drivers/nvdimm/region_devs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #include <linux/scatterlist.h>
   6 #include <linux/highmem.h>
   7 #include <linux/sched.h>
   8 #include <linux/slab.h>
   9 #include <linux/hash.h>
  10 #include <linux/sort.h>
  11 #include <linux/io.h>
  12 #include <linux/nd.h>
  13 #include "nd-core.h"
  14 #include "nd.h"
  15
  16 /*
  17  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  18  * irrelevant.
  19  */
  20 #include <linux/io-64-nonatomic-hi-lo.h>
  21
  22 static DEFINE_IDA(region_ida);
  23 static DEFINE_PER_CPU(int, flush_idx);
  24
  25 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
  26                 struct nd_region_data *ndrd)
  27 {
  28         int i, j;
  29
  30         dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
  31                         nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
  32         for (i = 0; i < (1 << ndrd->hints_shift); i++) {
  33                 struct resource *res = &nvdimm->flush_wpq[i];
  34                 unsigned long pfn = PHYS_PFN(res->start);
  35                 void __iomem *flush_page;
  36
  37                 /* check if flush hints share a page */
  38                 for (j = 0; j < i; j++) {
  39                         struct resource *res_j = &nvdimm->flush_wpq[j];
  40                         unsigned long pfn_j = PHYS_PFN(res_j->start);
  41
  42                         if (pfn == pfn_j)
  43                                 break;
  44                 }
  45
  46                 if (j < i)
  47                         flush_page = (void __iomem *) ((unsigned long)
  48                                         ndrd_get_flush_wpq(ndrd, dimm, j)
  49                                         & PAGE_MASK);
  50                 else
  51                         flush_page = devm_nvdimm_ioremap(dev,
  52                                         PFN_PHYS(pfn), PAGE_SIZE);
  53                 if (!flush_page)
  54                         return -ENXIO;
  55                 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
  56                                 + (res->start & ~PAGE_MASK));
  57         }
  58
  59         return 0;
  60 }
  61
  62 int nd_region_activate(struct nd_region *nd_region)
  63 {
  64         int i, j, num_flush = 0;
  65         struct nd_region_data *ndrd;
  66         struct device *dev = &nd_region->dev;
  67         size_t flush_data_size = sizeof(void *);
  68
  69         nvdimm_bus_lock(&nd_region->dev);
  70         for (i = 0; i < nd_region->ndr_mappings; i++) {
  71                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
  72                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
  73
  74                 if (test_bit(NDD_SECURITY_OVERWRITE, &nvdimm->flags)) {
  75                         nvdimm_bus_unlock(&nd_region->dev);
  76                         return -EBUSY;
  77                 }
  78
  79                 /* at least one null hint slot per-dimm for the "no-hint" case */
  80                 flush_data_size += sizeof(void *);
  81                 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
  82                 if (!nvdimm->num_flush)
  83                         continue;
  84                 flush_data_size += nvdimm->num_flush * sizeof(void *);
  85         }
  86         nvdimm_bus_unlock(&nd_region->dev);
  87
  88         ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
  89         if (!ndrd)
  90                 return -ENOMEM;
  91         dev_set_drvdata(dev, ndrd);
  92
  93         if (!num_flush)
  94                 return 0;
  95
  96         ndrd->hints_shift = ilog2(num_flush);
  97         for (i = 0; i < nd_region->ndr_mappings; i++) {
  98                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
  99                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 100                 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
 101
 102                 if (rc)
 103                         return rc;
 104         }
 105
 106         /*
 107          * Clear out entries that are duplicates. This should prevent the
 108          * extra flushings.
 109          */
 110         for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
 111                 /* ignore if NULL already */
 112                 if (!ndrd_get_flush_wpq(ndrd, i, 0))
 113                         continue;
 114
 115                 for (j = i + 1; j < nd_region->ndr_mappings; j++)
 116                         if (ndrd_get_flush_wpq(ndrd, i, 0) ==
 117                             ndrd_get_flush_wpq(ndrd, j, 0))
 118                                 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
 119         }
 120
 121         return 0;
 122 }
 123
 124 static void nd_region_release(struct device *dev)
 125 {
 126         struct nd_region *nd_region = to_nd_region(dev);
 127         u16 i;
 128
 129         for (i = 0; i < nd_region->ndr_mappings; i++) {
 130                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 131                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 132
 133                 put_device(&nvdimm->dev);
 134         }
 135         free_percpu(nd_region->lane);
 136         ida_simple_remove(&region_ida, nd_region->id);
 137         if (is_nd_blk(dev))
 138                 kfree(to_nd_blk_region(dev));
 139         else
 140                 kfree(nd_region);
 141 }
 142
 143 static struct device_type nd_blk_device_type = {
 144         .name = "nd_blk",
 145         .release = nd_region_release,
 146 };
 147
 148 static struct device_type nd_pmem_device_type = {
 149         .name = "nd_pmem",
 150         .release = nd_region_release,
 151 };
 152
 153 static struct device_type nd_volatile_device_type = {
 154         .name = "nd_volatile",
 155         .release = nd_region_release,
 156 };
 157
 158 bool is_nd_pmem(struct device *dev)
 159 {
 160         return dev ? dev->type == &nd_pmem_device_type : false;
 161 }
 162
 163 bool is_nd_blk(struct device *dev)
 164 {
 165         return dev ? dev->type == &nd_blk_device_type : false;
 166 }
 167
 168 bool is_nd_volatile(struct device *dev)
 169 {
 170         return dev ? dev->type == &nd_volatile_device_type : false;
 171 }
 172
 173 struct nd_region *to_nd_region(struct device *dev)
 174 {
 175         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
 176
 177         WARN_ON(dev->type->release != nd_region_release);
 178         return nd_region;
 179 }
 180 EXPORT_SYMBOL_GPL(to_nd_region);
 181
 182 struct device *nd_region_dev(struct nd_region *nd_region)
 183 {
 184         if (!nd_region)
 185                 return NULL;
 186         return &nd_region->dev;
 187 }
 188 EXPORT_SYMBOL_GPL(nd_region_dev);
 189
 190 struct nd_blk_region *to_nd_blk_region(struct device *dev)
 191 {
 192         struct nd_region *nd_region = to_nd_region(dev);
 193
 194         WARN_ON(!is_nd_blk(dev));
 195         return container_of(nd_region, struct nd_blk_region, nd_region);
 196 }
 197 EXPORT_SYMBOL_GPL(to_nd_blk_region);
 198
 199 void *nd_region_provider_data(struct nd_region *nd_region)
 200 {
 201         return nd_region->provider_data;
 202 }
 203 EXPORT_SYMBOL_GPL(nd_region_provider_data);
 204
 205 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
 206 {
 207         return ndbr->blk_provider_data;
 208 }
 209 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
 210
 211 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
 212 {
 213         ndbr->blk_provider_data = data;
 214 }
 215 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
 216
 217 /**
 218  * nd_region_to_nstype() - region to an integer namespace type
 219  * @nd_region: region-device to interrogate
 220  *
 221  * This is the 'nstype' attribute of a region as well, an input to the
 222  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
 223  * namespace devices with namespace drivers.
 224  */
 225 int nd_region_to_nstype(struct nd_region *nd_region)
 226 {
 227         if (is_memory(&nd_region->dev)) {
 228                 u16 i, alias;
 229
 230                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
 231                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 232                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 233
 234                         if (test_bit(NDD_ALIASING, &nvdimm->flags))
 235                                 alias++;
 236                 }
 237                 if (alias)
 238                         return ND_DEVICE_NAMESPACE_PMEM;
 239                 else
 240                         return ND_DEVICE_NAMESPACE_IO;
 241         } else if (is_nd_blk(&nd_region->dev)) {
 242                 return ND_DEVICE_NAMESPACE_BLK;
 243         }
 244
 245         return 0;
 246 }
 247 EXPORT_SYMBOL(nd_region_to_nstype);
 248
 249 static ssize_t size_show(struct device *dev,
 250                 struct device_attribute *attr, char *buf)
 251 {
 252         struct nd_region *nd_region = to_nd_region(dev);
 253         unsigned long long size = 0;
 254
 255         if (is_memory(dev)) {
 256                 size = nd_region->ndr_size;
 257         } else if (nd_region->ndr_mappings == 1) {
 258                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 259
 260                 size = nd_mapping->size;
 261         }
 262
 263         return sprintf(buf, "%llu\n", size);
 264 }
 265 static DEVICE_ATTR_RO(size);
 266
 267 static ssize_t deep_flush_show(struct device *dev,
 268                 struct device_attribute *attr, char *buf)
 269 {
 270         struct nd_region *nd_region = to_nd_region(dev);
 271
 272         /*
 273          * NOTE: in the nvdimm_has_flush() error case this attribute is
 274          * not visible.
 275          */
 276         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
 277 }
 278
 279 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
 280                 const char *buf, size_t len)
 281 {
 282         bool flush;
 283         int rc = strtobool(buf, &flush);
 284         struct nd_region *nd_region = to_nd_region(dev);
 285
 286         if (rc)
 287                 return rc;
 288         if (!flush)
 289                 return -EINVAL;
 290         rc = nvdimm_flush(nd_region, NULL);
 291         if (rc)
 292                 return rc;
 293
 294         return len;
 295 }
 296 static DEVICE_ATTR_RW(deep_flush);
 297
 298 static ssize_t mappings_show(struct device *dev,
 299                 struct device_attribute *attr, char *buf)
 300 {
 301         struct nd_region *nd_region = to_nd_region(dev);
 302
 303         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 304 }
 305 static DEVICE_ATTR_RO(mappings);
 306
 307 static ssize_t nstype_show(struct device *dev,
 308                 struct device_attribute *attr, char *buf)
 309 {
 310         struct nd_region *nd_region = to_nd_region(dev);
 311
 312         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 313 }
 314 static DEVICE_ATTR_RO(nstype);
 315
 316 static ssize_t set_cookie_show(struct device *dev,
 317                 struct device_attribute *attr, char *buf)
 318 {
 319         struct nd_region *nd_region = to_nd_region(dev);
 320         struct nd_interleave_set *nd_set = nd_region->nd_set;
 321         ssize_t rc = 0;
 322
 323         if (is_memory(dev) && nd_set)
 324                 /* pass, should be precluded by region_visible */;
 325         else
 326                 return -ENXIO;
 327
 328         /*
 329          * The cookie to show depends on which specification of the
 330          * labels we are using. If there are not labels then default to
 331          * the v1.1 namespace label cookie definition. To read all this
 332          * data we need to wait for probing to settle.
 333          */
 334         nd_device_lock(dev);
 335         nvdimm_bus_lock(dev);
 336         wait_nvdimm_bus_probe_idle(dev);
 337         if (nd_region->ndr_mappings) {
 338                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 339                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 340
 341                 if (ndd) {
 342                         struct nd_namespace_index *nsindex;
 343
 344                         nsindex = to_namespace_index(ndd, ndd->ns_current);
 345                         rc = sprintf(buf, "%#llx\n",
 346                                         nd_region_interleave_set_cookie(nd_region,
 347                                                 nsindex));
 348                 }
 349         }
 350         nvdimm_bus_unlock(dev);
 351         nd_device_unlock(dev);
 352
 353         if (rc)
 354                 return rc;
 355         return sprintf(buf, "%#llx\n", nd_set->cookie1);
 356 }
 357 static DEVICE_ATTR_RO(set_cookie);
 358
 359 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 360 {
 361         resource_size_t blk_max_overlap = 0, available, overlap;
 362         int i;
 363
 364         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 365
 366  retry:
 367         available = 0;
 368         overlap = blk_max_overlap;
 369         for (i = 0; i < nd_region->ndr_mappings; i++) {
 370                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 371                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 372
 373                 /* if a dimm is disabled the available capacity is zero */
 374                 if (!ndd)
 375                         return 0;
 376
 377                 if (is_memory(&nd_region->dev)) {
 378                         available += nd_pmem_available_dpa(nd_region,
 379                                         nd_mapping, &overlap);
 380                         if (overlap > blk_max_overlap) {
 381                                 blk_max_overlap = overlap;
 382                                 goto retry;
 383                         }
 384                 } else if (is_nd_blk(&nd_region->dev))
 385                         available += nd_blk_available_dpa(nd_region);
 386         }
 387
 388         return available;
 389 }
 390
 391 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
 392 {
 393         resource_size_t available = 0;
 394         int i;
 395
 396         if (is_memory(&nd_region->dev))
 397                 available = PHYS_ADDR_MAX;
 398
 399         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 400         for (i = 0; i < nd_region->ndr_mappings; i++) {
 401                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 402
 403                 if (is_memory(&nd_region->dev))
 404                         available = min(available,
 405                                         nd_pmem_max_contiguous_dpa(nd_region,
 406                                                                    nd_mapping));
 407                 else if (is_nd_blk(&nd_region->dev))
 408                         available += nd_blk_available_dpa(nd_region);
 409         }
 410         if (is_memory(&nd_region->dev))
 411                 return available * nd_region->ndr_mappings;
 412         return available;
 413 }
 414
 415 static ssize_t available_size_show(struct device *dev,
 416                 struct device_attribute *attr, char *buf)
 417 {
 418         struct nd_region *nd_region = to_nd_region(dev);
 419         unsigned long long available = 0;
 420
 421         /*
 422          * Flush in-flight updates and grab a snapshot of the available
 423          * size.  Of course, this value is potentially invalidated the
 424          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 425          * problem to not race itself.
 426          */
 427         nd_device_lock(dev);
 428         nvdimm_bus_lock(dev);
 429         wait_nvdimm_bus_probe_idle(dev);
 430         available = nd_region_available_dpa(nd_region);
 431         nvdimm_bus_unlock(dev);
 432         nd_device_unlock(dev);
 433
 434         return sprintf(buf, "%llu\n", available);
 435 }
 436 static DEVICE_ATTR_RO(available_size);
 437
 438 static ssize_t max_available_extent_show(struct device *dev,
 439                 struct device_attribute *attr, char *buf)
 440 {
 441         struct nd_region *nd_region = to_nd_region(dev);
 442         unsigned long long available = 0;
 443
 444         nd_device_lock(dev);
 445         nvdimm_bus_lock(dev);
 446         wait_nvdimm_bus_probe_idle(dev);
 447         available = nd_region_allocatable_dpa(nd_region);
 448         nvdimm_bus_unlock(dev);
 449         nd_device_unlock(dev);
 450
 451         return sprintf(buf, "%llu\n", available);
 452 }
 453 static DEVICE_ATTR_RO(max_available_extent);
 454
 455 static ssize_t init_namespaces_show(struct device *dev,
 456                 struct device_attribute *attr, char *buf)
 457 {
 458         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 459         ssize_t rc;
 460
 461         nvdimm_bus_lock(dev);
 462         if (ndrd)
 463                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 464         else
 465                 rc = -ENXIO;
 466         nvdimm_bus_unlock(dev);
 467
 468         return rc;
 469 }
 470 static DEVICE_ATTR_RO(init_namespaces);
 471
 472 static ssize_t namespace_seed_show(struct device *dev,
 473                 struct device_attribute *attr, char *buf)
 474 {
 475         struct nd_region *nd_region = to_nd_region(dev);
 476         ssize_t rc;
 477
 478         nvdimm_bus_lock(dev);
 479         if (nd_region->ns_seed)
 480                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 481         else
 482                 rc = sprintf(buf, "\n");
 483         nvdimm_bus_unlock(dev);
 484         return rc;
 485 }
 486 static DEVICE_ATTR_RO(namespace_seed);
 487
 488 static ssize_t btt_seed_show(struct device *dev,
 489                 struct device_attribute *attr, char *buf)
 490 {
 491         struct nd_region *nd_region = to_nd_region(dev);
 492         ssize_t rc;
 493
 494         nvdimm_bus_lock(dev);
 495         if (nd_region->btt_seed)
 496                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 497         else
 498                 rc = sprintf(buf, "\n");
 499         nvdimm_bus_unlock(dev);
 500
 501         return rc;
 502 }
 503 static DEVICE_ATTR_RO(btt_seed);
 504
 505 static ssize_t pfn_seed_show(struct device *dev,
 506                 struct device_attribute *attr, char *buf)
 507 {
 508         struct nd_region *nd_region = to_nd_region(dev);
 509         ssize_t rc;
 510
 511         nvdimm_bus_lock(dev);
 512         if (nd_region->pfn_seed)
 513                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 514         else
 515                 rc = sprintf(buf, "\n");
 516         nvdimm_bus_unlock(dev);
 517
 518         return rc;
 519 }
 520 static DEVICE_ATTR_RO(pfn_seed);
 521
 522 static ssize_t dax_seed_show(struct device *dev,
 523                 struct device_attribute *attr, char *buf)
 524 {
 525         struct nd_region *nd_region = to_nd_region(dev);
 526         ssize_t rc;
 527
 528         nvdimm_bus_lock(dev);
 529         if (nd_region->dax_seed)
 530                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 531         else
 532                 rc = sprintf(buf, "\n");
 533         nvdimm_bus_unlock(dev);
 534
 535         return rc;
 536 }
 537 static DEVICE_ATTR_RO(dax_seed);
 538
 539 static ssize_t read_only_show(struct device *dev,
 540                 struct device_attribute *attr, char *buf)
 541 {
 542         struct nd_region *nd_region = to_nd_region(dev);
 543
 544         return sprintf(buf, "%d\n", nd_region->ro);
 545 }
 546
 547 static ssize_t read_only_store(struct device *dev,
 548                 struct device_attribute *attr, const char *buf, size_t len)
 549 {
 550         bool ro;
 551         int rc = strtobool(buf, &ro);
 552         struct nd_region *nd_region = to_nd_region(dev);
 553
 554         if (rc)
 555                 return rc;
 556
 557         nd_region->ro = ro;
 558         return len;
 559 }
 560 static DEVICE_ATTR_RW(read_only);
 561
 562 static ssize_t region_badblocks_show(struct device *dev,
 563                 struct device_attribute *attr, char *buf)
 564 {
 565         struct nd_region *nd_region = to_nd_region(dev);
 566         ssize_t rc;
 567
 568         nd_device_lock(dev);
 569         if (dev->driver)
 570                 rc = badblocks_show(&nd_region->bb, buf, 0);
 571         else
 572                 rc = -ENXIO;
 573         nd_device_unlock(dev);
 574
 575         return rc;
 576 }
 577 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 578
 579 static ssize_t resource_show(struct device *dev,
 580                 struct device_attribute *attr, char *buf)
 581 {
 582         struct nd_region *nd_region = to_nd_region(dev);
 583
 584         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 585 }
 586 static DEVICE_ATTR_RO(resource);
 587
 588 static ssize_t persistence_domain_show(struct device *dev,
 589                 struct device_attribute *attr, char *buf)
 590 {
 591         struct nd_region *nd_region = to_nd_region(dev);
 592
 593         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
 594                 return sprintf(buf, "cpu_cache\n");
 595         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
 596                 return sprintf(buf, "memory_controller\n");
 597         else
 598                 return sprintf(buf, "\n");
 599 }
 600 static DEVICE_ATTR_RO(persistence_domain);
 601
 602 static struct attribute *nd_region_attributes[] = {
 603         &dev_attr_size.attr,
 604         &dev_attr_nstype.attr,
 605         &dev_attr_mappings.attr,
 606         &dev_attr_btt_seed.attr,
 607         &dev_attr_pfn_seed.attr,
 608         &dev_attr_dax_seed.attr,
 609         &dev_attr_deep_flush.attr,
 610         &dev_attr_read_only.attr,
 611         &dev_attr_set_cookie.attr,
 612         &dev_attr_available_size.attr,
 613         &dev_attr_max_available_extent.attr,
 614         &dev_attr_namespace_seed.attr,
 615         &dev_attr_init_namespaces.attr,
 616         &dev_attr_badblocks.attr,
 617         &dev_attr_resource.attr,
 618         &dev_attr_persistence_domain.attr,
 619         NULL,
 620 };
 621
 622 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 623 {
 624         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 625         struct nd_region *nd_region = to_nd_region(dev);
 626         struct nd_interleave_set *nd_set = nd_region->nd_set;
 627         int type = nd_region_to_nstype(nd_region);
 628
 629         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
 630                 return 0;
 631
 632         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
 633                 return 0;
 634
 635         if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
 636                 return 0;
 637
 638         if (a == &dev_attr_resource.attr) {
 639                 if (is_memory(dev))
 640                         return 0400;
 641                 else
 642                         return 0;
 643         }
 644
 645         if (a == &dev_attr_deep_flush.attr) {
 646                 int has_flush = nvdimm_has_flush(nd_region);
 647
 648                 if (has_flush == 1)
 649                         return a->mode;
 650                 else if (has_flush == 0)
 651                         return 0444;
 652                 else
 653                         return 0;
 654         }
 655
 656         if (a == &dev_attr_persistence_domain.attr) {
 657                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
 658                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
 659                         return 0;
 660                 return a->mode;
 661         }
 662
 663         if (a != &dev_attr_set_cookie.attr
 664                         && a != &dev_attr_available_size.attr)
 665                 return a->mode;
 666
 667         if ((type == ND_DEVICE_NAMESPACE_PMEM
 668                                 || type == ND_DEVICE_NAMESPACE_BLK)
 669                         && a == &dev_attr_available_size.attr)
 670                 return a->mode;
 671         else if (is_memory(dev) && nd_set)
 672                 return a->mode;
 673
 674         return 0;
 675 }
 676
 677 struct attribute_group nd_region_attribute_group = {
 678         .attrs = nd_region_attributes,
 679         .is_visible = region_visible,
 680 };
 681 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 682
 683 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 684                 struct nd_namespace_index *nsindex)
 685 {
 686         struct nd_interleave_set *nd_set = nd_region->nd_set;
 687
 688         if (!nd_set)
 689                 return 0;
 690
 691         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 692                         && __le16_to_cpu(nsindex->minor) == 1)
 693                 return nd_set->cookie1;
 694         return nd_set->cookie2;
 695 }
 696
 697 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 698 {
 699         struct nd_interleave_set *nd_set = nd_region->nd_set;
 700
 701         if (nd_set)
 702                 return nd_set->altcookie;
 703         return 0;
 704 }
 705
 706 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 707 {
 708         struct nd_label_ent *label_ent, *e;
 709
 710         lockdep_assert_held(&nd_mapping->lock);
 711         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 712                 list_del(&label_ent->list);
 713                 kfree(label_ent);
 714         }
 715 }
 716
 717 /*
 718  * When a namespace is activated create new seeds for the next
 719  * namespace, or namespace-personality to be configured.
 720  */
 721 void nd_region_advance_seeds(struct nd_region *nd_region, struct device *dev)
 722 {
 723         nvdimm_bus_lock(dev);
 724         if (nd_region->ns_seed == dev) {
 725                 nd_region_create_ns_seed(nd_region);
 726         } else if (is_nd_btt(dev)) {
 727                 struct nd_btt *nd_btt = to_nd_btt(dev);
 728
 729                 if (nd_region->btt_seed == dev)
 730                         nd_region_create_btt_seed(nd_region);
 731                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 732                         nd_region_create_ns_seed(nd_region);
 733         } else if (is_nd_pfn(dev)) {
 734                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 735
 736                 if (nd_region->pfn_seed == dev)
 737                         nd_region_create_pfn_seed(nd_region);
 738                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 739                         nd_region_create_ns_seed(nd_region);
 740         } else if (is_nd_dax(dev)) {
 741                 struct nd_dax *nd_dax = to_nd_dax(dev);
 742
 743                 if (nd_region->dax_seed == dev)
 744                         nd_region_create_dax_seed(nd_region);
 745                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 746                         nd_region_create_ns_seed(nd_region);
 747         }
 748         nvdimm_bus_unlock(dev);
 749 }
 750
 751 static ssize_t mappingN(struct device *dev, char *buf, int n)
 752 {
 753         struct nd_region *nd_region = to_nd_region(dev);
 754         struct nd_mapping *nd_mapping;
 755         struct nvdimm *nvdimm;
 756
 757         if (n >= nd_region->ndr_mappings)
 758                 return -ENXIO;
 759         nd_mapping = &nd_region->mapping[n];
 760         nvdimm = nd_mapping->nvdimm;
 761
 762         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
 763                         nd_mapping->start, nd_mapping->size,
 764                         nd_mapping->position);
 765 }
 766
 767 #define REGION_MAPPING(idx) \
 768 static ssize_t mapping##idx##_show(struct device *dev,          \
 769                 struct device_attribute *attr, char *buf)       \
 770 {                                                               \
 771         return mappingN(dev, buf, idx);                         \
 772 }                                                               \
 773 static DEVICE_ATTR_RO(mapping##idx)
 774
 775 /*
 776  * 32 should be enough for a while, even in the presence of socket
 777  * interleave a 32-way interleave set is a degenerate case.
 778  */
 779 REGION_MAPPING(0);
 780 REGION_MAPPING(1);
 781 REGION_MAPPING(2);
 782 REGION_MAPPING(3);
 783 REGION_MAPPING(4);
 784 REGION_MAPPING(5);
 785 REGION_MAPPING(6);
 786 REGION_MAPPING(7);
 787 REGION_MAPPING(8);
 788 REGION_MAPPING(9);
 789 REGION_MAPPING(10);
 790 REGION_MAPPING(11);
 791 REGION_MAPPING(12);
 792 REGION_MAPPING(13);
 793 REGION_MAPPING(14);
 794 REGION_MAPPING(15);
 795 REGION_MAPPING(16);
 796 REGION_MAPPING(17);
 797 REGION_MAPPING(18);
 798 REGION_MAPPING(19);
 799 REGION_MAPPING(20);
 800 REGION_MAPPING(21);
 801 REGION_MAPPING(22);
 802 REGION_MAPPING(23);
 803 REGION_MAPPING(24);
 804 REGION_MAPPING(25);
 805 REGION_MAPPING(26);
 806 REGION_MAPPING(27);
 807 REGION_MAPPING(28);
 808 REGION_MAPPING(29);
 809 REGION_MAPPING(30);
 810 REGION_MAPPING(31);
 811
 812 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 813 {
 814         struct device *dev = container_of(kobj, struct device, kobj);
 815         struct nd_region *nd_region = to_nd_region(dev);
 816
 817         if (n < nd_region->ndr_mappings)
 818                 return a->mode;
 819         return 0;
 820 }
 821
 822 static struct attribute *mapping_attributes[] = {
 823         &dev_attr_mapping0.attr,
 824         &dev_attr_mapping1.attr,
 825         &dev_attr_mapping2.attr,
 826         &dev_attr_mapping3.attr,
 827         &dev_attr_mapping4.attr,
 828         &dev_attr_mapping5.attr,
 829         &dev_attr_mapping6.attr,
 830         &dev_attr_mapping7.attr,
 831         &dev_attr_mapping8.attr,
 832         &dev_attr_mapping9.attr,
 833         &dev_attr_mapping10.attr,
 834         &dev_attr_mapping11.attr,
 835         &dev_attr_mapping12.attr,
 836         &dev_attr_mapping13.attr,
 837         &dev_attr_mapping14.attr,
 838         &dev_attr_mapping15.attr,
 839         &dev_attr_mapping16.attr,
 840         &dev_attr_mapping17.attr,
 841         &dev_attr_mapping18.attr,
 842         &dev_attr_mapping19.attr,
 843         &dev_attr_mapping20.attr,
 844         &dev_attr_mapping21.attr,
 845         &dev_attr_mapping22.attr,
 846         &dev_attr_mapping23.attr,
 847         &dev_attr_mapping24.attr,
 848         &dev_attr_mapping25.attr,
 849         &dev_attr_mapping26.attr,
 850         &dev_attr_mapping27.attr,
 851         &dev_attr_mapping28.attr,
 852         &dev_attr_mapping29.attr,
 853         &dev_attr_mapping30.attr,
 854         &dev_attr_mapping31.attr,
 855         NULL,
 856 };
 857
 858 struct attribute_group nd_mapping_attribute_group = {
 859         .is_visible = mapping_visible,
 860         .attrs = mapping_attributes,
 861 };
 862 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 863
 864 int nd_blk_region_init(struct nd_region *nd_region)
 865 {
 866         struct device *dev = &nd_region->dev;
 867         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 868
 869         if (!is_nd_blk(dev))
 870                 return 0;
 871
 872         if (nd_region->ndr_mappings < 1) {
 873                 dev_dbg(dev, "invalid BLK region\n");
 874                 return -ENXIO;
 875         }
 876
 877         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 878 }
 879
 880 /**
 881  * nd_region_acquire_lane - allocate and lock a lane
 882  * @nd_region: region id and number of lanes possible
 883  *
 884  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 885  * We optimize for the common case where there are 256 lanes, one
 886  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 887  * this implementation assumes the cost of maintaining an allocator for
 888  * free lanes is on the order of the lock hold time, so it implements a
 889  * static lane = cpu % num_lanes mapping.
 890  *
 891  * In the case of a BTT instance on top of a BLK namespace a lane may be
 892  * acquired recursively.  We lock on the first instance.
 893  *
 894  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 895  * for the BTT metadata updates.
 896  */
 897 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 898 {
 899         unsigned int cpu, lane;
 900
 901         cpu = get_cpu();
 902         if (nd_region->num_lanes < nr_cpu_ids) {
 903                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 904
 905                 lane = cpu % nd_region->num_lanes;
 906                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 907                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 908                 if (ndl_count->count++ == 0)
 909                         spin_lock(&ndl_lock->lock);
 910         } else
 911                 lane = cpu;
 912
 913         return lane;
 914 }
 915 EXPORT_SYMBOL(nd_region_acquire_lane);
 916
 917 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 918 {
 919         if (nd_region->num_lanes < nr_cpu_ids) {
 920                 unsigned int cpu = get_cpu();
 921                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 922
 923                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 924                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 925                 if (--ndl_count->count == 0)
 926                         spin_unlock(&ndl_lock->lock);
 927                 put_cpu();
 928         }
 929         put_cpu();
 930 }
 931 EXPORT_SYMBOL(nd_region_release_lane);
 932
 933 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 934                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 935                 const char *caller)
 936 {
 937         struct nd_region *nd_region;
 938         struct device *dev;
 939         void *region_buf;
 940         unsigned int i;
 941         int ro = 0;
 942
 943         for (i = 0; i < ndr_desc->num_mappings; i++) {
 944                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 945                 struct nvdimm *nvdimm = mapping->nvdimm;
 946
 947                 if ((mapping->start | mapping->size) % PAGE_SIZE) {
 948                         dev_err(&nvdimm_bus->dev,
 949                                 "%s: %s mapping%d is not %ld aligned\n",
 950                                 caller, dev_name(&nvdimm->dev), i, PAGE_SIZE);
 951                         return NULL;
 952                 }
 953
 954                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
 955                         ro = 1;
 956
 957                 if (test_bit(NDD_NOBLK, &nvdimm->flags)
 958                                 && dev_type == &nd_blk_device_type) {
 959                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
 960                                         caller, dev_name(&nvdimm->dev), i);
 961                         return NULL;
 962                 }
 963         }
 964
 965         if (dev_type == &nd_blk_device_type) {
 966                 struct nd_blk_region_desc *ndbr_desc;
 967                 struct nd_blk_region *ndbr;
 968
 969                 ndbr_desc = to_blk_region_desc(ndr_desc);
 970                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 971                                 * ndr_desc->num_mappings,
 972                                 GFP_KERNEL);
 973                 if (ndbr) {
 974                         nd_region = &ndbr->nd_region;
 975                         ndbr->enable = ndbr_desc->enable;
 976                         ndbr->do_io = ndbr_desc->do_io;
 977                 }
 978                 region_buf = ndbr;
 979         } else {
 980                 nd_region = kzalloc(struct_size(nd_region, mapping,
 981                                                 ndr_desc->num_mappings),
 982                                     GFP_KERNEL);
 983                 region_buf = nd_region;
 984         }
 985
 986         if (!region_buf)
 987                 return NULL;
 988         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 989         if (nd_region->id < 0)
 990                 goto err_id;
 991
 992         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 993         if (!nd_region->lane)
 994                 goto err_percpu;
 995
 996         for (i = 0; i < nr_cpu_ids; i++) {
 997                 struct nd_percpu_lane *ndl;
 998
 999                 ndl = per_cpu_ptr(nd_region->lane, i);
1000                 spin_lock_init(&ndl->lock);
1001                 ndl->count = 0;
1002         }
1003
1004         for (i = 0; i < ndr_desc->num_mappings; i++) {
1005                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1006                 struct nvdimm *nvdimm = mapping->nvdimm;
1007
1008                 nd_region->mapping[i].nvdimm = nvdimm;
1009                 nd_region->mapping[i].start = mapping->start;
1010                 nd_region->mapping[i].size = mapping->size;
1011                 nd_region->mapping[i].position = mapping->position;
1012                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1013                 mutex_init(&nd_region->mapping[i].lock);
1014
1015                 get_device(&nvdimm->dev);
1016         }
1017         nd_region->ndr_mappings = ndr_desc->num_mappings;
1018         nd_region->provider_data = ndr_desc->provider_data;
1019         nd_region->nd_set = ndr_desc->nd_set;
1020         nd_region->num_lanes = ndr_desc->num_lanes;
1021         nd_region->flags = ndr_desc->flags;
1022         nd_region->ro = ro;
1023         nd_region->numa_node = ndr_desc->numa_node;
1024         nd_region->target_node = ndr_desc->target_node;
1025         ida_init(&nd_region->ns_ida);
1026         ida_init(&nd_region->btt_ida);
1027         ida_init(&nd_region->pfn_ida);
1028         ida_init(&nd_region->dax_ida);
1029         dev = &nd_region->dev;
1030         dev_set_name(dev, "region%d", nd_region->id);
1031         dev->parent = &nvdimm_bus->dev;
1032         dev->type = dev_type;
1033         dev->groups = ndr_desc->attr_groups;
1034         dev->of_node = ndr_desc->of_node;
1035         nd_region->ndr_size = resource_size(ndr_desc->res);
1036         nd_region->ndr_start = ndr_desc->res->start;
1037         if (ndr_desc->flush)
1038                 nd_region->flush = ndr_desc->flush;
1039         else
1040                 nd_region->flush = NULL;
1041
1042         nd_device_register(dev);
1043
1044         return nd_region;
1045
1046  err_percpu:
1047         ida_simple_remove(&region_ida, nd_region->id);
1048  err_id:
1049         kfree(region_buf);
1050         return NULL;
1051 }
1052
1053 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1054                 struct nd_region_desc *ndr_desc)
1055 {
1056         ndr_desc->num_lanes = ND_MAX_LANES;
1057         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1058                         __func__);
1059 }
1060 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1061
1062 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1063                 struct nd_region_desc *ndr_desc)
1064 {
1065         if (ndr_desc->num_mappings > 1)
1066                 return NULL;
1067         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1068         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1069                         __func__);
1070 }
1071 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1072
1073 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1074                 struct nd_region_desc *ndr_desc)
1075 {
1076         ndr_desc->num_lanes = ND_MAX_LANES;
1077         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1078                         __func__);
1079 }
1080 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1081
1082 int nvdimm_flush(struct nd_region *nd_region, struct bio *bio)
1083 {
1084         int rc = 0;
1085
1086         if (!nd_region->flush)
1087                 rc = generic_nvdimm_flush(nd_region);
1088         else {
1089                 if (nd_region->flush(nd_region, bio))
1090                         rc = -EIO;
1091         }
1092
1093         return rc;
1094 }
1095 /**
1096  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1097  * @nd_region: blk or interleaved pmem region
1098  */
1099 int generic_nvdimm_flush(struct nd_region *nd_region)
1100 {
1101         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1102         int i, idx;
1103
1104         /*
1105          * Try to encourage some diversity in flush hint addresses
1106          * across cpus assuming a limited number of flush hints.
1107          */
1108         idx = this_cpu_read(flush_idx);
1109         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1110
1111         /*
1112          * The first wmb() is needed to 'sfence' all previous writes
1113          * such that they are architecturally visible for the platform
1114          * buffer flush.  Note that we've already arranged for pmem
1115          * writes to avoid the cache via memcpy_flushcache().  The final
1116          * wmb() ensures ordering for the NVDIMM flush write.
1117          */
1118         wmb();
1119         for (i = 0; i < nd_region->ndr_mappings; i++)
1120                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1121                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1122         wmb();
1123
1124         return 0;
1125 }
1126 EXPORT_SYMBOL_GPL(nvdimm_flush);
1127
1128 /**
1129  * nvdimm_has_flush - determine write flushing requirements
1130  * @nd_region: blk or interleaved pmem region
1131  *
1132  * Returns 1 if writes require flushing
1133  * Returns 0 if writes do not require flushing
1134  * Returns -ENXIO if flushing capability can not be determined
1135  */
1136 int nvdimm_has_flush(struct nd_region *nd_region)
1137 {
1138         int i;
1139
1140         /* no nvdimm or pmem api == flushing capability unknown */
1141         if (nd_region->ndr_mappings == 0
1142                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1143                 return -ENXIO;
1144
1145         for (i = 0; i < nd_region->ndr_mappings; i++) {
1146                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1147                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1148
1149                 /* flush hints present / available */
1150                 if (nvdimm->num_flush)
1151                         return 1;
1152         }
1153
1154         /*
1155          * The platform defines dimm devices without hints, assume
1156          * platform persistence mechanism like ADR
1157          */
1158         return 0;
1159 }
1160 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1161
1162 int nvdimm_has_cache(struct nd_region *nd_region)
1163 {
1164         return is_nd_pmem(&nd_region->dev) &&
1165                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1166 }
1167 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1168
1169 bool is_nvdimm_sync(struct nd_region *nd_region)
1170 {
1171         if (is_nd_volatile(&nd_region->dev))
1172                 return true;
1173
1174         return is_nd_pmem(&nd_region->dev) &&
1175                 !test_bit(ND_REGION_ASYNC, &nd_region->flags);
1176 }
1177 EXPORT_SYMBOL_GPL(is_nvdimm_sync);
1178
1179 struct conflict_context {
1180         struct nd_region *nd_region;
1181         resource_size_t start, size;
1182 };
1183
1184 static int region_conflict(struct device *dev, void *data)
1185 {
1186         struct nd_region *nd_region;
1187         struct conflict_context *ctx = data;
1188         resource_size_t res_end, region_end, region_start;
1189
1190         if (!is_memory(dev))
1191                 return 0;
1192
1193         nd_region = to_nd_region(dev);
1194         if (nd_region == ctx->nd_region)
1195                 return 0;
1196
1197         res_end = ctx->start + ctx->size;
1198         region_start = nd_region->ndr_start;
1199         region_end = region_start + nd_region->ndr_size;
1200         if (ctx->start >= region_start && ctx->start < region_end)
1201                 return -EBUSY;
1202         if (res_end > region_start && res_end <= region_end)
1203                 return -EBUSY;
1204         return 0;
1205 }
1206
1207 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1208                 resource_size_t size)
1209 {
1210         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1211         struct conflict_context ctx = {
1212                 .nd_region = nd_region,
1213                 .start = start,
1214                 .size = size,
1215         };
1216
1217         return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1218 }
1219
1220 void __exit nd_region_devs_exit(void)
1221 {
1222         ida_destroy(&region_ida);
1223 }