi_desc->dimms, nfit_mem_cmp);
1186
1187         return 0;
1188 }
1189
1190 static ssize_t bus_dsm_mask_show(struct device *dev,
1191                 struct device_attribute *attr, char *buf)
1192 {
1193         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1194         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1195
1196         return sprintf(buf, "%#lx\n", nd_desc->bus_dsm_mask);
1197 }
1198 static struct device_attribute dev_attr_bus_dsm_mask =
1199                 __ATTR(dsm_mask, 0444, bus_dsm_mask_show, NULL);
1200
1201 static ssize_t revision_show(struct device *dev,
1202                 struct device_attribute *attr, char *buf)
1203 {
1204         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1205         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1206         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1207
1208         return sprintf(buf, "%d\n", acpi_desc->acpi_header.revision);
1209 }
1210 static DEVICE_ATTR_RO(revision);
1211
1212 static ssize_t hw_error_scrub_show(struct device *dev,
1213                 struct device_attribute *attr, char *buf)
1214 {
1215         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1216         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1217         struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1218
1219         return sprintf(buf, "%d\n", acpi_desc->scrub_mode);
1220 }
1221
1222 /*
1223  * The 'hw_error_scrub' attribute can have the following values written to it:
1224  * '0': Switch to the default mode where an exception will only insert
1225  *      the address of the memory error into the poison and badblocks lists.
1226  * '1': Enable a full scrub to happen if an exception for a memory error is
1227  *      received.
1228  */
1229 static ssize_t hw_error_scrub_store(struct device *dev,
1230                 struct device_attribute *attr, const char *buf, size_t size)
1231 {
1232         struct nvdimm_bus_descriptor *nd_desc;
1233         ssize_t rc;
1234         long val;
1235
1236         rc = kstrtol(buf, 0, &val);
1237         if (rc)
1238                 return rc;
1239
1240         device_lock(dev);
1241         nd_desc = dev_get_drvdata(dev);
1242         if (nd_desc) {
1243                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1244
1245                 switch (val) {
1246                 case HW_ERROR_SCRUB_ON:
1247                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_ON;
1248                         break;
1249                 case HW_ERROR_SCRUB_OFF:
1250                         acpi_desc->scrub_mode = HW_ERROR_SCRUB_OFF;
1251                         break;
1252                 default:
1253                         rc = -EINVAL;
1254                         break;
1255                 }
1256         }
1257         device_unlock(dev);
1258         if (rc)
1259                 return rc;
1260         return size;
1261 }
1262 static DEVICE_ATTR_RW(hw_error_scrub);
1263
1264 /*
1265  * This shows the number of full Address Range Scrubs that have been
1266  * completed since driver load time. Userspace can wait on this using
1267  * select/poll etc. A '+' at the end indicates an ARS is in progress
1268  */
1269 static ssize_t scrub_show(struct device *dev,
1270                 struct device_attribute *attr, char *buf)
1271 {
1272         struct nvdimm_bus_descriptor *nd_desc;
1273         ssize_t rc = -ENXIO;
1274
1275         device_lock(dev);
1276         nd_desc = dev_get_drvdata(dev);
1277         if (nd_desc) {
1278                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1279
1280                 rc = sprintf(buf, "%d%s", acpi_desc->scrub_count,
1281                                 (work_busy(&acpi_desc->work)) ? "+\n" : "\n");
1282         }
1283         device_unlock(dev);
1284         return rc;
1285 }
1286
1287 static ssize_t scrub_store(struct device *dev,
1288                 struct device_attribute *attr, const char *buf, size_t size)
1289 {
1290         struct nvdimm_bus_descriptor *nd_desc;
1291         ssize_t rc;
1292         long val;
1293
1294         rc = kstrtol(buf, 0, &val);
1295         if (rc)
1296                 return rc;
1297         if (val != 1)
1298                 return -EINVAL;
1299
1300         device_lock(dev);
1301         nd_desc = dev_get_drvdata(dev);
1302         if (nd_desc) {
1303                 struct acpi_nfit_desc *acpi_desc = to_acpi_desc(nd_desc);
1304
1305                 rc = acpi_nfit_ars_rescan(acpi_desc, 0);
1306         }
1307         device_unlock(dev);
1308         if (rc)
1309                 return rc;
1310         return size;
1311 }
1312 static DEVICE_ATTR_RW(scrub);
1313
1314 static bool ars_supported(struct nvdimm_bus *nvdimm_bus)
1315 {
1316         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
1317         const unsigned long mask = 1 << ND_CMD_ARS_CAP | 1 << ND_CMD_ARS_START
1318                 | 1 << ND_CMD_ARS_STATUS;
1319
1320         return (nd_desc->cmd_mask & mask) == mask;
1321 }
1322
1323 static umode_t nfit_visible(struct kobject *kobj, struct attribute *a, int n)
1324 {
1325         struct device *dev = container_of(kobj, struct device, kobj);
1326         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
1327
1328         if (a == &dev_attr_scrub.attr && !ars_supported(nvdimm_bus))
1329                 return 0;
1330         return a->mode;
1331 }
1332
1333 static struct attribute *acpi_nfit_attributes[] = {
1334         &dev_attr_revision.attr,
1335         &dev_attr_scrub.attr,
1336         &dev_attr_hw_error_scrub.attr,
1337         &dev_attr_bus_dsm_mask.attr,
1338         NULL,
1339 };
1340
1341 static const struct attribute_group acpi_nfit_attribute_group = {
1342         .name = "nfit",
1343         .attrs = acpi_nfit_attributes,
1344         .is_visible = nfit_visible,
1345 };
1346
1347 static const struct attribute_group *acpi_nfit_attribute_groups[] = {
1348         &nvdimm_bus_attribute_group,
1349         &acpi_nfit_attribute_group,
1350         NULL,
1351 };
1352
1353 static struct acpi_nfit_memory_map *to_nfit_memdev(struct device *dev)
1354 {
1355         struct nvdimm *nvdimm = to_nvdimm(dev);
1356         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1357
1358         return __to_nfit_memdev(nfit_mem);
1359 }
1360
1361 static struct acpi_nfit_control_region *to_nfit_dcr(struct device *dev)
1362 {
1363         struct nvdimm *nvdimm = to_nvdimm(dev);
1364         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1365
1366         return nfit_mem->dcr;
1367 }
1368
1369 static ssize_t handle_show(struct device *dev,
1370                 struct device_attribute *attr, char *buf)
1371 {
1372         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1373
1374         return sprintf(buf, "%#x\n", memdev->device_handle);
1375 }
1376 static DEVICE_ATTR_RO(handle);
1377
1378 static ssize_t phys_id_show(struct device *dev,
1379                 struct device_attribute *attr, char *buf)
1380 {
1381         struct acpi_nfit_memory_map *memdev = to_nfit_memdev(dev);
1382
1383         return sprintf(buf, "%#x\n", memdev->physical_id);
1384 }
1385 static DEVICE_ATTR_RO(phys_id);
1386
1387 static ssize_t vendor_show(struct device *dev,
1388                 struct device_attribute *attr, char *buf)
1389 {
1390         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1391
1392         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->vendor_id));
1393 }
1394 static DEVICE_ATTR_RO(vendor);
1395
1396 static ssize_t rev_id_show(struct device *dev,
1397                 struct device_attribute *attr, char *buf)
1398 {
1399         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1400
1401         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->revision_id));
1402 }
1403 static DEVICE_ATTR_RO(rev_id);
1404
1405 static ssize_t device_show(struct device *dev,
1406                 struct device_attribute *attr, char *buf)
1407 {
1408         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1409
1410         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->device_id));
1411 }
1412 static DEVICE_ATTR_RO(device);
1413
1414 static ssize_t subsystem_vendor_show(struct device *dev,
1415                 struct device_attribute *attr, char *buf)
1416 {
1417         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1418
1419         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_vendor_id));
1420 }
1421 static DEVICE_ATTR_RO(subsystem_vendor);
1422
1423 static ssize_t subsystem_rev_id_show(struct device *dev,
1424                 struct device_attribute *attr, char *buf)
1425 {
1426         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1427
1428         return sprintf(buf, "0x%04x\n",
1429                         be16_to_cpu(dcr->subsystem_revision_id));
1430 }
1431 static DEVICE_ATTR_RO(subsystem_rev_id);
1432
1433 static ssize_t subsystem_device_show(struct device *dev,
1434                 struct device_attribute *attr, char *buf)
1435 {
1436         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1437
1438         return sprintf(buf, "0x%04x\n", be16_to_cpu(dcr->subsystem_device_id));
1439 }
1440 static DEVICE_ATTR_RO(subsystem_device);
1441
1442 static int num_nvdimm_formats(struct nvdimm *nvdimm)
1443 {
1444         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1445         int formats = 0;
1446
1447         if (nfit_mem->memdev_pmem)
1448                 formats++;
1449         if (nfit_mem->memdev_bdw)
1450                 formats++;
1451         return formats;
1452 }
1453
1454 static ssize_t format_show(struct device *dev,
1455                 struct device_attribute *attr, char *buf)
1456 {
1457         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1458
1459         return sprintf(buf, "0x%04x\n", le16_to_cpu(dcr->code));
1460 }
1461 static DEVICE_ATTR_RO(format);
1462
1463 static ssize_t format1_show(struct device *dev,
1464                 struct device_attribute *attr, char *buf)
1465 {
1466         u32 handle;
1467         ssize_t rc = -ENXIO;
1468         struct nfit_mem *nfit_mem;
1469         struct nfit_memdev *nfit_memdev;
1470         struct acpi_nfit_desc *acpi_desc;
1471         struct nvdimm *nvdimm = to_nvdimm(dev);
1472         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1473
1474         nfit_mem = nvdimm_provider_data(nvdimm);
1475         acpi_desc = nfit_mem->acpi_desc;
1476         handle = to_nfit_memdev(dev)->device_handle;
1477
1478         /* assumes DIMMs have at most 2 published interface codes */
1479         mutex_lock(&acpi_desc->init_mutex);
1480         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1481                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
1482                 struct nfit_dcr *nfit_dcr;
1483
1484                 if (memdev->device_handle != handle)
1485                         continue;
1486
1487                 list_for_each_entry(nfit_dcr, &acpi_desc->dcrs, list) {
1488                         if (nfit_dcr->dcr->region_index != memdev->region_index)
1489                                 continue;
1490                         if (nfit_dcr->dcr->code == dcr->code)
1491                                 continue;
1492                         rc = sprintf(buf, "0x%04x\n",
1493                                         le16_to_cpu(nfit_dcr->dcr->code));
1494                         break;
1495                 }
1496                 if (rc != ENXIO)
1497                         break;
1498         }
1499         mutex_unlock(&acpi_desc->init_mutex);
1500         return rc;
1501 }
1502 static DEVICE_ATTR_RO(format1);
1503
1504 static ssize_t formats_show(struct device *dev,
1505                 struct device_attribute *attr, char *buf)
1506 {
1507         struct nvdimm *nvdimm = to_nvdimm(dev);
1508
1509         return sprintf(buf, "%d\n", num_nvdimm_formats(nvdimm));
1510 }
1511 static DEVICE_ATTR_RO(formats);
1512
1513 static ssize_t serial_show(struct device *dev,
1514                 struct device_attribute *attr, char *buf)
1515 {
1516         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1517
1518         return sprintf(buf, "0x%08x\n", be32_to_cpu(dcr->serial_number));
1519 }
1520 static DEVICE_ATTR_RO(serial);
1521
1522 static ssize_t family_show(struct device *dev,
1523                 struct device_attribute *attr, char *buf)
1524 {
1525         struct nvdimm *nvdimm = to_nvdimm(dev);
1526         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1527
1528         if (nfit_mem->family < 0)
1529                 return -ENXIO;
1530         return sprintf(buf, "%d\n", nfit_mem->family);
1531 }
1532 static DEVICE_ATTR_RO(family);
1533
1534 static ssize_t dsm_mask_show(struct device *dev,
1535                 struct device_attribute *attr, char *buf)
1536 {
1537         struct nvdimm *nvdimm = to_nvdimm(dev);
1538         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
1539
1540         if (nfit_mem->family < 0)
1541                 return -ENXIO;
1542         return sprintf(buf, "%#lx\n", nfit_mem->dsm_mask);
1543 }
1544 static DEVICE_ATTR_RO(dsm_mask);
1545
1546 static ssize_t flags_show(struct device *dev,
1547                 struct device_attribute *attr, char *buf)
1548 {
1549         u16 flags = to_nfit_memdev(dev)->flags;
1550
1551         return sprintf(buf, "%s%s%s%s%s%s%s\n",
1552                 flags & ACPI_NFIT_MEM_SAVE_FAILED ? "save_fail " : "",
1553                 flags & ACPI_NFIT_MEM_RESTORE_FAILED ? "restore_fail " : "",
1554                 flags & ACPI_NFIT_MEM_FLUSH_FAILED ? "flush_fail " : "",
1555                 flags & ACPI_NFIT_MEM_NOT_ARMED ? "not_armed " : "",
1556                 flags & ACPI_NFIT_MEM_HEALTH_OBSERVED ? "smart_event " : "",
1557                 flags & ACPI_NFIT_MEM_MAP_FAILED ? "map_fail " : "",
1558                 flags & ACPI_NFIT_MEM_HEALTH_ENABLED ? "smart_notify " : "");
1559 }
1560 static DEVICE_ATTR_RO(flags);
1561
1562 static ssize_t id_show(struct device *dev,
1563                 struct device_attribute *attr, char *buf)
1564 {
1565         struct acpi_nfit_control_region *dcr = to_nfit_dcr(dev);
1566
1567         if (dcr->valid_fields & ACPI_NFIT_CONTROL_MFG_INFO_VALID)
1568                 return sprintf(buf, "%04x-%02x-%04x-%08x\n",
1569                                 be16_to_cpu(dcr->vendor_id),
1570                                 dcr->manufacturing_location,
1571                                 be16_to_cpu(dcr->manufacturing_date),
1572                                 be32_to_cpu(dcr->serial_number));
1573         else
1574                 return sprintf(buf, "%04x-%08x\n",
1575                                 be16_to_cpu(dcr->vendor_id),
1576                                 be32_to_cpu(dcr->serial_number));
1577 }
1578 static DEVICE_ATTR_RO(id);
1579
1580 static struct attribute *acpi_nfit_dimm_attributes[] = {
1581         &dev_attr_handle.attr,
1582         &dev_attr_phys_id.attr,
1583         &dev_attr_vendor.attr,
1584         &dev_attr_device.attr,
1585         &dev_attr_rev_id.attr,
1586         &dev_attr_subsystem_vendor.attr,
1587         &dev_attr_subsystem_device.attr,
1588         &dev_attr_subsystem_rev_id.attr,
1589         &dev_attr_format.attr,
1590         &dev_attr_formats.attr,
1591         &dev_attr_format1.attr,
1592         &dev_attr_serial.attr,
1593         &dev_attr_flags.attr,
1594         &dev_attr_id.attr,
1595         &dev_attr_family.attr,
1596         &dev_attr_dsm_mask.attr,
1597         NULL,
1598 };
1599
1600 static umode_t acpi_nfit_dimm_attr_visible(struct kobject *kobj,
1601                 struct attribute *a, int n)
1602 {
1603         struct device *dev = container_of(kobj, struct device, kobj);
1604         struct nvdimm *nvdimm = to_nvdimm(dev);
1605
1606         if (!to_nfit_dcr(dev)) {
1607                 /* Without a dcr only the memdev attributes can be surfaced */
1608                 if (a == &dev_attr_handle.attr || a == &dev_attr_phys_id.attr
1609                                 || a == &dev_attr_flags.attr
1610                                 || a == &dev_attr_family.attr
1611                                 || a == &dev_attr_dsm_mask.attr)
1612                         return a->mode;
1613                 return 0;
1614         }
1615
1616         if (a == &dev_attr_format1.attr && num_nvdimm_formats(nvdimm) <= 1)
1617                 return 0;
1618         return a->mode;
1619 }
1620
1621 static const struct attribute_group acpi_nfit_dimm_attribute_group = {
1622         .name = "nfit",
1623         .attrs = acpi_nfit_dimm_attributes,
1624         .is_visible = acpi_nfit_dimm_attr_visible,
1625 };
1626
1627 static const struct attribute_group *acpi_nfit_dimm_attribute_groups[] = {
1628         &nvdimm_attribute_group,
1629         &nd_device_attribute_group,
1630         &acpi_nfit_dimm_attribute_group,
1631         NULL,
1632 };
1633
1634 static struct nvdimm *acpi_nfit_dimm_by_handle(struct acpi_nfit_desc *acpi_desc,
1635                 u32 device_handle)
1636 {
1637         struct nfit_mem *nfit_mem;
1638
1639         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list)
1640                 if (__to_nfit_memdev(nfit_mem)->device_handle == device_handle)
1641                         return nfit_mem->nvdimm;
1642
1643         return NULL;
1644 }
1645
1646 void __acpi_nvdimm_notify(struct device *dev, u32 event)
1647 {
1648         struct nfit_mem *nfit_mem;
1649         struct acpi_nfit_desc *acpi_desc;
1650
1651         dev_dbg(dev->parent, "%s: %s: event: %d\n", dev_name(dev), __func__,
1652                         event);
1653
1654         if (event != NFIT_NOTIFY_DIMM_HEALTH) {
1655                 dev_dbg(dev->parent, "%s: unknown event: %d\n", dev_name(dev),
1656                                 event);
1657                 return;
1658         }
1659
1660         acpi_desc = dev_get_drvdata(dev->parent);
1661         if (!acpi_desc)
1662                 return;
1663
1664         /*
1665          * If we successfully retrieved acpi_desc, then we know nfit_mem data
1666          * is still valid.
1667          */
1668         nfit_mem = dev_get_drvdata(dev);
1669         if (nfit_mem && nfit_mem->flags_attr)
1670                 sysfs_notify_dirent(nfit_mem->flags_attr);
1671 }
1672 EXPORT_SYMBOL_GPL(__acpi_nvdimm_notify);
1673
1674 static void acpi_nvdimm_notify(acpi_handle handle, u32 event, void *data)
1675 {
1676         struct acpi_device *adev = data;
1677         struct device *dev = &adev->dev;
1678
1679         device_lock(dev->parent);
1680         __acpi_nvdimm_notify(dev, event);
1681         device_unlock(dev->parent);
1682 }
1683
1684 static int acpi_nfit_add_dimm(struct acpi_nfit_desc *acpi_desc,
1685                 struct nfit_mem *nfit_mem, u32 device_handle)
1686 {
1687         struct acpi_device *adev, *adev_dimm;
1688         struct device *dev = acpi_desc->dev;
1689         union acpi_object *obj;
1690         unsigned long dsm_mask;
1691         const guid_t *guid;
1692         int i;
1693         int family = -1;
1694
1695         /* nfit test assumes 1:1 relationship between commands and dsms */
1696         nfit_mem->dsm_mask = acpi_desc->dimm_cmd_force_en;
1697         nfit_mem->family = NVDIMM_FAMILY_INTEL;
1698         adev = to_acpi_dev(acpi_desc);
1699         if (!adev)
1700                 return 0;
1701
1702         adev_dimm = acpi_find_child_device(adev, device_handle, false);
1703         nfit_mem->adev = adev_dimm;
1704         if (!adev_dimm) {
1705                 dev_err(dev, "no ACPI.NFIT device with _ADR %#x, disabling...\n",
1706                                 device_handle);
1707                 return force_enable_dimms ? 0 : -ENODEV;
1708         }
1709
1710         if (ACPI_FAILURE(acpi_install_notify_handler(adev_dimm->handle,
1711                 ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify, adev_dimm))) {
1712                 dev_err(dev, "%s: notification registration failed\n",
1713                                 dev_name(&adev_dimm->dev));
1714                 return -ENXIO;
1715         }
1716         /*
1717          * Record nfit_mem for the notification path to track back to
1718          * the nfit sysfs attributes for this dimm device object.
1719          */
1720         dev_set_drvdata(&adev_dimm->dev, nfit_mem);
1721
1722         /*
1723          * Until standardization materializes we need to consider 4
1724          * different command sets.  Note, that checking for function0 (bit0)
1725          * tells us if any commands are reachable through this GUID.
1726          */
1727         for (i = 0; i <= NVDIMM_FAMILY_MAX; i++)
1728                 if (acpi_check_dsm(adev_dimm->handle, to_nfit_uuid(i), 1, 1))
1729                         if (family < 0 || i == default_dsm_family)
1730                                 family = i;
1731
1732         /* limit the supported commands to those that are publicly documented */
1733         nfit_mem->family = family;
1734         if (override_dsm_mask && !disable_vendor_specific)
1735                 dsm_mask = override_dsm_mask;
1736         else if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1737                 dsm_mask = NVDIMM_INTEL_CMDMASK;
1738                 if (disable_vendor_specific)
1739                         dsm_mask &= ~(1 << ND_CMD_VENDOR);
1740         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE1) {
1741                 dsm_mask = 0x1c3c76;
1742         } else if (nfit_mem->family == NVDIMM_FAMILY_HPE2) {
1743                 dsm_mask = 0x1fe;
1744                 if (disable_vendor_specific)
1745                         dsm_mask &= ~(1 << 8);
1746         } else if (nfit_mem->family == NVDIMM_FAMILY_MSFT) {
1747                 dsm_mask = 0xffffffff;
1748         } else {
1749                 dev_dbg(dev, "unknown dimm command family\n");
1750                 nfit_mem->family = -1;
1751                 /* DSMs are optional, continue loading the driver... */
1752                 return 0;
1753         }
1754
1755         guid = to_nfit_uuid(nfit_mem->family);
1756         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1757                 if (acpi_check_dsm(adev_dimm->handle, guid,
1758                                         nfit_dsm_revid(nfit_mem->family, i),
1759                                         1ULL << i))
1760                         set_bit(i, &nfit_mem->dsm_mask);
1761
1762         obj = acpi_label_info(adev_dimm->handle);
1763         if (obj) {
1764                 ACPI_FREE(obj);
1765                 nfit_mem->has_lsi = 1;
1766                 dev_dbg(dev, "%s: has _LSI\n", dev_name(&adev_dimm->dev));
1767         }
1768
1769         obj = acpi_label_read(adev_dimm->handle, 0, 0);
1770         if (obj) {
1771                 ACPI_FREE(obj);
1772                 nfit_mem->has_lsr = 1;
1773                 dev_dbg(dev, "%s: has _LSR\n", dev_name(&adev_dimm->dev));
1774         }
1775
1776         obj = acpi_label_write(adev_dimm->handle, 0, 0, NULL);
1777         if (obj) {
1778                 ACPI_FREE(obj);
1779                 nfit_mem->has_lsw = 1;
1780                 dev_dbg(dev, "%s: has _LSW\n", dev_name(&adev_dimm->dev));
1781         }
1782
1783         return 0;
1784 }
1785
1786 static void shutdown_dimm_notify(void *data)
1787 {
1788         struct acpi_nfit_desc *acpi_desc = data;
1789         struct nfit_mem *nfit_mem;
1790
1791         mutex_lock(&acpi_desc->init_mutex);
1792         /*
1793          * Clear out the nfit_mem->flags_attr and shut down dimm event
1794          * notifications.
1795          */
1796         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1797                 struct acpi_device *adev_dimm = nfit_mem->adev;
1798
1799                 if (nfit_mem->flags_attr) {
1800                         sysfs_put(nfit_mem->flags_attr);
1801                         nfit_mem->flags_attr = NULL;
1802                 }
1803                 if (adev_dimm) {
1804                         acpi_remove_notify_handler(adev_dimm->handle,
1805                                         ACPI_DEVICE_NOTIFY, acpi_nvdimm_notify);
1806                         dev_set_drvdata(&adev_dimm->dev, NULL);
1807                 }
1808         }
1809         mutex_unlock(&acpi_desc->init_mutex);
1810 }
1811
1812 static int acpi_nfit_register_dimms(struct acpi_nfit_desc *acpi_desc)
1813 {
1814         struct nfit_mem *nfit_mem;
1815         int dimm_count = 0, rc;
1816         struct nvdimm *nvdimm;
1817
1818         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1819                 struct acpi_nfit_flush_address *flush;
1820                 unsigned long flags = 0, cmd_mask;
1821                 struct nfit_memdev *nfit_memdev;
1822                 u32 device_handle;
1823                 u16 mem_flags;
1824
1825                 device_handle = __to_nfit_memdev(nfit_mem)->device_handle;
1826                 nvdimm = acpi_nfit_dimm_by_handle(acpi_desc, device_handle);
1827                 if (nvdimm) {
1828                         dimm_count++;
1829                         continue;
1830                 }
1831
1832                 if (nfit_mem->bdw && nfit_mem->memdev_pmem)
1833                         set_bit(NDD_ALIASING, &flags);
1834
1835                 /* collate flags across all memdevs for this dimm */
1836                 list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
1837                         struct acpi_nfit_memory_map *dimm_memdev;
1838
1839                         dimm_memdev = __to_nfit_memdev(nfit_mem);
1840                         if (dimm_memdev->device_handle
1841                                         != nfit_memdev->memdev->device_handle)
1842                                 continue;
1843                         dimm_memdev->flags |= nfit_memdev->memdev->flags;
1844                 }
1845
1846                 mem_flags = __to_nfit_memdev(nfit_mem)->flags;
1847                 if (mem_flags & ACPI_NFIT_MEM_NOT_ARMED)
1848                         set_bit(NDD_UNARMED, &flags);
1849
1850                 rc = acpi_nfit_add_dimm(acpi_desc, nfit_mem, device_handle);
1851                 if (rc)
1852                         continue;
1853
1854                 /*
1855                  * TODO: provide translation for non-NVDIMM_FAMILY_INTEL
1856                  * devices (i.e. from nd_cmd to acpi_dsm) to standardize the
1857                  * userspace interface.
1858                  */
1859                 cmd_mask = 1UL << ND_CMD_CALL;
1860                 if (nfit_mem->family == NVDIMM_FAMILY_INTEL) {
1861                         /*
1862                          * These commands have a 1:1 correspondence
1863                          * between DSM payload and libnvdimm ioctl
1864                          * payload format.
1865                          */
1866                         cmd_mask |= nfit_mem->dsm_mask & NVDIMM_STANDARD_CMDMASK;
1867                 }
1868
1869                 if (nfit_mem->has_lsi)
1870                         set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
1871                 if (nfit_mem->has_lsr)
1872                         set_bit(ND_CMD_GET_CONFIG_DATA, &cmd_mask);
1873                 if (nfit_mem->has_lsw)
1874                         set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
1875
1876                 flush = nfit_mem->nfit_flush ? nfit_mem->nfit_flush->flush
1877                         : NULL;
1878                 nvdimm = nvdimm_create(acpi_desc->nvdimm_bus, nfit_mem,
1879                                 acpi_nfit_dimm_attribute_groups,
1880                                 flags, cmd_mask, flush ? flush->hint_count : 0,
1881                                 nfit_mem->flush_wpq);
1882                 if (!nvdimm)
1883                         return -ENOMEM;
1884
1885                 nfit_mem->nvdimm = nvdimm;
1886                 dimm_count++;
1887
1888                 if ((mem_flags & ACPI_NFIT_MEM_FAILED_MASK) == 0)
1889                         continue;
1890
1891                 dev_info(acpi_desc->dev, "%s flags:%s%s%s%s%s\n",
1892                                 nvdimm_name(nvdimm),
1893                   mem_flags & ACPI_NFIT_MEM_SAVE_FAILED ? " save_fail" : "",
1894                   mem_flags & ACPI_NFIT_MEM_RESTORE_FAILED ? " restore_fail":"",
1895                   mem_flags & ACPI_NFIT_MEM_FLUSH_FAILED ? " flush_fail" : "",
1896                   mem_flags & ACPI_NFIT_MEM_NOT_ARMED ? " not_armed" : "",
1897                   mem_flags & ACPI_NFIT_MEM_MAP_FAILED ? " map_fail" : "");
1898
1899         }
1900
1901         rc = nvdimm_bus_check_dimm_count(acpi_desc->nvdimm_bus, dimm_count);
1902         if (rc)
1903                 return rc;
1904
1905         /*
1906          * Now that dimms are successfully registered, and async registration
1907          * is flushed, attempt to enable event notification.
1908          */
1909         list_for_each_entry(nfit_mem, &acpi_desc->dimms, list) {
1910                 struct kernfs_node *nfit_kernfs;
1911
1912                 nvdimm = nfit_mem->nvdimm;
1913                 if (!nvdimm)
1914                         continue;
1915
1916                 nfit_kernfs = sysfs_get_dirent(nvdimm_kobj(nvdimm)->sd, "nfit");
1917                 if (nfit_kernfs)
1918                         nfit_mem->flags_attr = sysfs_get_dirent(nfit_kernfs,
1919                                         "flags");
1920                 sysfs_put(nfit_kernfs);
1921                 if (!nfit_mem->flags_attr)
1922                         dev_warn(acpi_desc->dev, "%s: notifications disabled\n",
1923                                         nvdimm_name(nvdimm));
1924         }
1925
1926         return devm_add_action_or_reset(acpi_desc->dev, shutdown_dimm_notify,
1927                         acpi_desc);
1928 }
1929
1930 /*
1931  * These constants are private because there are no kernel consumers of
1932  * these commands.
1933  */
1934 enum nfit_aux_cmds {
1935         NFIT_CMD_TRANSLATE_SPA = 5,
1936         NFIT_CMD_ARS_INJECT_SET = 7,
1937         NFIT_CMD_ARS_INJECT_CLEAR = 8,
1938         NFIT_CMD_ARS_INJECT_GET = 9,
1939 };
1940
1941 static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
1942 {
1943         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
1944         const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);
1945         struct acpi_device *adev;
1946         unsigned long dsm_mask;
1947         int i;
1948
1949         nd_desc->cmd_mask = acpi_desc->bus_cmd_force_en;
1950         nd_desc->bus_dsm_mask = acpi_desc->bus_nfit_cmd_force_en;
1951         adev = to_acpi_dev(acpi_desc);
1952         if (!adev)
1953                 return;
1954
1955         for (i = ND_CMD_ARS_CAP; i <= ND_CMD_CLEAR_ERROR; i++)
1956                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1957                         set_bit(i, &nd_desc->cmd_mask);
1958         set_bit(ND_CMD_CALL, &nd_desc->cmd_mask);
1959
1960         dsm_mask =
1961                 (1 << ND_CMD_ARS_CAP) |
1962                 (1 << ND_CMD_ARS_START) |
1963                 (1 << ND_CMD_ARS_STATUS) |
1964                 (1 << ND_CMD_CLEAR_ERROR) |
1965                 (1 << NFIT_CMD_TRANSLATE_SPA) |
1966                 (1 << NFIT_CMD_ARS_INJECT_SET) |
1967                 (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
1968                 (1 << NFIT_CMD_ARS_INJECT_GET);
1969         for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
1970                 if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
1971                         set_bit(i, &nd_desc->bus_dsm_mask);
1972 }
1973
1974 static ssize_t range_index_show(struct device *dev,
1975                 struct device_attribute *attr, char *buf)
1976 {
1977         struct nd_region *nd_region = to_nd_region(dev);
1978         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1979
1980         return sprintf(buf, "%d\n", nfit_spa->spa->range_index);
1981 }
1982 static DEVICE_ATTR_RO(range_index);
1983
1984 static ssize_t ecc_unit_size_show(struct device *dev,
1985                 struct device_attribute *attr, char *buf)
1986 {
1987         struct nd_region *nd_region = to_nd_region(dev);
1988         struct nfit_spa *nfit_spa = nd_region_provider_data(nd_region);
1989
1990         return sprintf(buf, "%d\n", nfit_spa->clear_err_unit);
1991 }
1992 static DEVICE_ATTR_RO(ecc_unit_size);
1993
1994 static struct attribute *acpi_nfit_region_attributes[] = {
1995         &dev_attr_range_index.attr,
1996         &dev_attr_ecc_unit_size.attr,
1997         NULL,
1998 };
1999
2000 static const struct attribute_group acpi_nfit_region_attribute_group = {
2001         .name = "nfit",
2002         .attrs = acpi_nfit_region_attributes,
2003 };
2004
2005 static const struct attribute_group *acpi_nfit_region_attribute_groups[] = {
2006         &nd_region_attribute_group,
2007         &nd_mapping_attribute_group,
2008         &nd_device_attribute_group,
2009         &nd_numa_attribute_group,
2010         &acpi_nfit_region_attribute_group,
2011         NULL,
2012 };
2013
2014 /* enough info to uniquely specify an interleave set */
2015 struct nfit_set_info {
2016         struct nfit_set_info_map {
2017                 u64 region_offset;
2018                 u32 serial_number;
2019                 u32 pad;
2020         } mapping[0];
2021 };
2022
2023 struct nfit_set_info2 {
2024         struct nfit_set_info_map2 {
2025                 u64 region_offset;
2026                 u32 serial_number;
2027                 u16 vendor_id;
2028                 u16 manufacturing_date;
2029                 u8  manufacturing_location;
2030                 u8  reserved[31];
2031         } mapping[0];
2032 };
2033
2034 static size_t sizeof_nfit_set_info(int num_mappings)
2035 {
2036         return sizeof(struct nfit_set_info)
2037                 + num_mappings * sizeof(struct nfit_set_info_map);
2038 }
2039
2040 static size_t sizeof_nfit_set_info2(int num_mappings)
2041 {
2042         return sizeof(struct nfit_set_info2)
2043                 + num_mappings * sizeof(struct nfit_set_info_map2);
2044 }
2045
2046 static int cmp_map_compat(const void *m0, const void *m1)
2047 {
2048         const struct nfit_set_info_map *map0 = m0;
2049         const struct nfit_set_info_map *map1 = m1;
2050
2051         return memcmp(&map0->region_offset, &map1->region_offset,
2052                         sizeof(u64));
2053 }
2054
2055 static int cmp_map(const void *m0, const void *m1)
2056 {
2057         const struct nfit_set_info_map *map0 = m0;
2058         const struct nfit_set_info_map *map1 = m1;
2059
2060         if (map0->region_offset < map1->region_offset)
2061                 return -1;
2062         else if (map0->region_offset > map1->region_offset)
2063                 return 1;
2064         return 0;
2065 }
2066
2067 static int cmp_map2(const void *m0, const void *m1)
2068 {
2069         const struct nfit_set_info_map2 *map0 = m0;
2070         const struct nfit_set_info_map2 *map1 = m1;
2071
2072         if (map0->region_offset < map1->region_offset)
2073                 return -1;
2074         else if (map0->region_offset > map1->region_offset)
2075                 return 1;
2076         return 0;
2077 }
2078
2079 /* Retrieve the nth entry referencing this spa */
2080 static struct acpi_nfit_memory_map *memdev_from_spa(
2081                 struct acpi_nfit_desc *acpi_desc, u16 range_index, int n)
2082 {
2083         struct nfit_memdev *nfit_memdev;
2084
2085         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list)
2086                 if (nfit_memdev->memdev->range_index == range_index)
2087                         if (n-- == 0)
2088                                 return nfit_memdev->memdev;
2089         return NULL;
2090 }
2091
2092 static int acpi_nfit_init_interleave_set(struct acpi_nfit_desc *acpi_desc,
2093                 struct nd_region_desc *ndr_desc,
2094                 struct acpi_nfit_system_address *spa)
2095 {
2096         struct device *dev = acpi_desc->dev;
2097         struct nd_interleave_set *nd_set;
2098         u16 nr = ndr_desc->num_mappings;
2099         struct nfit_set_info2 *info2;
2100         struct nfit_set_info *info;
2101         int i;
2102
2103         nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
2104         if (!nd_set)
2105                 return -ENOMEM;
2106         ndr_desc->nd_set = nd_set;
2107         guid_copy(&nd_set->type_guid, (guid_t *) spa->range_guid);
2108
2109         info = devm_kzalloc(dev, sizeof_nfit_set_info(nr), GFP_KERNEL);
2110         if (!info)
2111                 return -ENOMEM;
2112
2113         info2 = devm_kzalloc(dev, sizeof_nfit_set_info2(nr), GFP_KERNEL);
2114         if (!info2)
2115                 return -ENOMEM;
2116
2117         for (i = 0; i < nr; i++) {
2118                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
2119                 struct nfit_set_info_map *map = &info->mapping[i];
2120                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2121                 struct nvdimm *nvdimm = mapping->nvdimm;
2122                 struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2123                 struct acpi_nfit_memory_map *memdev = memdev_from_spa(acpi_desc,
2124                                 spa->range_index, i);
2125                 struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2126
2127                 if (!memdev || !nfit_mem->dcr) {
2128                         dev_err(dev, "%s: failed to find DCR\n", __func__);
2129                         return -ENODEV;
2130                 }
2131
2132                 map->region_offset = memdev->region_offset;
2133                 map->serial_number = dcr->serial_number;
2134
2135                 map2->region_offset = memdev->region_offset;
2136                 map2->serial_number = dcr->serial_number;
2137                 map2->vendor_id = dcr->vendor_id;
2138                 map2->manufacturing_date = dcr->manufacturing_date;
2139                 map2->manufacturing_location = dcr->manufacturing_location;
2140         }
2141
2142         /* v1.1 namespaces */
2143         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2144                         cmp_map, NULL);
2145         nd_set->cookie1 = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2146
2147         /* v1.2 namespaces */
2148         sort(&info2->mapping[0], nr, sizeof(struct nfit_set_info_map2),
2149                         cmp_map2, NULL);
2150         nd_set->cookie2 = nd_fletcher64(info2, sizeof_nfit_set_info2(nr), 0);
2151
2152         /* support v1.1 namespaces created with the wrong sort order */
2153         sort(&info->mapping[0], nr, sizeof(struct nfit_set_info_map),
2154                         cmp_map_compat, NULL);
2155         nd_set->altcookie = nd_fletcher64(info, sizeof_nfit_set_info(nr), 0);
2156
2157         /* record the result of the sort for the mapping position */
2158         for (i = 0; i < nr; i++) {
2159                 struct nfit_set_info_map2 *map2 = &info2->mapping[i];
2160                 int j;
2161
2162                 for (j = 0; j < nr; j++) {
2163                         struct nd_mapping_desc *mapping = &ndr_desc->mapping[j];
2164                         struct nvdimm *nvdimm = mapping->nvdimm;
2165                         struct nfit_mem *nfit_mem = nvdimm_provider_data(nvdimm);
2166                         struct acpi_nfit_control_region *dcr = nfit_mem->dcr;
2167
2168                         if (map2->serial_number == dcr->serial_number &&
2169                             map2->vendor_id == dcr->vendor_id &&
2170                             map2->manufacturing_date == dcr->manufacturing_date &&
2171                             map2->manufacturing_location
2172                                     == dcr->manufacturing_location) {
2173                                 mapping->position = i;
2174                                 break;
2175                         }
2176                 }
2177         }
2178
2179         ndr_desc->nd_set = nd_set;
2180         devm_kfree(dev, info);
2181         devm_kfree(dev, info2);
2182
2183         return 0;
2184 }
2185
2186 static u64 to_interleave_offset(u64 offset, struct nfit_blk_mmio *mmio)
2187 {
2188         struct acpi_nfit_interleave *idt = mmio->idt;
2189         u32 sub_line_offset, line_index, line_offset;
2190         u64 line_no, table_skip_count, table_offset;
2191
2192         line_no = div_u64_rem(offset, mmio->line_size, &sub_line_offset);
2193         table_skip_count = div_u64_rem(line_no, mmio->num_lines, &line_index);
2194         line_offset = idt->line_offset[line_index]
2195                 * mmio->line_size;
2196         table_offset = table_skip_count * mmio->table_size;
2197
2198         return mmio->base_offset + line_offset + table_offset + sub_line_offset;
2199 }
2200
2201 static u32 read_blk_stat(struct nfit_blk *nfit_blk, unsigned int bw)
2202 {
2203         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2204         u64 offset = nfit_blk->stat_offset + mmio->size * bw;
2205         const u32 STATUS_MASK = 0x80000037;
2206
2207         if (mmio->num_lines)
2208                 offset = to_interleave_offset(offset, mmio);
2209
2210         return readl(mmio->addr.base + offset) & STATUS_MASK;
2211 }
2212
2213 static void write_blk_ctl(struct nfit_blk *nfit_blk, unsigned int bw,
2214                 resource_size_t dpa, unsigned int len, unsigned int write)
2215 {
2216         u64 cmd, offset;
2217         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[DCR];
2218
2219         enum {
2220                 BCW_OFFSET_MASK = (1ULL << 48)-1,
2221                 BCW_LEN_SHIFT = 48,
2222                 BCW_LEN_MASK = (1ULL << 8) - 1,
2223                 BCW_CMD_SHIFT = 56,
2224         };
2225
2226         cmd = (dpa >> L1_CACHE_SHIFT) & BCW_OFFSET_MASK;
2227         len = len >> L1_CACHE_SHIFT;
2228         cmd |= ((u64) len & BCW_LEN_MASK) << BCW_LEN_SHIFT;
2229         cmd |= ((u64) write) << BCW_CMD_SHIFT;
2230
2231         offset = nfit_blk->cmd_offset + mmio->size * bw;
2232         if (mmio->num_lines)
2233                 offset = to_interleave_offset(offset, mmio);
2234
2235         writeq(cmd, mmio->addr.base + offset);
2236         nvdimm_flush(nfit_blk->nd_region);
2237
2238         if (nfit_blk->dimm_flags & NFIT_BLK_DCR_LATCH)
2239                 readq(mmio->addr.base + offset);
2240 }
2241
2242 static int acpi_nfit_blk_single_io(struct nfit_blk *nfit_blk,
2243                 resource_size_t dpa, void *iobuf, size_t len, int rw,
2244                 unsigned int lane)
2245 {
2246         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2247         unsigned int copied = 0;
2248         u64 base_offset;
2249         int rc;
2250
2251         base_offset = nfit_blk->bdw_offset + dpa % L1_CACHE_BYTES
2252                 + lane * mmio->size;
2253         write_blk_ctl(nfit_blk, lane, dpa, len, rw);
2254         while (len) {
2255                 unsigned int c;
2256                 u64 offset;
2257
2258                 if (mmio->num_lines) {
2259                         u32 line_offset;
2260
2261                         offset = to_interleave_offset(base_offset + copied,
2262                                         mmio);
2263                         div_u64_rem(offset, mmio->line_size, &line_offset);
2264                         c = min_t(size_t, len, mmio->line_size - line_offset);
2265                 } else {
2266                         offset = base_offset + nfit_blk->bdw_offset;
2267                         c = len;
2268                 }
2269
2270                 if (rw)
2271                         memcpy_flushcache(mmio->addr.aperture + offset, iobuf + copied, c);
2272                 else {
2273                         if (nfit_blk->dimm_flags & NFIT_BLK_READ_FLUSH)
2274                                 arch_invalidate_pmem((void __force *)
2275                                         mmio->addr.aperture + offset, c);
2276
2277                         memcpy(iobuf + copied, mmio->addr.aperture + offset, c);
2278                 }
2279
2280                 copied += c;
2281                 len -= c;
2282         }
2283
2284         if (rw)
2285                 nvdimm_flush(nfit_blk->nd_region);
2286
2287         rc = read_blk_stat(nfit_blk, lane) ? -EIO : 0;
2288         return rc;
2289 }
2290
2291 static int acpi_nfit_blk_region_do_io(struct nd_blk_region *ndbr,
2292                 resource_size_t dpa, void *iobuf, u64 len, int rw)
2293 {
2294         struct nfit_blk *nfit_blk = nd_blk_region_provider_data(ndbr);
2295         struct nfit_blk_mmio *mmio = &nfit_blk->mmio[BDW];
2296         struct nd_region *nd_region = nfit_blk->nd_region;
2297         unsigned int lane, copied = 0;
2298         int rc = 0;
2299
2300         lane = nd_region_acquire_lane(nd_region);
2301         while (len) {
2302                 u64 c = min(len, mmio->size);
2303
2304                 rc = acpi_nfit_blk_single_io(nfit_blk, dpa + copied,
2305                                 iobuf + copied, c, rw, lane);
2306                 if (rc)
2307                         break;
2308
2309                 copied += c;
2310                 len -= c;
2311         }
2312         nd_region_release_lane(nd_region, lane);
2313
2314         return rc;
2315 }
2316
2317 static int nfit_blk_init_interleave(struct nfit_blk_mmio *mmio,
2318                 struct acpi_nfit_interleave *idt, u16 interleave_ways)
2319 {
2320         if (idt) {
2321                 mmio->num_lines = idt->line_count;
2322                 mmio->line_size = idt->line_size;
2323                 if (interleave_ways == 0)
2324                         return -ENXIO;
2325                 mmio->table_size = mmio->num_lines * interleave_ways
2326                         * mmio->line_size;
2327         }
2328
2329         return 0;
2330 }
2331
2332 static int acpi_nfit_blk_get_flags(struct nvdimm_bus_descriptor *nd_desc,
2333                 struct nvdimm *nvdimm, struct nfit_blk *nfit_blk)
2334 {
2335         struct nd_cmd_dimm_flags flags;
2336         int rc;
2337
2338         memset(&flags, 0, sizeof(flags));
2339         rc = nd_desc->ndctl(nd_desc, nvdimm, ND_CMD_DIMM_FLAGS, &flags,
2340                         sizeof(flags), NULL);
2341
2342         if (rc >= 0 && flags.status == 0)
2343                 nfit_blk->dimm_flags = flags.flags;
2344         else if (rc == -ENOTTY) {
2345                 /* fall back to a conservative default */
2346                 nfit_blk->dimm_flags = NFIT_BLK_DCR_LATCH | NFIT_BLK_READ_FLUSH;
2347                 rc = 0;
2348         } else
2349                 rc = -ENXIO;
2350
2351         return rc;
2352 }
2353
2354 static int acpi_nfit_blk_region_enable(struct nvdimm_bus *nvdimm_bus,
2355                 struct device *dev)
2356 {
2357         struct nvdimm_bus_descriptor *nd_desc = to_nd_desc(nvdimm_bus);
2358         struct nd_blk_region *ndbr = to_nd_blk_region(dev);
2359         struct nfit_blk_mmio *mmio;
2360         struct nfit_blk *nfit_blk;
2361         struct nfit_mem *nfit_mem;
2362         struct nvdimm *nvdimm;
2363         int rc;
2364
2365         nvdimm = nd_blk_region_to_dimm(ndbr);
2366         nfit_mem = nvdimm_provider_data(nvdimm);
2367         if (!nfit_mem || !nfit_mem->dcr || !nfit_mem->bdw) {
2368                 dev_dbg(dev, "%s: missing%s%s%s\n", __func__,
2369                                 nfit_mem ? "" : " nfit_mem",
2370                                 (nfit_mem && nfit_mem->dcr) ? "" : " dcr",
2371                                 (nfit_mem && nfit_mem->bdw) ? "" : " bdw");
2372                 return -ENXIO;
2373         }
2374
2375         nfit_blk = devm_kzalloc(dev, sizeof(*nfit_blk), GFP_KERNEL);
2376         if (!nfit_blk)
2377                 return -ENOMEM;
2378         nd_blk_region_set_provider_data(ndbr, nfit_blk);
2379         nfit_blk->nd_region = to_nd_region(dev);
2380
2381         /* map block aperture memory */
2382         nfit_blk->bdw_offset = nfit_mem->bdw->offset;
2383         mmio = &nfit_blk->mmio[BDW];
2384         mmio->addr.base = devm_nvdimm_memremap(dev, nfit_mem->spa_bdw->address,
2385                         nfit_mem->spa_bdw->length, nd_blk_memremap_flags(ndbr));
2386         if (!mmio->addr.base) {
2387                 dev_dbg(dev, "%s: %s failed to map bdw\n", __func__,
2388                                 nvdimm_name(nvdimm));
2389                 return -ENOMEM;
2390         }
2391         mmio->size = nfit_mem->bdw->size;
2392         mmio->base_offset = nfit_mem->memdev_bdw->region_offset;
2393         mmio->idt = nfit_mem->idt_bdw;
2394         mmio->spa = nfit_mem->spa_bdw;
2395         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_bdw,
2396                         nfit_mem->memdev_bdw->interleave_ways);
2397         if (rc) {
2398                 dev_dbg(dev, "%s: %s failed to init bdw interleave\n",
2399                                 __func__, nvdimm_name(nvdimm));
2400                 return rc;
2401         }
2402
2403         /* map block control memory */
2404         nfit_blk->cmd_offset = nfit_mem->dcr->command_offset;
2405         nfit_blk->stat_offset = nfit_mem->dcr->status_offset;
2406         mmio = &nfit_blk->mmio[DCR];
2407         mmio->addr.base = devm_nvdimm_ioremap(dev, nfit_mem->spa_dcr->address,
2408                         nfit_mem->spa_dcr->length);
2409         if (!mmio->addr.base) {
2410                 dev_dbg(dev, "%s: %s failed to map dcr\n", __func__,
2411                                 nvdimm_name(nvdimm));
2412                 return -ENOMEM;
2413         }
2414         mmio->size = nfit_mem->dcr->window_size;
2415         mmio->base_offset = nfit_mem->memdev_dcr->region_offset;
2416         mmio->idt = nfit_mem->idt_dcr;
2417         mmio->spa = nfit_mem->spa_dcr;
2418         rc = nfit_blk_init_interleave(mmio, nfit_mem->idt_dcr,
2419                         nfit_mem->memdev_dcr->interleave_ways);
2420         if (rc) {
2421                 dev_dbg(dev, "%s: %s failed to init dcr interleave\n",
2422                                 __func__, nvdimm_name(nvdimm));
2423                 return rc;
2424         }
2425
2426         rc = acpi_nfit_blk_get_flags(nd_desc, nvdimm, nfit_blk);
2427         if (rc < 0) {
2428                 dev_dbg(dev, "%s: %s failed get DIMM flags\n",
2429                                 __func__, nvdimm_name(nvdimm));
2430                 return rc;
2431         }
2432
2433         if (nvdimm_has_flush(nfit_blk->nd_region) < 0)
2434                 dev_warn(dev, "unable to guarantee persistence of writes\n");
2435
2436         if (mmio->line_size == 0)
2437                 return 0;
2438
2439         if ((u32) nfit_blk->cmd_offset % mmio->line_size
2440                         + 8 > mmio->line_size) {
2441                 dev_dbg(dev, "cmd_offset crosses interleave boundary\n");
2442                 return -ENXIO;
2443         } else if ((u32) nfit_blk->stat_offset % mmio->line_size
2444                         + 8 > mmio->line_size) {
2445                 dev_dbg(dev, "stat_offset crosses interleave boundary\n");
2446                 return -ENXIO;
2447         }
2448
2449         return 0;
2450 }
2451
2452 static int ars_get_cap(struct acpi_nfit_desc *acpi_desc,
2453                 struct nd_cmd_ars_cap *cmd, struct nfit_spa *nfit_spa)
2454 {
2455         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2456         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2457         int cmd_rc, rc;
2458
2459         cmd->address = spa->address;
2460         cmd->length = spa->length;
2461         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, cmd,
2462                         sizeof(*cmd), &cmd_rc);
2463         if (rc < 0)
2464                 return rc;
2465         return cmd_rc;
2466 }
2467
2468 static int ars_start(struct acpi_nfit_desc *acpi_desc, struct nfit_spa *nfit_spa)
2469 {
2470         int rc;
2471         int cmd_rc;
2472         struct nd_cmd_ars_start ars_start;
2473         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2474         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2475
2476         memset(&ars_start, 0, sizeof(ars_start));
2477         ars_start.address = spa->address;
2478         ars_start.length = spa->length;
2479         ars_start.flags = acpi_desc->ars_start_flags;
2480         if (nfit_spa_type(spa) == NFIT_SPA_PM)
2481                 ars_start.type = ND_ARS_PERSISTENT;
2482         else if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE)
2483                 ars_start.type = ND_ARS_VOLATILE;
2484         else
2485                 return -ENOTTY;
2486
2487         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2488                         sizeof(ars_start), &cmd_rc);
2489
2490         if (rc < 0)
2491                 return rc;
2492         return cmd_rc;
2493 }
2494
2495 static int ars_continue(struct acpi_nfit_desc *acpi_desc)
2496 {
2497         int rc, cmd_rc;
2498         struct nd_cmd_ars_start ars_start;
2499         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2500         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2501
2502         memset(&ars_start, 0, sizeof(ars_start));
2503         ars_start.address = ars_status->restart_address;
2504         ars_start.length = ars_status->restart_length;
2505         ars_start.type = ars_status->type;
2506         ars_start.flags = acpi_desc->ars_start_flags;
2507         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_START, &ars_start,
2508                         sizeof(ars_start), &cmd_rc);
2509         if (rc < 0)
2510                 return rc;
2511         return cmd_rc;
2512 }
2513
2514 static int ars_get_status(struct acpi_nfit_desc *acpi_desc)
2515 {
2516         struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
2517         struct nd_cmd_ars_status *ars_status = acpi_desc->ars_status;
2518         int rc, cmd_rc;
2519
2520         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_STATUS, ars_status,
2521                         acpi_desc->ars_status_size, &cmd_rc);
2522         if (rc < 0)
2523                 return rc;
2524         return cmd_rc;
2525 }
2526
2527 static int ars_status_process_records(struct acpi_nfit_desc *acpi_desc,
2528                 struct nd_cmd_ars_status *ars_status)
2529 {
2530         struct nvdimm_bus *nvdimm_bus = acpi_desc->nvdimm_bus;
2531         int rc;
2532         u32 i;
2533
2534         /*
2535          * First record starts at 44 byte offset from the start of the
2536          * payload.
2537          */
2538         if (ars_status->out_length < 44)
2539                 return 0;
2540         for (i = 0; i < ars_status->num_records; i++) {
2541                 /* only process full records */
2542                 if (ars_status->out_length
2543                                 < 44 + sizeof(struct nd_ars_record) * (i + 1))
2544                         break;
2545                 rc = nvdimm_bus_add_badrange(nvdimm_bus,
2546                                 ars_status->records[i].err_address,
2547                                 ars_status->records[i].length);
2548                 if (rc)
2549                         return rc;
2550         }
2551         if (i < ars_status->num_records)
2552                 dev_warn(acpi_desc->dev, "detected truncated ars results\n");
2553
2554         return 0;
2555 }
2556
2557 static void acpi_nfit_remove_resource(void *data)
2558 {
2559         struct resource *res = data;
2560
2561         remove_resource(res);
2562 }
2563
2564 static int acpi_nfit_insert_resource(struct acpi_nfit_desc *acpi_desc,
2565                 struct nd_region_desc *ndr_desc)
2566 {
2567         struct resource *res, *nd_res = ndr_desc->res;
2568         int is_pmem, ret;
2569
2570         /* No operation if the region is already registered as PMEM */
2571         is_pmem = region_intersects(nd_res->start, resource_size(nd_res),
2572                                 IORESOURCE_MEM, IORES_DESC_PERSISTENT_MEMORY);
2573         if (is_pmem == REGION_INTERSECTS)
2574                 return 0;
2575
2576         res = devm_kzalloc(acpi_desc->dev, sizeof(*res), GFP_KERNEL);
2577         if (!res)
2578                 return -ENOMEM;
2579
2580         res->name = "Persistent Memory";
2581         res->start = nd_res->start;
2582         res->end = nd_res->end;
2583         res->flags = IORESOURCE_MEM;
2584         res->desc = IORES_DESC_PERSISTENT_MEMORY;
2585
2586         ret = insert_resource(&iomem_resource, res);
2587         if (ret)
2588                 return ret;
2589
2590         ret = devm_add_action_or_reset(acpi_desc->dev,
2591                                         acpi_nfit_remove_resource,
2592                                         res);
2593         if (ret)
2594                 return ret;
2595
2596         return 0;
2597 }
2598
2599 static int acpi_nfit_init_mapping(struct acpi_nfit_desc *acpi_desc,
2600                 struct nd_mapping_desc *mapping, struct nd_region_desc *ndr_desc,
2601                 struct acpi_nfit_memory_map *memdev,
2602                 struct nfit_spa *nfit_spa)
2603 {
2604         struct nvdimm *nvdimm = acpi_nfit_dimm_by_handle(acpi_desc,
2605                         memdev->device_handle);
2606         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2607         struct nd_blk_region_desc *ndbr_desc;
2608         struct nfit_mem *nfit_mem;
2609         int blk_valid = 0, rc;
2610
2611         if (!nvdimm) {
2612                 dev_err(acpi_desc->dev, "spa%d dimm: %#x not found\n",
2613                                 spa->range_index, memdev->device_handle);
2614                 return -ENODEV;
2615         }
2616
2617         mapping->nvdimm = nvdimm;
2618         switch (nfit_spa_type(spa)) {
2619         case NFIT_SPA_PM:
2620         case NFIT_SPA_VOLATILE:
2621                 mapping->start = memdev->address;
2622                 mapping->size = memdev->region_size;
2623                 break;
2624         case NFIT_SPA_DCR:
2625                 nfit_mem = nvdimm_provider_data(nvdimm);
2626                 if (!nfit_mem || !nfit_mem->bdw) {
2627                         dev_dbg(acpi_desc->dev, "spa%d %s missing bdw\n",
2628                                         spa->range_index, nvdimm_name(nvdimm));
2629                 } else {
2630                         mapping->size = nfit_mem->bdw->capacity;
2631                         mapping->start = nfit_mem->bdw->start_address;
2632                         ndr_desc->num_lanes = nfit_mem->bdw->windows;
2633                         blk_valid = 1;
2634                 }
2635
2636                 ndr_desc->mapping = mapping;
2637                 ndr_desc->num_mappings = blk_valid;
2638                 ndbr_desc = to_blk_region_desc(ndr_desc);
2639                 ndbr_desc->enable = acpi_nfit_blk_region_enable;
2640                 ndbr_desc->do_io = acpi_desc->blk_do_io;
2641                 rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2642                 if (rc)
2643                         return rc;
2644                 nfit_spa->nd_region = nvdimm_blk_region_create(acpi_desc->nvdimm_bus,
2645                                 ndr_desc);
2646                 if (!nfit_spa->nd_region)
2647                         return -ENOMEM;
2648                 break;
2649         }
2650
2651         return 0;
2652 }
2653
2654 static bool nfit_spa_is_virtual(struct acpi_nfit_system_address *spa)
2655 {
2656         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2657                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2658                 nfit_spa_type(spa) == NFIT_SPA_PDISK ||
2659                 nfit_spa_type(spa) == NFIT_SPA_PCD);
2660 }
2661
2662 static bool nfit_spa_is_volatile(struct acpi_nfit_system_address *spa)
2663 {
2664         return (nfit_spa_type(spa) == NFIT_SPA_VDISK ||
2665                 nfit_spa_type(spa) == NFIT_SPA_VCD   ||
2666                 nfit_spa_type(spa) == NFIT_SPA_VOLATILE);
2667 }
2668
2669 static int acpi_nfit_register_region(struct acpi_nfit_desc *acpi_desc,
2670                 struct nfit_spa *nfit_spa)
2671 {
2672         static struct nd_mapping_desc mappings[ND_MAX_MAPPINGS];
2673         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2674         struct nd_blk_region_desc ndbr_desc;
2675         struct nd_region_desc *ndr_desc;
2676         struct nfit_memdev *nfit_memdev;
2677         struct nvdimm_bus *nvdimm_bus;
2678         struct resource res;
2679         int count = 0, rc;
2680
2681         if (nfit_spa->nd_region)
2682                 return 0;
2683
2684         if (spa->range_index == 0 && !nfit_spa_is_virtual(spa)) {
2685                 dev_dbg(acpi_desc->dev, "%s: detected invalid spa index\n",
2686                                 __func__);
2687                 return 0;
2688         }
2689
2690         memset(&res, 0, sizeof(res));
2691         memset(&mappings, 0, sizeof(mappings));
2692         memset(&ndbr_desc, 0, sizeof(ndbr_desc));
2693         res.start = spa->address;
2694         res.end = res.start + spa->length - 1;
2695         ndr_desc = &ndbr_desc.ndr_desc;
2696         ndr_desc->res = &res;
2697         ndr_desc->provider_data = nfit_spa;
2698         ndr_desc->attr_groups = acpi_nfit_region_attribute_groups;
2699         if (spa->flags & ACPI_NFIT_PROXIMITY_VALID)
2700                 ndr_desc->numa_node = acpi_map_pxm_to_online_node(
2701                                                 spa->proximity_domain);
2702         else
2703                 ndr_desc->numa_node = NUMA_NO_NODE;
2704
2705         /*
2706          * Persistence domain bits are hierarchical, if
2707          * ACPI_NFIT_CAPABILITY_CACHE_FLUSH is set then
2708          * ACPI_NFIT_CAPABILITY_MEM_FLUSH is implied.
2709          */
2710         if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_CACHE_FLUSH)
2711                 set_bit(ND_REGION_PERSIST_CACHE, &ndr_desc->flags);
2712         else if (acpi_desc->platform_cap & ACPI_NFIT_CAPABILITY_MEM_FLUSH)
2713                 set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc->flags);
2714
2715         list_for_each_entry(nfit_memdev, &acpi_desc->memdevs, list) {
2716                 struct acpi_nfit_memory_map *memdev = nfit_memdev->memdev;
2717                 struct nd_mapping_desc *mapping;
2718
2719                 if (memdev->range_index != spa->range_index)
2720                         continue;
2721                 if (count >= ND_MAX_MAPPINGS) {
2722                         dev_err(acpi_desc->dev, "spa%d exceeds max mappings %d\n",
2723                                         spa->range_index, ND_MAX_MAPPINGS);
2724                         return -ENXIO;
2725                 }
2726                 mapping = &mappings[count++];
2727                 rc = acpi_nfit_init_mapping(acpi_desc, mapping, ndr_desc,
2728                                 memdev, nfit_spa);
2729                 if (rc)
2730                         goto out;
2731         }
2732
2733         ndr_desc->mapping = mappings;
2734         ndr_desc->num_mappings = count;
2735         rc = acpi_nfit_init_interleave_set(acpi_desc, ndr_desc, spa);
2736         if (rc)
2737                 goto out;
2738
2739         nvdimm_bus = acpi_desc->nvdimm_bus;
2740         if (nfit_spa_type(spa) == NFIT_SPA_PM) {
2741                 rc = acpi_nfit_insert_resource(acpi_desc, ndr_desc);
2742                 if (rc) {
2743                         dev_warn(acpi_desc->dev,
2744                                 "failed to insert pmem resource to iomem: %d\n",
2745                                 rc);
2746                         goto out;
2747                 }
2748
2749                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2750                                 ndr_desc);
2751                 if (!nfit_spa->nd_region)
2752                         rc = -ENOMEM;
2753         } else if (nfit_spa_is_volatile(spa)) {
2754                 nfit_spa->nd_region = nvdimm_volatile_region_create(nvdimm_bus,
2755                                 ndr_desc);
2756                 if (!nfit_spa->nd_region)
2757                         rc = -ENOMEM;
2758         } else if (nfit_spa_is_virtual(spa)) {
2759                 nfit_spa->nd_region = nvdimm_pmem_region_create(nvdimm_bus,
2760                                 ndr_desc);
2761                 if (!nfit_spa->nd_region)
2762                         rc = -ENOMEM;
2763         }
2764
2765  out:
2766         if (rc)
2767                 dev_err(acpi_desc->dev, "failed to register spa range %d\n",
2768                                 nfit_spa->spa->range_index);
2769         return rc;
2770 }
2771
2772 static int ars_status_alloc(struct acpi_nfit_desc *acpi_desc,
2773                 u32 max_ars)
2774 {
2775         struct device *dev = acpi_desc->dev;
2776         struct nd_cmd_ars_status *ars_status;
2777
2778         if (acpi_desc->ars_status && acpi_desc->ars_status_size >= max_ars) {
2779                 memset(acpi_desc->ars_status, 0, acpi_desc->ars_status_size);
2780                 return 0;
2781         }
2782
2783         if (acpi_desc->ars_status)
2784                 devm_kfree(dev, acpi_desc->ars_status);
2785         acpi_desc->ars_status = NULL;
2786         ars_status = devm_kzalloc(dev, max_ars, GFP_KERNEL);
2787         if (!ars_status)
2788                 return -ENOMEM;
2789         acpi_desc->ars_status = ars_status;
2790         acpi_desc->ars_status_size = max_ars;
2791         return 0;
2792 }
2793
2794 static int acpi_nfit_query_poison(struct acpi_nfit_desc *acpi_desc,
2795                 struct nfit_spa *nfit_spa)
2796 {
2797         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2798         int rc;
2799
2800         if (!nfit_spa->max_ars) {
2801                 struct nd_cmd_ars_cap ars_cap;
2802
2803                 memset(&ars_cap, 0, sizeof(ars_cap));
2804                 rc = ars_get_cap(acpi_desc, &ars_cap, nfit_spa);
2805                 if (rc < 0)
2806                         return rc;
2807                 nfit_spa->max_ars = ars_cap.max_ars_out;
2808                 nfit_spa->clear_err_unit = ars_cap.clear_err_unit;
2809                 /* check that the supported scrub types match the spa type */
2810                 if (nfit_spa_type(spa) == NFIT_SPA_VOLATILE &&
2811                                 ((ars_cap.status >> 16) & ND_ARS_VOLATILE) == 0)
2812                         return -ENOTTY;
2813                 else if (nfit_spa_type(spa) == NFIT_SPA_PM &&
2814                                 ((ars_cap.status >> 16) & ND_ARS_PERSISTENT) == 0)
2815                         return -ENOTTY;
2816         }
2817
2818         if (ars_status_alloc(acpi_desc, nfit_spa->max_ars))
2819                 return -ENOMEM;
2820
2821         rc = ars_get_status(acpi_desc);
2822         if (rc < 0 && rc != -ENOSPC)
2823                 return rc;
2824
2825         if (ars_status_process_records(acpi_desc, acpi_desc->ars_status))
2826                 return -ENOMEM;
2827
2828         return 0;
2829 }
2830
2831 static void acpi_nfit_async_scrub(struct acpi_nfit_desc *acpi_desc,
2832                 struct nfit_spa *nfit_spa)
2833 {
2834         struct acpi_nfit_system_address *spa = nfit_spa->spa;
2835         unsigned int overflow_retry = scrub_overflow_abort;
2836         u64 init_ars_start = 0, init_ars_len = 0;
2837         struct device *dev = acpi_desc->dev;
2838         unsigned int tmo = scrub_timeout;
2839         int rc;
2840
2841         if (!nfit_spa->ars_required || !nfit_spa->nd_region)
2842                 return;
2843
2844         rc = ars_start(acpi_desc, nfit_spa);
2845         /*
2846          * If we timed out the initial scan we'll still be busy here,
2847          * and will wait another timeout before giving up permanently.
2848          */
2849         if (rc < 0 && rc != -EBUSY)
2850                 return;
2851
2852         do {
2853                 u64 ars_start, ars_len;
2854
2855                 if (acpi_desc->cancel)
2856                         break;
2857                 rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2858                 if (rc == -ENOTTY)
2859                         break;
2860                 if (rc == -EBUSY && !tmo) {
2861                         dev_warn(dev, "range %d ars timeout, aborting\n",
2862                                         spa->range_index);
2863                         break;
2864                 }
2865
2866                 if (rc == -EBUSY) {
2867                         /*
2868                          * Note, entries may be appended to the list
2869                          * while the lock is dropped, but the workqueue
2870                          * being active prevents entries being deleted /
2871                          * freed.
2872                          */
2873                         mutex_unlock(&acpi_desc->init_mutex);
2874                         ssleep(1);
2875                         tmo--;
2876                         mutex_lock(&acpi_desc->init_mutex);
2877                         continue;
2878                 }
2879
2880                 /* we got some results, but there are more pending... */
2881                 if (rc == -ENOSPC && overflow_retry--) {
2882                         if (!init_ars_len) {
2883                                 init_ars_len = acpi_desc->ars_status->length;
2884                                 init_ars_start = acpi_desc->ars_status->address;
2885                         }
2886                         rc = ars_continue(acpi_desc);
2887                 }
2888
2889                 if (rc < 0) {
2890                         dev_warn(dev, "range %d ars continuation failed\n",
2891                                         spa->range_index);
2892                         break;
2893                 }
2894
2895                 if (init_ars_len) {
2896                         ars_start = init_ars_start;
2897                         ars_len = init_ars_len;
2898                 } else {
2899                         ars_start = acpi_desc->ars_status->address;
2900                         ars_len = acpi_desc->ars_status->length;
2901                 }
2902                 dev_dbg(dev, "spa range: %d ars from %#llx + %#llx complete\n",
2903                                 spa->range_index, ars_start, ars_len);
2904                 /* notify the region about new poison entries */
2905                 nvdimm_region_notify(nfit_spa->nd_region,
2906                                 NVDIMM_REVALIDATE_POISON);
2907                 break;
2908         } while (1);
2909 }
2910
2911 static void acpi_nfit_scrub(struct work_struct *work)
2912 {
2913         struct device *dev;
2914         u64 init_scrub_length = 0;
2915         struct nfit_spa *nfit_spa;
2916         u64 init_scrub_address = 0;
2917         bool init_ars_done = false;
2918         struct acpi_nfit_desc *acpi_desc;
2919         unsigned int tmo = scrub_timeout;
2920         unsigned int overflow_retry = scrub_overflow_abort;
2921
2922         acpi_desc = container_of(work, typeof(*acpi_desc), work);
2923         dev = acpi_desc->dev;
2924
2925         /*
2926          * We scrub in 2 phases.  The first phase waits for any platform
2927          * firmware initiated scrubs to complete and then we go search for the
2928          * affected spa regions to mark them scanned.  In the second phase we
2929          * initiate a directed scrub for every range that was not scrubbed in
2930          * phase 1. If we're called for a 'rescan', we harmlessly pass through
2931          * the first phase, but really only care about running phase 2, where
2932          * regions can be notified of new poison.
2933          */
2934
2935         /* process platform firmware initiated scrubs */
2936  retry:
2937         mutex_lock(&acpi_desc->init_mutex);
2938         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
2939                 struct nd_cmd_ars_status *ars_status;
2940                 struct acpi_nfit_system_address *spa;
2941                 u64 ars_start, ars_len;
2942                 int rc;
2943
2944                 if (acpi_desc->cancel)
2945                         break;
2946
2947                 if (nfit_spa->nd_region)
2948                         continue;
2949
2950                 if (init_ars_done) {
2951                         /*
2952                          * No need to re-query, we're now just
2953                          * reconciling all the ranges covered by the
2954                          * initial scrub
2955                          */
2956                         rc = 0;
2957                 } else
2958                         rc = acpi_nfit_query_poison(acpi_desc, nfit_spa);
2959
2960                 if (rc == -ENOTTY) {
2961                         /* no ars capability, just register spa and move on */
2962                         acpi_nfit_register_region(acpi_desc, nfit_spa);
2963                         continue;
2964                 }
2965
2966                 if (rc == -EBUSY && !tmo) {
2967                         /* fallthrough to directed scrub in phase 2 */
2968                         dev_warn(dev, "timeout awaiting ars results, continuing...\n");
2969                         break;
2970                 } else if (rc == -EBUSY) {
2971                         mutex_unlock(&acpi_desc->init_mutex);
2972                         ssleep(1);
2973                         tmo--;
2974                         goto retry;
2975                 }
2976
2977                 /* we got some results, but there are more pending... */
2978                 if (rc == -ENOSPC && overflow_retry--) {
2979                         ars_status = acpi_desc->ars_status;
2980                         /*
2981                          * Record the original scrub range, so that we
2982                          * can recall all the ranges impacted by the
2983                          * initial scrub.
2984                          */
2985                         if (!init_scrub_length) {
2986                                 init_scrub_length = ars_status->length;
2987                                 init_scrub_address = ars_status->address;
2988                         }
2989                         rc = ars_continue(acpi_desc);
2990                         if (rc == 0) {
2991                                 mutex_unlock(&acpi_desc->init_mutex);
2992                                 goto retry;
2993                         }
2994                 }
2995
2996                 if (rc < 0) {
2997                         /*
2998                          * Initial scrub failed, we'll give it one more
2999                          * try below...
3000                          */
3001                         break;
3002                 }
3003
3004                 /* We got some final results, record completed ranges */
3005                 ars_status = acpi_desc->ars_status;
3006                 if (init_scrub_length) {
3007                         ars_start = init_scrub_address;
3008                         ars_len = ars_start + init_scrub_length;
3009                 } else {
3010                         ars_start = ars_status->address;
3011                         ars_len = ars_status->length;
3012                 }
3013                 spa = nfit_spa->spa;
3014
3015                 if (!init_ars_done) {
3016                         init_ars_done = true;
3017                         dev_dbg(dev, "init scrub %#llx + %#llx complete\n",
3018                                         ars_start, ars_len);
3019                 }
3020                 if (ars_start <= spa->address && ars_start + ars_len
3021                                 >= spa->address + spa->length)
3022                         acpi_nfit_register_region(acpi_desc, nfit_spa);
3023         }
3024
3025         /*
3026          * For all the ranges not covered by an initial scrub we still
3027          * want to see if there are errors, but it's ok to discover them
3028          * asynchronously.
3029          */
3030         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3031                 /*
3032                  * Flag all the ranges that still need scrubbing, but
3033                  * register them now to make data available.
3034                  */
3035                 if (!nfit_spa->nd_region) {
3036                         nfit_spa->ars_required = 1;
3037                         acpi_nfit_register_region(acpi_desc, nfit_spa);
3038                 }
3039         }
3040         acpi_desc->init_complete = 1;
3041
3042         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3043                 acpi_nfit_async_scrub(acpi_desc, nfit_spa);
3044         acpi_desc->scrub_count++;
3045         acpi_desc->ars_start_flags = 0;
3046         if (acpi_desc->scrub_count_state)
3047                 sysfs_notify_dirent(acpi_desc->scrub_count_state);
3048         mutex_unlock(&acpi_desc->init_mutex);
3049 }
3050
3051 static int acpi_nfit_register_regions(struct acpi_nfit_desc *acpi_desc)
3052 {
3053         struct nfit_spa *nfit_spa;
3054         int rc;
3055
3056         list_for_each_entry(nfit_spa, &acpi_desc->spas, list)
3057                 if (nfit_spa_type(nfit_spa->spa) == NFIT_SPA_DCR) {
3058                         /* BLK regions don't need to wait for ars results */
3059                         rc = acpi_nfit_register_region(acpi_desc, nfit_spa);
3060                         if (rc)
3061                                 return rc;
3062                 }
3063
3064         acpi_desc->ars_start_flags = 0;
3065         if (!acpi_desc->cancel)
3066                 queue_work(nfit_wq, &acpi_desc->work);
3067         return 0;
3068 }
3069
3070 static int acpi_nfit_check_deletions(struct acpi_nfit_desc *acpi_desc,
3071                 struct nfit_table_prev *prev)
3072 {
3073         struct device *dev = acpi_desc->dev;
3074
3075         if (!list_empty(&prev->spas) ||
3076                         !list_empty(&prev->memdevs) ||
3077                         !list_empty(&prev->dcrs) ||
3078                         !list_empty(&prev->bdws) ||
3079                         !list_empty(&prev->idts) ||
3080                         !list_empty(&prev->flushes)) {
3081                 dev_err(dev, "new nfit deletes entries (unsupported)\n");
3082                 return -ENXIO;
3083         }
3084         return 0;
3085 }
3086
3087 static int acpi_nfit_desc_init_scrub_attr(struct acpi_nfit_desc *acpi_desc)
3088 {
3089         struct device *dev = acpi_desc->dev;
3090         struct kernfs_node *nfit;
3091         struct device *bus_dev;
3092
3093         if (!ars_supported(acpi_desc->nvdimm_bus))
3094                 return 0;
3095
3096         bus_dev = to_nvdimm_bus_dev(acpi_desc->nvdimm_bus);
3097         nfit = sysfs_get_dirent(bus_dev->kobj.sd, "nfit");
3098         if (!nfit) {
3099                 dev_err(dev, "sysfs_get_dirent 'nfit' failed\n");
3100                 return -ENODEV;
3101         }
3102         acpi_desc->scrub_count_state = sysfs_get_dirent(nfit, "scrub");
3103         sysfs_put(nfit);
3104         if (!acpi_desc->scrub_count_state) {
3105                 dev_err(dev, "sysfs_get_dirent 'scrub' failed\n");
3106                 return -ENODEV;
3107         }
3108
3109         return 0;
3110 }
3111
3112 static void acpi_nfit_unregister(void *data)
3113 {
3114         struct acpi_nfit_desc *acpi_desc = data;
3115
3116         nvdimm_bus_unregister(acpi_desc->nvdimm_bus);
3117 }
3118
3119 int acpi_nfit_init(struct acpi_nfit_desc *acpi_desc, void *data, acpi_size sz)
3120 {
3121         struct device *dev = acpi_desc->dev;
3122         struct nfit_table_prev prev;
3123         const void *end;
3124         int rc;
3125
3126         if (!acpi_desc->nvdimm_bus) {
3127                 acpi_nfit_init_dsms(acpi_desc);
3128
3129                 acpi_desc->nvdimm_bus = nvdimm_bus_register(dev,
3130                                 &acpi_desc->nd_desc);
3131                 if (!acpi_desc->nvdimm_bus)
3132                         return -ENOMEM;
3133
3134                 rc = devm_add_action_or_reset(dev, acpi_nfit_unregister,
3135                                 acpi_desc);
3136                 if (rc)
3137                         return rc;
3138
3139                 rc = acpi_nfit_desc_init_scrub_attr(acpi_desc);
3140                 if (rc)
3141                         return rc;
3142
3143                 /* register this acpi_desc for mce notifications */
3144                 mutex_lock(&acpi_desc_lock);
3145                 list_add_tail(&acpi_desc->list, &acpi_descs);
3146                 mutex_unlock(&acpi_desc_lock);
3147         }
3148
3149         mutex_lock(&acpi_desc->init_mutex);
3150
3151         INIT_LIST_HEAD(&prev.spas);
3152         INIT_LIST_HEAD(&prev.memdevs);
3153         INIT_LIST_HEAD(&prev.dcrs);
3154         INIT_LIST_HEAD(&prev.bdws);
3155         INIT_LIST_HEAD(&prev.idts);
3156         INIT_LIST_HEAD(&prev.flushes);
3157
3158         list_cut_position(&prev.spas, &acpi_desc->spas,
3159                                 acpi_desc->spas.prev);
3160         list_cut_position(&prev.memdevs, &acpi_desc->memdevs,
3161                                 acpi_desc->memdevs.prev);
3162         list_cut_position(&prev.dcrs, &acpi_desc->dcrs,
3163                                 acpi_desc->dcrs.prev);
3164         list_cut_position(&prev.bdws, &acpi_desc->bdws,
3165                                 acpi_desc->bdws.prev);
3166         list_cut_position(&prev.idts, &acpi_desc->idts,
3167                                 acpi_desc->idts.prev);
3168         list_cut_position(&prev.flushes, &acpi_desc->flushes,
3169                                 acpi_desc->flushes.prev);
3170
3171         end = data + sz;
3172         while (!IS_ERR_OR_NULL(data))
3173                 data = add_table(acpi_desc, &prev, data, end);
3174
3175         if (IS_ERR(data)) {
3176                 dev_dbg(dev, "%s: nfit table parsing error: %ld\n", __func__,
3177                                 PTR_ERR(data));
3178                 rc = PTR_ERR(data);
3179                 goto out_unlock;
3180         }
3181
3182         rc = acpi_nfit_check_deletions(acpi_desc, &prev);
3183         if (rc)
3184                 goto out_unlock;
3185
3186         rc = nfit_mem_init(acpi_desc);
3187         if (rc)
3188                 goto out_unlock;
3189
3190         rc = acpi_nfit_register_dimms(acpi_desc);
3191         if (rc)
3192                 goto out_unlock;
3193
3194         rc = acpi_nfit_register_regions(acpi_desc);
3195
3196  out_unlock:
3197         mutex_unlock(&acpi_desc->init_mutex);
3198         return rc;
3199 }
3200 EXPORT_SYMBOL_GPL(acpi_nfit_init);
3201
3202 struct acpi_nfit_flush_work {
3203         struct work_struct work;
3204         struct completion cmp;
3205 };
3206
3207 static void flush_probe(struct work_struct *work)
3208 {
3209         struct acpi_nfit_flush_work *flush;
3210
3211         flush = container_of(work, typeof(*flush), work);
3212         complete(&flush->cmp);
3213 }
3214
3215 static int acpi_nfit_flush_probe(struct nvdimm_bus_descriptor *nd_desc)
3216 {
3217         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3218         struct device *dev = acpi_desc->dev;
3219         struct acpi_nfit_flush_work flush;
3220         int rc;
3221
3222         /* bounce the device lock to flush acpi_nfit_add / acpi_nfit_notify */
3223         device_lock(dev);
3224         device_unlock(dev);
3225
3226         /* bounce the init_mutex to make init_complete valid */
3227         mutex_lock(&acpi_desc->init_mutex);
3228         if (acpi_desc->cancel || acpi_desc->init_complete) {
3229                 mutex_unlock(&acpi_desc->init_mutex);
3230                 return 0;
3231         }
3232
3233         /*
3234          * Scrub work could take 10s of seconds, userspace may give up so we
3235          * need to be interruptible while waiting.
3236          */
3237         INIT_WORK_ONSTACK(&flush.work, flush_probe);
3238         init_completion(&flush.cmp);
3239         queue_work(nfit_wq, &flush.work);
3240         mutex_unlock(&acpi_desc->init_mutex);
3241
3242         rc = wait_for_completion_interruptible(&flush.cmp);
3243         cancel_work_sync(&flush.work);
3244         return rc;
3245 }
3246
3247 static int acpi_nfit_clear_to_send(struct nvdimm_bus_descriptor *nd_desc,
3248                 struct nvdimm *nvdimm, unsigned int cmd)
3249 {
3250         struct acpi_nfit_desc *acpi_desc = to_acpi_nfit_desc(nd_desc);
3251
3252         if (nvdimm)
3253                 return 0;
3254         if (cmd != ND_CMD_ARS_START)
3255                 return 0;
3256
3257         /*
3258          * The kernel and userspace may race to initiate a scrub, but
3259          * the scrub thread is prepared to lose that initial race.  It
3260          * just needs guarantees that any ars it initiates are not
3261          * interrupted by any intervening start reqeusts from userspace.
3262          */
3263         if (work_busy(&acpi_desc->work))
3264                 return -EBUSY;
3265
3266         return 0;
3267 }
3268
3269 int acpi_nfit_ars_rescan(struct acpi_nfit_desc *acpi_desc, u8 flags)
3270 {
3271         struct device *dev = acpi_desc->dev;
3272         struct nfit_spa *nfit_spa;
3273
3274         if (work_busy(&acpi_desc->work))
3275                 return -EBUSY;
3276
3277         mutex_lock(&acpi_desc->init_mutex);
3278         if (acpi_desc->cancel) {
3279                 mutex_unlock(&acpi_desc->init_mutex);
3280                 return 0;
3281         }
3282
3283         list_for_each_entry(nfit_spa, &acpi_desc->spas, list) {
3284                 struct acpi_nfit_system_address *spa = nfit_spa->spa;
3285
3286                 if (nfit_spa_type(spa) != NFIT_SPA_PM)
3287                         continue;
3288
3289                 nfit_spa->ars_required = 1;
3290         }
3291         acpi_desc->ars_start_flags = flags;
3292         queue_work(nfit_wq, &acpi_desc->work);
3293         dev_dbg(dev, "%s: ars_scan triggered\n", __func__);
3294         mutex_unlock(&acpi_desc->init_mutex);
3295
3296         return 0;
3297 }
3298
3299 void acpi_nfit_desc_init(struct acpi_nfit_desc *acpi_desc, struct device *dev)
3300 {
3301         struct nvdimm_bus_descriptor *nd_desc;
3302
3303         dev_set_drvdata(dev, acpi_desc);
3304         acpi_desc->dev = dev;
3305         acpi_desc->blk_do_io = acpi_nfit_blk_region_do_io;
3306         nd_desc = &acpi_desc->nd_desc;
3307         nd_desc->provider_name = "ACPI.NFIT";
3308         nd_desc->module = THIS_MODULE;
3309         nd_desc->ndctl = acpi_nfit_ctl;
3310         nd_desc->flush_probe = acpi_nfit_flush_probe;
3311         nd_desc->clear_to_send = acpi_nfit_clear_to_send;
3312         nd_desc->attr_groups = acpi_nfit_attribute_groups;
3313
3314         INIT_LIST_HEAD(&acpi_desc->spas);
3315         INIT_LIST_HEAD(&acpi_desc->dcrs);
3316         INIT_LIST_HEAD(&acpi_desc->bdws);
3317         INIT_LIST_HEAD(&acpi_desc->idts);
3318         INIT_LIST_HEAD(&acpi_desc->flushes);
3319         INIT_LIST_HEAD(&acpi_desc->memdevs);
3320         INIT_LIST_HEAD(&acpi_desc->dimms);
3321         INIT_LIST_HEAD(&acpi_desc->list);
3322         mutex_init(&acpi_desc->init_mutex);
3323         INIT_WORK(&acpi_desc->work, acpi_nfit_scrub);
3324 }