drivers/nvdimm/dimm_devs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   6 #include <linux/moduleparam.h>
   7 #include <linux/vmalloc.h>
   8 #include <linux/device.h>
   9 #include <linux/ndctl.h>
  10 #include <linux/slab.h>
  11 #include <linux/io.h>
  12 #include <linux/fs.h>
  13 #include <linux/mm.h>
  14 #include "nd-core.h"
  15 #include "label.h"
  16 #include "pmem.h"
  17 #include "nd.h"
  18
  19 static DEFINE_IDA(dimm_ida);
  20
  21 static bool noblk;
  22 module_param(noblk, bool, 0444);
  23 MODULE_PARM_DESC(noblk, "force disable BLK / local alias support");
  24
  25 /*
  26  * Retrieve bus and dimm handle and return if this bus supports
  27  * get_config_data commands
  28  */
  29 int nvdimm_check_config_data(struct device *dev)
  30 {
  31         struct nvdimm *nvdimm = to_nvdimm(dev);
  32
  33         if (!nvdimm->cmd_mask ||
  34             !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
  35                 if (test_bit(NDD_LABELING, &nvdimm->flags))
  36                         return -ENXIO;
  37                 else
  38                         return -ENOTTY;
  39         }
  40
  41         return 0;
  42 }
  43
  44 static int validate_dimm(struct nvdimm_drvdata *ndd)
  45 {
  46         int rc;
  47
  48         if (!ndd)
  49                 return -EINVAL;
  50
  51         rc = nvdimm_check_config_data(ndd->dev);
  52         if (rc)
  53                 dev_dbg(ndd->dev, "%ps: %s error: %d\n",
  54                                 __builtin_return_address(0), __func__, rc);
  55         return rc;
  56 }
  57
  58 /**
  59  * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
  60  * @nvdimm: dimm to initialize
  61  */
  62 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
  63 {
  64         struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
  65         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
  66         struct nvdimm_bus_descriptor *nd_desc;
  67         int rc = validate_dimm(ndd);
  68         int cmd_rc = 0;
  69
  70         if (rc)
  71                 return rc;
  72
  73         if (cmd->config_size)
  74                 return 0; /* already valid */
  75
  76         memset(cmd, 0, sizeof(*cmd));
  77         nd_desc = nvdimm_bus->nd_desc;
  78         rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
  79                         ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
  80         if (rc < 0)
  81                 return rc;
  82         return cmd_rc;
  83 }
  84
  85 int nvdimm_get_config_data(struct nvdimm_drvdata *ndd, void *buf,
  86                            size_t offset, size_t len)
  87 {
  88         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
  89         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
  90         int rc = validate_dimm(ndd), cmd_rc = 0;
  91         struct nd_cmd_get_config_data_hdr *cmd;
  92         size_t max_cmd_size, buf_offset;
  93
  94         if (rc)
  95                 return rc;
  96
  97         if (offset + len > ndd->nsarea.config_size)
  98                 return -ENXIO;
  99
 100         max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
 101         cmd = kvzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
 102         if (!cmd)
 103                 return -ENOMEM;
 104
 105         for (buf_offset = 0; len;
 106              len -= cmd->in_length, buf_offset += cmd->in_length) {
 107                 size_t cmd_size;
 108
 109                 cmd->in_offset = offset + buf_offset;
 110                 cmd->in_length = min(max_cmd_size, len);
 111
 112                 cmd_size = sizeof(*cmd) + cmd->in_length;
 113
 114                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
 115                                 ND_CMD_GET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
 116                 if (rc < 0)
 117                         break;
 118                 if (cmd_rc < 0) {
 119                         rc = cmd_rc;
 120                         break;
 121                 }
 122
 123                 /* out_buf should be valid, copy it into our output buffer */
 124                 memcpy(buf + buf_offset, cmd->out_buf, cmd->in_length);
 125         }
 126         kvfree(cmd);
 127
 128         return rc;
 129 }
 130
 131 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
 132                 void *buf, size_t len)
 133 {
 134         size_t max_cmd_size, buf_offset;
 135         struct nd_cmd_set_config_hdr *cmd;
 136         int rc = validate_dimm(ndd), cmd_rc = 0;
 137         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
 138         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 139
 140         if (rc)
 141                 return rc;
 142
 143         if (offset + len > ndd->nsarea.config_size)
 144                 return -ENXIO;
 145
 146         max_cmd_size = min_t(u32, len, ndd->nsarea.max_xfer);
 147         cmd = kvzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
 148         if (!cmd)
 149                 return -ENOMEM;
 150
 151         for (buf_offset = 0; len; len -= cmd->in_length,
 152                         buf_offset += cmd->in_length) {
 153                 size_t cmd_size;
 154
 155                 cmd->in_offset = offset + buf_offset;
 156                 cmd->in_length = min(max_cmd_size, len);
 157                 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
 158
 159                 /* status is output in the last 4-bytes of the command buffer */
 160                 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
 161
 162                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
 163                                 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, &cmd_rc);
 164                 if (rc < 0)
 165                         break;
 166                 if (cmd_rc < 0) {
 167                         rc = cmd_rc;
 168                         break;
 169                 }
 170         }
 171         kvfree(cmd);
 172
 173         return rc;
 174 }
 175
 176 void nvdimm_set_labeling(struct device *dev)
 177 {
 178         struct nvdimm *nvdimm = to_nvdimm(dev);
 179
 180         set_bit(NDD_LABELING, &nvdimm->flags);
 181 }
 182
 183 void nvdimm_set_locked(struct device *dev)
 184 {
 185         struct nvdimm *nvdimm = to_nvdimm(dev);
 186
 187         set_bit(NDD_LOCKED, &nvdimm->flags);
 188 }
 189
 190 void nvdimm_clear_locked(struct device *dev)
 191 {
 192         struct nvdimm *nvdimm = to_nvdimm(dev);
 193
 194         clear_bit(NDD_LOCKED, &nvdimm->flags);
 195 }
 196
 197 static void nvdimm_release(struct device *dev)
 198 {
 199         struct nvdimm *nvdimm = to_nvdimm(dev);
 200
 201         ida_simple_remove(&dimm_ida, nvdimm->id);
 202         kfree(nvdimm);
 203 }
 204
 205 struct nvdimm *to_nvdimm(struct device *dev)
 206 {
 207         struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
 208
 209         WARN_ON(!is_nvdimm(dev));
 210         return nvdimm;
 211 }
 212 EXPORT_SYMBOL_GPL(to_nvdimm);
 213
 214 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
 215 {
 216         struct nd_region *nd_region = &ndbr->nd_region;
 217         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 218
 219         return nd_mapping->nvdimm;
 220 }
 221 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
 222
 223 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
 224 {
 225         /* pmem mapping properties are private to libnvdimm */
 226         return ARCH_MEMREMAP_PMEM;
 227 }
 228 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
 229
 230 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
 231 {
 232         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 233
 234         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
 235
 236         return dev_get_drvdata(&nvdimm->dev);
 237 }
 238 EXPORT_SYMBOL(to_ndd);
 239
 240 void nvdimm_drvdata_release(struct kref *kref)
 241 {
 242         struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
 243         struct device *dev = ndd->dev;
 244         struct resource *res, *_r;
 245
 246         dev_dbg(dev, "trace\n");
 247         nvdimm_bus_lock(dev);
 248         for_each_dpa_resource_safe(ndd, res, _r)
 249                 nvdimm_free_dpa(ndd, res);
 250         nvdimm_bus_unlock(dev);
 251
 252         kvfree(ndd->data);
 253         kfree(ndd);
 254         put_device(dev);
 255 }
 256
 257 void get_ndd(struct nvdimm_drvdata *ndd)
 258 {
 259         kref_get(&ndd->kref);
 260 }
 261
 262 void put_ndd(struct nvdimm_drvdata *ndd)
 263 {
 264         if (ndd)
 265                 kref_put(&ndd->kref, nvdimm_drvdata_release);
 266 }
 267
 268 const char *nvdimm_name(struct nvdimm *nvdimm)
 269 {
 270         return dev_name(&nvdimm->dev);
 271 }
 272 EXPORT_SYMBOL_GPL(nvdimm_name);
 273
 274 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
 275 {
 276         return &nvdimm->dev.kobj;
 277 }
 278 EXPORT_SYMBOL_GPL(nvdimm_kobj);
 279
 280 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
 281 {
 282         return nvdimm->cmd_mask;
 283 }
 284 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
 285
 286 void *nvdimm_provider_data(struct nvdimm *nvdimm)
 287 {
 288         if (nvdimm)
 289                 return nvdimm->provider_data;
 290         return NULL;
 291 }
 292 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
 293
 294 static ssize_t commands_show(struct device *dev,
 295                 struct device_attribute *attr, char *buf)
 296 {
 297         struct nvdimm *nvdimm = to_nvdimm(dev);
 298         int cmd, len = 0;
 299
 300         if (!nvdimm->cmd_mask)
 301                 return sprintf(buf, "\n");
 302
 303         for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
 304                 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
 305         len += sprintf(buf + len, "\n");
 306         return len;
 307 }
 308 static DEVICE_ATTR_RO(commands);
 309
 310 static ssize_t flags_show(struct device *dev,
 311                 struct device_attribute *attr, char *buf)
 312 {
 313         struct nvdimm *nvdimm = to_nvdimm(dev);
 314
 315         return sprintf(buf, "%s%s%s\n",
 316                         test_bit(NDD_ALIASING, &nvdimm->flags) ? "alias " : "",
 317                         test_bit(NDD_LABELING, &nvdimm->flags) ? "label " : "",
 318                         test_bit(NDD_LOCKED, &nvdimm->flags) ? "lock " : "");
 319 }
 320 static DEVICE_ATTR_RO(flags);
 321
 322 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 323                 char *buf)
 324 {
 325         struct nvdimm *nvdimm = to_nvdimm(dev);
 326
 327         /*
 328          * The state may be in the process of changing, userspace should
 329          * quiesce probing if it wants a static answer
 330          */
 331         nvdimm_bus_lock(dev);
 332         nvdimm_bus_unlock(dev);
 333         return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
 334                         ? "active" : "idle");
 335 }
 336 static DEVICE_ATTR_RO(state);
 337
 338 static ssize_t available_slots_show(struct device *dev,
 339                 struct device_attribute *attr, char *buf)
 340 {
 341         struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
 342         ssize_t rc;
 343         u32 nfree;
 344
 345         if (!ndd)
 346                 return -ENXIO;
 347
 348         nvdimm_bus_lock(dev);
 349         nfree = nd_label_nfree(ndd);
 350         if (nfree - 1 > nfree) {
 351                 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
 352                 nfree = 0;
 353         } else
 354                 nfree--;
 355         rc = sprintf(buf, "%d\n", nfree);
 356         nvdimm_bus_unlock(dev);
 357         return rc;
 358 }
 359 static DEVICE_ATTR_RO(available_slots);
 360
 361 __weak ssize_t security_show(struct device *dev,
 362                 struct device_attribute *attr, char *buf)
 363 {
 364         struct nvdimm *nvdimm = to_nvdimm(dev);
 365
 366         if (test_bit(NVDIMM_SECURITY_DISABLED, &nvdimm->sec.flags))
 367                 return sprintf(buf, "disabled\n");
 368         if (test_bit(NVDIMM_SECURITY_UNLOCKED, &nvdimm->sec.flags))
 369                 return sprintf(buf, "unlocked\n");
 370         if (test_bit(NVDIMM_SECURITY_LOCKED, &nvdimm->sec.flags))
 371                 return sprintf(buf, "locked\n");
 372         if (test_bit(NVDIMM_SECURITY_OVERWRITE, &nvdimm->sec.flags))
 373                 return sprintf(buf, "overwrite\n");
 374         return -ENOTTY;
 375 }
 376
 377 static ssize_t frozen_show(struct device *dev,
 378                 struct device_attribute *attr, char *buf)
 379 {
 380         struct nvdimm *nvdimm = to_nvdimm(dev);
 381
 382         return sprintf(buf, "%d\n", test_bit(NVDIMM_SECURITY_FROZEN,
 383                                 &nvdimm->sec.flags));
 384 }
 385 static DEVICE_ATTR_RO(frozen);
 386
 387 static ssize_t security_store(struct device *dev,
 388                 struct device_attribute *attr, const char *buf, size_t len)
 389
 390 {
 391         ssize_t rc;
 392
 393         /*
 394          * Require all userspace triggered security management to be
 395          * done while probing is idle and the DIMM is not in active use
 396          * in any region.
 397          */
 398         nd_device_lock(dev);
 399         nvdimm_bus_lock(dev);
 400         wait_nvdimm_bus_probe_idle(dev);
 401         rc = nvdimm_security_store(dev, buf, len);
 402         nvdimm_bus_unlock(dev);
 403         nd_device_unlock(dev);
 404
 405         return rc;
 406 }
 407 static DEVICE_ATTR_RW(security);
 408
 409 static struct attribute *nvdimm_attributes[] = {
 410         &dev_attr_state.attr,
 411         &dev_attr_flags.attr,
 412         &dev_attr_commands.attr,
 413         &dev_attr_available_slots.attr,
 414         &dev_attr_security.attr,
 415         &dev_attr_frozen.attr,
 416         NULL,
 417 };
 418
 419 static umode_t nvdimm_visible(struct kobject *kobj, struct attribute *a, int n)
 420 {
 421         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 422         struct nvdimm *nvdimm = to_nvdimm(dev);
 423
 424         if (a != &dev_attr_security.attr && a != &dev_attr_frozen.attr)
 425                 return a->mode;
 426         if (!nvdimm->sec.flags)
 427                 return 0;
 428
 429         if (a == &dev_attr_security.attr) {
 430                 /* Are there any state mutation ops (make writable)? */
 431                 if (nvdimm->sec.ops->freeze || nvdimm->sec.ops->disable
 432                                 || nvdimm->sec.ops->change_key
 433                                 || nvdimm->sec.ops->erase
 434                                 || nvdimm->sec.ops->overwrite)
 435                         return a->mode;
 436                 return 0444;
 437         }
 438
 439         if (nvdimm->sec.ops->freeze)
 440                 return a->mode;
 441         return 0;
 442 }
 443
 444 static const struct attribute_group nvdimm_attribute_group = {
 445         .attrs = nvdimm_attributes,
 446         .is_visible = nvdimm_visible,
 447 };
 448
 449 static const struct attribute_group *nvdimm_attribute_groups[] = {
 450         &nd_device_attribute_group,
 451         &nvdimm_attribute_group,
 452         NULL,
 453 };
 454
 455 static const struct device_type nvdimm_device_type = {
 456         .name = "nvdimm",
 457         .release = nvdimm_release,
 458         .groups = nvdimm_attribute_groups,
 459 };
 460
 461 bool is_nvdimm(struct device *dev)
 462 {
 463         return dev->type == &nvdimm_device_type;
 464 }
 465
 466 struct nvdimm *__nvdimm_create(struct nvdimm_bus *nvdimm_bus,
 467                 void *provider_data, const struct attribute_group **groups,
 468                 unsigned long flags, unsigned long cmd_mask, int num_flush,
 469                 struct resource *flush_wpq, const char *dimm_id,
 470                 const struct nvdimm_security_ops *sec_ops)
 471 {
 472         struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
 473         struct device *dev;
 474
 475         if (!nvdimm)
 476                 return NULL;
 477
 478         nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
 479         if (nvdimm->id < 0) {
 480                 kfree(nvdimm);
 481                 return NULL;
 482         }
 483
 484         nvdimm->dimm_id = dimm_id;
 485         nvdimm->provider_data = provider_data;
 486         if (noblk)
 487                 flags |= 1 << NDD_NOBLK;
 488         nvdimm->flags = flags;
 489         nvdimm->cmd_mask = cmd_mask;
 490         nvdimm->num_flush = num_flush;
 491         nvdimm->flush_wpq = flush_wpq;
 492         atomic_set(&nvdimm->busy, 0);
 493         dev = &nvdimm->dev;
 494         dev_set_name(dev, "nmem%d", nvdimm->id);
 495         dev->parent = &nvdimm_bus->dev;
 496         dev->type = &nvdimm_device_type;
 497         dev->devt = MKDEV(nvdimm_major, nvdimm->id);
 498         dev->groups = groups;
 499         nvdimm->sec.ops = sec_ops;
 500         nvdimm->sec.overwrite_tmo = 0;
 501         INIT_DELAYED_WORK(&nvdimm->dwork, nvdimm_security_overwrite_query);
 502         /*
 503          * Security state must be initialized before device_add() for
 504          * attribute visibility.
 505          */
 506         /* get security state and extended (master) state */
 507         nvdimm->sec.flags = nvdimm_security_flags(nvdimm, NVDIMM_USER);
 508         nvdimm->sec.ext_flags = nvdimm_security_flags(nvdimm, NVDIMM_MASTER);
 509         nd_device_register(dev);
 510
 511         return nvdimm;
 512 }
 513 EXPORT_SYMBOL_GPL(__nvdimm_create);
 514
 515 static void shutdown_security_notify(void *data)
 516 {
 517         struct nvdimm *nvdimm = data;
 518
 519         sysfs_put(nvdimm->sec.overwrite_state);
 520 }
 521
 522 int nvdimm_security_setup_events(struct device *dev)
 523 {
 524         struct nvdimm *nvdimm = to_nvdimm(dev);
 525
 526         if (!nvdimm->sec.flags || !nvdimm->sec.ops
 527                         || !nvdimm->sec.ops->overwrite)
 528                 return 0;
 529         nvdimm->sec.overwrite_state = sysfs_get_dirent(dev->kobj.sd, "security");
 530         if (!nvdimm->sec.overwrite_state)
 531                 return -ENOMEM;
 532
 533         return devm_add_action_or_reset(dev, shutdown_security_notify, nvdimm);
 534 }
 535 EXPORT_SYMBOL_GPL(nvdimm_security_setup_events);
 536
 537 int nvdimm_in_overwrite(struct nvdimm *nvdimm)
 538 {
 539         return test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags);
 540 }
 541 EXPORT_SYMBOL_GPL(nvdimm_in_overwrite);
 542
 543 int nvdimm_security_freeze(struct nvdimm *nvdimm)
 544 {
 545         int rc;
 546
 547         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
 548
 549         if (!nvdimm->sec.ops || !nvdimm->sec.ops->freeze)
 550                 return -EOPNOTSUPP;
 551
 552         if (!nvdimm->sec.flags)
 553                 return -EIO;
 554
 555         if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
 556                 dev_warn(&nvdimm->dev, "Overwrite operation in progress.\n");
 557                 return -EBUSY;
 558         }
 559
 560         rc = nvdimm->sec.ops->freeze(nvdimm);
 561         nvdimm->sec.flags = nvdimm_security_flags(nvdimm, NVDIMM_USER);
 562
 563         return rc;
 564 }
 565
 566 static unsigned long dpa_align(struct nd_region *nd_region)
 567 {
 568         struct device *dev = &nd_region->dev;
 569
 570         if (dev_WARN_ONCE(dev, !is_nvdimm_bus_locked(dev),
 571                                 "bus lock required for capacity provision\n"))
 572                 return 0;
 573         if (dev_WARN_ONCE(dev, !nd_region->ndr_mappings || nd_region->align
 574                                 % nd_region->ndr_mappings,
 575                                 "invalid region align %#lx mappings: %d\n",
 576                                 nd_region->align, nd_region->ndr_mappings))
 577                 return 0;
 578         return nd_region->align / nd_region->ndr_mappings;
 579 }
 580
 581 int alias_dpa_busy(struct device *dev, void *data)
 582 {
 583         resource_size_t map_end, blk_start, new;
 584         struct blk_alloc_info *info = data;
 585         struct nd_mapping *nd_mapping;
 586         struct nd_region *nd_region;
 587         struct nvdimm_drvdata *ndd;
 588         struct resource *res;
 589         unsigned long align;
 590         int i;
 591
 592         if (!is_memory(dev))
 593                 return 0;
 594
 595         nd_region = to_nd_region(dev);
 596         for (i = 0; i < nd_region->ndr_mappings; i++) {
 597                 nd_mapping  = &nd_region->mapping[i];
 598                 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
 599                         break;
 600         }
 601
 602         if (i >= nd_region->ndr_mappings)
 603                 return 0;
 604
 605         ndd = to_ndd(nd_mapping);
 606         map_end = nd_mapping->start + nd_mapping->size - 1;
 607         blk_start = nd_mapping->start;
 608
 609         /*
 610          * In the allocation case ->res is set to free space that we are
 611          * looking to validate against PMEM aliasing collision rules
 612          * (i.e. BLK is allocated after all aliased PMEM).
 613          */
 614         if (info->res) {
 615                 if (info->res->start >= nd_mapping->start
 616                                 && info->res->start < map_end)
 617                         /* pass */;
 618                 else
 619                         return 0;
 620         }
 621
 622  retry:
 623         /*
 624          * Find the free dpa from the end of the last pmem allocation to
 625          * the end of the interleave-set mapping.
 626          */
 627         align = dpa_align(nd_region);
 628         if (!align)
 629                 return 0;
 630
 631         for_each_dpa_resource(ndd, res) {
 632                 resource_size_t start, end;
 633
 634                 if (strncmp(res->name, "pmem", 4) != 0)
 635                         continue;
 636
 637                 start = ALIGN_DOWN(res->start, align);
 638                 end = ALIGN(res->end + 1, align) - 1;
 639                 if ((start >= blk_start && start < map_end)
 640                                 || (end >= blk_start && end <= map_end)) {
 641                         new = max(blk_start, min(map_end, end) + 1);
 642                         if (new != blk_start) {
 643                                 blk_start = new;
 644                                 goto retry;
 645                         }
 646                 }
 647         }
 648
 649         /* update the free space range with the probed blk_start */
 650         if (info->res && blk_start > info->res->start) {
 651                 info->res->start = max(info->res->start, blk_start);
 652                 if (info->res->start > info->res->end)
 653                         info->res->end = info->res->start - 1;
 654                 return 1;
 655         }
 656
 657         info->available -= blk_start - nd_mapping->start;
 658
 659         return 0;
 660 }
 661
 662 /**
 663  * nd_blk_available_dpa - account the unused dpa of BLK region
 664  * @nd_mapping: container of dpa-resource-root + labels
 665  *
 666  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
 667  * we arrange for them to never start at an lower dpa than the last
 668  * PMEM allocation in an aliased region.
 669  */
 670 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
 671 {
 672         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
 673         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 674         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 675         struct blk_alloc_info info = {
 676                 .nd_mapping = nd_mapping,
 677                 .available = nd_mapping->size,
 678                 .res = NULL,
 679         };
 680         struct resource *res;
 681         unsigned long align;
 682
 683         if (!ndd)
 684                 return 0;
 685
 686         device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
 687
 688         /* now account for busy blk allocations in unaliased dpa */
 689         align = dpa_align(nd_region);
 690         if (!align)
 691                 return 0;
 692         for_each_dpa_resource(ndd, res) {
 693                 resource_size_t start, end, size;
 694
 695                 if (strncmp(res->name, "blk", 3) != 0)
 696                         continue;
 697                 start = ALIGN_DOWN(res->start, align);
 698                 end = ALIGN(res->end + 1, align) - 1;
 699                 size = end - start + 1;
 700                 if (size >= info.available)
 701                         return 0;
 702                 info.available -= size;
 703         }
 704
 705         return info.available;
 706 }
 707
 708 /**
 709  * nd_pmem_max_contiguous_dpa - For the given dimm+region, return the max
 710  *                         contiguous unallocated dpa range.
 711  * @nd_region: constrain available space check to this reference region
 712  * @nd_mapping: container of dpa-resource-root + labels
 713  */
 714 resource_size_t nd_pmem_max_contiguous_dpa(struct nd_region *nd_region,
 715                                            struct nd_mapping *nd_mapping)
 716 {
 717         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 718         struct nvdimm_bus *nvdimm_bus;
 719         resource_size_t max = 0;
 720         struct resource *res;
 721         unsigned long align;
 722
 723         /* if a dimm is disabled the available capacity is zero */
 724         if (!ndd)
 725                 return 0;
 726
 727         align = dpa_align(nd_region);
 728         if (!align)
 729                 return 0;
 730
 731         nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
 732         if (__reserve_free_pmem(&nd_region->dev, nd_mapping->nvdimm))
 733                 return 0;
 734         for_each_dpa_resource(ndd, res) {
 735                 resource_size_t start, end;
 736
 737                 if (strcmp(res->name, "pmem-reserve") != 0)
 738                         continue;
 739                 /* trim free space relative to current alignment setting */
 740                 start = ALIGN(res->start, align);
 741                 end = ALIGN_DOWN(res->end + 1, align) - 1;
 742                 if (end < start)
 743                         continue;
 744                 if (end - start + 1 > max)
 745                         max = end - start + 1;
 746         }
 747         release_free_pmem(nvdimm_bus, nd_mapping);
 748         return max;
 749 }
 750
 751 /**
 752  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
 753  * @nd_mapping: container of dpa-resource-root + labels
 754  * @nd_region: constrain available space check to this reference region
 755  * @overlap: calculate available space assuming this level of overlap
 756  *
 757  * Validate that a PMEM label, if present, aligns with the start of an
 758  * interleave set and truncate the available size at the lowest BLK
 759  * overlap point.
 760  *
 761  * The expectation is that this routine is called multiple times as it
 762  * probes for the largest BLK encroachment for any single member DIMM of
 763  * the interleave set.  Once that value is determined the PMEM-limit for
 764  * the set can be established.
 765  */
 766 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
 767                 struct nd_mapping *nd_mapping, resource_size_t *overlap)
 768 {
 769         resource_size_t map_start, map_end, busy = 0, available, blk_start;
 770         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 771         struct resource *res;
 772         const char *reason;
 773         unsigned long align;
 774
 775         if (!ndd)
 776                 return 0;
 777
 778         align = dpa_align(nd_region);
 779         if (!align)
 780                 return 0;
 781
 782         map_start = nd_mapping->start;
 783         map_end = map_start + nd_mapping->size - 1;
 784         blk_start = max(map_start, map_end + 1 - *overlap);
 785         for_each_dpa_resource(ndd, res) {
 786                 resource_size_t start, end;
 787
 788                 start = ALIGN_DOWN(res->start, align);
 789                 end = ALIGN(res->end + 1, align) - 1;
 790                 if (start >= map_start && start < map_end) {
 791                         if (strncmp(res->name, "blk", 3) == 0)
 792                                 blk_start = min(blk_start,
 793                                                 max(map_start, start));
 794                         else if (end > map_end) {
 795                                 reason = "misaligned to iset";
 796                                 goto err;
 797                         } else
 798                                 busy += end - start + 1;
 799                 } else if (end >= map_start && end <= map_end) {
 800                         if (strncmp(res->name, "blk", 3) == 0) {
 801                                 /*
 802                                  * If a BLK allocation overlaps the start of
 803                                  * PMEM the entire interleave set may now only
 804                                  * be used for BLK.
 805                                  */
 806                                 blk_start = map_start;
 807                         } else
 808                                 busy += end - start + 1;
 809                 } else if (map_start > start && map_start < end) {
 810                         /* total eclipse of the mapping */
 811                         busy += nd_mapping->size;
 812                         blk_start = map_start;
 813                 }
 814         }
 815
 816         *overlap = map_end + 1 - blk_start;
 817         available = blk_start - map_start;
 818         if (busy < available)
 819                 return ALIGN_DOWN(available - busy, align);
 820         return 0;
 821
 822  err:
 823         nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
 824         return 0;
 825 }
 826
 827 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
 828 {
 829         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
 830         kfree(res->name);
 831         __release_region(&ndd->dpa, res->start, resource_size(res));
 832 }
 833
 834 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
 835                 struct nd_label_id *label_id, resource_size_t start,
 836                 resource_size_t n)
 837 {
 838         char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
 839         struct resource *res;
 840
 841         if (!name)
 842                 return NULL;
 843
 844         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
 845         res = __request_region(&ndd->dpa, start, n, name, 0);
 846         if (!res)
 847                 kfree(name);
 848         return res;
 849 }
 850
 851 /**
 852  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
 853  * @nvdimm: container of dpa-resource-root + labels
 854  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
 855  */
 856 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
 857                 struct nd_label_id *label_id)
 858 {
 859         resource_size_t allocated = 0;
 860         struct resource *res;
 861
 862         for_each_dpa_resource(ndd, res)
 863                 if (strcmp(res->name, label_id->id) == 0)
 864                         allocated += resource_size(res);
 865
 866         return allocated;
 867 }
 868
 869 static int count_dimms(struct device *dev, void *c)
 870 {
 871         int *count = c;
 872
 873         if (is_nvdimm(dev))
 874                 (*count)++;
 875         return 0;
 876 }
 877
 878 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
 879 {
 880         int count = 0;
 881         /* Flush any possible dimm registration failures */
 882         nd_synchronize();
 883
 884         device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
 885         dev_dbg(&nvdimm_bus->dev, "count: %d\n", count);
 886         if (count != dimm_count)
 887                 return -ENXIO;
 888         return 0;
 889 }
 890 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
 891
 892 void __exit nvdimm_devs_exit(void)
 893 {
 894         ida_destroy(&dimm_ida);
 895 }