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         nvdimm_flush(nd_region);
 291
 292         return len;
 293 }
 294 static DEVICE_ATTR_RW(deep_flush);
 295
 296 static ssize_t mappings_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->ndr_mappings);
 302 }
 303 static DEVICE_ATTR_RO(mappings);
 304
 305 static ssize_t nstype_show(struct device *dev,
 306                 struct device_attribute *attr, char *buf)
 307 {
 308         struct nd_region *nd_region = to_nd_region(dev);
 309
 310         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 311 }
 312 static DEVICE_ATTR_RO(nstype);
 313
 314 static ssize_t set_cookie_show(struct device *dev,
 315                 struct device_attribute *attr, char *buf)
 316 {
 317         struct nd_region *nd_region = to_nd_region(dev);
 318         struct nd_interleave_set *nd_set = nd_region->nd_set;
 319         ssize_t rc = 0;
 320
 321         if (is_memory(dev) && nd_set)
 322                 /* pass, should be precluded by region_visible */;
 323         else
 324                 return -ENXIO;
 325
 326         /*
 327          * The cookie to show depends on which specification of the
 328          * labels we are using. If there are not labels then default to
 329          * the v1.1 namespace label cookie definition. To read all this
 330          * data we need to wait for probing to settle.
 331          */
 332         device_lock(dev);
 333         nvdimm_bus_lock(dev);
 334         wait_nvdimm_bus_probe_idle(dev);
 335         if (nd_region->ndr_mappings) {
 336                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 337                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 338
 339                 if (ndd) {
 340                         struct nd_namespace_index *nsindex;
 341
 342                         nsindex = to_namespace_index(ndd, ndd->ns_current);
 343                         rc = sprintf(buf, "%#llx\n",
 344                                         nd_region_interleave_set_cookie(nd_region,
 345                                                 nsindex));
 346                 }
 347         }
 348         nvdimm_bus_unlock(dev);
 349         device_unlock(dev);
 350
 351         if (rc)
 352                 return rc;
 353         return sprintf(buf, "%#llx\n", nd_set->cookie1);
 354 }
 355 static DEVICE_ATTR_RO(set_cookie);
 356
 357 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 358 {
 359         resource_size_t blk_max_overlap = 0, available, overlap;
 360         int i;
 361
 362         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 363
 364  retry:
 365         available = 0;
 366         overlap = blk_max_overlap;
 367         for (i = 0; i < nd_region->ndr_mappings; i++) {
 368                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 369                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 370
 371                 /* if a dimm is disabled the available capacity is zero */
 372                 if (!ndd)
 373                         return 0;
 374
 375                 if (is_memory(&nd_region->dev)) {
 376                         available += nd_pmem_available_dpa(nd_region,
 377                                         nd_mapping, &overlap);
 378                         if (overlap > blk_max_overlap) {
 379                                 blk_max_overlap = overlap;
 380                                 goto retry;
 381                         }
 382                 } else if (is_nd_blk(&nd_region->dev))
 383                         available += nd_blk_available_dpa(nd_region);
 384         }
 385
 386         return available;
 387 }
 388
 389 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
 390 {
 391         resource_size_t available = 0;
 392         int i;
 393
 394         if (is_memory(&nd_region->dev))
 395                 available = PHYS_ADDR_MAX;
 396
 397         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 398         for (i = 0; i < nd_region->ndr_mappings; i++) {
 399                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 400
 401                 if (is_memory(&nd_region->dev))
 402                         available = min(available,
 403                                         nd_pmem_max_contiguous_dpa(nd_region,
 404                                                                    nd_mapping));
 405                 else if (is_nd_blk(&nd_region->dev))
 406                         available += nd_blk_available_dpa(nd_region);
 407         }
 408         if (is_memory(&nd_region->dev))
 409                 return available * nd_region->ndr_mappings;
 410         return available;
 411 }
 412
 413 static ssize_t available_size_show(struct device *dev,
 414                 struct device_attribute *attr, char *buf)
 415 {
 416         struct nd_region *nd_region = to_nd_region(dev);
 417         unsigned long long available = 0;
 418
 419         /*
 420          * Flush in-flight updates and grab a snapshot of the available
 421          * size.  Of course, this value is potentially invalidated the
 422          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 423          * problem to not race itself.
 424          */
 425         nvdimm_bus_lock(dev);
 426         wait_nvdimm_bus_probe_idle(dev);
 427         available = nd_region_available_dpa(nd_region);
 428         nvdimm_bus_unlock(dev);
 429
 430         return sprintf(buf, "%llu\n", available);
 431 }
 432 static DEVICE_ATTR_RO(available_size);
 433
 434 static ssize_t max_available_extent_show(struct device *dev,
 435                 struct device_attribute *attr, char *buf)
 436 {
 437         struct nd_region *nd_region = to_nd_region(dev);
 438         unsigned long long available = 0;
 439
 440         nvdimm_bus_lock(dev);
 441         wait_nvdimm_bus_probe_idle(dev);
 442         available = nd_region_allocatable_dpa(nd_region);
 443         nvdimm_bus_unlock(dev);
 444
 445         return sprintf(buf, "%llu\n", available);
 446 }
 447 static DEVICE_ATTR_RO(max_available_extent);
 448
 449 static ssize_t init_namespaces_show(struct device *dev,
 450                 struct device_attribute *attr, char *buf)
 451 {
 452         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 453         ssize_t rc;
 454
 455         nvdimm_bus_lock(dev);
 456         if (ndrd)
 457                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 458         else
 459                 rc = -ENXIO;
 460         nvdimm_bus_unlock(dev);
 461
 462         return rc;
 463 }
 464 static DEVICE_ATTR_RO(init_namespaces);
 465
 466 static ssize_t namespace_seed_show(struct device *dev,
 467                 struct device_attribute *attr, char *buf)
 468 {
 469         struct nd_region *nd_region = to_nd_region(dev);
 470         ssize_t rc;
 471
 472         nvdimm_bus_lock(dev);
 473         if (nd_region->ns_seed)
 474                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 475         else
 476                 rc = sprintf(buf, "\n");
 477         nvdimm_bus_unlock(dev);
 478         return rc;
 479 }
 480 static DEVICE_ATTR_RO(namespace_seed);
 481
 482 static ssize_t btt_seed_show(struct device *dev,
 483                 struct device_attribute *attr, char *buf)
 484 {
 485         struct nd_region *nd_region = to_nd_region(dev);
 486         ssize_t rc;
 487
 488         nvdimm_bus_lock(dev);
 489         if (nd_region->btt_seed)
 490                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 491         else
 492                 rc = sprintf(buf, "\n");
 493         nvdimm_bus_unlock(dev);
 494
 495         return rc;
 496 }
 497 static DEVICE_ATTR_RO(btt_seed);
 498
 499 static ssize_t pfn_seed_show(struct device *dev,
 500                 struct device_attribute *attr, char *buf)
 501 {
 502         struct nd_region *nd_region = to_nd_region(dev);
 503         ssize_t rc;
 504
 505         nvdimm_bus_lock(dev);
 506         if (nd_region->pfn_seed)
 507                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 508         else
 509                 rc = sprintf(buf, "\n");
 510         nvdimm_bus_unlock(dev);
 511
 512         return rc;
 513 }
 514 static DEVICE_ATTR_RO(pfn_seed);
 515
 516 static ssize_t dax_seed_show(struct device *dev,
 517                 struct device_attribute *attr, char *buf)
 518 {
 519         struct nd_region *nd_region = to_nd_region(dev);
 520         ssize_t rc;
 521
 522         nvdimm_bus_lock(dev);
 523         if (nd_region->dax_seed)
 524                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 525         else
 526                 rc = sprintf(buf, "\n");
 527         nvdimm_bus_unlock(dev);
 528
 529         return rc;
 530 }
 531 static DEVICE_ATTR_RO(dax_seed);
 532
 533 static ssize_t read_only_show(struct device *dev,
 534                 struct device_attribute *attr, char *buf)
 535 {
 536         struct nd_region *nd_region = to_nd_region(dev);
 537
 538         return sprintf(buf, "%d\n", nd_region->ro);
 539 }
 540
 541 static ssize_t read_only_store(struct device *dev,
 542                 struct device_attribute *attr, const char *buf, size_t len)
 543 {
 544         bool ro;
 545         int rc = strtobool(buf, &ro);
 546         struct nd_region *nd_region = to_nd_region(dev);
 547
 548         if (rc)
 549                 return rc;
 550
 551         nd_region->ro = ro;
 552         return len;
 553 }
 554 static DEVICE_ATTR_RW(read_only);
 555
 556 static ssize_t region_badblocks_show(struct device *dev,
 557                 struct device_attribute *attr, char *buf)
 558 {
 559         struct nd_region *nd_region = to_nd_region(dev);
 560         ssize_t rc;
 561
 562         device_lock(dev);
 563         if (dev->driver)
 564                 rc = badblocks_show(&nd_region->bb, buf, 0);
 565         else
 566                 rc = -ENXIO;
 567         device_unlock(dev);
 568
 569         return rc;
 570 }
 571 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 572
 573 static ssize_t resource_show(struct device *dev,
 574                 struct device_attribute *attr, char *buf)
 575 {
 576         struct nd_region *nd_region = to_nd_region(dev);
 577
 578         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 579 }
 580 static DEVICE_ATTR_RO(resource);
 581
 582 static ssize_t persistence_domain_show(struct device *dev,
 583                 struct device_attribute *attr, char *buf)
 584 {
 585         struct nd_region *nd_region = to_nd_region(dev);
 586
 587         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
 588                 return sprintf(buf, "cpu_cache\n");
 589         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
 590                 return sprintf(buf, "memory_controller\n");
 591         else
 592                 return sprintf(buf, "\n");
 593 }
 594 static DEVICE_ATTR_RO(persistence_domain);
 595
 596 static struct attribute *nd_region_attributes[] = {
 597         &dev_attr_size.attr,
 598         &dev_attr_nstype.attr,
 599         &dev_attr_mappings.attr,
 600         &dev_attr_btt_seed.attr,
 601         &dev_attr_pfn_seed.attr,
 602         &dev_attr_dax_seed.attr,
 603         &dev_attr_deep_flush.attr,
 604         &dev_attr_read_only.attr,
 605         &dev_attr_set_cookie.attr,
 606         &dev_attr_available_size.attr,
 607         &dev_attr_max_available_extent.attr,
 608         &dev_attr_namespace_seed.attr,
 609         &dev_attr_init_namespaces.attr,
 610         &dev_attr_badblocks.attr,
 611         &dev_attr_resource.attr,
 612         &dev_attr_persistence_domain.attr,
 613         NULL,
 614 };
 615
 616 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 617 {
 618         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 619         struct nd_region *nd_region = to_nd_region(dev);
 620         struct nd_interleave_set *nd_set = nd_region->nd_set;
 621         int type = nd_region_to_nstype(nd_region);
 622
 623         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
 624                 return 0;
 625
 626         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
 627                 return 0;
 628
 629         if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
 630                 return 0;
 631
 632         if (a == &dev_attr_resource.attr) {
 633                 if (is_nd_pmem(dev))
 634                         return 0400;
 635                 else
 636                         return 0;
 637         }
 638
 639         if (a == &dev_attr_deep_flush.attr) {
 640                 int has_flush = nvdimm_has_flush(nd_region);
 641
 642                 if (has_flush == 1)
 643                         return a->mode;
 644                 else if (has_flush == 0)
 645                         return 0444;
 646                 else
 647                         return 0;
 648         }
 649
 650         if (a == &dev_attr_persistence_domain.attr) {
 651                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
 652                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
 653                         return 0;
 654                 return a->mode;
 655         }
 656
 657         if (a != &dev_attr_set_cookie.attr
 658                         && a != &dev_attr_available_size.attr)
 659                 return a->mode;
 660
 661         if ((type == ND_DEVICE_NAMESPACE_PMEM
 662                                 || type == ND_DEVICE_NAMESPACE_BLK)
 663                         && a == &dev_attr_available_size.attr)
 664                 return a->mode;
 665         else if (is_memory(dev) && nd_set)
 666                 return a->mode;
 667
 668         return 0;
 669 }
 670
 671 struct attribute_group nd_region_attribute_group = {
 672         .attrs = nd_region_attributes,
 673         .is_visible = region_visible,
 674 };
 675 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 676
 677 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 678                 struct nd_namespace_index *nsindex)
 679 {
 680         struct nd_interleave_set *nd_set = nd_region->nd_set;
 681
 682         if (!nd_set)
 683                 return 0;
 684
 685         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 686                         && __le16_to_cpu(nsindex->minor) == 1)
 687                 return nd_set->cookie1;
 688         return nd_set->cookie2;
 689 }
 690
 691 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 692 {
 693         struct nd_interleave_set *nd_set = nd_region->nd_set;
 694
 695         if (nd_set)
 696                 return nd_set->altcookie;
 697         return 0;
 698 }
 699
 700 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 701 {
 702         struct nd_label_ent *label_ent, *e;
 703
 704         lockdep_assert_held(&nd_mapping->lock);
 705         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 706                 list_del(&label_ent->list);
 707                 kfree(label_ent);
 708         }
 709 }
 710
 711 /*
 712  * Upon successful probe/remove, take/release a reference on the
 713  * associated interleave set (if present), and plant new btt + namespace
 714  * seeds.  Also, on the removal of a BLK region, notify the provider to
 715  * disable the region.
 716  */
 717 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 718                 struct device *dev, bool probe)
 719 {
 720         struct nd_region *nd_region;
 721
 722         if (!probe && is_nd_region(dev)) {
 723                 int i;
 724
 725                 nd_region = to_nd_region(dev);
 726                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 727                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 728                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 729                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 730
 731                         mutex_lock(&nd_mapping->lock);
 732                         nd_mapping_free_labels(nd_mapping);
 733                         mutex_unlock(&nd_mapping->lock);
 734
 735                         put_ndd(ndd);
 736                         nd_mapping->ndd = NULL;
 737                         if (ndd)
 738                                 atomic_dec(&nvdimm->busy);
 739                 }
 740         }
 741         if (dev->parent && is_nd_region(dev->parent) && probe) {
 742                 nd_region = to_nd_region(dev->parent);
 743                 nvdimm_bus_lock(dev);
 744                 if (nd_region->ns_seed == dev)
 745                         nd_region_create_ns_seed(nd_region);
 746                 nvdimm_bus_unlock(dev);
 747         }
 748         if (is_nd_btt(dev) && probe) {
 749                 struct nd_btt *nd_btt = to_nd_btt(dev);
 750
 751                 nd_region = to_nd_region(dev->parent);
 752                 nvdimm_bus_lock(dev);
 753                 if (nd_region->btt_seed == dev)
 754                         nd_region_create_btt_seed(nd_region);
 755                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 756                         nd_region_create_ns_seed(nd_region);
 757                 nvdimm_bus_unlock(dev);
 758         }
 759         if (is_nd_pfn(dev) && probe) {
 760                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 761
 762                 nd_region = to_nd_region(dev->parent);
 763                 nvdimm_bus_lock(dev);
 764                 if (nd_region->pfn_seed == dev)
 765                         nd_region_create_pfn_seed(nd_region);
 766                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 767                         nd_region_create_ns_seed(nd_region);
 768                 nvdimm_bus_unlock(dev);
 769         }
 770         if (is_nd_dax(dev) && probe) {
 771                 struct nd_dax *nd_dax = to_nd_dax(dev);
 772
 773                 nd_region = to_nd_region(dev->parent);
 774                 nvdimm_bus_lock(dev);
 775                 if (nd_region->dax_seed == dev)
 776                         nd_region_create_dax_seed(nd_region);
 777                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 778                         nd_region_create_ns_seed(nd_region);
 779                 nvdimm_bus_unlock(dev);
 780         }
 781 }
 782
 783 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 784 {
 785         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 786 }
 787
 788 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 789 {
 790         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 791 }
 792
 793 static ssize_t mappingN(struct device *dev, char *buf, int n)
 794 {
 795         struct nd_region *nd_region = to_nd_region(dev);
 796         struct nd_mapping *nd_mapping;
 797         struct nvdimm *nvdimm;
 798
 799         if (n >= nd_region->ndr_mappings)
 800                 return -ENXIO;
 801         nd_mapping = &nd_region->mapping[n];
 802         nvdimm = nd_mapping->nvdimm;
 803
 804         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
 805                         nd_mapping->start, nd_mapping->size,
 806                         nd_mapping->position);
 807 }
 808
 809 #define REGION_MAPPING(idx) \
 810 static ssize_t mapping##idx##_show(struct device *dev,          \
 811                 struct device_attribute *attr, char *buf)       \
 812 {                                                               \
 813         return mappingN(dev, buf, idx);                         \
 814 }                                                               \
 815 static DEVICE_ATTR_RO(mapping##idx)
 816
 817 /*
 818  * 32 should be enough for a while, even in the presence of socket
 819  * interleave a 32-way interleave set is a degenerate case.
 820  */
 821 REGION_MAPPING(0);
 822 REGION_MAPPING(1);
 823 REGION_MAPPING(2);
 824 REGION_MAPPING(3);
 825 REGION_MAPPING(4);
 826 REGION_MAPPING(5);
 827 REGION_MAPPING(6);
 828 REGION_MAPPING(7);
 829 REGION_MAPPING(8);
 830 REGION_MAPPING(9);
 831 REGION_MAPPING(10);
 832 REGION_MAPPING(11);
 833 REGION_MAPPING(12);
 834 REGION_MAPPING(13);
 835 REGION_MAPPING(14);
 836 REGION_MAPPING(15);
 837 REGION_MAPPING(16);
 838 REGION_MAPPING(17);
 839 REGION_MAPPING(18);
 840 REGION_MAPPING(19);
 841 REGION_MAPPING(20);
 842 REGION_MAPPING(21);
 843 REGION_MAPPING(22);
 844 REGION_MAPPING(23);
 845 REGION_MAPPING(24);
 846 REGION_MAPPING(25);
 847 REGION_MAPPING(26);
 848 REGION_MAPPING(27);
 849 REGION_MAPPING(28);
 850 REGION_MAPPING(29);
 851 REGION_MAPPING(30);
 852 REGION_MAPPING(31);
 853
 854 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 855 {
 856         struct device *dev = container_of(kobj, struct device, kobj);
 857         struct nd_region *nd_region = to_nd_region(dev);
 858
 859         if (n < nd_region->ndr_mappings)
 860                 return a->mode;
 861         return 0;
 862 }
 863
 864 static struct attribute *mapping_attributes[] = {
 865         &dev_attr_mapping0.attr,
 866         &dev_attr_mapping1.attr,
 867         &dev_attr_mapping2.attr,
 868         &dev_attr_mapping3.attr,
 869         &dev_attr_mapping4.attr,
 870         &dev_attr_mapping5.attr,
 871         &dev_attr_mapping6.attr,
 872         &dev_attr_mapping7.attr,
 873         &dev_attr_mapping8.attr,
 874         &dev_attr_mapping9.attr,
 875         &dev_attr_mapping10.attr,
 876         &dev_attr_mapping11.attr,
 877         &dev_attr_mapping12.attr,
 878         &dev_attr_mapping13.attr,
 879         &dev_attr_mapping14.attr,
 880         &dev_attr_mapping15.attr,
 881         &dev_attr_mapping16.attr,
 882         &dev_attr_mapping17.attr,
 883         &dev_attr_mapping18.attr,
 884         &dev_attr_mapping19.attr,
 885         &dev_attr_mapping20.attr,
 886         &dev_attr_mapping21.attr,
 887         &dev_attr_mapping22.attr,
 888         &dev_attr_mapping23.attr,
 889         &dev_attr_mapping24.attr,
 890         &dev_attr_mapping25.attr,
 891         &dev_attr_mapping26.attr,
 892         &dev_attr_mapping27.attr,
 893         &dev_attr_mapping28.attr,
 894         &dev_attr_mapping29.attr,
 895         &dev_attr_mapping30.attr,
 896         &dev_attr_mapping31.attr,
 897         NULL,
 898 };
 899
 900 struct attribute_group nd_mapping_attribute_group = {
 901         .is_visible = mapping_visible,
 902         .attrs = mapping_attributes,
 903 };
 904 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 905
 906 int nd_blk_region_init(struct nd_region *nd_region)
 907 {
 908         struct device *dev = &nd_region->dev;
 909         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 910
 911         if (!is_nd_blk(dev))
 912                 return 0;
 913
 914         if (nd_region->ndr_mappings < 1) {
 915                 dev_dbg(dev, "invalid BLK region\n");
 916                 return -ENXIO;
 917         }
 918
 919         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 920 }
 921
 922 /**
 923  * nd_region_acquire_lane - allocate and lock a lane
 924  * @nd_region: region id and number of lanes possible
 925  *
 926  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 927  * We optimize for the common case where there are 256 lanes, one
 928  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 929  * this implementation assumes the cost of maintaining an allocator for
 930  * free lanes is on the order of the lock hold time, so it implements a
 931  * static lane = cpu % num_lanes mapping.
 932  *
 933  * In the case of a BTT instance on top of a BLK namespace a lane may be
 934  * acquired recursively.  We lock on the first instance.
 935  *
 936  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 937  * for the BTT metadata updates.
 938  */
 939 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 940 {
 941         unsigned int cpu, lane;
 942
 943         cpu = get_cpu();
 944         if (nd_region->num_lanes < nr_cpu_ids) {
 945                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 946
 947                 lane = cpu % nd_region->num_lanes;
 948                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 949                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 950                 if (ndl_count->count++ == 0)
 951                         spin_lock(&ndl_lock->lock);
 952         } else
 953                 lane = cpu;
 954
 955         return lane;
 956 }
 957 EXPORT_SYMBOL(nd_region_acquire_lane);
 958
 959 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 960 {
 961         if (nd_region->num_lanes < nr_cpu_ids) {
 962                 unsigned int cpu = get_cpu();
 963                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 964
 965                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 966                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 967                 if (--ndl_count->count == 0)
 968                         spin_unlock(&ndl_lock->lock);
 969                 put_cpu();
 970         }
 971         put_cpu();
 972 }
 973 EXPORT_SYMBOL(nd_region_release_lane);
 974
 975 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 976                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 977                 const char *caller)
 978 {
 979         struct nd_region *nd_region;
 980         struct device *dev;
 981         void *region_buf;
 982         unsigned int i;
 983         int ro = 0;
 984
 985         for (i = 0; i < ndr_desc->num_mappings; i++) {
 986                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 987                 struct nvdimm *nvdimm = mapping->nvdimm;
 988
 989                 if ((mapping->start | mapping->size) % SZ_4K) {
 990                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 991                                         caller, dev_name(&nvdimm->dev), i);
 992
 993                         return NULL;
 994                 }
 995
 996                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
 997                         ro = 1;
 998
 999                 if (test_bit(NDD_NOBLK, &nvdimm->flags)
1000                                 && dev_type == &nd_blk_device_type) {
1001                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not BLK capable\n",
1002                                         caller, dev_name(&nvdimm->dev), i);
1003                         return NULL;
1004                 }
1005         }
1006
1007         if (dev_type == &nd_blk_device_type) {
1008                 struct nd_blk_region_desc *ndbr_desc;
1009                 struct nd_blk_region *ndbr;
1010
1011                 ndbr_desc = to_blk_region_desc(ndr_desc);
1012                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1013                                 * ndr_desc->num_mappings,
1014                                 GFP_KERNEL);
1015                 if (ndbr) {
1016                         nd_region = &ndbr->nd_region;
1017                         ndbr->enable = ndbr_desc->enable;
1018                         ndbr->do_io = ndbr_desc->do_io;
1019                 }
1020                 region_buf = ndbr;
1021         } else {
1022                 nd_region = kzalloc(sizeof(struct nd_region)
1023                                 + sizeof(struct nd_mapping)
1024                                 * ndr_desc->num_mappings,
1025                                 GFP_KERNEL);
1026                 region_buf = nd_region;
1027         }
1028
1029         if (!region_buf)
1030                 return NULL;
1031         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
1032         if (nd_region->id < 0)
1033                 goto err_id;
1034
1035         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1036         if (!nd_region->lane)
1037                 goto err_percpu;
1038
1039         for (i = 0; i < nr_cpu_ids; i++) {
1040                 struct nd_percpu_lane *ndl;
1041
1042                 ndl = per_cpu_ptr(nd_region->lane, i);
1043                 spin_lock_init(&ndl->lock);
1044                 ndl->count = 0;
1045         }
1046
1047         for (i = 0; i < ndr_desc->num_mappings; i++) {
1048                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1049                 struct nvdimm *nvdimm = mapping->nvdimm;
1050
1051                 nd_region->mapping[i].nvdimm = nvdimm;
1052                 nd_region->mapping[i].start = mapping->start;
1053                 nd_region->mapping[i].size = mapping->size;
1054                 nd_region->mapping[i].position = mapping->position;
1055                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1056                 mutex_init(&nd_region->mapping[i].lock);
1057
1058                 get_device(&nvdimm->dev);
1059         }
1060         nd_region->ndr_mappings = ndr_desc->num_mappings;
1061         nd_region->provider_data = ndr_desc->provider_data;
1062         nd_region->nd_set = ndr_desc->nd_set;
1063         nd_region->num_lanes = ndr_desc->num_lanes;
1064         nd_region->flags = ndr_desc->flags;
1065         nd_region->ro = ro;
1066         nd_region->numa_node = ndr_desc->numa_node;
1067         nd_region->target_node = ndr_desc->target_node;
1068         ida_init(&nd_region->ns_ida);
1069         ida_init(&nd_region->btt_ida);
1070         ida_init(&nd_region->pfn_ida);
1071         ida_init(&nd_region->dax_ida);
1072         dev = &nd_region->dev;
1073         dev_set_name(dev, "region%d", nd_region->id);
1074         dev->parent = &nvdimm_bus->dev;
1075         dev->type = dev_type;
1076         dev->groups = ndr_desc->attr_groups;
1077         dev->of_node = ndr_desc->of_node;
1078         nd_region->ndr_size = resource_size(ndr_desc->res);
1079         nd_region->ndr_start = ndr_desc->res->start;
1080         nd_device_register(dev);
1081
1082         return nd_region;
1083
1084  err_percpu:
1085         ida_simple_remove(&region_ida, nd_region->id);
1086  err_id:
1087         kfree(region_buf);
1088         return NULL;
1089 }
1090
1091 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1092                 struct nd_region_desc *ndr_desc)
1093 {
1094         ndr_desc->num_lanes = ND_MAX_LANES;
1095         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1096                         __func__);
1097 }
1098 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1099
1100 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1101                 struct nd_region_desc *ndr_desc)
1102 {
1103         if (ndr_desc->num_mappings > 1)
1104                 return NULL;
1105         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1106         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1107                         __func__);
1108 }
1109 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1110
1111 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1112                 struct nd_region_desc *ndr_desc)
1113 {
1114         ndr_desc->num_lanes = ND_MAX_LANES;
1115         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1116                         __func__);
1117 }
1118 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1119
1120 /**
1121  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1122  * @nd_region: blk or interleaved pmem region
1123  */
1124 void nvdimm_flush(struct nd_region *nd_region)
1125 {
1126         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1127         int i, idx;
1128
1129         /*
1130          * Try to encourage some diversity in flush hint addresses
1131          * across cpus assuming a limited number of flush hints.
1132          */
1133         idx = this_cpu_read(flush_idx);
1134         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1135
1136         /*
1137          * The first wmb() is needed to 'sfence' all previous writes
1138          * such that they are architecturally visible for the platform
1139          * buffer flush.  Note that we've already arranged for pmem
1140          * writes to avoid the cache via memcpy_flushcache().  The final
1141          * wmb() ensures ordering for the NVDIMM flush write.
1142          */
1143         wmb();
1144         for (i = 0; i < nd_region->ndr_mappings; i++)
1145                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1146                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1147         wmb();
1148 }
1149 EXPORT_SYMBOL_GPL(nvdimm_flush);
1150
1151 /**
1152  * nvdimm_has_flush - determine write flushing requirements
1153  * @nd_region: blk or interleaved pmem region
1154  *
1155  * Returns 1 if writes require flushing
1156  * Returns 0 if writes do not require flushing
1157  * Returns -ENXIO if flushing capability can not be determined
1158  */
1159 int nvdimm_has_flush(struct nd_region *nd_region)
1160 {
1161         int i;
1162
1163         /* no nvdimm or pmem api == flushing capability unknown */
1164         if (nd_region->ndr_mappings == 0
1165                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1166                 return -ENXIO;
1167
1168         for (i = 0; i < nd_region->ndr_mappings; i++) {
1169                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1170                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1171
1172                 /* flush hints present / available */
1173                 if (nvdimm->num_flush)
1174                         return 1;
1175         }
1176
1177         /*
1178          * The platform defines dimm devices without hints, assume
1179          * platform persistence mechanism like ADR
1180          */
1181         return 0;
1182 }
1183 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1184
1185 int nvdimm_has_cache(struct nd_region *nd_region)
1186 {
1187         return is_nd_pmem(&nd_region->dev) &&
1188                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1189 }
1190 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1191
1192 struct conflict_context {
1193         struct nd_region *nd_region;
1194         resource_size_t start, size;
1195 };
1196
1197 static int region_conflict(struct device *dev, void *data)
1198 {
1199         struct nd_region *nd_region;
1200         struct conflict_context *ctx = data;
1201         resource_size_t res_end, region_end, region_start;
1202
1203         if (!is_memory(dev))
1204                 return 0;
1205
1206         nd_region = to_nd_region(dev);
1207         if (nd_region == ctx->nd_region)
1208                 return 0;
1209
1210         res_end = ctx->start + ctx->size;
1211         region_start = nd_region->ndr_start;
1212         region_end = region_start + nd_region->ndr_size;
1213         if (ctx->start >= region_start && ctx->start < region_end)
1214                 return -EBUSY;
1215         if (res_end > region_start && res_end <= region_end)
1216                 return -EBUSY;
1217         return 0;
1218 }
1219
1220 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1221                 resource_size_t size)
1222 {
1223         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1224         struct conflict_context ctx = {
1225                 .nd_region = nd_region,
1226                 .start = start,
1227                 .size = size,
1228         };
1229
1230         return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1231 }
1232
1233 void __exit nd_region_devs_exit(void)
1234 {
1235         ida_destroy(&region_ida);
1236 }