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 bool is_nd_volatile(struct device *dev)
 172 {
 173         return dev ? dev->type == &nd_volatile_device_type : false;
 174 }
 175
 176 struct nd_region *to_nd_region(struct device *dev)
 177 {
 178         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
 179
 180         WARN_ON(dev->type->release != nd_region_release);
 181         return nd_region;
 182 }
 183 EXPORT_SYMBOL_GPL(to_nd_region);
 184
 185 struct device *nd_region_dev(struct nd_region *nd_region)
 186 {
 187         if (!nd_region)
 188                 return NULL;
 189         return &nd_region->dev;
 190 }
 191 EXPORT_SYMBOL_GPL(nd_region_dev);
 192
 193 struct nd_blk_region *to_nd_blk_region(struct device *dev)
 194 {
 195         struct nd_region *nd_region = to_nd_region(dev);
 196
 197         WARN_ON(!is_nd_blk(dev));
 198         return container_of(nd_region, struct nd_blk_region, nd_region);
 199 }
 200 EXPORT_SYMBOL_GPL(to_nd_blk_region);
 201
 202 void *nd_region_provider_data(struct nd_region *nd_region)
 203 {
 204         return nd_region->provider_data;
 205 }
 206 EXPORT_SYMBOL_GPL(nd_region_provider_data);
 207
 208 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
 209 {
 210         return ndbr->blk_provider_data;
 211 }
 212 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
 213
 214 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
 215 {
 216         ndbr->blk_provider_data = data;
 217 }
 218 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
 219
 220 /**
 221  * nd_region_to_nstype() - region to an integer namespace type
 222  * @nd_region: region-device to interrogate
 223  *
 224  * This is the 'nstype' attribute of a region as well, an input to the
 225  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
 226  * namespace devices with namespace drivers.
 227  */
 228 int nd_region_to_nstype(struct nd_region *nd_region)
 229 {
 230         if (is_memory(&nd_region->dev)) {
 231                 u16 i, alias;
 232
 233                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
 234                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 235                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 236
 237                         if (test_bit(NDD_ALIASING, &nvdimm->flags))
 238                                 alias++;
 239                 }
 240                 if (alias)
 241                         return ND_DEVICE_NAMESPACE_PMEM;
 242                 else
 243                         return ND_DEVICE_NAMESPACE_IO;
 244         } else if (is_nd_blk(&nd_region->dev)) {
 245                 return ND_DEVICE_NAMESPACE_BLK;
 246         }
 247
 248         return 0;
 249 }
 250 EXPORT_SYMBOL(nd_region_to_nstype);
 251
 252 static ssize_t size_show(struct device *dev,
 253                 struct device_attribute *attr, char *buf)
 254 {
 255         struct nd_region *nd_region = to_nd_region(dev);
 256         unsigned long long size = 0;
 257
 258         if (is_memory(dev)) {
 259                 size = nd_region->ndr_size;
 260         } else if (nd_region->ndr_mappings == 1) {
 261                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 262
 263                 size = nd_mapping->size;
 264         }
 265
 266         return sprintf(buf, "%llu\n", size);
 267 }
 268 static DEVICE_ATTR_RO(size);
 269
 270 static ssize_t deep_flush_show(struct device *dev,
 271                 struct device_attribute *attr, char *buf)
 272 {
 273         struct nd_region *nd_region = to_nd_region(dev);
 274
 275         /*
 276          * NOTE: in the nvdimm_has_flush() error case this attribute is
 277          * not visible.
 278          */
 279         return sprintf(buf, "%d\n", nvdimm_has_flush(nd_region));
 280 }
 281
 282 static ssize_t deep_flush_store(struct device *dev, struct device_attribute *attr,
 283                 const char *buf, size_t len)
 284 {
 285         bool flush;
 286         int rc = strtobool(buf, &flush);
 287         struct nd_region *nd_region = to_nd_region(dev);
 288
 289         if (rc)
 290                 return rc;
 291         if (!flush)
 292                 return -EINVAL;
 293         nvdimm_flush(nd_region);
 294
 295         return len;
 296 }
 297 static DEVICE_ATTR_RW(deep_flush);
 298
 299 static ssize_t mappings_show(struct device *dev,
 300                 struct device_attribute *attr, char *buf)
 301 {
 302         struct nd_region *nd_region = to_nd_region(dev);
 303
 304         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 305 }
 306 static DEVICE_ATTR_RO(mappings);
 307
 308 static ssize_t nstype_show(struct device *dev,
 309                 struct device_attribute *attr, char *buf)
 310 {
 311         struct nd_region *nd_region = to_nd_region(dev);
 312
 313         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 314 }
 315 static DEVICE_ATTR_RO(nstype);
 316
 317 static ssize_t set_cookie_show(struct device *dev,
 318                 struct device_attribute *attr, char *buf)
 319 {
 320         struct nd_region *nd_region = to_nd_region(dev);
 321         struct nd_interleave_set *nd_set = nd_region->nd_set;
 322         ssize_t rc = 0;
 323
 324         if (is_memory(dev) && nd_set)
 325                 /* pass, should be precluded by region_visible */;
 326         else
 327                 return -ENXIO;
 328
 329         /*
 330          * The cookie to show depends on which specification of the
 331          * labels we are using. If there are not labels then default to
 332          * the v1.1 namespace label cookie definition. To read all this
 333          * data we need to wait for probing to settle.
 334          */
 335         device_lock(dev);
 336         nvdimm_bus_lock(dev);
 337         wait_nvdimm_bus_probe_idle(dev);
 338         if (nd_region->ndr_mappings) {
 339                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 340                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 341
 342                 if (ndd) {
 343                         struct nd_namespace_index *nsindex;
 344
 345                         nsindex = to_namespace_index(ndd, ndd->ns_current);
 346                         rc = sprintf(buf, "%#llx\n",
 347                                         nd_region_interleave_set_cookie(nd_region,
 348                                                 nsindex));
 349                 }
 350         }
 351         nvdimm_bus_unlock(dev);
 352         device_unlock(dev);
 353
 354         if (rc)
 355                 return rc;
 356         return sprintf(buf, "%#llx\n", nd_set->cookie1);
 357 }
 358 static DEVICE_ATTR_RO(set_cookie);
 359
 360 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 361 {
 362         resource_size_t blk_max_overlap = 0, available, overlap;
 363         int i;
 364
 365         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 366
 367  retry:
 368         available = 0;
 369         overlap = blk_max_overlap;
 370         for (i = 0; i < nd_region->ndr_mappings; i++) {
 371                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 372                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 373
 374                 /* if a dimm is disabled the available capacity is zero */
 375                 if (!ndd)
 376                         return 0;
 377
 378                 if (is_memory(&nd_region->dev)) {
 379                         available += nd_pmem_available_dpa(nd_region,
 380                                         nd_mapping, &overlap);
 381                         if (overlap > blk_max_overlap) {
 382                                 blk_max_overlap = overlap;
 383                                 goto retry;
 384                         }
 385                 } else if (is_nd_blk(&nd_region->dev))
 386                         available += nd_blk_available_dpa(nd_region);
 387         }
 388
 389         return available;
 390 }
 391
 392 static ssize_t available_size_show(struct device *dev,
 393                 struct device_attribute *attr, char *buf)
 394 {
 395         struct nd_region *nd_region = to_nd_region(dev);
 396         unsigned long long available = 0;
 397
 398         /*
 399          * Flush in-flight updates and grab a snapshot of the available
 400          * size.  Of course, this value is potentially invalidated the
 401          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 402          * problem to not race itself.
 403          */
 404         nvdimm_bus_lock(dev);
 405         wait_nvdimm_bus_probe_idle(dev);
 406         available = nd_region_available_dpa(nd_region);
 407         nvdimm_bus_unlock(dev);
 408
 409         return sprintf(buf, "%llu\n", available);
 410 }
 411 static DEVICE_ATTR_RO(available_size);
 412
 413 static ssize_t init_namespaces_show(struct device *dev,
 414                 struct device_attribute *attr, char *buf)
 415 {
 416         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 417         ssize_t rc;
 418
 419         nvdimm_bus_lock(dev);
 420         if (ndrd)
 421                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 422         else
 423                 rc = -ENXIO;
 424         nvdimm_bus_unlock(dev);
 425
 426         return rc;
 427 }
 428 static DEVICE_ATTR_RO(init_namespaces);
 429
 430 static ssize_t namespace_seed_show(struct device *dev,
 431                 struct device_attribute *attr, char *buf)
 432 {
 433         struct nd_region *nd_region = to_nd_region(dev);
 434         ssize_t rc;
 435
 436         nvdimm_bus_lock(dev);
 437         if (nd_region->ns_seed)
 438                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 439         else
 440                 rc = sprintf(buf, "\n");
 441         nvdimm_bus_unlock(dev);
 442         return rc;
 443 }
 444 static DEVICE_ATTR_RO(namespace_seed);
 445
 446 static ssize_t btt_seed_show(struct device *dev,
 447                 struct device_attribute *attr, char *buf)
 448 {
 449         struct nd_region *nd_region = to_nd_region(dev);
 450         ssize_t rc;
 451
 452         nvdimm_bus_lock(dev);
 453         if (nd_region->btt_seed)
 454                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 455         else
 456                 rc = sprintf(buf, "\n");
 457         nvdimm_bus_unlock(dev);
 458
 459         return rc;
 460 }
 461 static DEVICE_ATTR_RO(btt_seed);
 462
 463 static ssize_t pfn_seed_show(struct device *dev,
 464                 struct device_attribute *attr, char *buf)
 465 {
 466         struct nd_region *nd_region = to_nd_region(dev);
 467         ssize_t rc;
 468
 469         nvdimm_bus_lock(dev);
 470         if (nd_region->pfn_seed)
 471                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 472         else
 473                 rc = sprintf(buf, "\n");
 474         nvdimm_bus_unlock(dev);
 475
 476         return rc;
 477 }
 478 static DEVICE_ATTR_RO(pfn_seed);
 479
 480 static ssize_t dax_seed_show(struct device *dev,
 481                 struct device_attribute *attr, char *buf)
 482 {
 483         struct nd_region *nd_region = to_nd_region(dev);
 484         ssize_t rc;
 485
 486         nvdimm_bus_lock(dev);
 487         if (nd_region->dax_seed)
 488                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 489         else
 490                 rc = sprintf(buf, "\n");
 491         nvdimm_bus_unlock(dev);
 492
 493         return rc;
 494 }
 495 static DEVICE_ATTR_RO(dax_seed);
 496
 497 static ssize_t read_only_show(struct device *dev,
 498                 struct device_attribute *attr, char *buf)
 499 {
 500         struct nd_region *nd_region = to_nd_region(dev);
 501
 502         return sprintf(buf, "%d\n", nd_region->ro);
 503 }
 504
 505 static ssize_t read_only_store(struct device *dev,
 506                 struct device_attribute *attr, const char *buf, size_t len)
 507 {
 508         bool ro;
 509         int rc = strtobool(buf, &ro);
 510         struct nd_region *nd_region = to_nd_region(dev);
 511
 512         if (rc)
 513                 return rc;
 514
 515         nd_region->ro = ro;
 516         return len;
 517 }
 518 static DEVICE_ATTR_RW(read_only);
 519
 520 static ssize_t region_badblocks_show(struct device *dev,
 521                 struct device_attribute *attr, char *buf)
 522 {
 523         struct nd_region *nd_region = to_nd_region(dev);
 524
 525         return badblocks_show(&nd_region->bb, buf, 0);
 526 }
 527
 528 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 529
 530 static ssize_t resource_show(struct device *dev,
 531                 struct device_attribute *attr, char *buf)
 532 {
 533         struct nd_region *nd_region = to_nd_region(dev);
 534
 535         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 536 }
 537 static DEVICE_ATTR_RO(resource);
 538
 539 static ssize_t persistence_domain_show(struct device *dev,
 540                 struct device_attribute *attr, char *buf)
 541 {
 542         struct nd_region *nd_region = to_nd_region(dev);
 543
 544         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
 545                 return sprintf(buf, "cpu_cache\n");
 546         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
 547                 return sprintf(buf, "memory_controller\n");
 548         else
 549                 return sprintf(buf, "\n");
 550 }
 551 static DEVICE_ATTR_RO(persistence_domain);
 552
 553 static struct attribute *nd_region_attributes[] = {
 554         &dev_attr_size.attr,
 555         &dev_attr_nstype.attr,
 556         &dev_attr_mappings.attr,
 557         &dev_attr_btt_seed.attr,
 558         &dev_attr_pfn_seed.attr,
 559         &dev_attr_dax_seed.attr,
 560         &dev_attr_deep_flush.attr,
 561         &dev_attr_read_only.attr,
 562         &dev_attr_set_cookie.attr,
 563         &dev_attr_available_size.attr,
 564         &dev_attr_namespace_seed.attr,
 565         &dev_attr_init_namespaces.attr,
 566         &dev_attr_badblocks.attr,
 567         &dev_attr_resource.attr,
 568         &dev_attr_persistence_domain.attr,
 569         NULL,
 570 };
 571
 572 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 573 {
 574         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 575         struct nd_region *nd_region = to_nd_region(dev);
 576         struct nd_interleave_set *nd_set = nd_region->nd_set;
 577         int type = nd_region_to_nstype(nd_region);
 578
 579         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
 580                 return 0;
 581
 582         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
 583                 return 0;
 584
 585         if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
 586                 return 0;
 587
 588         if (a == &dev_attr_resource.attr) {
 589                 if (is_nd_pmem(dev))
 590                         return 0400;
 591                 else
 592                         return 0;
 593         }
 594
 595         if (a == &dev_attr_deep_flush.attr) {
 596                 int has_flush = nvdimm_has_flush(nd_region);
 597
 598                 if (has_flush == 1)
 599                         return a->mode;
 600                 else if (has_flush == 0)
 601                         return 0444;
 602                 else
 603                         return 0;
 604         }
 605
 606         if (a == &dev_attr_persistence_domain.attr) {
 607                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
 608                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
 609                         return 0;
 610                 return a->mode;
 611         }
 612
 613         if (a != &dev_attr_set_cookie.attr
 614                         && a != &dev_attr_available_size.attr)
 615                 return a->mode;
 616
 617         if ((type == ND_DEVICE_NAMESPACE_PMEM
 618                                 || type == ND_DEVICE_NAMESPACE_BLK)
 619                         && a == &dev_attr_available_size.attr)
 620                 return a->mode;
 621         else if (is_memory(dev) && nd_set)
 622                 return a->mode;
 623
 624         return 0;
 625 }
 626
 627 struct attribute_group nd_region_attribute_group = {
 628         .attrs = nd_region_attributes,
 629         .is_visible = region_visible,
 630 };
 631 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 632
 633 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 634                 struct nd_namespace_index *nsindex)
 635 {
 636         struct nd_interleave_set *nd_set = nd_region->nd_set;
 637
 638         if (!nd_set)
 639                 return 0;
 640
 641         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 642                         && __le16_to_cpu(nsindex->minor) == 1)
 643                 return nd_set->cookie1;
 644         return nd_set->cookie2;
 645 }
 646
 647 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 648 {
 649         struct nd_interleave_set *nd_set = nd_region->nd_set;
 650
 651         if (nd_set)
 652                 return nd_set->altcookie;
 653         return 0;
 654 }
 655
 656 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 657 {
 658         struct nd_label_ent *label_ent, *e;
 659
 660         lockdep_assert_held(&nd_mapping->lock);
 661         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 662                 list_del(&label_ent->list);
 663                 kfree(label_ent);
 664         }
 665 }
 666
 667 /*
 668  * Upon successful probe/remove, take/release a reference on the
 669  * associated interleave set (if present), and plant new btt + namespace
 670  * seeds.  Also, on the removal of a BLK region, notify the provider to
 671  * disable the region.
 672  */
 673 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 674                 struct device *dev, bool probe)
 675 {
 676         struct nd_region *nd_region;
 677
 678         if (!probe && is_nd_region(dev)) {
 679                 int i;
 680
 681                 nd_region = to_nd_region(dev);
 682                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 683                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 684                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 685                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 686
 687                         mutex_lock(&nd_mapping->lock);
 688                         nd_mapping_free_labels(nd_mapping);
 689                         mutex_unlock(&nd_mapping->lock);
 690
 691                         put_ndd(ndd);
 692                         nd_mapping->ndd = NULL;
 693                         if (ndd)
 694                                 atomic_dec(&nvdimm->busy);
 695                 }
 696         }
 697         if (dev->parent && is_nd_region(dev->parent) && probe) {
 698                 nd_region = to_nd_region(dev->parent);
 699                 nvdimm_bus_lock(dev);
 700                 if (nd_region->ns_seed == dev)
 701                         nd_region_create_ns_seed(nd_region);
 702                 nvdimm_bus_unlock(dev);
 703         }
 704         if (is_nd_btt(dev) && probe) {
 705                 struct nd_btt *nd_btt = to_nd_btt(dev);
 706
 707                 nd_region = to_nd_region(dev->parent);
 708                 nvdimm_bus_lock(dev);
 709                 if (nd_region->btt_seed == dev)
 710                         nd_region_create_btt_seed(nd_region);
 711                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 712                         nd_region_create_ns_seed(nd_region);
 713                 nvdimm_bus_unlock(dev);
 714         }
 715         if (is_nd_pfn(dev) && probe) {
 716                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 717
 718                 nd_region = to_nd_region(dev->parent);
 719                 nvdimm_bus_lock(dev);
 720                 if (nd_region->pfn_seed == dev)
 721                         nd_region_create_pfn_seed(nd_region);
 722                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 723                         nd_region_create_ns_seed(nd_region);
 724                 nvdimm_bus_unlock(dev);
 725         }
 726         if (is_nd_dax(dev) && probe) {
 727                 struct nd_dax *nd_dax = to_nd_dax(dev);
 728
 729                 nd_region = to_nd_region(dev->parent);
 730                 nvdimm_bus_lock(dev);
 731                 if (nd_region->dax_seed == dev)
 732                         nd_region_create_dax_seed(nd_region);
 733                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 734                         nd_region_create_ns_seed(nd_region);
 735                 nvdimm_bus_unlock(dev);
 736         }
 737 }
 738
 739 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 740 {
 741         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 742 }
 743
 744 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 745 {
 746         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 747 }
 748
 749 static ssize_t mappingN(struct device *dev, char *buf, int n)
 750 {
 751         struct nd_region *nd_region = to_nd_region(dev);
 752         struct nd_mapping *nd_mapping;
 753         struct nvdimm *nvdimm;
 754
 755         if (n >= nd_region->ndr_mappings)
 756                 return -ENXIO;
 757         nd_mapping = &nd_region->mapping[n];
 758         nvdimm = nd_mapping->nvdimm;
 759
 760         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
 761                         nd_mapping->start, nd_mapping->size,
 762                         nd_mapping->position);
 763 }
 764
 765 #define REGION_MAPPING(idx) \
 766 static ssize_t mapping##idx##_show(struct device *dev,          \
 767                 struct device_attribute *attr, char *buf)       \
 768 {                                                               \
 769         return mappingN(dev, buf, idx);                         \
 770 }                                                               \
 771 static DEVICE_ATTR_RO(mapping##idx)
 772
 773 /*
 774  * 32 should be enough for a while, even in the presence of socket
 775  * interleave a 32-way interleave set is a degenerate case.
 776  */
 777 REGION_MAPPING(0);
 778 REGION_MAPPING(1);
 779 REGION_MAPPING(2);
 780 REGION_MAPPING(3);
 781 REGION_MAPPING(4);
 782 REGION_MAPPING(5);
 783 REGION_MAPPING(6);
 784 REGION_MAPPING(7);
 785 REGION_MAPPING(8);
 786 REGION_MAPPING(9);
 787 REGION_MAPPING(10);
 788 REGION_MAPPING(11);
 789 REGION_MAPPING(12);
 790 REGION_MAPPING(13);
 791 REGION_MAPPING(14);
 792 REGION_MAPPING(15);
 793 REGION_MAPPING(16);
 794 REGION_MAPPING(17);
 795 REGION_MAPPING(18);
 796 REGION_MAPPING(19);
 797 REGION_MAPPING(20);
 798 REGION_MAPPING(21);
 799 REGION_MAPPING(22);
 800 REGION_MAPPING(23);
 801 REGION_MAPPING(24);
 802 REGION_MAPPING(25);
 803 REGION_MAPPING(26);
 804 REGION_MAPPING(27);
 805 REGION_MAPPING(28);
 806 REGION_MAPPING(29);
 807 REGION_MAPPING(30);
 808 REGION_MAPPING(31);
 809
 810 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 811 {
 812         struct device *dev = container_of(kobj, struct device, kobj);
 813         struct nd_region *nd_region = to_nd_region(dev);
 814
 815         if (n < nd_region->ndr_mappings)
 816                 return a->mode;
 817         return 0;
 818 }
 819
 820 static struct attribute *mapping_attributes[] = {
 821         &dev_attr_mapping0.attr,
 822         &dev_attr_mapping1.attr,
 823         &dev_attr_mapping2.attr,
 824         &dev_attr_mapping3.attr,
 825         &dev_attr_mapping4.attr,
 826         &dev_attr_mapping5.attr,
 827         &dev_attr_mapping6.attr,
 828         &dev_attr_mapping7.attr,
 829         &dev_attr_mapping8.attr,
 830         &dev_attr_mapping9.attr,
 831         &dev_attr_mapping10.attr,
 832         &dev_attr_mapping11.attr,
 833         &dev_attr_mapping12.attr,
 834         &dev_attr_mapping13.attr,
 835         &dev_attr_mapping14.attr,
 836         &dev_attr_mapping15.attr,
 837         &dev_attr_mapping16.attr,
 838         &dev_attr_mapping17.attr,
 839         &dev_attr_mapping18.attr,
 840         &dev_attr_mapping19.attr,
 841         &dev_attr_mapping20.attr,
 842         &dev_attr_mapping21.attr,
 843         &dev_attr_mapping22.attr,
 844         &dev_attr_mapping23.attr,
 845         &dev_attr_mapping24.attr,
 846         &dev_attr_mapping25.attr,
 847         &dev_attr_mapping26.attr,
 848         &dev_attr_mapping27.attr,
 849         &dev_attr_mapping28.attr,
 850         &dev_attr_mapping29.attr,
 851         &dev_attr_mapping30.attr,
 852         &dev_attr_mapping31.attr,
 853         NULL,
 854 };
 855
 856 struct attribute_group nd_mapping_attribute_group = {
 857         .is_visible = mapping_visible,
 858         .attrs = mapping_attributes,
 859 };
 860 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 861
 862 int nd_blk_region_init(struct nd_region *nd_region)
 863 {
 864         struct device *dev = &nd_region->dev;
 865         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 866
 867         if (!is_nd_blk(dev))
 868                 return 0;
 869
 870         if (nd_region->ndr_mappings < 1) {
 871                 dev_dbg(dev, "invalid BLK region\n");
 872                 return -ENXIO;
 873         }
 874
 875         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 876 }
 877
 878 /**
 879  * nd_region_acquire_lane - allocate and lock a lane
 880  * @nd_region: region id and number of lanes possible
 881  *
 882  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 883  * We optimize for the common case where there are 256 lanes, one
 884  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 885  * this implementation assumes the cost of maintaining an allocator for
 886  * free lanes is on the order of the lock hold time, so it implements a
 887  * static lane = cpu % num_lanes mapping.
 888  *
 889  * In the case of a BTT instance on top of a BLK namespace a lane may be
 890  * acquired recursively.  We lock on the first instance.
 891  *
 892  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 893  * for the BTT metadata updates.
 894  */
 895 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 896 {
 897         unsigned int cpu, lane;
 898
 899         cpu = get_cpu();
 900         if (nd_region->num_lanes < nr_cpu_ids) {
 901                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 902
 903                 lane = cpu % nd_region->num_lanes;
 904                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 905                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 906                 if (ndl_count->count++ == 0)
 907                         spin_lock(&ndl_lock->lock);
 908         } else
 909                 lane = cpu;
 910
 911         return lane;
 912 }
 913 EXPORT_SYMBOL(nd_region_acquire_lane);
 914
 915 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 916 {
 917         if (nd_region->num_lanes < nr_cpu_ids) {
 918                 unsigned int cpu = get_cpu();
 919                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 920
 921                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 922                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 923                 if (--ndl_count->count == 0)
 924                         spin_unlock(&ndl_lock->lock);
 925                 put_cpu();
 926         }
 927         put_cpu();
 928 }
 929 EXPORT_SYMBOL(nd_region_release_lane);
 930
 931 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 932                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 933                 const char *caller)
 934 {
 935         struct nd_region *nd_region;
 936         struct device *dev;
 937         void *region_buf;
 938         unsigned int i;
 939         int ro = 0;
 940
 941         for (i = 0; i < ndr_desc->num_mappings; i++) {
 942                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 943                 struct nvdimm *nvdimm = mapping->nvdimm;
 944
 945                 if ((mapping->start | mapping->size) % SZ_4K) {
 946                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 947                                         caller, dev_name(&nvdimm->dev), i);
 948
 949                         return NULL;
 950                 }
 951
 952                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
 953                         ro = 1;
 954         }
 955
 956         if (dev_type == &nd_blk_device_type) {
 957                 struct nd_blk_region_desc *ndbr_desc;
 958                 struct nd_blk_region *ndbr;
 959
 960                 ndbr_desc = to_blk_region_desc(ndr_desc);
 961                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 962                                 * ndr_desc->num_mappings,
 963                                 GFP_KERNEL);
 964                 if (ndbr) {
 965                         nd_region = &ndbr->nd_region;
 966                         ndbr->enable = ndbr_desc->enable;
 967                         ndbr->do_io = ndbr_desc->do_io;
 968                 }
 969                 region_buf = ndbr;
 970         } else {
 971                 nd_region = kzalloc(sizeof(struct nd_region)
 972                                 + sizeof(struct nd_mapping)
 973                                 * ndr_desc->num_mappings,
 974                                 GFP_KERNEL);
 975                 region_buf = nd_region;
 976         }
 977
 978         if (!region_buf)
 979                 return NULL;
 980         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 981         if (nd_region->id < 0)
 982                 goto err_id;
 983
 984         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 985         if (!nd_region->lane)
 986                 goto err_percpu;
 987
 988         for (i = 0; i < nr_cpu_ids; i++) {
 989                 struct nd_percpu_lane *ndl;
 990
 991                 ndl = per_cpu_ptr(nd_region->lane, i);
 992                 spin_lock_init(&ndl->lock);
 993                 ndl->count = 0;
 994         }
 995
 996         for (i = 0; i < ndr_desc->num_mappings; i++) {
 997                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 998                 struct nvdimm *nvdimm = mapping->nvdimm;
 999
1000                 nd_region->mapping[i].nvdimm = nvdimm;
1001                 nd_region->mapping[i].start = mapping->start;
1002                 nd_region->mapping[i].size = mapping->size;
1003                 nd_region->mapping[i].position = mapping->position;
1004                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1005                 mutex_init(&nd_region->mapping[i].lock);
1006
1007                 get_device(&nvdimm->dev);
1008         }
1009         nd_region->ndr_mappings = ndr_desc->num_mappings;
1010         nd_region->provider_data = ndr_desc->provider_data;
1011         nd_region->nd_set = ndr_desc->nd_set;
1012         nd_region->num_lanes = ndr_desc->num_lanes;
1013         nd_region->flags = ndr_desc->flags;
1014         nd_region->ro = ro;
1015         nd_region->numa_node = ndr_desc->numa_node;
1016         ida_init(&nd_region->ns_ida);
1017         ida_init(&nd_region->btt_ida);
1018         ida_init(&nd_region->pfn_ida);
1019         ida_init(&nd_region->dax_ida);
1020         dev = &nd_region->dev;
1021         dev_set_name(dev, "region%d", nd_region->id);
1022         dev->parent = &nvdimm_bus->dev;
1023         dev->type = dev_type;
1024         dev->groups = ndr_desc->attr_groups;
1025         dev->of_node = ndr_desc->of_node;
1026         nd_region->ndr_size = resource_size(ndr_desc->res);
1027         nd_region->ndr_start = ndr_desc->res->start;
1028         nd_device_register(dev);
1029
1030         return nd_region;
1031
1032  err_percpu:
1033         ida_simple_remove(&region_ida, nd_region->id);
1034  err_id:
1035         kfree(region_buf);
1036         return NULL;
1037 }
1038
1039 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1040                 struct nd_region_desc *ndr_desc)
1041 {
1042         ndr_desc->num_lanes = ND_MAX_LANES;
1043         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1044                         __func__);
1045 }
1046 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1047
1048 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1049                 struct nd_region_desc *ndr_desc)
1050 {
1051         if (ndr_desc->num_mappings > 1)
1052                 return NULL;
1053         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1054         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1055                         __func__);
1056 }
1057 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1058
1059 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1060                 struct nd_region_desc *ndr_desc)
1061 {
1062         ndr_desc->num_lanes = ND_MAX_LANES;
1063         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1064                         __func__);
1065 }
1066 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1067
1068 /**
1069  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1070  * @nd_region: blk or interleaved pmem region
1071  */
1072 void nvdimm_flush(struct nd_region *nd_region)
1073 {
1074         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1075         int i, idx;
1076
1077         /*
1078          * Try to encourage some diversity in flush hint addresses
1079          * across cpus assuming a limited number of flush hints.
1080          */
1081         idx = this_cpu_read(flush_idx);
1082         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1083
1084         /*
1085          * The first wmb() is needed to 'sfence' all previous writes
1086          * such that they are architecturally visible for the platform
1087          * buffer flush.  Note that we've already arranged for pmem
1088          * writes to avoid the cache via memcpy_flushcache().  The final
1089          * wmb() ensures ordering for the NVDIMM flush write.
1090          */
1091         wmb();
1092         for (i = 0; i < nd_region->ndr_mappings; i++)
1093                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1094                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1095         wmb();
1096 }
1097 EXPORT_SYMBOL_GPL(nvdimm_flush);
1098
1099 /**
1100  * nvdimm_has_flush - determine write flushing requirements
1101  * @nd_region: blk or interleaved pmem region
1102  *
1103  * Returns 1 if writes require flushing
1104  * Returns 0 if writes do not require flushing
1105  * Returns -ENXIO if flushing capability can not be determined
1106  */
1107 int nvdimm_has_flush(struct nd_region *nd_region)
1108 {
1109         int i;
1110
1111         /* no nvdimm or pmem api == flushing capability unknown */
1112         if (nd_region->ndr_mappings == 0
1113                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1114                 return -ENXIO;
1115
1116         for (i = 0; i < nd_region->ndr_mappings; i++) {
1117                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1118                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1119
1120                 /* flush hints present / available */
1121                 if (nvdimm->num_flush)
1122                         return 1;
1123         }
1124
1125         /*
1126          * The platform defines dimm devices without hints, assume
1127          * platform persistence mechanism like ADR
1128          */
1129         return 0;
1130 }
1131 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1132
1133 int nvdimm_has_cache(struct nd_region *nd_region)
1134 {
1135         return is_nd_pmem(&nd_region->dev) &&
1136                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1137 }
1138 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1139
1140 void __exit nd_region_devs_exit(void)
1141 {
1142         ida_destroy(&region_ida);
1143 }