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