drivers/nvdimm/region_devs.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/scatterlist.h>
  14 #include <linux/highmem.h>
  15 #include <linux/sched.h>
  16 #include <linux/slab.h>
  17 #include <linux/hash.h>
  18 #include <linux/pmem.h>
  19 #include <linux/sort.h>
  20 #include <linux/io.h>
  21 #include <linux/nd.h>
  22 #include "nd-core.h"
  23 #include "nd.h"
  24
  25 /*
  26  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  27  * irrelevant.
  28  */
  29 #include <linux/io-64-nonatomic-hi-lo.h>
  30
  31 static DEFINE_IDA(region_ida);
  32 static DEFINE_PER_CPU(int, flush_idx);
  33
  34 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
  35                 struct nd_region_data *ndrd)
  36 {
  37         int i, j;
  38
  39         dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
  40                         nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
  41         for (i = 0; i < (1 << ndrd->hints_shift); i++) {
  42                 struct resource *res = &nvdimm->flush_wpq[i];
  43                 unsigned long pfn = PHYS_PFN(res->start);
  44                 void __iomem *flush_page;
  45
  46                 /* check if flush hints share a page */
  47                 for (j = 0; j < i; j++) {
  48                         struct resource *res_j = &nvdimm->flush_wpq[j];
  49                         unsigned long pfn_j = PHYS_PFN(res_j->start);
  50
  51                         if (pfn == pfn_j)
  52                                 break;
  53                 }
  54
  55                 if (j < i)
  56                         flush_page = (void __iomem *) ((unsigned long)
  57                                         ndrd_get_flush_wpq(ndrd, dimm, j)
  58                                         & PAGE_MASK);
  59                 else
  60                         flush_page = devm_nvdimm_ioremap(dev,
  61                                         PFN_PHYS(pfn), PAGE_SIZE);
  62                 if (!flush_page)
  63                         return -ENXIO;
  64                 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
  65                                 + (res->start & ~PAGE_MASK));
  66         }
  67
  68         return 0;
  69 }
  70
  71 int nd_region_activate(struct nd_region *nd_region)
  72 {
  73         int i, j, num_flush = 0;
  74         struct nd_region_data *ndrd;
  75         struct device *dev = &nd_region->dev;
  76         size_t flush_data_size = sizeof(void *);
  77
  78         nvdimm_bus_lock(&nd_region->dev);
  79         for (i = 0; i < nd_region->ndr_mappings; i++) {
  80                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
  81                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
  82
  83                 /* at least one null hint slot per-dimm for the "no-hint" case */
  84                 flush_data_size += sizeof(void *);
  85                 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
  86                 if (!nvdimm->num_flush)
  87                         continue;
  88                 flush_data_size += nvdimm->num_flush * sizeof(void *);
  89         }
  90         nvdimm_bus_unlock(&nd_region->dev);
  91
  92         ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
  93         if (!ndrd)
  94                 return -ENOMEM;
  95         dev_set_drvdata(dev, ndrd);
  96
  97         if (!num_flush)
  98                 return 0;
  99
 100         ndrd->hints_shift = ilog2(num_flush);
 101         for (i = 0; i < nd_region->ndr_mappings; i++) {
 102                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 103                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 104                 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
 105
 106                 if (rc)
 107                         return rc;
 108         }
 109
 110         /*
 111          * Clear out entries that are duplicates. This should prevent the
 112          * extra flushings.
 113          */
 114         for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
 115                 /* ignore if NULL already */
 116                 if (!ndrd_get_flush_wpq(ndrd, i, 0))
 117                         continue;
 118
 119                 for (j = i + 1; j < nd_region->ndr_mappings; j++)
 120                         if (ndrd_get_flush_wpq(ndrd, i, 0) ==
 121                             ndrd_get_flush_wpq(ndrd, j, 0))
 122                                 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
 123         }
 124
 125         return 0;
 126 }
 127
 128 static void nd_region_release(struct device *dev)
 129 {
 130         struct nd_region *nd_region = to_nd_region(dev);
 131         u16 i;
 132
 133         for (i = 0; i < nd_region->ndr_mappings; i++) {
 134                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 135                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 136
 137                 put_device(&nvdimm->dev);
 138         }
 139         free_percpu(nd_region->lane);
 140         ida_simple_remove(&region_ida, nd_region->id);
 141         if (is_nd_blk(dev))
 142                 kfree(to_nd_blk_region(dev));
 143         else
 144                 kfree(nd_region);
 145 }
 146
 147 static struct device_type nd_blk_device_type = {
 148         .name = "nd_blk",
 149         .release = nd_region_release,
 150 };
 151
 152 static struct device_type nd_pmem_device_type = {
 153         .name = "nd_pmem",
 154         .release = nd_region_release,
 155 };
 156
 157 static struct device_type nd_volatile_device_type = {
 158         .name = "nd_volatile",
 159         .release = nd_region_release,
 160 };
 161
 162 bool is_nd_pmem(struct device *dev)
 163 {
 164         return dev ? dev->type == &nd_pmem_device_type : false;
 165 }
 166
 167 bool is_nd_blk(struct device *dev)
 168 {
 169         return dev ? dev->type == &nd_blk_device_type : false;
 170 }
 171
 172 struct nd_region *to_nd_region(struct device *dev)
 173 {
 174         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
 175
 176         WARN_ON(dev->type->release != nd_region_release);
 177         return nd_region;
 178 }
 179 EXPORT_SYMBOL_GPL(to_nd_region);
 180
 181 struct nd_blk_region *to_nd_blk_region(struct device *dev)
 182 {
 183         struct nd_region *nd_region = to_nd_region(dev);
 184
 185         WARN_ON(!is_nd_blk(dev));
 186         return container_of(nd_region, struct nd_blk_region, nd_region);
 187 }
 188 EXPORT_SYMBOL_GPL(to_nd_blk_region);
 189
 190 void *nd_region_provider_data(struct nd_region *nd_region)
 191 {
 192         return nd_region->provider_data;
 193 }
 194 EXPORT_SYMBOL_GPL(nd_region_provider_data);
 195
 196 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
 197 {
 198         return ndbr->blk_provider_data;
 199 }
 200 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
 201
 202 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
 203 {
 204         ndbr->blk_provider_data = data;
 205 }
 206 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
 207
 208 /**
 209  * nd_region_to_nstype() - region to an integer namespace type
 210  * @nd_region: region-device to interrogate
 211  *
 212  * This is the 'nstype' attribute of a region as well, an input to the
 213  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
 214  * namespace devices with namespace drivers.
 215  */
 216 int nd_region_to_nstype(struct nd_region *nd_region)
 217 {
 218         if (is_nd_pmem(&nd_region->dev)) {
 219                 u16 i, alias;
 220
 221                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
 222                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 223                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 224
 225                         if (test_bit(NDD_ALIASING, &nvdimm->flags))
 226                                 alias++;
 227                 }
 228                 if (alias)
 229                         return ND_DEVICE_NAMESPACE_PMEM;
 230                 else
 231                         return ND_DEVICE_NAMESPACE_IO;
 232         } else if (is_nd_blk(&nd_region->dev)) {
 233                 return ND_DEVICE_NAMESPACE_BLK;
 234         }
 235
 236         return 0;
 237 }
 238 EXPORT_SYMBOL(nd_region_to_nstype);
 239
 240 static ssize_t size_show(struct device *dev,
 241                 struct device_attribute *attr, char *buf)
 242 {
 243         struct nd_region *nd_region = to_nd_region(dev);
 244         unsigned long long size = 0;
 245
 246         if (is_nd_pmem(dev)) {
 247                 size = nd_region->ndr_size;
 248         } else if (nd_region->ndr_mappings == 1) {
 249                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 250
 251                 size = nd_mapping->size;
 252         }
 253
 254         return sprintf(buf, "%llu\n", size);
 255 }
 256 static DEVICE_ATTR_RO(size);
 257
 258 static ssize_t deep_flush_show(struct device *dev,
 259                 struct device_attribute *attr, char *buf)
 260 {
 261         struct nd_region *nd_region = to_nd_region(dev);
 262
 263         /*
 264          * NOTE: in the nvdimm_has_flush() error case this attribute is
 265          * not visible.
 266          */
 267         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
 268 }
 269
 270 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
 271                 const char *buf, size_t len)
 272 {
 273         bool flush;
 274         int rc = strtobool(buf, &flush);
 275         struct nd_region *nd_region = to_nd_region(dev);
 276
 277         if (rc)
 278                 return rc;
 279         if (!flush)
 280                 return -EINVAL;
 281         nvdimm_flush(nd_region);
 282
 283         return len;
 284 }
 285 static DEVICE_ATTR_RW(deep_flush);
 286
 287 static ssize_t mappings_show(struct device *dev,
 288                 struct device_attribute *attr, char *buf)
 289 {
 290         struct nd_region *nd_region = to_nd_region(dev);
 291
 292         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 293 }
 294 static DEVICE_ATTR_RO(mappings);
 295
 296 static ssize_t nstype_show(struct device *dev,
 297                 struct device_attribute *attr, char *buf)
 298 {
 299         struct nd_region *nd_region = to_nd_region(dev);
 300
 301         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 302 }
 303 static DEVICE_ATTR_RO(nstype);
 304
 305 static ssize_t set_cookie_show(struct device *dev,
 306                 struct device_attribute *attr, char *buf)
 307 {
 308         struct nd_region *nd_region = to_nd_region(dev);
 309         struct nd_interleave_set *nd_set = nd_region->nd_set;
 310         ssize_t rc = 0;
 311
 312         if (is_nd_pmem(dev) && nd_set)
 313                 /* pass, should be precluded by region_visible */;
 314         else
 315                 return -ENXIO;
 316
 317         /*
 318          * The cookie to show depends on which specification of the
 319          * labels we are using. If there are not labels then default to
 320          * the v1.1 namespace label cookie definition. To read all this
 321          * data we need to wait for probing to settle.
 322          */
 323         device_lock(dev);
 324         nvdimm_bus_lock(dev);
 325         wait_nvdimm_bus_probe_idle(dev);
 326         if (nd_region->ndr_mappings) {
 327                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 328                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 329
 330                 if (ndd) {
 331                         struct nd_namespace_index *nsindex;
 332
 333                         nsindex = to_namespace_index(ndd, ndd->ns_current);
 334                         rc = sprintf(buf, "%#llx\n",
 335                                         nd_region_interleave_set_cookie(nd_region,
 336                                                 nsindex));
 337                 }
 338         }
 339         nvdimm_bus_unlock(dev);
 340         device_unlock(dev);
 341
 342         if (rc)
 343                 return rc;
 344         return sprintf(buf, "%#llx\n", nd_set->cookie1);
 345 }
 346 static DEVICE_ATTR_RO(set_cookie);
 347
 348 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 349 {
 350         resource_size_t blk_max_overlap = 0, available, overlap;
 351         int i;
 352
 353         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 354
 355  retry:
 356         available = 0;
 357         overlap = blk_max_overlap;
 358         for (i = 0; i < nd_region->ndr_mappings; i++) {
 359                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 360                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 361
 362                 /* if a dimm is disabled the available capacity is zero */
 363                 if (!ndd)
 364                         return 0;
 365
 366                 if (is_nd_pmem(&nd_region->dev)) {
 367                         available += nd_pmem_available_dpa(nd_region,
 368                                         nd_mapping, &overlap);
 369                         if (overlap > blk_max_overlap) {
 370                                 blk_max_overlap = overlap;
 371                                 goto retry;
 372                         }
 373                 } else if (is_nd_blk(&nd_region->dev))
 374                         available += nd_blk_available_dpa(nd_region);
 375         }
 376
 377         return available;
 378 }
 379
 380 static ssize_t available_size_show(struct device *dev,
 381                 struct device_attribute *attr, char *buf)
 382 {
 383         struct nd_region *nd_region = to_nd_region(dev);
 384         unsigned long long available = 0;
 385
 386         /*
 387          * Flush in-flight updates and grab a snapshot of the available
 388          * size.  Of course, this value is potentially invalidated the
 389          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 390          * problem to not race itself.
 391          */
 392         nvdimm_bus_lock(dev);
 393         wait_nvdimm_bus_probe_idle(dev);
 394         available = nd_region_available_dpa(nd_region);
 395         nvdimm_bus_unlock(dev);
 396
 397         return sprintf(buf, "%llu\n", available);
 398 }
 399 static DEVICE_ATTR_RO(available_size);
 400
 401 static ssize_t init_namespaces_show(struct device *dev,
 402                 struct device_attribute *attr, char *buf)
 403 {
 404         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 405         ssize_t rc;
 406
 407         nvdimm_bus_lock(dev);
 408         if (ndrd)
 409                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 410         else
 411                 rc = -ENXIO;
 412         nvdimm_bus_unlock(dev);
 413
 414         return rc;
 415 }
 416 static DEVICE_ATTR_RO(init_namespaces);
 417
 418 static ssize_t namespace_seed_show(struct device *dev,
 419                 struct device_attribute *attr, char *buf)
 420 {
 421         struct nd_region *nd_region = to_nd_region(dev);
 422         ssize_t rc;
 423
 424         nvdimm_bus_lock(dev);
 425         if (nd_region->ns_seed)
 426                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 427         else
 428                 rc = sprintf(buf, "\n");
 429         nvdimm_bus_unlock(dev);
 430         return rc;
 431 }
 432 static DEVICE_ATTR_RO(namespace_seed);
 433
 434 static ssize_t btt_seed_show(struct device *dev,
 435                 struct device_attribute *attr, char *buf)
 436 {
 437         struct nd_region *nd_region = to_nd_region(dev);
 438         ssize_t rc;
 439
 440         nvdimm_bus_lock(dev);
 441         if (nd_region->btt_seed)
 442                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 443         else
 444                 rc = sprintf(buf, "\n");
 445         nvdimm_bus_unlock(dev);
 446
 447         return rc;
 448 }
 449 static DEVICE_ATTR_RO(btt_seed);
 450
 451 static ssize_t pfn_seed_show(struct device *dev,
 452                 struct device_attribute *attr, char *buf)
 453 {
 454         struct nd_region *nd_region = to_nd_region(dev);
 455         ssize_t rc;
 456
 457         nvdimm_bus_lock(dev);
 458         if (nd_region->pfn_seed)
 459                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 460         else
 461                 rc = sprintf(buf, "\n");
 462         nvdimm_bus_unlock(dev);
 463
 464         return rc;
 465 }
 466 static DEVICE_ATTR_RO(pfn_seed);
 467
 468 static ssize_t dax_seed_show(struct device *dev,
 469                 struct device_attribute *attr, char *buf)
 470 {
 471         struct nd_region *nd_region = to_nd_region(dev);
 472         ssize_t rc;
 473
 474         nvdimm_bus_lock(dev);
 475         if (nd_region->dax_seed)
 476                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 477         else
 478                 rc = sprintf(buf, "\n");
 479         nvdimm_bus_unlock(dev);
 480
 481         return rc;
 482 }
 483 static DEVICE_ATTR_RO(dax_seed);
 484
 485 static ssize_t read_only_show(struct device *dev,
 486                 struct device_attribute *attr, char *buf)
 487 {
 488         struct nd_region *nd_region = to_nd_region(dev);
 489
 490         return sprintf(buf, "%d\n", nd_region->ro);
 491 }
 492
 493 static ssize_t read_only_store(struct device *dev,
 494                 struct device_attribute *attr, const char *buf, size_t len)
 495 {
 496         bool ro;
 497         int rc = strtobool(buf, &ro);
 498         struct nd_region *nd_region = to_nd_region(dev);
 499
 500         if (rc)
 501                 return rc;
 502
 503         nd_region->ro = ro;
 504         return len;
 505 }
 506 static DEVICE_ATTR_RW(read_only);
 507
 508 static ssize_t region_badblocks_show(struct device *dev,
 509                 struct device_attribute *attr, char *buf)
 510 {
 511         struct nd_region *nd_region = to_nd_region(dev);
 512
 513         return badblocks_show(&nd_region->bb, buf, 0);
 514 }
 515
 516 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 517
 518 static ssize_t resource_show(struct device *dev,
 519                 struct device_attribute *attr, char *buf)
 520 {
 521         struct nd_region *nd_region = to_nd_region(dev);
 522
 523         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 524 }
 525 static DEVICE_ATTR_RO(resource);
 526
 527 static struct attribute *nd_region_attributes[] = {
 528         &dev_attr_size.attr,
 529         &dev_attr_nstype.attr,
 530         &dev_attr_mappings.attr,
 531         &dev_attr_btt_seed.attr,
 532         &dev_attr_pfn_seed.attr,
 533         &dev_attr_dax_seed.attr,
 534         &dev_attr_deep_flush.attr,
 535         &dev_attr_read_only.attr,
 536         &dev_attr_set_cookie.attr,
 537         &dev_attr_available_size.attr,
 538         &dev_attr_namespace_seed.attr,
 539         &dev_attr_init_namespaces.attr,
 540         &dev_attr_badblocks.attr,
 541         &dev_attr_resource.attr,
 542         NULL,
 543 };
 544
 545 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 546 {
 547         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 548         struct nd_region *nd_region = to_nd_region(dev);
 549         struct nd_interleave_set *nd_set = nd_region->nd_set;
 550         int type = nd_region_to_nstype(nd_region);
 551
 552         if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
 553                 return 0;
 554
 555         if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
 556                 return 0;
 557
 558         if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
 559                 return 0;
 560
 561         if (!is_nd_pmem(dev) && a == &dev_attr_resource.attr)
 562                 return 0;
 563
 564         if (a == &dev_attr_deep_flush.attr) {
 565                 int has_flush = nvdimm_has_flush(nd_region);
 566
 567                 if (has_flush == 1)
 568                         return a->mode;
 569                 else if (has_flush == 0)
 570                         return 0444;
 571                 else
 572                         return 0;
 573         }
 574
 575         if (a != &dev_attr_set_cookie.attr
 576                         && a != &dev_attr_available_size.attr)
 577                 return a->mode;
 578
 579         if ((type == ND_DEVICE_NAMESPACE_PMEM
 580                                 || type == ND_DEVICE_NAMESPACE_BLK)
 581                         && a == &dev_attr_available_size.attr)
 582                 return a->mode;
 583         else if (is_nd_pmem(dev) && nd_set)
 584                 return a->mode;
 585
 586         return 0;
 587 }
 588
 589 struct attribute_group nd_region_attribute_group = {
 590         .attrs = nd_region_attributes,
 591         .is_visible = region_visible,
 592 };
 593 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 594
 595 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 596                 struct nd_namespace_index *nsindex)
 597 {
 598         struct nd_interleave_set *nd_set = nd_region->nd_set;
 599
 600         if (!nd_set)
 601                 return 0;
 602
 603         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 604                         && __le16_to_cpu(nsindex->minor) == 1)
 605                 return nd_set->cookie1;
 606         return nd_set->cookie2;
 607 }
 608
 609 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 610 {
 611         struct nd_interleave_set *nd_set = nd_region->nd_set;
 612
 613         if (nd_set)
 614                 return nd_set->altcookie;
 615         return 0;
 616 }
 617
 618 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 619 {
 620         struct nd_label_ent *label_ent, *e;
 621
 622         lockdep_assert_held(&nd_mapping->lock);
 623         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 624                 list_del(&label_ent->list);
 625                 kfree(label_ent);
 626         }
 627 }
 628
 629 /*
 630  * Upon successful probe/remove, take/release a reference on the
 631  * associated interleave set (if present), and plant new btt + namespace
 632  * seeds.  Also, on the removal of a BLK region, notify the provider to
 633  * disable the region.
 634  */
 635 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 636                 struct device *dev, bool probe)
 637 {
 638         struct nd_region *nd_region;
 639
 640         if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
 641                 int i;
 642
 643                 nd_region = to_nd_region(dev);
 644                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 645                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 646                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 647                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 648
 649                         mutex_lock(&nd_mapping->lock);
 650                         nd_mapping_free_labels(nd_mapping);
 651                         mutex_unlock(&nd_mapping->lock);
 652
 653                         put_ndd(ndd);
 654                         nd_mapping->ndd = NULL;
 655                         if (ndd)
 656                                 atomic_dec(&nvdimm->busy);
 657                 }
 658
 659                 if (is_nd_pmem(dev))
 660                         return;
 661         }
 662         if (dev->parent && (is_nd_blk(dev->parent) || is_nd_pmem(dev->parent))
 663                         && probe) {
 664                 nd_region = to_nd_region(dev->parent);
 665                 nvdimm_bus_lock(dev);
 666                 if (nd_region->ns_seed == dev)
 667                         nd_region_create_ns_seed(nd_region);
 668                 nvdimm_bus_unlock(dev);
 669         }
 670         if (is_nd_btt(dev) && probe) {
 671                 struct nd_btt *nd_btt = to_nd_btt(dev);
 672
 673                 nd_region = to_nd_region(dev->parent);
 674                 nvdimm_bus_lock(dev);
 675                 if (nd_region->btt_seed == dev)
 676                         nd_region_create_btt_seed(nd_region);
 677                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 678                         nd_region_create_ns_seed(nd_region);
 679                 nvdimm_bus_unlock(dev);
 680         }
 681         if (is_nd_pfn(dev) && probe) {
 682                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 683
 684                 nd_region = to_nd_region(dev->parent);
 685                 nvdimm_bus_lock(dev);
 686                 if (nd_region->pfn_seed == dev)
 687                         nd_region_create_pfn_seed(nd_region);
 688                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 689                         nd_region_create_ns_seed(nd_region);
 690                 nvdimm_bus_unlock(dev);
 691         }
 692         if (is_nd_dax(dev) && probe) {
 693                 struct nd_dax *nd_dax = to_nd_dax(dev);
 694
 695                 nd_region = to_nd_region(dev->parent);
 696                 nvdimm_bus_lock(dev);
 697                 if (nd_region->dax_seed == dev)
 698                         nd_region_create_dax_seed(nd_region);
 699                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 700                         nd_region_create_ns_seed(nd_region);
 701                 nvdimm_bus_unlock(dev);
 702         }
 703 }
 704
 705 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 706 {
 707         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 708 }
 709
 710 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 711 {
 712         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 713 }
 714
 715 static ssize_t mappingN(struct device *dev, char *buf, int n)
 716 {
 717         struct nd_region *nd_region = to_nd_region(dev);
 718         struct nd_mapping *nd_mapping;
 719         struct nvdimm *nvdimm;
 720
 721         if (n >= nd_region->ndr_mappings)
 722                 return -ENXIO;
 723         nd_mapping = &nd_region->mapping[n];
 724         nvdimm = nd_mapping->nvdimm;
 725
 726         return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
 727                         nd_mapping->start, nd_mapping->size);
 728 }
 729
 730 #define REGION_MAPPING(idx) \
 731 static ssize_t mapping##idx##_show(struct device *dev,          \
 732                 struct device_attribute *attr, char *buf)       \
 733 {                                                               \
 734         return mappingN(dev, buf, idx);                         \
 735 }                                                               \
 736 static DEVICE_ATTR_RO(mapping##idx)
 737
 738 /*
 739  * 32 should be enough for a while, even in the presence of socket
 740  * interleave a 32-way interleave set is a degenerate case.
 741  */
 742 REGION_MAPPING(0);
 743 REGION_MAPPING(1);
 744 REGION_MAPPING(2);
 745 REGION_MAPPING(3);
 746 REGION_MAPPING(4);
 747 REGION_MAPPING(5);
 748 REGION_MAPPING(6);
 749 REGION_MAPPING(7);
 750 REGION_MAPPING(8);
 751 REGION_MAPPING(9);
 752 REGION_MAPPING(10);
 753 REGION_MAPPING(11);
 754 REGION_MAPPING(12);
 755 REGION_MAPPING(13);
 756 REGION_MAPPING(14);
 757 REGION_MAPPING(15);
 758 REGION_MAPPING(16);
 759 REGION_MAPPING(17);
 760 REGION_MAPPING(18);
 761 REGION_MAPPING(19);
 762 REGION_MAPPING(20);
 763 REGION_MAPPING(21);
 764 REGION_MAPPING(22);
 765 REGION_MAPPING(23);
 766 REGION_MAPPING(24);
 767 REGION_MAPPING(25);
 768 REGION_MAPPING(26);
 769 REGION_MAPPING(27);
 770 REGION_MAPPING(28);
 771 REGION_MAPPING(29);
 772 REGION_MAPPING(30);
 773 REGION_MAPPING(31);
 774
 775 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 776 {
 777         struct device *dev = container_of(kobj, struct device, kobj);
 778         struct nd_region *nd_region = to_nd_region(dev);
 779
 780         if (n < nd_region->ndr_mappings)
 781                 return a->mode;
 782         return 0;
 783 }
 784
 785 static struct attribute *mapping_attributes[] = {
 786         &dev_attr_mapping0.attr,
 787         &dev_attr_mapping1.attr,
 788         &dev_attr_mapping2.attr,
 789         &dev_attr_mapping3.attr,
 790         &dev_attr_mapping4.attr,
 791         &dev_attr_mapping5.attr,
 792         &dev_attr_mapping6.attr,
 793         &dev_attr_mapping7.attr,
 794         &dev_attr_mapping8.attr,
 795         &dev_attr_mapping9.attr,
 796         &dev_attr_mapping10.attr,
 797         &dev_attr_mapping11.attr,
 798         &dev_attr_mapping12.attr,
 799         &dev_attr_mapping13.attr,
 800         &dev_attr_mapping14.attr,
 801         &dev_attr_mapping15.attr,
 802         &dev_attr_mapping16.attr,
 803         &dev_attr_mapping17.attr,
 804         &dev_attr_mapping18.attr,
 805         &dev_attr_mapping19.attr,
 806         &dev_attr_mapping20.attr,
 807         &dev_attr_mapping21.attr,
 808         &dev_attr_mapping22.attr,
 809         &dev_attr_mapping23.attr,
 810         &dev_attr_mapping24.attr,
 811         &dev_attr_mapping25.attr,
 812         &dev_attr_mapping26.attr,
 813         &dev_attr_mapping27.attr,
 814         &dev_attr_mapping28.attr,
 815         &dev_attr_mapping29.attr,
 816         &dev_attr_mapping30.attr,
 817         &dev_attr_mapping31.attr,
 818         NULL,
 819 };
 820
 821 struct attribute_group nd_mapping_attribute_group = {
 822         .is_visible = mapping_visible,
 823         .attrs = mapping_attributes,
 824 };
 825 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 826
 827 int nd_blk_region_init(struct nd_region *nd_region)
 828 {
 829         struct device *dev = &nd_region->dev;
 830         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 831
 832         if (!is_nd_blk(dev))
 833                 return 0;
 834
 835         if (nd_region->ndr_mappings < 1) {
 836                 dev_dbg(dev, "invalid BLK region\n");
 837                 return -ENXIO;
 838         }
 839
 840         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 841 }
 842
 843 /**
 844  * nd_region_acquire_lane - allocate and lock a lane
 845  * @nd_region: region id and number of lanes possible
 846  *
 847  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 848  * We optimize for the common case where there are 256 lanes, one
 849  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 850  * this implementation assumes the cost of maintaining an allocator for
 851  * free lanes is on the order of the lock hold time, so it implements a
 852  * static lane = cpu % num_lanes mapping.
 853  *
 854  * In the case of a BTT instance on top of a BLK namespace a lane may be
 855  * acquired recursively.  We lock on the first instance.
 856  *
 857  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 858  * for the BTT metadata updates.
 859  */
 860 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 861 {
 862         unsigned int cpu, lane;
 863
 864         cpu = get_cpu();
 865         if (nd_region->num_lanes < nr_cpu_ids) {
 866                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 867
 868                 lane = cpu % nd_region->num_lanes;
 869                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 870                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 871                 if (ndl_count->count++ == 0)
 872                         spin_lock(&ndl_lock->lock);
 873         } else
 874                 lane = cpu;
 875
 876         return lane;
 877 }
 878 EXPORT_SYMBOL(nd_region_acquire_lane);
 879
 880 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 881 {
 882         if (nd_region->num_lanes < nr_cpu_ids) {
 883                 unsigned int cpu = get_cpu();
 884                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 885
 886                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 887                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 888                 if (--ndl_count->count == 0)
 889                         spin_unlock(&ndl_lock->lock);
 890                 put_cpu();
 891         }
 892         put_cpu();
 893 }
 894 EXPORT_SYMBOL(nd_region_release_lane);
 895
 896 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 897                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 898                 const char *caller)
 899 {
 900         struct nd_region *nd_region;
 901         struct device *dev;
 902         void *region_buf;
 903         unsigned int i;
 904         int ro = 0;
 905
 906         for (i = 0; i < ndr_desc->num_mappings; i++) {
 907                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 908                 struct nvdimm *nvdimm = mapping->nvdimm;
 909
 910                 if ((mapping->start | mapping->size) % SZ_4K) {
 911                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 912                                         caller, dev_name(&nvdimm->dev), i);
 913
 914                         return NULL;
 915                 }
 916
 917                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
 918                         ro = 1;
 919         }
 920
 921         if (dev_type == &nd_blk_device_type) {
 922                 struct nd_blk_region_desc *ndbr_desc;
 923                 struct nd_blk_region *ndbr;
 924
 925                 ndbr_desc = to_blk_region_desc(ndr_desc);
 926                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 927                                 * ndr_desc->num_mappings,
 928                                 GFP_KERNEL);
 929                 if (ndbr) {
 930                         nd_region = &ndbr->nd_region;
 931                         ndbr->enable = ndbr_desc->enable;
 932                         ndbr->do_io = ndbr_desc->do_io;
 933                 }
 934                 region_buf = ndbr;
 935         } else {
 936                 nd_region = kzalloc(sizeof(struct nd_region)
 937                                 + sizeof(struct nd_mapping)
 938                                 * ndr_desc->num_mappings,
 939                                 GFP_KERNEL);
 940                 region_buf = nd_region;
 941         }
 942
 943         if (!region_buf)
 944                 return NULL;
 945         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 946         if (nd_region->id < 0)
 947                 goto err_id;
 948
 949         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 950         if (!nd_region->lane)
 951                 goto err_percpu;
 952
 953         for (i = 0; i < nr_cpu_ids; i++) {
 954                 struct nd_percpu_lane *ndl;
 955
 956                 ndl = per_cpu_ptr(nd_region->lane, i);
 957                 spin_lock_init(&ndl->lock);
 958                 ndl->count = 0;
 959         }
 960
 961         for (i = 0; i < ndr_desc->num_mappings; i++) {
 962                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 963                 struct nvdimm *nvdimm = mapping->nvdimm;
 964
 965                 nd_region->mapping[i].nvdimm = nvdimm;
 966                 nd_region->mapping[i].start = mapping->start;
 967                 nd_region->mapping[i].size = mapping->size;
 968                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
 969                 mutex_init(&nd_region->mapping[i].lock);
 970
 971                 get_device(&nvdimm->dev);
 972         }
 973         nd_region->ndr_mappings = ndr_desc->num_mappings;
 974         nd_region->provider_data = ndr_desc->provider_data;
 975         nd_region->nd_set = ndr_desc->nd_set;
 976         nd_region->num_lanes = ndr_desc->num_lanes;
 977         nd_region->flags = ndr_desc->flags;
 978         nd_region->ro = ro;
 979         nd_region->numa_node = ndr_desc->numa_node;
 980         ida_init(&nd_region->ns_ida);
 981         ida_init(&nd_region->btt_ida);
 982         ida_init(&nd_region->pfn_ida);
 983         ida_init(&nd_region->dax_ida);
 984         dev = &nd_region->dev;
 985         dev_set_name(dev, "region%d", nd_region->id);
 986         dev->parent = &nvdimm_bus->dev;
 987         dev->type = dev_type;
 988         dev->groups = ndr_desc->attr_groups;
 989         nd_region->ndr_size = resource_size(ndr_desc->res);
 990         nd_region->ndr_start = ndr_desc->res->start;
 991         nd_device_register(dev);
 992
 993         return nd_region;
 994
 995  err_percpu:
 996         ida_simple_remove(&region_ida, nd_region->id);
 997  err_id:
 998         kfree(region_buf);
 999         return NULL;
1000 }
1001
1002 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1003                 struct nd_region_desc *ndr_desc)
1004 {
1005         ndr_desc->num_lanes = ND_MAX_LANES;
1006         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1007                         __func__);
1008 }
1009 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1010
1011 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1012                 struct nd_region_desc *ndr_desc)
1013 {
1014         if (ndr_desc->num_mappings > 1)
1015                 return NULL;
1016         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1017         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1018                         __func__);
1019 }
1020 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1021
1022 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1023                 struct nd_region_desc *ndr_desc)
1024 {
1025         ndr_desc->num_lanes = ND_MAX_LANES;
1026         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1027                         __func__);
1028 }
1029 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1030
1031 /**
1032  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1033  * @nd_region: blk or interleaved pmem region
1034  */
1035 void nvdimm_flush(struct nd_region *nd_region)
1036 {
1037         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1038         int i, idx;
1039
1040         /*
1041          * Try to encourage some diversity in flush hint addresses
1042          * across cpus assuming a limited number of flush hints.
1043          */
1044         idx = this_cpu_read(flush_idx);
1045         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1046
1047         /*
1048          * The first wmb() is needed to 'sfence' all previous writes
1049          * such that they are architecturally visible for the platform
1050          * buffer flush.  Note that we've already arranged for pmem
1051          * writes to avoid the cache via arch_memcpy_to_pmem().  The
1052          * final wmb() ensures ordering for the NVDIMM flush write.
1053          */
1054         wmb();
1055         for (i = 0; i < nd_region->ndr_mappings; i++)
1056                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1057                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1058         wmb();
1059 }
1060 EXPORT_SYMBOL_GPL(nvdimm_flush);
1061
1062 /**
1063  * nvdimm_has_flush - determine write flushing requirements
1064  * @nd_region: blk or interleaved pmem region
1065  *
1066  * Returns 1 if writes require flushing
1067  * Returns 0 if writes do not require flushing
1068  * Returns -ENXIO if flushing capability can not be determined
1069  */
1070 int nvdimm_has_flush(struct nd_region *nd_region)
1071 {
1072         int i;
1073
1074         /* no nvdimm == flushing capability unknown */
1075         if (nd_region->ndr_mappings == 0)
1076                 return -ENXIO;
1077
1078         for (i = 0; i < nd_region->ndr_mappings; i++) {
1079                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1080                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1081
1082                 /* flush hints present / available */
1083                 if (nvdimm->num_flush)
1084                         return 1;
1085         }
1086
1087         /*
1088          * The platform defines dimm devices without hints, assume
1089          * platform persistence mechanism like ADR
1090          */
1091         return 0;
1092 }
1093 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1094
1095 void __exit nd_region_devs_exit(void)
1096 {
1097         ida_destroy(&region_ida);
1098 }