drivers/nvdimm/region_devs.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/scatterlist.h>
  14 #include <linux/highmem.h>
  15 #include <linux/sched.h>
  16 #include <linux/slab.h>
  17 #include <linux/hash.h>
  18 #include <linux/pmem.h>
  19 #include <linux/sort.h>
  20 #include <linux/io.h>
  21 #include <linux/nd.h>
  22 #include "nd-core.h"
  23 #include "nd.h"
  24
  25 /*
  26  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  27  * irrelevant.
  28  */
  29 #include <linux/io-64-nonatomic-hi-lo.h>
  30
  31 static DEFINE_IDA(region_ida);
  32 static DEFINE_PER_CPU(int, flush_idx);
  33
  34 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
  35                 struct nd_region_data *ndrd)
  36 {
  37         int i, j;
  38
  39         dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
  40                         nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
  41         for (i = 0; i < (1 << ndrd->hints_shift); i++) {
  42                 struct resource *res = &nvdimm->flush_wpq[i];
  43                 unsigned long pfn = PHYS_PFN(res->start);
  44                 void __iomem *flush_page;
  45
  46                 /* check if flush hints share a page */
  47                 for (j = 0; j < i; j++) {
  48                         struct resource *res_j = &nvdimm->flush_wpq[j];
  49                         unsigned long pfn_j = PHYS_PFN(res_j->start);
  50
  51                         if (pfn == pfn_j)
  52                                 break;
  53                 }
  54
  55                 if (j < i)
  56                         flush_page = (void __iomem *) ((unsigned long)
  57                                         ndrd_get_flush_wpq(ndrd, dimm, j)
  58                                         & PAGE_MASK);
  59                 else
  60                         flush_page = devm_nvdimm_ioremap(dev,
  61                                         PFN_PHYS(pfn), PAGE_SIZE);
  62                 if (!flush_page)
  63                         return -ENXIO;
  64                 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
  65                                 + (res->start & ~PAGE_MASK));
  66         }
  67
  68         return 0;
  69 }
  70
  71 int nd_region_activate(struct nd_region *nd_region)
  72 {
  73         int i, j, num_flush = 0;
  74         struct nd_region_data *ndrd;
  75         struct device *dev = &nd_region->dev;
  76         size_t flush_data_size = sizeof(void *);
  77
  78         nvdimm_bus_lock(&nd_region->dev);
  79         for (i = 0; i < nd_region->ndr_mappings; i++) {
  80                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
  81                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
  82
  83                 /* at least one null hint slot per-dimm for the "no-hint" case */
  84                 flush_data_size += sizeof(void *);
  85                 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
  86                 if (!nvdimm->num_flush)
  87                         continue;
  88                 flush_data_size += nvdimm->num_flush * sizeof(void *);
  89         }
  90         nvdimm_bus_unlock(&nd_region->dev);
  91
  92         ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
  93         if (!ndrd)
  94                 return -ENOMEM;
  95         dev_set_drvdata(dev, ndrd);
  96
  97         if (!num_flush)
  98                 return 0;
  99
 100         ndrd->hints_shift = ilog2(num_flush);
 101         for (i = 0; i < nd_region->ndr_mappings; i++) {
 102                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 103                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 104                 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
 105
 106                 if (rc)
 107                         return rc;
 108         }
 109
 110         /*
 111          * Clear out entries that are duplicates. This should prevent the
 112          * extra flushings.
 113          */
 114         for (i = 0; i < nd_region->ndr_mappings - 1; i++) {
 115                 /* ignore if NULL already */
 116                 if (!ndrd_get_flush_wpq(ndrd, i, 0))
 117                         continue;
 118
 119                 for (j = i + 1; j < nd_region->ndr_mappings; j++)
 120                         if (ndrd_get_flush_wpq(ndrd, i, 0) ==
 121                             ndrd_get_flush_wpq(ndrd, j, 0))
 122                                 ndrd_set_flush_wpq(ndrd, j, 0, NULL);
 123         }
 124
 125         return 0;
 126 }
 127
 128 static void nd_region_release(struct device *dev)
 129 {
 130         struct nd_region *nd_region = to_nd_region(dev);
 131         u16 i;
 132
 133         for (i = 0; i < nd_region->ndr_mappings; i++) {
 134                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 135                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 136
 137                 put_device(&nvdimm->dev);
 138         }
 139         free_percpu(nd_region->lane);
 140         ida_simple_remove(&region_ida, nd_region->id);
 141         if (is_nd_blk(dev))
 142                 kfree(to_nd_blk_region(dev));
 143         else
 144                 kfree(nd_region);
 145 }
 146
 147 static struct device_type nd_blk_device_type = {
 148         .name = "nd_blk",
 149         .release = nd_region_release,
 150 };
 151
 152 static struct device_type nd_pmem_device_type = {
 153         .name = "nd_pmem",
 154         .release = nd_region_release,
 155 };
 156
 157 static struct device_type nd_volatile_device_type = {
 158         .name = "nd_volatile",
 159         .release = nd_region_release,
 160 };
 161
 162 bool is_nd_pmem(struct device *dev)
 163 {
 164         return dev ? dev->type == &nd_pmem_device_type : false;
 165 }
 166
 167 bool is_nd_blk(struct device *dev)
 168 {
 169         return dev ? dev->type == &nd_blk_device_type : false;
 170 }
 171
 172 struct nd_region *to_nd_region(struct device *dev)
 173 {
 174         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
 175
 176         WARN_ON(dev->type->release != nd_region_release);
 177         return nd_region;
 178 }
 179 EXPORT_SYMBOL_GPL(to_nd_region);
 180
 181 struct nd_blk_region *to_nd_blk_region(struct device *dev)
 182 {
 183         struct nd_region *nd_region = to_nd_region(dev);
 184
 185         WARN_ON(!is_nd_blk(dev));
 186         return container_of(nd_region, struct nd_blk_region, nd_region);
 187 }
 188 EXPORT_SYMBOL_GPL(to_nd_blk_region);
 189
 190 void *nd_region_provider_data(struct nd_region *nd_region)
 191 {
 192         return nd_region->provider_data;
 193 }
 194 EXPORT_SYMBOL_GPL(nd_region_provider_data);
 195
 196 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
 197 {
 198         return ndbr->blk_provider_data;
 199 }
 200 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
 201
 202 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
 203 {
 204         ndbr->blk_provider_data = data;
 205 }
 206 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
 207
 208 /**
 209  * nd_region_to_nstype() - region to an integer namespace type
 210  * @nd_region: region-device to interrogate
 211  *
 212  * This is the 'nstype' attribute of a region as well, an input to the
 213  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
 214  * namespace devices with namespace drivers.
 215  */
 216 int nd_region_to_nstype(struct nd_region *nd_region)
 217 {
 218         if (is_nd_pmem(&nd_region->dev)) {
 219                 u16 i, alias;
 220
 221                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
 222                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 223                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 224
 225                         if (nvdimm->flags & NDD_ALIASING)
 226                                 alias++;
 227                 }
 228                 if (alias)
 229                         return ND_DEVICE_NAMESPACE_PMEM;
 230                 else
 231                         return ND_DEVICE_NAMESPACE_IO;
 232         } else if (is_nd_blk(&nd_region->dev)) {
 233                 return ND_DEVICE_NAMESPACE_BLK;
 234         }
 235
 236         return 0;
 237 }
 238 EXPORT_SYMBOL(nd_region_to_nstype);
 239
 240 static ssize_t size_show(struct device *dev,
 241                 struct device_attribute *attr, char *buf)
 242 {
 243         struct nd_region *nd_region = to_nd_region(dev);
 244         unsigned long long size = 0;
 245
 246         if (is_nd_pmem(dev)) {
 247                 size = nd_region->ndr_size;
 248         } else if (nd_region->ndr_mappings == 1) {
 249                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 250
 251                 size = nd_mapping->size;
 252         }
 253
 254         return sprintf(buf, "%llu\n", size);
 255 }
 256 static DEVICE_ATTR_RO(size);
 257
 258 static ssize_t deep_flush_show(struct device *dev,
 259                 struct device_attribute *attr, char *buf)
 260 {
 261         struct nd_region *nd_region = to_nd_region(dev);
 262
 263         /*
 264          * NOTE: in the nvdimm_has_flush() error case this attribute is
 265          * not visible.
 266          */
 267         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
 268 }
 269
 270 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
 271                 const char *buf, size_t len)
 272 {
 273         bool flush;
 274         int rc = strtobool(buf, &flush);
 275         struct nd_region *nd_region = to_nd_region(dev);
 276
 277         if (rc)
 278                 return rc;
 279         if (!flush)
 280                 return -EINVAL;
 281         nvdimm_flush(nd_region);
 282
 283         return len;
 284 }
 285 static DEVICE_ATTR_RW(deep_flush);
 286
 287 static ssize_t mappings_show(struct device *dev,
 288                 struct device_attribute *attr, char *buf)
 289 {
 290         struct nd_region *nd_region = to_nd_region(dev);
 291
 292         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 293 }
 294 static DEVICE_ATTR_RO(mappings);
 295
 296 static ssize_t nstype_show(struct device *dev,
 297                 struct device_attribute *attr, char *buf)
 298 {
 299         struct nd_region *nd_region = to_nd_region(dev);
 300
 301         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 302 }
 303 static DEVICE_ATTR_RO(nstype);
 304
 305 static ssize_t set_cookie_show(struct device *dev,
 306                 struct device_attribute *attr, char *buf)
 307 {
 308         struct nd_region *nd_region = to_nd_region(dev);
 309         struct nd_interleave_set *nd_set = nd_region->nd_set;
 310
 311         if (is_nd_pmem(dev) && nd_set)
 312                 /* pass, should be precluded by region_visible */;
 313         else
 314                 return -ENXIO;
 315
 316         return sprintf(buf, "%#llx\n", nd_set->cookie);
 317 }
 318 static DEVICE_ATTR_RO(set_cookie);
 319
 320 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 321 {
 322         resource_size_t blk_max_overlap = 0, available, overlap;
 323         int i;
 324
 325         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 326
 327  retry:
 328         available = 0;
 329         overlap = blk_max_overlap;
 330         for (i = 0; i < nd_region->ndr_mappings; i++) {
 331                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 332                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 333
 334                 /* if a dimm is disabled the available capacity is zero */
 335                 if (!ndd)
 336                         return 0;
 337
 338                 if (is_nd_pmem(&nd_region->dev)) {
 339                         available += nd_pmem_available_dpa(nd_region,
 340                                         nd_mapping, &overlap);
 341                         if (overlap > blk_max_overlap) {
 342                                 blk_max_overlap = overlap;
 343                                 goto retry;
 344                         }
 345                 } else if (is_nd_blk(&nd_region->dev))
 346                         available += nd_blk_available_dpa(nd_region);
 347         }
 348
 349         return available;
 350 }
 351
 352 static ssize_t available_size_show(struct device *dev,
 353                 struct device_attribute *attr, char *buf)
 354 {
 355         struct nd_region *nd_region = to_nd_region(dev);
 356         unsigned long long available = 0;
 357
 358         /*
 359          * Flush in-flight updates and grab a snapshot of the available
 360          * size.  Of course, this value is potentially invalidated the
 361          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 362          * problem to not race itself.
 363          */
 364         nvdimm_bus_lock(dev);
 365         wait_nvdimm_bus_probe_idle(dev);
 366         available = nd_region_available_dpa(nd_region);
 367         nvdimm_bus_unlock(dev);
 368
 369         return sprintf(buf, "%llu\n", available);
 370 }
 371 static DEVICE_ATTR_RO(available_size);
 372
 373 static ssize_t init_namespaces_show(struct device *dev,
 374                 struct device_attribute *attr, char *buf)
 375 {
 376         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 377         ssize_t rc;
 378
 379         nvdimm_bus_lock(dev);
 380         if (ndrd)
 381                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 382         else
 383                 rc = -ENXIO;
 384         nvdimm_bus_unlock(dev);
 385
 386         return rc;
 387 }
 388 static DEVICE_ATTR_RO(init_namespaces);
 389
 390 static ssize_t namespace_seed_show(struct device *dev,
 391                 struct device_attribute *attr, char *buf)
 392 {
 393         struct nd_region *nd_region = to_nd_region(dev);
 394         ssize_t rc;
 395
 396         nvdimm_bus_lock(dev);
 397         if (nd_region->ns_seed)
 398                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 399         else
 400                 rc = sprintf(buf, "\n");
 401         nvdimm_bus_unlock(dev);
 402         return rc;
 403 }
 404 static DEVICE_ATTR_RO(namespace_seed);
 405
 406 static ssize_t btt_seed_show(struct device *dev,
 407                 struct device_attribute *attr, char *buf)
 408 {
 409         struct nd_region *nd_region = to_nd_region(dev);
 410         ssize_t rc;
 411
 412         nvdimm_bus_lock(dev);
 413         if (nd_region->btt_seed)
 414                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 415         else
 416                 rc = sprintf(buf, "\n");
 417         nvdimm_bus_unlock(dev);
 418
 419         return rc;
 420 }
 421 static DEVICE_ATTR_RO(btt_seed);
 422
 423 static ssize_t pfn_seed_show(struct device *dev,
 424                 struct device_attribute *attr, char *buf)
 425 {
 426         struct nd_region *nd_region = to_nd_region(dev);
 427         ssize_t rc;
 428
 429         nvdimm_bus_lock(dev);
 430         if (nd_region->pfn_seed)
 431                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 432         else
 433                 rc = sprintf(buf, "\n");
 434         nvdimm_bus_unlock(dev);
 435
 436         return rc;
 437 }
 438 static DEVICE_ATTR_RO(pfn_seed);
 439
 440 static ssize_t dax_seed_show(struct device *dev,
 441                 struct device_attribute *attr, char *buf)
 442 {
 443         struct nd_region *nd_region = to_nd_region(dev);
 444         ssize_t rc;
 445
 446         nvdimm_bus_lock(dev);
 447         if (nd_region->dax_seed)
 448                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 449         else
 450                 rc = sprintf(buf, "\n");
 451         nvdimm_bus_unlock(dev);
 452
 453         return rc;
 454 }
 455 static DEVICE_ATTR_RO(dax_seed);
 456
 457 static ssize_t read_only_show(struct device *dev,
 458                 struct device_attribute *attr, char *buf)
 459 {
 460         struct nd_region *nd_region = to_nd_region(dev);
 461
 462         return sprintf(buf, "%d\n", nd_region->ro);
 463 }
 464
 465 static ssize_t read_only_store(struct device *dev,
 466                 struct device_attribute *attr, const char *buf, size_t len)
 467 {
 468         bool ro;
 469         int rc = strtobool(buf, &ro);
 470         struct nd_region *nd_region = to_nd_region(dev);
 471
 472         if (rc)
 473                 return rc;
 474
 475         nd_region->ro = ro;
 476         return len;
 477 }
 478 static DEVICE_ATTR_RW(read_only);
 479
 480 static ssize_t nd_badblocks_show(struct device *dev,
 481                 struct device_attribute *attr, char *buf)
 482 {
 483         struct nd_region *nd_region = to_nd_region(dev);
 484
 485         return badblocks_show(&nd_region->bb, buf, 0);
 486 }
 487 static struct device_attribute dev_attr_nd_badblocks = {
 488         .attr = {
 489                 .name = "badblocks",
 490                 .mode = S_IRUGO
 491         },
 492         .show = nd_badblocks_show,
 493 };
 494
 495 static ssize_t resource_show(struct device *dev,
 496                 struct device_attribute *attr, char *buf)
 497 {
 498         struct nd_region *nd_region = to_nd_region(dev);
 499
 500         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 501 }
 502 static DEVICE_ATTR_RO(resource);
 503
 504 static struct attribute *nd_region_attributes[] = {
 505         &dev_attr_size.attr,
 506         &dev_attr_nstype.attr,
 507         &dev_attr_mappings.attr,
 508         &dev_attr_btt_seed.attr,
 509         &dev_attr_pfn_seed.attr,
 510         &dev_attr_dax_seed.attr,
 511         &dev_attr_deep_flush.attr,
 512         &dev_attr_read_only.attr,
 513         &dev_attr_set_cookie.attr,
 514         &dev_attr_available_size.attr,
 515         &dev_attr_namespace_seed.attr,
 516         &dev_attr_init_namespaces.attr,
 517         &dev_attr_nd_badblocks.attr,
 518         &dev_attr_resource.attr,
 519         NULL,
 520 };
 521
 522 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 523 {
 524         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 525         struct nd_region *nd_region = to_nd_region(dev);
 526         struct nd_interleave_set *nd_set = nd_region->nd_set;
 527         int type = nd_region_to_nstype(nd_region);
 528
 529         if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
 530                 return 0;
 531
 532         if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
 533                 return 0;
 534
 535         if (!is_nd_pmem(dev) && a == &dev_attr_nd_badblocks.attr)
 536                 return 0;
 537
 538         if (!is_nd_pmem(dev) && a == &dev_attr_resource.attr)
 539                 return 0;
 540
 541         if (a == &dev_attr_deep_flush.attr) {
 542                 int has_flush = nvdimm_has_flush(nd_region);
 543
 544                 if (has_flush == 1)
 545                         return a->mode;
 546                 else if (has_flush == 0)
 547                         return 0444;
 548                 else
 549                         return 0;
 550         }
 551
 552         if (a != &dev_attr_set_cookie.attr
 553                         && a != &dev_attr_available_size.attr)
 554                 return a->mode;
 555
 556         if ((type == ND_DEVICE_NAMESPACE_PMEM
 557                                 || type == ND_DEVICE_NAMESPACE_BLK)
 558                         && a == &dev_attr_available_size.attr)
 559                 return a->mode;
 560         else if (is_nd_pmem(dev) && nd_set)
 561                 return a->mode;
 562
 563         return 0;
 564 }
 565
 566 struct attribute_group nd_region_attribute_group = {
 567         .attrs = nd_region_attributes,
 568         .is_visible = region_visible,
 569 };
 570 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 571
 572 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
 573 {
 574         struct nd_interleave_set *nd_set = nd_region->nd_set;
 575
 576         if (nd_set)
 577                 return nd_set->cookie;
 578         return 0;
 579 }
 580
 581 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 582 {
 583         struct nd_interleave_set *nd_set = nd_region->nd_set;
 584
 585         if (nd_set)
 586                 return nd_set->altcookie;
 587         return 0;
 588 }
 589
 590 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 591 {
 592         struct nd_label_ent *label_ent, *e;
 593
 594         lockdep_assert_held(&nd_mapping->lock);
 595         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 596                 list_del(&label_ent->list);
 597                 kfree(label_ent);
 598         }
 599 }
 600
 601 /*
 602  * Upon successful probe/remove, take/release a reference on the
 603  * associated interleave set (if present), and plant new btt + namespace
 604  * seeds.  Also, on the removal of a BLK region, notify the provider to
 605  * disable the region.
 606  */
 607 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 608                 struct device *dev, bool probe)
 609 {
 610         struct nd_region *nd_region;
 611
 612         if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
 613                 int i;
 614
 615                 nd_region = to_nd_region(dev);
 616                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 617                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 618                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 619                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 620
 621                         mutex_lock(&nd_mapping->lock);
 622                         nd_mapping_free_labels(nd_mapping);
 623                         mutex_unlock(&nd_mapping->lock);
 624
 625                         put_ndd(ndd);
 626                         nd_mapping->ndd = NULL;
 627                         if (ndd)
 628                                 atomic_dec(&nvdimm->busy);
 629                 }
 630
 631                 if (is_nd_pmem(dev))
 632                         return;
 633         }
 634         if (dev->parent && (is_nd_blk(dev->parent) || is_nd_pmem(dev->parent))
 635                         && probe) {
 636                 nd_region = to_nd_region(dev->parent);
 637                 nvdimm_bus_lock(dev);
 638                 if (nd_region->ns_seed == dev)
 639                         nd_region_create_ns_seed(nd_region);
 640                 nvdimm_bus_unlock(dev);
 641         }
 642         if (is_nd_btt(dev) && probe) {
 643                 struct nd_btt *nd_btt = to_nd_btt(dev);
 644
 645                 nd_region = to_nd_region(dev->parent);
 646                 nvdimm_bus_lock(dev);
 647                 if (nd_region->btt_seed == dev)
 648                         nd_region_create_btt_seed(nd_region);
 649                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 650                         nd_region_create_ns_seed(nd_region);
 651                 nvdimm_bus_unlock(dev);
 652         }
 653         if (is_nd_pfn(dev) && probe) {
 654                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 655
 656                 nd_region = to_nd_region(dev->parent);
 657                 nvdimm_bus_lock(dev);
 658                 if (nd_region->pfn_seed == dev)
 659                         nd_region_create_pfn_seed(nd_region);
 660                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 661                         nd_region_create_ns_seed(nd_region);
 662                 nvdimm_bus_unlock(dev);
 663         }
 664         if (is_nd_dax(dev) && probe) {
 665                 struct nd_dax *nd_dax = to_nd_dax(dev);
 666
 667                 nd_region = to_nd_region(dev->parent);
 668                 nvdimm_bus_lock(dev);
 669                 if (nd_region->dax_seed == dev)
 670                         nd_region_create_dax_seed(nd_region);
 671                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 672                         nd_region_create_ns_seed(nd_region);
 673                 nvdimm_bus_unlock(dev);
 674         }
 675 }
 676
 677 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 678 {
 679         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 680 }
 681
 682 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 683 {
 684         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 685 }
 686
 687 static ssize_t mappingN(struct device *dev, char *buf, int n)
 688 {
 689         struct nd_region *nd_region = to_nd_region(dev);
 690         struct nd_mapping *nd_mapping;
 691         struct nvdimm *nvdimm;
 692
 693         if (n >= nd_region->ndr_mappings)
 694                 return -ENXIO;
 695         nd_mapping = &nd_region->mapping[n];
 696         nvdimm = nd_mapping->nvdimm;
 697
 698         return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
 699                         nd_mapping->start, nd_mapping->size);
 700 }
 701
 702 #define REGION_MAPPING(idx) \
 703 static ssize_t mapping##idx##_show(struct device *dev,          \
 704                 struct device_attribute *attr, char *buf)       \
 705 {                                                               \
 706         return mappingN(dev, buf, idx);                         \
 707 }                                                               \
 708 static DEVICE_ATTR_RO(mapping##idx)
 709
 710 /*
 711  * 32 should be enough for a while, even in the presence of socket
 712  * interleave a 32-way interleave set is a degenerate case.
 713  */
 714 REGION_MAPPING(0);
 715 REGION_MAPPING(1);
 716 REGION_MAPPING(2);
 717 REGION_MAPPING(3);
 718 REGION_MAPPING(4);
 719 REGION_MAPPING(5);
 720 REGION_MAPPING(6);
 721 REGION_MAPPING(7);
 722 REGION_MAPPING(8);
 723 REGION_MAPPING(9);
 724 REGION_MAPPING(10);
 725 REGION_MAPPING(11);
 726 REGION_MAPPING(12);
 727 REGION_MAPPING(13);
 728 REGION_MAPPING(14);
 729 REGION_MAPPING(15);
 730 REGION_MAPPING(16);
 731 REGION_MAPPING(17);
 732 REGION_MAPPING(18);
 733 REGION_MAPPING(19);
 734 REGION_MAPPING(20);
 735 REGION_MAPPING(21);
 736 REGION_MAPPING(22);
 737 REGION_MAPPING(23);
 738 REGION_MAPPING(24);
 739 REGION_MAPPING(25);
 740 REGION_MAPPING(26);
 741 REGION_MAPPING(27);
 742 REGION_MAPPING(28);
 743 REGION_MAPPING(29);
 744 REGION_MAPPING(30);
 745 REGION_MAPPING(31);
 746
 747 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 748 {
 749         struct device *dev = container_of(kobj, struct device, kobj);
 750         struct nd_region *nd_region = to_nd_region(dev);
 751
 752         if (n < nd_region->ndr_mappings)
 753                 return a->mode;
 754         return 0;
 755 }
 756
 757 static struct attribute *mapping_attributes[] = {
 758         &dev_attr_mapping0.attr,
 759         &dev_attr_mapping1.attr,
 760         &dev_attr_mapping2.attr,
 761         &dev_attr_mapping3.attr,
 762         &dev_attr_mapping4.attr,
 763         &dev_attr_mapping5.attr,
 764         &dev_attr_mapping6.attr,
 765         &dev_attr_mapping7.attr,
 766         &dev_attr_mapping8.attr,
 767         &dev_attr_mapping9.attr,
 768         &dev_attr_mapping10.attr,
 769         &dev_attr_mapping11.attr,
 770         &dev_attr_mapping12.attr,
 771         &dev_attr_mapping13.attr,
 772         &dev_attr_mapping14.attr,
 773         &dev_attr_mapping15.attr,
 774         &dev_attr_mapping16.attr,
 775         &dev_attr_mapping17.attr,
 776         &dev_attr_mapping18.attr,
 777         &dev_attr_mapping19.attr,
 778         &dev_attr_mapping20.attr,
 779         &dev_attr_mapping21.attr,
 780         &dev_attr_mapping22.attr,
 781         &dev_attr_mapping23.attr,
 782         &dev_attr_mapping24.attr,
 783         &dev_attr_mapping25.attr,
 784         &dev_attr_mapping26.attr,
 785         &dev_attr_mapping27.attr,
 786         &dev_attr_mapping28.attr,
 787         &dev_attr_mapping29.attr,
 788         &dev_attr_mapping30.attr,
 789         &dev_attr_mapping31.attr,
 790         NULL,
 791 };
 792
 793 struct attribute_group nd_mapping_attribute_group = {
 794         .is_visible = mapping_visible,
 795         .attrs = mapping_attributes,
 796 };
 797 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 798
 799 int nd_blk_region_init(struct nd_region *nd_region)
 800 {
 801         struct device *dev = &nd_region->dev;
 802         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 803
 804         if (!is_nd_blk(dev))
 805                 return 0;
 806
 807         if (nd_region->ndr_mappings < 1) {
 808                 dev_err(dev, "invalid BLK region\n");
 809                 return -ENXIO;
 810         }
 811
 812         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 813 }
 814
 815 /**
 816  * nd_region_acquire_lane - allocate and lock a lane
 817  * @nd_region: region id and number of lanes possible
 818  *
 819  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 820  * We optimize for the common case where there are 256 lanes, one
 821  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 822  * this implementation assumes the cost of maintaining an allocator for
 823  * free lanes is on the order of the lock hold time, so it implements a
 824  * static lane = cpu % num_lanes mapping.
 825  *
 826  * In the case of a BTT instance on top of a BLK namespace a lane may be
 827  * acquired recursively.  We lock on the first instance.
 828  *
 829  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 830  * for the BTT metadata updates.
 831  */
 832 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 833 {
 834         unsigned int cpu, lane;
 835
 836         cpu = get_cpu();
 837         if (nd_region->num_lanes < nr_cpu_ids) {
 838                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 839
 840                 lane = cpu % nd_region->num_lanes;
 841                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 842                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 843                 if (ndl_count->count++ == 0)
 844                         spin_lock(&ndl_lock->lock);
 845         } else
 846                 lane = cpu;
 847
 848         return lane;
 849 }
 850 EXPORT_SYMBOL(nd_region_acquire_lane);
 851
 852 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 853 {
 854         if (nd_region->num_lanes < nr_cpu_ids) {
 855                 unsigned int cpu = get_cpu();
 856                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 857
 858                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 859                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 860                 if (--ndl_count->count == 0)
 861                         spin_unlock(&ndl_lock->lock);
 862                 put_cpu();
 863         }
 864         put_cpu();
 865 }
 866 EXPORT_SYMBOL(nd_region_release_lane);
 867
 868 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 869                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 870                 const char *caller)
 871 {
 872         struct nd_region *nd_region;
 873         struct device *dev;
 874         void *region_buf;
 875         unsigned int i;
 876         int ro = 0;
 877
 878         for (i = 0; i < ndr_desc->num_mappings; i++) {
 879                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 880                 struct nvdimm *nvdimm = mapping->nvdimm;
 881
 882                 if ((mapping->start | mapping->size) % SZ_4K) {
 883                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 884                                         caller, dev_name(&nvdimm->dev), i);
 885
 886                         return NULL;
 887                 }
 888
 889                 if (nvdimm->flags & NDD_UNARMED)
 890                         ro = 1;
 891         }
 892
 893         if (dev_type == &nd_blk_device_type) {
 894                 struct nd_blk_region_desc *ndbr_desc;
 895                 struct nd_blk_region *ndbr;
 896
 897                 ndbr_desc = to_blk_region_desc(ndr_desc);
 898                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 899                                 * ndr_desc->num_mappings,
 900                                 GFP_KERNEL);
 901                 if (ndbr) {
 902                         nd_region = &ndbr->nd_region;
 903                         ndbr->enable = ndbr_desc->enable;
 904                         ndbr->do_io = ndbr_desc->do_io;
 905                 }
 906                 region_buf = ndbr;
 907         } else {
 908                 nd_region = kzalloc(sizeof(struct nd_region)
 909                                 + sizeof(struct nd_mapping)
 910                                 * ndr_desc->num_mappings,
 911                                 GFP_KERNEL);
 912                 region_buf = nd_region;
 913         }
 914
 915         if (!region_buf)
 916                 return NULL;
 917         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 918         if (nd_region->id < 0)
 919                 goto err_id;
 920
 921         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 922         if (!nd_region->lane)
 923                 goto err_percpu;
 924
 925         for (i = 0; i < nr_cpu_ids; i++) {
 926                 struct nd_percpu_lane *ndl;
 927
 928                 ndl = per_cpu_ptr(nd_region->lane, i);
 929                 spin_lock_init(&ndl->lock);
 930                 ndl->count = 0;
 931         }
 932
 933         for (i = 0; i < ndr_desc->num_mappings; i++) {
 934                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 935                 struct nvdimm *nvdimm = mapping->nvdimm;
 936
 937                 nd_region->mapping[i].nvdimm = nvdimm;
 938                 nd_region->mapping[i].start = mapping->start;
 939                 nd_region->mapping[i].size = mapping->size;
 940                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
 941                 mutex_init(&nd_region->mapping[i].lock);
 942
 943                 get_device(&nvdimm->dev);
 944         }
 945         nd_region->ndr_mappings = ndr_desc->num_mappings;
 946         nd_region->provider_data = ndr_desc->provider_data;
 947         nd_region->nd_set = ndr_desc->nd_set;
 948         nd_region->num_lanes = ndr_desc->num_lanes;
 949         nd_region->flags = ndr_desc->flags;
 950         nd_region->ro = ro;
 951         nd_region->numa_node = ndr_desc->numa_node;
 952         ida_init(&nd_region->ns_ida);
 953         ida_init(&nd_region->btt_ida);
 954         ida_init(&nd_region->pfn_ida);
 955         ida_init(&nd_region->dax_ida);
 956         dev = &nd_region->dev;
 957         dev_set_name(dev, "region%d", nd_region->id);
 958         dev->parent = &nvdimm_bus->dev;
 959         dev->type = dev_type;
 960         dev->groups = ndr_desc->attr_groups;
 961         nd_region->ndr_size = resource_size(ndr_desc->res);
 962         nd_region->ndr_start = ndr_desc->res->start;
 963         nd_device_register(dev);
 964
 965         return nd_region;
 966
 967  err_percpu:
 968         ida_simple_remove(&region_ida, nd_region->id);
 969  err_id:
 970         kfree(region_buf);
 971         return NULL;
 972 }
 973
 974 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
 975                 struct nd_region_desc *ndr_desc)
 976 {
 977         ndr_desc->num_lanes = ND_MAX_LANES;
 978         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
 979                         __func__);
 980 }
 981 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
 982
 983 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
 984                 struct nd_region_desc *ndr_desc)
 985 {
 986         if (ndr_desc->num_mappings > 1)
 987                 return NULL;
 988         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
 989         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
 990                         __func__);
 991 }
 992 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
 993
 994 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
 995                 struct nd_region_desc *ndr_desc)
 996 {
 997         ndr_desc->num_lanes = ND_MAX_LANES;
 998         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
 999                         __func__);
1000 }
1001 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1002
1003 /**
1004  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1005  * @nd_region: blk or interleaved pmem region
1006  */
1007 void nvdimm_flush(struct nd_region *nd_region)
1008 {
1009         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1010         int i, idx;
1011
1012         /*
1013          * Try to encourage some diversity in flush hint addresses
1014          * across cpus assuming a limited number of flush hints.
1015          */
1016         idx = this_cpu_read(flush_idx);
1017         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1018
1019         /*
1020          * The first wmb() is needed to 'sfence' all previous writes
1021          * such that they are architecturally visible for the platform
1022          * buffer flush.  Note that we've already arranged for pmem
1023          * writes to avoid the cache via arch_memcpy_to_pmem().  The
1024          * final wmb() ensures ordering for the NVDIMM flush write.
1025          */
1026         wmb();
1027         for (i = 0; i < nd_region->ndr_mappings; i++)
1028                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1029                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1030         wmb();
1031 }
1032 EXPORT_SYMBOL_GPL(nvdimm_flush);
1033
1034 /**
1035  * nvdimm_has_flush - determine write flushing requirements
1036  * @nd_region: blk or interleaved pmem region
1037  *
1038  * Returns 1 if writes require flushing
1039  * Returns 0 if writes do not require flushing
1040  * Returns -ENXIO if flushing capability can not be determined
1041  */
1042 int nvdimm_has_flush(struct nd_region *nd_region)
1043 {
1044         int i;
1045
1046         /* no nvdimm == flushing capability unknown */
1047         if (nd_region->ndr_mappings == 0)
1048                 return -ENXIO;
1049
1050         for (i = 0; i < nd_region->ndr_mappings; i++) {
1051                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1052                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1053
1054                 /* flush hints present / available */
1055                 if (nvdimm->num_flush)
1056                         return 1;
1057         }
1058
1059         /*
1060          * The platform defines dimm devices without hints, assume
1061          * platform persistence mechanism like ADR
1062          */
1063         return 0;
1064 }
1065 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1066
1067 void __exit nd_region_devs_exit(void)
1068 {
1069         ida_destroy(&region_ida);
1070 }