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 resource_size_t nd_region_allocatable_dpa(struct nd_region *nd_region)
 393 {
 394         resource_size_t available = 0;
 395         int i;
 396
 397         if (is_memory(&nd_region->dev))
 398                 available = PHYS_ADDR_MAX;
 399
 400         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 401         for (i = 0; i < nd_region->ndr_mappings; i++) {
 402                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 403
 404                 if (is_memory(&nd_region->dev))
 405                         available = min(available,
 406                                         nd_pmem_max_contiguous_dpa(nd_region,
 407                                                                    nd_mapping));
 408                 else if (is_nd_blk(&nd_region->dev))
 409                         available += nd_blk_available_dpa(nd_region);
 410         }
 411         if (is_memory(&nd_region->dev))
 412                 return available * nd_region->ndr_mappings;
 413         return available;
 414 }
 415
 416 static ssize_t available_size_show(struct device *dev,
 417                 struct device_attribute *attr, char *buf)
 418 {
 419         struct nd_region *nd_region = to_nd_region(dev);
 420         unsigned long long available = 0;
 421
 422         /*
 423          * Flush in-flight updates and grab a snapshot of the available
 424          * size.  Of course, this value is potentially invalidated the
 425          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 426          * problem to not race itself.
 427          */
 428         nvdimm_bus_lock(dev);
 429         wait_nvdimm_bus_probe_idle(dev);
 430         available = nd_region_available_dpa(nd_region);
 431         nvdimm_bus_unlock(dev);
 432
 433         return sprintf(buf, "%llu\n", available);
 434 }
 435 static DEVICE_ATTR_RO(available_size);
 436
 437 static ssize_t max_available_extent_show(struct device *dev,
 438                 struct device_attribute *attr, char *buf)
 439 {
 440         struct nd_region *nd_region = to_nd_region(dev);
 441         unsigned long long available = 0;
 442
 443         nvdimm_bus_lock(dev);
 444         wait_nvdimm_bus_probe_idle(dev);
 445         available = nd_region_allocatable_dpa(nd_region);
 446         nvdimm_bus_unlock(dev);
 447
 448         return sprintf(buf, "%llu\n", available);
 449 }
 450 static DEVICE_ATTR_RO(max_available_extent);
 451
 452 static ssize_t init_namespaces_show(struct device *dev,
 453                 struct device_attribute *attr, char *buf)
 454 {
 455         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 456         ssize_t rc;
 457
 458         nvdimm_bus_lock(dev);
 459         if (ndrd)
 460                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 461         else
 462                 rc = -ENXIO;
 463         nvdimm_bus_unlock(dev);
 464
 465         return rc;
 466 }
 467 static DEVICE_ATTR_RO(init_namespaces);
 468
 469 static ssize_t namespace_seed_show(struct device *dev,
 470                 struct device_attribute *attr, char *buf)
 471 {
 472         struct nd_region *nd_region = to_nd_region(dev);
 473         ssize_t rc;
 474
 475         nvdimm_bus_lock(dev);
 476         if (nd_region->ns_seed)
 477                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 478         else
 479                 rc = sprintf(buf, "\n");
 480         nvdimm_bus_unlock(dev);
 481         return rc;
 482 }
 483 static DEVICE_ATTR_RO(namespace_seed);
 484
 485 static ssize_t btt_seed_show(struct device *dev,
 486                 struct device_attribute *attr, char *buf)
 487 {
 488         struct nd_region *nd_region = to_nd_region(dev);
 489         ssize_t rc;
 490
 491         nvdimm_bus_lock(dev);
 492         if (nd_region->btt_seed)
 493                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 494         else
 495                 rc = sprintf(buf, "\n");
 496         nvdimm_bus_unlock(dev);
 497
 498         return rc;
 499 }
 500 static DEVICE_ATTR_RO(btt_seed);
 501
 502 static ssize_t pfn_seed_show(struct device *dev,
 503                 struct device_attribute *attr, char *buf)
 504 {
 505         struct nd_region *nd_region = to_nd_region(dev);
 506         ssize_t rc;
 507
 508         nvdimm_bus_lock(dev);
 509         if (nd_region->pfn_seed)
 510                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 511         else
 512                 rc = sprintf(buf, "\n");
 513         nvdimm_bus_unlock(dev);
 514
 515         return rc;
 516 }
 517 static DEVICE_ATTR_RO(pfn_seed);
 518
 519 static ssize_t dax_seed_show(struct device *dev,
 520                 struct device_attribute *attr, char *buf)
 521 {
 522         struct nd_region *nd_region = to_nd_region(dev);
 523         ssize_t rc;
 524
 525         nvdimm_bus_lock(dev);
 526         if (nd_region->dax_seed)
 527                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 528         else
 529                 rc = sprintf(buf, "\n");
 530         nvdimm_bus_unlock(dev);
 531
 532         return rc;
 533 }
 534 static DEVICE_ATTR_RO(dax_seed);
 535
 536 static ssize_t read_only_show(struct device *dev,
 537                 struct device_attribute *attr, char *buf)
 538 {
 539         struct nd_region *nd_region = to_nd_region(dev);
 540
 541         return sprintf(buf, "%d\n", nd_region->ro);
 542 }
 543
 544 static ssize_t read_only_store(struct device *dev,
 545                 struct device_attribute *attr, const char *buf, size_t len)
 546 {
 547         bool ro;
 548         int rc = strtobool(buf, &ro);
 549         struct nd_region *nd_region = to_nd_region(dev);
 550
 551         if (rc)
 552                 return rc;
 553
 554         nd_region->ro = ro;
 555         return len;
 556 }
 557 static DEVICE_ATTR_RW(read_only);
 558
 559 static ssize_t region_badblocks_show(struct device *dev,
 560                 struct device_attribute *attr, char *buf)
 561 {
 562         struct nd_region *nd_region = to_nd_region(dev);
 563         ssize_t rc;
 564
 565         device_lock(dev);
 566         if (dev->driver)
 567                 rc = badblocks_show(&nd_region->bb, buf, 0);
 568         else
 569                 rc = -ENXIO;
 570         device_unlock(dev);
 571
 572         return rc;
 573 }
 574 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 575
 576 static ssize_t resource_show(struct device *dev,
 577                 struct device_attribute *attr, char *buf)
 578 {
 579         struct nd_region *nd_region = to_nd_region(dev);
 580
 581         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 582 }
 583 static DEVICE_ATTR_RO(resource);
 584
 585 static ssize_t persistence_domain_show(struct device *dev,
 586                 struct device_attribute *attr, char *buf)
 587 {
 588         struct nd_region *nd_region = to_nd_region(dev);
 589
 590         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
 591                 return sprintf(buf, "cpu_cache\n");
 592         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
 593                 return sprintf(buf, "memory_controller\n");
 594         else
 595                 return sprintf(buf, "\n");
 596 }
 597 static DEVICE_ATTR_RO(persistence_domain);
 598
 599 static struct attribute *nd_region_attributes[] = {
 600         &dev_attr_size.attr,
 601         &dev_attr_nstype.attr,
 602         &dev_attr_mappings.attr,
 603         &dev_attr_btt_seed.attr,
 604         &dev_attr_pfn_seed.attr,
 605         &dev_attr_dax_seed.attr,
 606         &dev_attr_deep_flush.attr,
 607         &dev_attr_read_only.attr,
 608         &dev_attr_set_cookie.attr,
 609         &dev_attr_available_size.attr,
 610         &dev_attr_max_available_extent.attr,
 611         &dev_attr_namespace_seed.attr,
 612         &dev_attr_init_namespaces.attr,
 613         &dev_attr_badblocks.attr,
 614         &dev_attr_resource.attr,
 615         &dev_attr_persistence_domain.attr,
 616         NULL,
 617 };
 618
 619 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 620 {
 621         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 622         struct nd_region *nd_region = to_nd_region(dev);
 623         struct nd_interleave_set *nd_set = nd_region->nd_set;
 624         int type = nd_region_to_nstype(nd_region);
 625
 626         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
 627                 return 0;
 628
 629         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
 630                 return 0;
 631
 632         if (!is_nd_pmem(dev) && a == &dev_attr_badblocks.attr)
 633                 return 0;
 634
 635         if (a == &dev_attr_resource.attr) {
 636                 if (is_nd_pmem(dev))
 637                         return 0400;
 638                 else
 639                         return 0;
 640         }
 641
 642         if (a == &dev_attr_deep_flush.attr) {
 643                 int has_flush = nvdimm_has_flush(nd_region);
 644
 645                 if (has_flush == 1)
 646                         return a->mode;
 647                 else if (has_flush == 0)
 648                         return 0444;
 649                 else
 650                         return 0;
 651         }
 652
 653         if (a == &dev_attr_persistence_domain.attr) {
 654                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
 655                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
 656                         return 0;
 657                 return a->mode;
 658         }
 659
 660         if (a != &dev_attr_set_cookie.attr
 661                         && a != &dev_attr_available_size.attr)
 662                 return a->mode;
 663
 664         if ((type == ND_DEVICE_NAMESPACE_PMEM
 665                                 || type == ND_DEVICE_NAMESPACE_BLK)
 666                         && a == &dev_attr_available_size.attr)
 667                 return a->mode;
 668         else if (is_memory(dev) && nd_set)
 669                 return a->mode;
 670
 671         return 0;
 672 }
 673
 674 struct attribute_group nd_region_attribute_group = {
 675         .attrs = nd_region_attributes,
 676         .is_visible = region_visible,
 677 };
 678 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 679
 680 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 681                 struct nd_namespace_index *nsindex)
 682 {
 683         struct nd_interleave_set *nd_set = nd_region->nd_set;
 684
 685         if (!nd_set)
 686                 return 0;
 687
 688         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 689                         && __le16_to_cpu(nsindex->minor) == 1)
 690                 return nd_set->cookie1;
 691         return nd_set->cookie2;
 692 }
 693
 694 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 695 {
 696         struct nd_interleave_set *nd_set = nd_region->nd_set;
 697
 698         if (nd_set)
 699                 return nd_set->altcookie;
 700         return 0;
 701 }
 702
 703 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 704 {
 705         struct nd_label_ent *label_ent, *e;
 706
 707         lockdep_assert_held(&nd_mapping->lock);
 708         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 709                 list_del(&label_ent->list);
 710                 kfree(label_ent);
 711         }
 712 }
 713
 714 /*
 715  * Upon successful probe/remove, take/release a reference on the
 716  * associated interleave set (if present), and plant new btt + namespace
 717  * seeds.  Also, on the removal of a BLK region, notify the provider to
 718  * disable the region.
 719  */
 720 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 721                 struct device *dev, bool probe)
 722 {
 723         struct nd_region *nd_region;
 724
 725         if (!probe && is_nd_region(dev)) {
 726                 int i;
 727
 728                 nd_region = to_nd_region(dev);
 729                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 730                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 731                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 732                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 733
 734                         mutex_lock(&nd_mapping->lock);
 735                         nd_mapping_free_labels(nd_mapping);
 736                         mutex_unlock(&nd_mapping->lock);
 737
 738                         put_ndd(ndd);
 739                         nd_mapping->ndd = NULL;
 740                         if (ndd)
 741                                 atomic_dec(&nvdimm->busy);
 742                 }
 743         }
 744         if (dev->parent && is_nd_region(dev->parent) && probe) {
 745                 nd_region = to_nd_region(dev->parent);
 746                 nvdimm_bus_lock(dev);
 747                 if (nd_region->ns_seed == dev)
 748                         nd_region_create_ns_seed(nd_region);
 749                 nvdimm_bus_unlock(dev);
 750         }
 751         if (is_nd_btt(dev) && probe) {
 752                 struct nd_btt *nd_btt = to_nd_btt(dev);
 753
 754                 nd_region = to_nd_region(dev->parent);
 755                 nvdimm_bus_lock(dev);
 756                 if (nd_region->btt_seed == dev)
 757                         nd_region_create_btt_seed(nd_region);
 758                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 759                         nd_region_create_ns_seed(nd_region);
 760                 nvdimm_bus_unlock(dev);
 761         }
 762         if (is_nd_pfn(dev) && probe) {
 763                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 764
 765                 nd_region = to_nd_region(dev->parent);
 766                 nvdimm_bus_lock(dev);
 767                 if (nd_region->pfn_seed == dev)
 768                         nd_region_create_pfn_seed(nd_region);
 769                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 770                         nd_region_create_ns_seed(nd_region);
 771                 nvdimm_bus_unlock(dev);
 772         }
 773         if (is_nd_dax(dev) && probe) {
 774                 struct nd_dax *nd_dax = to_nd_dax(dev);
 775
 776                 nd_region = to_nd_region(dev->parent);
 777                 nvdimm_bus_lock(dev);
 778                 if (nd_region->dax_seed == dev)
 779                         nd_region_create_dax_seed(nd_region);
 780                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 781                         nd_region_create_ns_seed(nd_region);
 782                 nvdimm_bus_unlock(dev);
 783         }
 784 }
 785
 786 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 787 {
 788         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 789 }
 790
 791 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 792 {
 793         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 794 }
 795
 796 static ssize_t mappingN(struct device *dev, char *buf, int n)
 797 {
 798         struct nd_region *nd_region = to_nd_region(dev);
 799         struct nd_mapping *nd_mapping;
 800         struct nvdimm *nvdimm;
 801
 802         if (n >= nd_region->ndr_mappings)
 803                 return -ENXIO;
 804         nd_mapping = &nd_region->mapping[n];
 805         nvdimm = nd_mapping->nvdimm;
 806
 807         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
 808                         nd_mapping->start, nd_mapping->size,
 809                         nd_mapping->position);
 810 }
 811
 812 #define REGION_MAPPING(idx) \
 813 static ssize_t mapping##idx##_show(struct device *dev,          \
 814                 struct device_attribute *attr, char *buf)       \
 815 {                                                               \
 816         return mappingN(dev, buf, idx);                         \
 817 }                                                               \
 818 static DEVICE_ATTR_RO(mapping##idx)
 819
 820 /*
 821  * 32 should be enough for a while, even in the presence of socket
 822  * interleave a 32-way interleave set is a degenerate case.
 823  */
 824 REGION_MAPPING(0);
 825 REGION_MAPPING(1);
 826 REGION_MAPPING(2);
 827 REGION_MAPPING(3);
 828 REGION_MAPPING(4);
 829 REGION_MAPPING(5);
 830 REGION_MAPPING(6);
 831 REGION_MAPPING(7);
 832 REGION_MAPPING(8);
 833 REGION_MAPPING(9);
 834 REGION_MAPPING(10);
 835 REGION_MAPPING(11);
 836 REGION_MAPPING(12);
 837 REGION_MAPPING(13);
 838 REGION_MAPPING(14);
 839 REGION_MAPPING(15);
 840 REGION_MAPPING(16);
 841 REGION_MAPPING(17);
 842 REGION_MAPPING(18);
 843 REGION_MAPPING(19);
 844 REGION_MAPPING(20);
 845 REGION_MAPPING(21);
 846 REGION_MAPPING(22);
 847 REGION_MAPPING(23);
 848 REGION_MAPPING(24);
 849 REGION_MAPPING(25);
 850 REGION_MAPPING(26);
 851 REGION_MAPPING(27);
 852 REGION_MAPPING(28);
 853 REGION_MAPPING(29);
 854 REGION_MAPPING(30);
 855 REGION_MAPPING(31);
 856
 857 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 858 {
 859         struct device *dev = container_of(kobj, struct device, kobj);
 860         struct nd_region *nd_region = to_nd_region(dev);
 861
 862         if (n < nd_region->ndr_mappings)
 863                 return a->mode;
 864         return 0;
 865 }
 866
 867 static struct attribute *mapping_attributes[] = {
 868         &dev_attr_mapping0.attr,
 869         &dev_attr_mapping1.attr,
 870         &dev_attr_mapping2.attr,
 871         &dev_attr_mapping3.attr,
 872         &dev_attr_mapping4.attr,
 873         &dev_attr_mapping5.attr,
 874         &dev_attr_mapping6.attr,
 875         &dev_attr_mapping7.attr,
 876         &dev_attr_mapping8.attr,
 877         &dev_attr_mapping9.attr,
 878         &dev_attr_mapping10.attr,
 879         &dev_attr_mapping11.attr,
 880         &dev_attr_mapping12.attr,
 881         &dev_attr_mapping13.attr,
 882         &dev_attr_mapping14.attr,
 883         &dev_attr_mapping15.attr,
 884         &dev_attr_mapping16.attr,
 885         &dev_attr_mapping17.attr,
 886         &dev_attr_mapping18.attr,
 887         &dev_attr_mapping19.attr,
 888         &dev_attr_mapping20.attr,
 889         &dev_attr_mapping21.attr,
 890         &dev_attr_mapping22.attr,
 891         &dev_attr_mapping23.attr,
 892         &dev_attr_mapping24.attr,
 893         &dev_attr_mapping25.attr,
 894         &dev_attr_mapping26.attr,
 895         &dev_attr_mapping27.attr,
 896         &dev_attr_mapping28.attr,
 897         &dev_attr_mapping29.attr,
 898         &dev_attr_mapping30.attr,
 899         &dev_attr_mapping31.attr,
 900         NULL,
 901 };
 902
 903 struct attribute_group nd_mapping_attribute_group = {
 904         .is_visible = mapping_visible,
 905         .attrs = mapping_attributes,
 906 };
 907 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 908
 909 int nd_blk_region_init(struct nd_region *nd_region)
 910 {
 911         struct device *dev = &nd_region->dev;
 912         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 913
 914         if (!is_nd_blk(dev))
 915                 return 0;
 916
 917         if (nd_region->ndr_mappings < 1) {
 918                 dev_dbg(dev, "invalid BLK region\n");
 919                 return -ENXIO;
 920         }
 921
 922         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 923 }
 924
 925 /**
 926  * nd_region_acquire_lane - allocate and lock a lane
 927  * @nd_region: region id and number of lanes possible
 928  *
 929  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 930  * We optimize for the common case where there are 256 lanes, one
 931  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 932  * this implementation assumes the cost of maintaining an allocator for
 933  * free lanes is on the order of the lock hold time, so it implements a
 934  * static lane = cpu % num_lanes mapping.
 935  *
 936  * In the case of a BTT instance on top of a BLK namespace a lane may be
 937  * acquired recursively.  We lock on the first instance.
 938  *
 939  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 940  * for the BTT metadata updates.
 941  */
 942 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 943 {
 944         unsigned int cpu, lane;
 945
 946         cpu = get_cpu();
 947         if (nd_region->num_lanes < nr_cpu_ids) {
 948                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 949
 950                 lane = cpu % nd_region->num_lanes;
 951                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 952                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 953                 if (ndl_count->count++ == 0)
 954                         spin_lock(&ndl_lock->lock);
 955         } else
 956                 lane = cpu;
 957
 958         return lane;
 959 }
 960 EXPORT_SYMBOL(nd_region_acquire_lane);
 961
 962 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 963 {
 964         if (nd_region->num_lanes < nr_cpu_ids) {
 965                 unsigned int cpu = get_cpu();
 966                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 967
 968                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 969                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 970                 if (--ndl_count->count == 0)
 971                         spin_unlock(&ndl_lock->lock);
 972                 put_cpu();
 973         }
 974         put_cpu();
 975 }
 976 EXPORT_SYMBOL(nd_region_release_lane);
 977
 978 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 979                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 980                 const char *caller)
 981 {
 982         struct nd_region *nd_region;
 983         struct device *dev;
 984         void *region_buf;
 985         unsigned int i;
 986         int ro = 0;
 987
 988         for (i = 0; i < ndr_desc->num_mappings; i++) {
 989                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 990                 struct nvdimm *nvdimm = mapping->nvdimm;
 991
 992                 if ((mapping->start | mapping->size) % SZ_4K) {
 993                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 994                                         caller, dev_name(&nvdimm->dev), i);
 995
 996                         return NULL;
 997                 }
 998
 999                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1000                         ro = 1;
1001         }
1002
1003         if (dev_type == &nd_blk_device_type) {
1004                 struct nd_blk_region_desc *ndbr_desc;
1005                 struct nd_blk_region *ndbr;
1006
1007                 ndbr_desc = to_blk_region_desc(ndr_desc);
1008                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1009                                 * ndr_desc->num_mappings,
1010                                 GFP_KERNEL);
1011                 if (ndbr) {
1012                         nd_region = &ndbr->nd_region;
1013                         ndbr->enable = ndbr_desc->enable;
1014                         ndbr->do_io = ndbr_desc->do_io;
1015                 }
1016                 region_buf = ndbr;
1017         } else {
1018                 nd_region = kzalloc(sizeof(struct nd_region)
1019                                 + sizeof(struct nd_mapping)
1020                                 * ndr_desc->num_mappings,
1021                                 GFP_KERNEL);
1022                 region_buf = nd_region;
1023         }
1024
1025         if (!region_buf)
1026                 return NULL;
1027         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
1028         if (nd_region->id < 0)
1029                 goto err_id;
1030
1031         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1032         if (!nd_region->lane)
1033                 goto err_percpu;
1034
1035         for (i = 0; i < nr_cpu_ids; i++) {
1036                 struct nd_percpu_lane *ndl;
1037
1038                 ndl = per_cpu_ptr(nd_region->lane, i);
1039                 spin_lock_init(&ndl->lock);
1040                 ndl->count = 0;
1041         }
1042
1043         for (i = 0; i < ndr_desc->num_mappings; i++) {
1044                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1045                 struct nvdimm *nvdimm = mapping->nvdimm;
1046
1047                 nd_region->mapping[i].nvdimm = nvdimm;
1048                 nd_region->mapping[i].start = mapping->start;
1049                 nd_region->mapping[i].size = mapping->size;
1050                 nd_region->mapping[i].position = mapping->position;
1051                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1052                 mutex_init(&nd_region->mapping[i].lock);
1053
1054                 get_device(&nvdimm->dev);
1055         }
1056         nd_region->ndr_mappings = ndr_desc->num_mappings;
1057         nd_region->provider_data = ndr_desc->provider_data;
1058         nd_region->nd_set = ndr_desc->nd_set;
1059         nd_region->num_lanes = ndr_desc->num_lanes;
1060         nd_region->flags = ndr_desc->flags;
1061         nd_region->ro = ro;
1062         nd_region->numa_node = ndr_desc->numa_node;
1063         ida_init(&nd_region->ns_ida);
1064         ida_init(&nd_region->btt_ida);
1065         ida_init(&nd_region->pfn_ida);
1066         ida_init(&nd_region->dax_ida);
1067         dev = &nd_region->dev;
1068         dev_set_name(dev, "region%d", nd_region->id);
1069         dev->parent = &nvdimm_bus->dev;
1070         dev->type = dev_type;
1071         dev->groups = ndr_desc->attr_groups;
1072         dev->of_node = ndr_desc->of_node;
1073         nd_region->ndr_size = resource_size(ndr_desc->res);
1074         nd_region->ndr_start = ndr_desc->res->start;
1075         nd_device_register(dev);
1076
1077         return nd_region;
1078
1079  err_percpu:
1080         ida_simple_remove(&region_ida, nd_region->id);
1081  err_id:
1082         kfree(region_buf);
1083         return NULL;
1084 }
1085
1086 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1087                 struct nd_region_desc *ndr_desc)
1088 {
1089         ndr_desc->num_lanes = ND_MAX_LANES;
1090         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1091                         __func__);
1092 }
1093 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1094
1095 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1096                 struct nd_region_desc *ndr_desc)
1097 {
1098         if (ndr_desc->num_mappings > 1)
1099                 return NULL;
1100         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1101         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1102                         __func__);
1103 }
1104 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1105
1106 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1107                 struct nd_region_desc *ndr_desc)
1108 {
1109         ndr_desc->num_lanes = ND_MAX_LANES;
1110         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1111                         __func__);
1112 }
1113 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1114
1115 /**
1116  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1117  * @nd_region: blk or interleaved pmem region
1118  */
1119 void nvdimm_flush(struct nd_region *nd_region)
1120 {
1121         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1122         int i, idx;
1123
1124         /*
1125          * Try to encourage some diversity in flush hint addresses
1126          * across cpus assuming a limited number of flush hints.
1127          */
1128         idx = this_cpu_read(flush_idx);
1129         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1130
1131         /*
1132          * The first wmb() is needed to 'sfence' all previous writes
1133          * such that they are architecturally visible for the platform
1134          * buffer flush.  Note that we've already arranged for pmem
1135          * writes to avoid the cache via memcpy_flushcache().  The final
1136          * wmb() ensures ordering for the NVDIMM flush write.
1137          */
1138         wmb();
1139         for (i = 0; i < nd_region->ndr_mappings; i++)
1140                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1141                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1142         wmb();
1143 }
1144 EXPORT_SYMBOL_GPL(nvdimm_flush);
1145
1146 /**
1147  * nvdimm_has_flush - determine write flushing requirements
1148  * @nd_region: blk or interleaved pmem region
1149  *
1150  * Returns 1 if writes require flushing
1151  * Returns 0 if writes do not require flushing
1152  * Returns -ENXIO if flushing capability can not be determined
1153  */
1154 int nvdimm_has_flush(struct nd_region *nd_region)
1155 {
1156         int i;
1157
1158         /* no nvdimm or pmem api == flushing capability unknown */
1159         if (nd_region->ndr_mappings == 0
1160                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1161                 return -ENXIO;
1162
1163         for (i = 0; i < nd_region->ndr_mappings; i++) {
1164                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1165                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1166
1167                 /* flush hints present / available */
1168                 if (nvdimm->num_flush)
1169                         return 1;
1170         }
1171
1172         /*
1173          * The platform defines dimm devices without hints, assume
1174          * platform persistence mechanism like ADR
1175          */
1176         return 0;
1177 }
1178 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1179
1180 int nvdimm_has_cache(struct nd_region *nd_region)
1181 {
1182         return is_nd_pmem(&nd_region->dev) &&
1183                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1184 }
1185 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1186
1187 void __exit nd_region_devs_exit(void)
1188 {
1189         ida_destroy(&region_ida);
1190 }