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