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 mappings_show(struct device *dev,
 259                 struct device_attribute *attr, char *buf)
 260 {
 261         struct nd_region *nd_region = to_nd_region(dev);
 262
 263         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 264 }
 265 static DEVICE_ATTR_RO(mappings);
 266
 267 static ssize_t nstype_show(struct device *dev,
 268                 struct device_attribute *attr, char *buf)
 269 {
 270         struct nd_region *nd_region = to_nd_region(dev);
 271
 272         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 273 }
 274 static DEVICE_ATTR_RO(nstype);
 275
 276 static ssize_t set_cookie_show(struct device *dev,
 277                 struct device_attribute *attr, char *buf)
 278 {
 279         struct nd_region *nd_region = to_nd_region(dev);
 280         struct nd_interleave_set *nd_set = nd_region->nd_set;
 281
 282         if (is_nd_pmem(dev) && nd_set)
 283                 /* pass, should be precluded by region_visible */;
 284         else
 285                 return -ENXIO;
 286
 287         return sprintf(buf, "%#llx\n", nd_set->cookie);
 288 }
 289 static DEVICE_ATTR_RO(set_cookie);
 290
 291 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 292 {
 293         resource_size_t blk_max_overlap = 0, available, overlap;
 294         int i;
 295
 296         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 297
 298  retry:
 299         available = 0;
 300         overlap = blk_max_overlap;
 301         for (i = 0; i < nd_region->ndr_mappings; i++) {
 302                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 303                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 304
 305                 /* if a dimm is disabled the available capacity is zero */
 306                 if (!ndd)
 307                         return 0;
 308
 309                 if (is_nd_pmem(&nd_region->dev)) {
 310                         available += nd_pmem_available_dpa(nd_region,
 311                                         nd_mapping, &overlap);
 312                         if (overlap > blk_max_overlap) {
 313                                 blk_max_overlap = overlap;
 314                                 goto retry;
 315                         }
 316                 } else if (is_nd_blk(&nd_region->dev))
 317                         available += nd_blk_available_dpa(nd_region);
 318         }
 319
 320         return available;
 321 }
 322
 323 static ssize_t available_size_show(struct device *dev,
 324                 struct device_attribute *attr, char *buf)
 325 {
 326         struct nd_region *nd_region = to_nd_region(dev);
 327         unsigned long long available = 0;
 328
 329         /*
 330          * Flush in-flight updates and grab a snapshot of the available
 331          * size.  Of course, this value is potentially invalidated the
 332          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 333          * problem to not race itself.
 334          */
 335         nvdimm_bus_lock(dev);
 336         wait_nvdimm_bus_probe_idle(dev);
 337         available = nd_region_available_dpa(nd_region);
 338         nvdimm_bus_unlock(dev);
 339
 340         return sprintf(buf, "%llu\n", available);
 341 }
 342 static DEVICE_ATTR_RO(available_size);
 343
 344 static ssize_t init_namespaces_show(struct device *dev,
 345                 struct device_attribute *attr, char *buf)
 346 {
 347         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 348         ssize_t rc;
 349
 350         nvdimm_bus_lock(dev);
 351         if (ndrd)
 352                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 353         else
 354                 rc = -ENXIO;
 355         nvdimm_bus_unlock(dev);
 356
 357         return rc;
 358 }
 359 static DEVICE_ATTR_RO(init_namespaces);
 360
 361 static ssize_t namespace_seed_show(struct device *dev,
 362                 struct device_attribute *attr, char *buf)
 363 {
 364         struct nd_region *nd_region = to_nd_region(dev);
 365         ssize_t rc;
 366
 367         nvdimm_bus_lock(dev);
 368         if (nd_region->ns_seed)
 369                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 370         else
 371                 rc = sprintf(buf, "\n");
 372         nvdimm_bus_unlock(dev);
 373         return rc;
 374 }
 375 static DEVICE_ATTR_RO(namespace_seed);
 376
 377 static ssize_t btt_seed_show(struct device *dev,
 378                 struct device_attribute *attr, char *buf)
 379 {
 380         struct nd_region *nd_region = to_nd_region(dev);
 381         ssize_t rc;
 382
 383         nvdimm_bus_lock(dev);
 384         if (nd_region->btt_seed)
 385                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 386         else
 387                 rc = sprintf(buf, "\n");
 388         nvdimm_bus_unlock(dev);
 389
 390         return rc;
 391 }
 392 static DEVICE_ATTR_RO(btt_seed);
 393
 394 static ssize_t pfn_seed_show(struct device *dev,
 395                 struct device_attribute *attr, char *buf)
 396 {
 397         struct nd_region *nd_region = to_nd_region(dev);
 398         ssize_t rc;
 399
 400         nvdimm_bus_lock(dev);
 401         if (nd_region->pfn_seed)
 402                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 403         else
 404                 rc = sprintf(buf, "\n");
 405         nvdimm_bus_unlock(dev);
 406
 407         return rc;
 408 }
 409 static DEVICE_ATTR_RO(pfn_seed);
 410
 411 static ssize_t dax_seed_show(struct device *dev,
 412                 struct device_attribute *attr, char *buf)
 413 {
 414         struct nd_region *nd_region = to_nd_region(dev);
 415         ssize_t rc;
 416
 417         nvdimm_bus_lock(dev);
 418         if (nd_region->dax_seed)
 419                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 420         else
 421                 rc = sprintf(buf, "\n");
 422         nvdimm_bus_unlock(dev);
 423
 424         return rc;
 425 }
 426 static DEVICE_ATTR_RO(dax_seed);
 427
 428 static ssize_t read_only_show(struct device *dev,
 429                 struct device_attribute *attr, char *buf)
 430 {
 431         struct nd_region *nd_region = to_nd_region(dev);
 432
 433         return sprintf(buf, "%d\n", nd_region->ro);
 434 }
 435
 436 static ssize_t read_only_store(struct device *dev,
 437                 struct device_attribute *attr, const char *buf, size_t len)
 438 {
 439         bool ro;
 440         int rc = strtobool(buf, &ro);
 441         struct nd_region *nd_region = to_nd_region(dev);
 442
 443         if (rc)
 444                 return rc;
 445
 446         nd_region->ro = ro;
 447         return len;
 448 }
 449 static DEVICE_ATTR_RW(read_only);
 450
 451 static ssize_t nd_badblocks_show(struct device *dev,
 452                 struct device_attribute *attr, char *buf)
 453 {
 454         struct nd_region *nd_region = to_nd_region(dev);
 455
 456         return badblocks_show(&nd_region->bb, buf, 0);
 457 }
 458 static struct device_attribute dev_attr_nd_badblocks = {
 459         .attr = {
 460                 .name = "badblocks",
 461                 .mode = S_IRUGO
 462         },
 463         .show = nd_badblocks_show,
 464 };
 465
 466 static ssize_t resource_show(struct device *dev,
 467                 struct device_attribute *attr, char *buf)
 468 {
 469         struct nd_region *nd_region = to_nd_region(dev);
 470
 471         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 472 }
 473 static DEVICE_ATTR_RO(resource);
 474
 475 static struct attribute *nd_region_attributes[] = {
 476         &dev_attr_size.attr,
 477         &dev_attr_nstype.attr,
 478         &dev_attr_mappings.attr,
 479         &dev_attr_btt_seed.attr,
 480         &dev_attr_pfn_seed.attr,
 481         &dev_attr_dax_seed.attr,
 482         &dev_attr_read_only.attr,
 483         &dev_attr_set_cookie.attr,
 484         &dev_attr_available_size.attr,
 485         &dev_attr_namespace_seed.attr,
 486         &dev_attr_init_namespaces.attr,
 487         &dev_attr_nd_badblocks.attr,
 488         &dev_attr_resource.attr,
 489         NULL,
 490 };
 491
 492 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 493 {
 494         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 495         struct nd_region *nd_region = to_nd_region(dev);
 496         struct nd_interleave_set *nd_set = nd_region->nd_set;
 497         int type = nd_region_to_nstype(nd_region);
 498
 499         if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
 500                 return 0;
 501
 502         if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
 503                 return 0;
 504
 505         if (!is_nd_pmem(dev) && a == &dev_attr_nd_badblocks.attr)
 506                 return 0;
 507
 508         if (!is_nd_pmem(dev) && a == &dev_attr_resource.attr)
 509                 return 0;
 510
 511         if (a != &dev_attr_set_cookie.attr
 512                         && a != &dev_attr_available_size.attr)
 513                 return a->mode;
 514
 515         if ((type == ND_DEVICE_NAMESPACE_PMEM
 516                                 || type == ND_DEVICE_NAMESPACE_BLK)
 517                         && a == &dev_attr_available_size.attr)
 518                 return a->mode;
 519         else if (is_nd_pmem(dev) && nd_set)
 520                 return a->mode;
 521
 522         return 0;
 523 }
 524
 525 struct attribute_group nd_region_attribute_group = {
 526         .attrs = nd_region_attributes,
 527         .is_visible = region_visible,
 528 };
 529 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 530
 531 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
 532 {
 533         struct nd_interleave_set *nd_set = nd_region->nd_set;
 534
 535         if (nd_set)
 536                 return nd_set->cookie;
 537         return 0;
 538 }
 539
 540 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 541 {
 542         struct nd_interleave_set *nd_set = nd_region->nd_set;
 543
 544         if (nd_set)
 545                 return nd_set->altcookie;
 546         return 0;
 547 }
 548
 549 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 550 {
 551         struct nd_label_ent *label_ent, *e;
 552
 553         lockdep_assert_held(&nd_mapping->lock);
 554         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 555                 list_del(&label_ent->list);
 556                 kfree(label_ent);
 557         }
 558 }
 559
 560 /*
 561  * Upon successful probe/remove, take/release a reference on the
 562  * associated interleave set (if present), and plant new btt + namespace
 563  * seeds.  Also, on the removal of a BLK region, notify the provider to
 564  * disable the region.
 565  */
 566 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 567                 struct device *dev, bool probe)
 568 {
 569         struct nd_region *nd_region;
 570
 571         if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
 572                 int i;
 573
 574                 nd_region = to_nd_region(dev);
 575                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 576                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 577                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 578                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 579
 580                         mutex_lock(&nd_mapping->lock);
 581                         nd_mapping_free_labels(nd_mapping);
 582                         mutex_unlock(&nd_mapping->lock);
 583
 584                         put_ndd(ndd);
 585                         nd_mapping->ndd = NULL;
 586                         if (ndd)
 587                                 atomic_dec(&nvdimm->busy);
 588                 }
 589
 590                 if (is_nd_pmem(dev))
 591                         return;
 592         }
 593         if (dev->parent && (is_nd_blk(dev->parent) || is_nd_pmem(dev->parent))
 594                         && probe) {
 595                 nd_region = to_nd_region(dev->parent);
 596                 nvdimm_bus_lock(dev);
 597                 if (nd_region->ns_seed == dev)
 598                         nd_region_create_ns_seed(nd_region);
 599                 nvdimm_bus_unlock(dev);
 600         }
 601         if (is_nd_btt(dev) && probe) {
 602                 struct nd_btt *nd_btt = to_nd_btt(dev);
 603
 604                 nd_region = to_nd_region(dev->parent);
 605                 nvdimm_bus_lock(dev);
 606                 if (nd_region->btt_seed == dev)
 607                         nd_region_create_btt_seed(nd_region);
 608                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 609                         nd_region_create_ns_seed(nd_region);
 610                 nvdimm_bus_unlock(dev);
 611         }
 612         if (is_nd_pfn(dev) && probe) {
 613                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 614
 615                 nd_region = to_nd_region(dev->parent);
 616                 nvdimm_bus_lock(dev);
 617                 if (nd_region->pfn_seed == dev)
 618                         nd_region_create_pfn_seed(nd_region);
 619                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 620                         nd_region_create_ns_seed(nd_region);
 621                 nvdimm_bus_unlock(dev);
 622         }
 623         if (is_nd_dax(dev) && probe) {
 624                 struct nd_dax *nd_dax = to_nd_dax(dev);
 625
 626                 nd_region = to_nd_region(dev->parent);
 627                 nvdimm_bus_lock(dev);
 628                 if (nd_region->dax_seed == dev)
 629                         nd_region_create_dax_seed(nd_region);
 630                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 631                         nd_region_create_ns_seed(nd_region);
 632                 nvdimm_bus_unlock(dev);
 633         }
 634 }
 635
 636 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 637 {
 638         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 639 }
 640
 641 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 642 {
 643         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 644 }
 645
 646 static ssize_t mappingN(struct device *dev, char *buf, int n)
 647 {
 648         struct nd_region *nd_region = to_nd_region(dev);
 649         struct nd_mapping *nd_mapping;
 650         struct nvdimm *nvdimm;
 651
 652         if (n >= nd_region->ndr_mappings)
 653                 return -ENXIO;
 654         nd_mapping = &nd_region->mapping[n];
 655         nvdimm = nd_mapping->nvdimm;
 656
 657         return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
 658                         nd_mapping->start, nd_mapping->size);
 659 }
 660
 661 #define REGION_MAPPING(idx) \
 662 static ssize_t mapping##idx##_show(struct device *dev,          \
 663                 struct device_attribute *attr, char *buf)       \
 664 {                                                               \
 665         return mappingN(dev, buf, idx);                         \
 666 }                                                               \
 667 static DEVICE_ATTR_RO(mapping##idx)
 668
 669 /*
 670  * 32 should be enough for a while, even in the presence of socket
 671  * interleave a 32-way interleave set is a degenerate case.
 672  */
 673 REGION_MAPPING(0);
 674 REGION_MAPPING(1);
 675 REGION_MAPPING(2);
 676 REGION_MAPPING(3);
 677 REGION_MAPPING(4);
 678 REGION_MAPPING(5);
 679 REGION_MAPPING(6);
 680 REGION_MAPPING(7);
 681 REGION_MAPPING(8);
 682 REGION_MAPPING(9);
 683 REGION_MAPPING(10);
 684 REGION_MAPPING(11);
 685 REGION_MAPPING(12);
 686 REGION_MAPPING(13);
 687 REGION_MAPPING(14);
 688 REGION_MAPPING(15);
 689 REGION_MAPPING(16);
 690 REGION_MAPPING(17);
 691 REGION_MAPPING(18);
 692 REGION_MAPPING(19);
 693 REGION_MAPPING(20);
 694 REGION_MAPPING(21);
 695 REGION_MAPPING(22);
 696 REGION_MAPPING(23);
 697 REGION_MAPPING(24);
 698 REGION_MAPPING(25);
 699 REGION_MAPPING(26);
 700 REGION_MAPPING(27);
 701 REGION_MAPPING(28);
 702 REGION_MAPPING(29);
 703 REGION_MAPPING(30);
 704 REGION_MAPPING(31);
 705
 706 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 707 {
 708         struct device *dev = container_of(kobj, struct device, kobj);
 709         struct nd_region *nd_region = to_nd_region(dev);
 710
 711         if (n < nd_region->ndr_mappings)
 712                 return a->mode;
 713         return 0;
 714 }
 715
 716 static struct attribute *mapping_attributes[] = {
 717         &dev_attr_mapping0.attr,
 718         &dev_attr_mapping1.attr,
 719         &dev_attr_mapping2.attr,
 720         &dev_attr_mapping3.attr,
 721         &dev_attr_mapping4.attr,
 722         &dev_attr_mapping5.attr,
 723         &dev_attr_mapping6.attr,
 724         &dev_attr_mapping7.attr,
 725         &dev_attr_mapping8.attr,
 726         &dev_attr_mapping9.attr,
 727         &dev_attr_mapping10.attr,
 728         &dev_attr_mapping11.attr,
 729         &dev_attr_mapping12.attr,
 730         &dev_attr_mapping13.attr,
 731         &dev_attr_mapping14.attr,
 732         &dev_attr_mapping15.attr,
 733         &dev_attr_mapping16.attr,
 734         &dev_attr_mapping17.attr,
 735         &dev_attr_mapping18.attr,
 736         &dev_attr_mapping19.attr,
 737         &dev_attr_mapping20.attr,
 738         &dev_attr_mapping21.attr,
 739         &dev_attr_mapping22.attr,
 740         &dev_attr_mapping23.attr,
 741         &dev_attr_mapping24.attr,
 742         &dev_attr_mapping25.attr,
 743         &dev_attr_mapping26.attr,
 744         &dev_attr_mapping27.attr,
 745         &dev_attr_mapping28.attr,
 746         &dev_attr_mapping29.attr,
 747         &dev_attr_mapping30.attr,
 748         &dev_attr_mapping31.attr,
 749         NULL,
 750 };
 751
 752 struct attribute_group nd_mapping_attribute_group = {
 753         .is_visible = mapping_visible,
 754         .attrs = mapping_attributes,
 755 };
 756 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 757
 758 int nd_blk_region_init(struct nd_region *nd_region)
 759 {
 760         struct device *dev = &nd_region->dev;
 761         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 762
 763         if (!is_nd_blk(dev))
 764                 return 0;
 765
 766         if (nd_region->ndr_mappings < 1) {
 767                 dev_err(dev, "invalid BLK region\n");
 768                 return -ENXIO;
 769         }
 770
 771         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 772 }
 773
 774 /**
 775  * nd_region_acquire_lane - allocate and lock a lane
 776  * @nd_region: region id and number of lanes possible
 777  *
 778  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 779  * We optimize for the common case where there are 256 lanes, one
 780  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 781  * this implementation assumes the cost of maintaining an allocator for
 782  * free lanes is on the order of the lock hold time, so it implements a
 783  * static lane = cpu % num_lanes mapping.
 784  *
 785  * In the case of a BTT instance on top of a BLK namespace a lane may be
 786  * acquired recursively.  We lock on the first instance.
 787  *
 788  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 789  * for the BTT metadata updates.
 790  */
 791 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 792 {
 793         unsigned int cpu, lane;
 794
 795         cpu = get_cpu();
 796         if (nd_region->num_lanes < nr_cpu_ids) {
 797                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 798
 799                 lane = cpu % nd_region->num_lanes;
 800                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 801                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 802                 if (ndl_count->count++ == 0)
 803                         spin_lock(&ndl_lock->lock);
 804         } else
 805                 lane = cpu;
 806
 807         return lane;
 808 }
 809 EXPORT_SYMBOL(nd_region_acquire_lane);
 810
 811 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 812 {
 813         if (nd_region->num_lanes < nr_cpu_ids) {
 814                 unsigned int cpu = get_cpu();
 815                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 816
 817                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 818                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 819                 if (--ndl_count->count == 0)
 820                         spin_unlock(&ndl_lock->lock);
 821                 put_cpu();
 822         }
 823         put_cpu();
 824 }
 825 EXPORT_SYMBOL(nd_region_release_lane);
 826
 827 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 828                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 829                 const char *caller)
 830 {
 831         struct nd_region *nd_region;
 832         struct device *dev;
 833         void *region_buf;
 834         unsigned int i;
 835         int ro = 0;
 836
 837         for (i = 0; i < ndr_desc->num_mappings; i++) {
 838                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 839                 struct nvdimm *nvdimm = mapping->nvdimm;
 840
 841                 if ((mapping->start | mapping->size) % SZ_4K) {
 842                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 843                                         caller, dev_name(&nvdimm->dev), i);
 844
 845                         return NULL;
 846                 }
 847
 848                 if (nvdimm->flags & NDD_UNARMED)
 849                         ro = 1;
 850         }
 851
 852         if (dev_type == &nd_blk_device_type) {
 853                 struct nd_blk_region_desc *ndbr_desc;
 854                 struct nd_blk_region *ndbr;
 855
 856                 ndbr_desc = to_blk_region_desc(ndr_desc);
 857                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 858                                 * ndr_desc->num_mappings,
 859                                 GFP_KERNEL);
 860                 if (ndbr) {
 861                         nd_region = &ndbr->nd_region;
 862                         ndbr->enable = ndbr_desc->enable;
 863                         ndbr->do_io = ndbr_desc->do_io;
 864                 }
 865                 region_buf = ndbr;
 866         } else {
 867                 nd_region = kzalloc(sizeof(struct nd_region)
 868                                 + sizeof(struct nd_mapping)
 869                                 * ndr_desc->num_mappings,
 870                                 GFP_KERNEL);
 871                 region_buf = nd_region;
 872         }
 873
 874         if (!region_buf)
 875                 return NULL;
 876         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 877         if (nd_region->id < 0)
 878                 goto err_id;
 879
 880         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 881         if (!nd_region->lane)
 882                 goto err_percpu;
 883
 884         for (i = 0; i < nr_cpu_ids; i++) {
 885                 struct nd_percpu_lane *ndl;
 886
 887                 ndl = per_cpu_ptr(nd_region->lane, i);
 888                 spin_lock_init(&ndl->lock);
 889                 ndl->count = 0;
 890         }
 891
 892         for (i = 0; i < ndr_desc->num_mappings; i++) {
 893                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 894                 struct nvdimm *nvdimm = mapping->nvdimm;
 895
 896                 nd_region->mapping[i].nvdimm = nvdimm;
 897                 nd_region->mapping[i].start = mapping->start;
 898                 nd_region->mapping[i].size = mapping->size;
 899                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
 900                 mutex_init(&nd_region->mapping[i].lock);
 901
 902                 get_device(&nvdimm->dev);
 903         }
 904         nd_region->ndr_mappings = ndr_desc->num_mappings;
 905         nd_region->provider_data = ndr_desc->provider_data;
 906         nd_region->nd_set = ndr_desc->nd_set;
 907         nd_region->num_lanes = ndr_desc->num_lanes;
 908         nd_region->flags = ndr_desc->flags;
 909         nd_region->ro = ro;
 910         nd_region->numa_node = ndr_desc->numa_node;
 911         ida_init(&nd_region->ns_ida);
 912         ida_init(&nd_region->btt_ida);
 913         ida_init(&nd_region->pfn_ida);
 914         ida_init(&nd_region->dax_ida);
 915         dev = &nd_region->dev;
 916         dev_set_name(dev, "region%d", nd_region->id);
 917         dev->parent = &nvdimm_bus->dev;
 918         dev->type = dev_type;
 919         dev->groups = ndr_desc->attr_groups;
 920         nd_region->ndr_size = resource_size(ndr_desc->res);
 921         nd_region->ndr_start = ndr_desc->res->start;
 922         nd_device_register(dev);
 923
 924         return nd_region;
 925
 926  err_percpu:
 927         ida_simple_remove(&region_ida, nd_region->id);
 928  err_id:
 929         kfree(region_buf);
 930         return NULL;
 931 }
 932
 933 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
 934                 struct nd_region_desc *ndr_desc)
 935 {
 936         ndr_desc->num_lanes = ND_MAX_LANES;
 937         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
 938                         __func__);
 939 }
 940 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
 941
 942 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
 943                 struct nd_region_desc *ndr_desc)
 944 {
 945         if (ndr_desc->num_mappings > 1)
 946                 return NULL;
 947         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
 948         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
 949                         __func__);
 950 }
 951 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
 952
 953 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
 954                 struct nd_region_desc *ndr_desc)
 955 {
 956         ndr_desc->num_lanes = ND_MAX_LANES;
 957         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
 958                         __func__);
 959 }
 960 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
 961
 962 /**
 963  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
 964  * @nd_region: blk or interleaved pmem region
 965  */
 966 void nvdimm_flush(struct nd_region *nd_region)
 967 {
 968         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
 969         int i, idx;
 970
 971         /*
 972          * Try to encourage some diversity in flush hint addresses
 973          * across cpus assuming a limited number of flush hints.
 974          */
 975         idx = this_cpu_read(flush_idx);
 976         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
 977
 978         /*
 979          * The first wmb() is needed to 'sfence' all previous writes
 980          * such that they are architecturally visible for the platform
 981          * buffer flush.  Note that we've already arranged for pmem
 982          * writes to avoid the cache via arch_memcpy_to_pmem().  The
 983          * final wmb() ensures ordering for the NVDIMM flush write.
 984          */
 985         wmb();
 986         for (i = 0; i < nd_region->ndr_mappings; i++)
 987                 if (ndrd_get_flush_wpq(ndrd, i, 0))
 988                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
 989         wmb();
 990 }
 991 EXPORT_SYMBOL_GPL(nvdimm_flush);
 992
 993 /**
 994  * nvdimm_has_flush - determine write flushing requirements
 995  * @nd_region: blk or interleaved pmem region
 996  *
 997  * Returns 1 if writes require flushing
 998  * Returns 0 if writes do not require flushing
 999  * Returns -ENXIO if flushing capability can not be determined
1000  */
1001 int nvdimm_has_flush(struct nd_region *nd_region)
1002 {
1003         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1004         int i;
1005
1006         /* no nvdimm == flushing capability unknown */
1007         if (nd_region->ndr_mappings == 0)
1008                 return -ENXIO;
1009
1010         for (i = 0; i < nd_region->ndr_mappings; i++)
1011                 /* flush hints present, flushing required */
1012                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1013                         return 1;
1014
1015         /*
1016          * The platform defines dimm devices without hints, assume
1017          * platform persistence mechanism like ADR
1018          */
1019         return 0;
1020 }
1021 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1022
1023 void __exit nd_region_devs_exit(void)
1024 {
1025         ida_destroy(&region_ida);
1026 }