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