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         device_lock(dev);
 429         nvdimm_bus_lock(dev);
 430         wait_nvdimm_bus_probe_idle(dev);
 431         available = nd_region_available_dpa(nd_region);
 432         nvdimm_bus_unlock(dev);
 433         device_unlock(dev);
 434
 435         return sprintf(buf, "%llu\n", available);
 436 }
 437 static DEVICE_ATTR_RO(available_size);
 438
 439 static ssize_t max_available_extent_show(struct device *dev,
 440                 struct device_attribute *attr, char *buf)
 441 {
 442         struct nd_region *nd_region = to_nd_region(dev);
 443         unsigned long long available = 0;
 444
 445         device_lock(dev);
 446         nvdimm_bus_lock(dev);
 447         wait_nvdimm_bus_probe_idle(dev);
 448         available = nd_region_allocatable_dpa(nd_region);
 449         nvdimm_bus_unlock(dev);
 450         device_unlock(dev);
 451
 452         return sprintf(buf, "%llu\n", available);
 453 }
 454 static DEVICE_ATTR_RO(max_available_extent);
 455
 456 static ssize_t init_namespaces_show(struct device *dev,
 457                 struct device_attribute *attr, char *buf)
 458 {
 459         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 460         ssize_t rc;
 461
 462         nvdimm_bus_lock(dev);
 463         if (ndrd)
 464                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 465         else
 466                 rc = -ENXIO;
 467         nvdimm_bus_unlock(dev);
 468
 469         return rc;
 470 }
 471 static DEVICE_ATTR_RO(init_namespaces);
 472
 473 static ssize_t namespace_seed_show(struct device *dev,
 474                 struct device_attribute *attr, char *buf)
 475 {
 476         struct nd_region *nd_region = to_nd_region(dev);
 477         ssize_t rc;
 478
 479         nvdimm_bus_lock(dev);
 480         if (nd_region->ns_seed)
 481                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 482         else
 483                 rc = sprintf(buf, "\n");
 484         nvdimm_bus_unlock(dev);
 485         return rc;
 486 }
 487 static DEVICE_ATTR_RO(namespace_seed);
 488
 489 static ssize_t btt_seed_show(struct device *dev,
 490                 struct device_attribute *attr, char *buf)
 491 {
 492         struct nd_region *nd_region = to_nd_region(dev);
 493         ssize_t rc;
 494
 495         nvdimm_bus_lock(dev);
 496         if (nd_region->btt_seed)
 497                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 498         else
 499                 rc = sprintf(buf, "\n");
 500         nvdimm_bus_unlock(dev);
 501
 502         return rc;
 503 }
 504 static DEVICE_ATTR_RO(btt_seed);
 505
 506 static ssize_t pfn_seed_show(struct device *dev,
 507                 struct device_attribute *attr, char *buf)
 508 {
 509         struct nd_region *nd_region = to_nd_region(dev);
 510         ssize_t rc;
 511
 512         nvdimm_bus_lock(dev);
 513         if (nd_region->pfn_seed)
 514                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 515         else
 516                 rc = sprintf(buf, "\n");
 517         nvdimm_bus_unlock(dev);
 518
 519         return rc;
 520 }
 521 static DEVICE_ATTR_RO(pfn_seed);
 522
 523 static ssize_t dax_seed_show(struct device *dev,
 524                 struct device_attribute *attr, char *buf)
 525 {
 526         struct nd_region *nd_region = to_nd_region(dev);
 527         ssize_t rc;
 528
 529         nvdimm_bus_lock(dev);
 530         if (nd_region->dax_seed)
 531                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 532         else
 533                 rc = sprintf(buf, "\n");
 534         nvdimm_bus_unlock(dev);
 535
 536         return rc;
 537 }
 538 static DEVICE_ATTR_RO(dax_seed);
 539
 540 static ssize_t read_only_show(struct device *dev,
 541                 struct device_attribute *attr, char *buf)
 542 {
 543         struct nd_region *nd_region = to_nd_region(dev);
 544
 545         return sprintf(buf, "%d\n", nd_region->ro);
 546 }
 547
 548 static ssize_t read_only_store(struct device *dev,
 549                 struct device_attribute *attr, const char *buf, size_t len)
 550 {
 551         bool ro;
 552         int rc = strtobool(buf, &ro);
 553         struct nd_region *nd_region = to_nd_region(dev);
 554
 555         if (rc)
 556                 return rc;
 557
 558         nd_region->ro = ro;
 559         return len;
 560 }
 561 static DEVICE_ATTR_RW(read_only);
 562
 563 static ssize_t region_badblocks_show(struct device *dev,
 564                 struct device_attribute *attr, char *buf)
 565 {
 566         struct nd_region *nd_region = to_nd_region(dev);
 567         ssize_t rc;
 568
 569         device_lock(dev);
 570         if (dev->driver)
 571                 rc = badblocks_show(&nd_region->bb, buf, 0);
 572         else
 573                 rc = -ENXIO;
 574         device_unlock(dev);
 575
 576         return rc;
 577 }
 578 static DEVICE_ATTR(badblocks, 0444, region_badblocks_show, NULL);
 579
 580 static ssize_t resource_show(struct device *dev,
 581                 struct device_attribute *attr, char *buf)
 582 {
 583         struct nd_region *nd_region = to_nd_region(dev);
 584
 585         return sprintf(buf, "%#llx\n", nd_region->ndr_start);
 586 }
 587 static DEVICE_ATTR_RO(resource);
 588
 589 static ssize_t persistence_domain_show(struct device *dev,
 590                 struct device_attribute *attr, char *buf)
 591 {
 592         struct nd_region *nd_region = to_nd_region(dev);
 593
 594         if (test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags))
 595                 return sprintf(buf, "cpu_cache\n");
 596         else if (test_bit(ND_REGION_PERSIST_MEMCTRL, &nd_region->flags))
 597                 return sprintf(buf, "memory_controller\n");
 598         else
 599                 return sprintf(buf, "\n");
 600 }
 601 static DEVICE_ATTR_RO(persistence_domain);
 602
 603 static struct attribute *nd_region_attributes[] = {
 604         &dev_attr_size.attr,
 605         &dev_attr_nstype.attr,
 606         &dev_attr_mappings.attr,
 607         &dev_attr_btt_seed.attr,
 608         &dev_attr_pfn_seed.attr,
 609         &dev_attr_dax_seed.attr,
 610         &dev_attr_deep_flush.attr,
 611         &dev_attr_read_only.attr,
 612         &dev_attr_set_cookie.attr,
 613         &dev_attr_available_size.attr,
 614         &dev_attr_max_available_extent.attr,
 615         &dev_attr_namespace_seed.attr,
 616         &dev_attr_init_namespaces.attr,
 617         &dev_attr_badblocks.attr,
 618         &dev_attr_resource.attr,
 619         &dev_attr_persistence_domain.attr,
 620         NULL,
 621 };
 622
 623 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 624 {
 625         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 626         struct nd_region *nd_region = to_nd_region(dev);
 627         struct nd_interleave_set *nd_set = nd_region->nd_set;
 628         int type = nd_region_to_nstype(nd_region);
 629
 630         if (!is_memory(dev) && a == &dev_attr_pfn_seed.attr)
 631                 return 0;
 632
 633         if (!is_memory(dev) && a == &dev_attr_dax_seed.attr)
 634                 return 0;
 635
 636         if (!is_memory(dev) && a == &dev_attr_badblocks.attr)
 637                 return 0;
 638
 639         if (a == &dev_attr_resource.attr) {
 640                 if (is_memory(dev))
 641                         return 0400;
 642                 else
 643                         return 0;
 644         }
 645
 646         if (a == &dev_attr_deep_flush.attr) {
 647                 int has_flush = nvdimm_has_flush(nd_region);
 648
 649                 if (has_flush == 1)
 650                         return a->mode;
 651                 else if (has_flush == 0)
 652                         return 0444;
 653                 else
 654                         return 0;
 655         }
 656
 657         if (a == &dev_attr_persistence_domain.attr) {
 658                 if ((nd_region->flags & (BIT(ND_REGION_PERSIST_CACHE)
 659                                         | BIT(ND_REGION_PERSIST_MEMCTRL))) == 0)
 660                         return 0;
 661                 return a->mode;
 662         }
 663
 664         if (a != &dev_attr_set_cookie.attr
 665                         && a != &dev_attr_available_size.attr)
 666                 return a->mode;
 667
 668         if ((type == ND_DEVICE_NAMESPACE_PMEM
 669                                 || type == ND_DEVICE_NAMESPACE_BLK)
 670                         && a == &dev_attr_available_size.attr)
 671                 return a->mode;
 672         else if (is_memory(dev) && nd_set)
 673                 return a->mode;
 674
 675         return 0;
 676 }
 677
 678 struct attribute_group nd_region_attribute_group = {
 679         .attrs = nd_region_attributes,
 680         .is_visible = region_visible,
 681 };
 682 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 683
 684 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region,
 685                 struct nd_namespace_index *nsindex)
 686 {
 687         struct nd_interleave_set *nd_set = nd_region->nd_set;
 688
 689         if (!nd_set)
 690                 return 0;
 691
 692         if (nsindex && __le16_to_cpu(nsindex->major) == 1
 693                         && __le16_to_cpu(nsindex->minor) == 1)
 694                 return nd_set->cookie1;
 695         return nd_set->cookie2;
 696 }
 697
 698 u64 nd_region_interleave_set_altcookie(struct nd_region *nd_region)
 699 {
 700         struct nd_interleave_set *nd_set = nd_region->nd_set;
 701
 702         if (nd_set)
 703                 return nd_set->altcookie;
 704         return 0;
 705 }
 706
 707 void nd_mapping_free_labels(struct nd_mapping *nd_mapping)
 708 {
 709         struct nd_label_ent *label_ent, *e;
 710
 711         lockdep_assert_held(&nd_mapping->lock);
 712         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
 713                 list_del(&label_ent->list);
 714                 kfree(label_ent);
 715         }
 716 }
 717
 718 /*
 719  * Upon successful probe/remove, take/release a reference on the
 720  * associated interleave set (if present), and plant new btt + namespace
 721  * seeds.  Also, on the removal of a BLK region, notify the provider to
 722  * disable the region.
 723  */
 724 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 725                 struct device *dev, bool probe)
 726 {
 727         struct nd_region *nd_region;
 728
 729         if (!probe && is_nd_region(dev)) {
 730                 int i;
 731
 732                 nd_region = to_nd_region(dev);
 733                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 734                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 735                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 736                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 737
 738                         mutex_lock(&nd_mapping->lock);
 739                         nd_mapping_free_labels(nd_mapping);
 740                         mutex_unlock(&nd_mapping->lock);
 741
 742                         put_ndd(ndd);
 743                         nd_mapping->ndd = NULL;
 744                         if (ndd)
 745                                 atomic_dec(&nvdimm->busy);
 746                 }
 747         }
 748         if (dev->parent && is_nd_region(dev->parent) && probe) {
 749                 nd_region = to_nd_region(dev->parent);
 750                 nvdimm_bus_lock(dev);
 751                 if (nd_region->ns_seed == dev)
 752                         nd_region_create_ns_seed(nd_region);
 753                 nvdimm_bus_unlock(dev);
 754         }
 755         if (is_nd_btt(dev) && probe) {
 756                 struct nd_btt *nd_btt = to_nd_btt(dev);
 757
 758                 nd_region = to_nd_region(dev->parent);
 759                 nvdimm_bus_lock(dev);
 760                 if (nd_region->btt_seed == dev)
 761                         nd_region_create_btt_seed(nd_region);
 762                 if (nd_region->ns_seed == &nd_btt->ndns->dev)
 763                         nd_region_create_ns_seed(nd_region);
 764                 nvdimm_bus_unlock(dev);
 765         }
 766         if (is_nd_pfn(dev) && probe) {
 767                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 768
 769                 nd_region = to_nd_region(dev->parent);
 770                 nvdimm_bus_lock(dev);
 771                 if (nd_region->pfn_seed == dev)
 772                         nd_region_create_pfn_seed(nd_region);
 773                 if (nd_region->ns_seed == &nd_pfn->ndns->dev)
 774                         nd_region_create_ns_seed(nd_region);
 775                 nvdimm_bus_unlock(dev);
 776         }
 777         if (is_nd_dax(dev) && probe) {
 778                 struct nd_dax *nd_dax = to_nd_dax(dev);
 779
 780                 nd_region = to_nd_region(dev->parent);
 781                 nvdimm_bus_lock(dev);
 782                 if (nd_region->dax_seed == dev)
 783                         nd_region_create_dax_seed(nd_region);
 784                 if (nd_region->ns_seed == &nd_dax->nd_pfn.ndns->dev)
 785                         nd_region_create_ns_seed(nd_region);
 786                 nvdimm_bus_unlock(dev);
 787         }
 788 }
 789
 790 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 791 {
 792         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 793 }
 794
 795 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 796 {
 797         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 798 }
 799
 800 static ssize_t mappingN(struct device *dev, char *buf, int n)
 801 {
 802         struct nd_region *nd_region = to_nd_region(dev);
 803         struct nd_mapping *nd_mapping;
 804         struct nvdimm *nvdimm;
 805
 806         if (n >= nd_region->ndr_mappings)
 807                 return -ENXIO;
 808         nd_mapping = &nd_region->mapping[n];
 809         nvdimm = nd_mapping->nvdimm;
 810
 811         return sprintf(buf, "%s,%llu,%llu,%d\n", dev_name(&nvdimm->dev),
 812                         nd_mapping->start, nd_mapping->size,
 813                         nd_mapping->position);
 814 }
 815
 816 #define REGION_MAPPING(idx) \
 817 static ssize_t mapping##idx##_show(struct device *dev,          \
 818                 struct device_attribute *attr, char *buf)       \
 819 {                                                               \
 820         return mappingN(dev, buf, idx);                         \
 821 }                                                               \
 822 static DEVICE_ATTR_RO(mapping##idx)
 823
 824 /*
 825  * 32 should be enough for a while, even in the presence of socket
 826  * interleave a 32-way interleave set is a degenerate case.
 827  */
 828 REGION_MAPPING(0);
 829 REGION_MAPPING(1);
 830 REGION_MAPPING(2);
 831 REGION_MAPPING(3);
 832 REGION_MAPPING(4);
 833 REGION_MAPPING(5);
 834 REGION_MAPPING(6);
 835 REGION_MAPPING(7);
 836 REGION_MAPPING(8);
 837 REGION_MAPPING(9);
 838 REGION_MAPPING(10);
 839 REGION_MAPPING(11);
 840 REGION_MAPPING(12);
 841 REGION_MAPPING(13);
 842 REGION_MAPPING(14);
 843 REGION_MAPPING(15);
 844 REGION_MAPPING(16);
 845 REGION_MAPPING(17);
 846 REGION_MAPPING(18);
 847 REGION_MAPPING(19);
 848 REGION_MAPPING(20);
 849 REGION_MAPPING(21);
 850 REGION_MAPPING(22);
 851 REGION_MAPPING(23);
 852 REGION_MAPPING(24);
 853 REGION_MAPPING(25);
 854 REGION_MAPPING(26);
 855 REGION_MAPPING(27);
 856 REGION_MAPPING(28);
 857 REGION_MAPPING(29);
 858 REGION_MAPPING(30);
 859 REGION_MAPPING(31);
 860
 861 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 862 {
 863         struct device *dev = container_of(kobj, struct device, kobj);
 864         struct nd_region *nd_region = to_nd_region(dev);
 865
 866         if (n < nd_region->ndr_mappings)
 867                 return a->mode;
 868         return 0;
 869 }
 870
 871 static struct attribute *mapping_attributes[] = {
 872         &dev_attr_mapping0.attr,
 873         &dev_attr_mapping1.attr,
 874         &dev_attr_mapping2.attr,
 875         &dev_attr_mapping3.attr,
 876         &dev_attr_mapping4.attr,
 877         &dev_attr_mapping5.attr,
 878         &dev_attr_mapping6.attr,
 879         &dev_attr_mapping7.attr,
 880         &dev_attr_mapping8.attr,
 881         &dev_attr_mapping9.attr,
 882         &dev_attr_mapping10.attr,
 883         &dev_attr_mapping11.attr,
 884         &dev_attr_mapping12.attr,
 885         &dev_attr_mapping13.attr,
 886         &dev_attr_mapping14.attr,
 887         &dev_attr_mapping15.attr,
 888         &dev_attr_mapping16.attr,
 889         &dev_attr_mapping17.attr,
 890         &dev_attr_mapping18.attr,
 891         &dev_attr_mapping19.attr,
 892         &dev_attr_mapping20.attr,
 893         &dev_attr_mapping21.attr,
 894         &dev_attr_mapping22.attr,
 895         &dev_attr_mapping23.attr,
 896         &dev_attr_mapping24.attr,
 897         &dev_attr_mapping25.attr,
 898         &dev_attr_mapping26.attr,
 899         &dev_attr_mapping27.attr,
 900         &dev_attr_mapping28.attr,
 901         &dev_attr_mapping29.attr,
 902         &dev_attr_mapping30.attr,
 903         &dev_attr_mapping31.attr,
 904         NULL,
 905 };
 906
 907 struct attribute_group nd_mapping_attribute_group = {
 908         .is_visible = mapping_visible,
 909         .attrs = mapping_attributes,
 910 };
 911 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 912
 913 int nd_blk_region_init(struct nd_region *nd_region)
 914 {
 915         struct device *dev = &nd_region->dev;
 916         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 917
 918         if (!is_nd_blk(dev))
 919                 return 0;
 920
 921         if (nd_region->ndr_mappings < 1) {
 922                 dev_dbg(dev, "invalid BLK region\n");
 923                 return -ENXIO;
 924         }
 925
 926         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 927 }
 928
 929 /**
 930  * nd_region_acquire_lane - allocate and lock a lane
 931  * @nd_region: region id and number of lanes possible
 932  *
 933  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 934  * We optimize for the common case where there are 256 lanes, one
 935  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 936  * this implementation assumes the cost of maintaining an allocator for
 937  * free lanes is on the order of the lock hold time, so it implements a
 938  * static lane = cpu % num_lanes mapping.
 939  *
 940  * In the case of a BTT instance on top of a BLK namespace a lane may be
 941  * acquired recursively.  We lock on the first instance.
 942  *
 943  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 944  * for the BTT metadata updates.
 945  */
 946 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 947 {
 948         unsigned int cpu, lane;
 949
 950         cpu = get_cpu();
 951         if (nd_region->num_lanes < nr_cpu_ids) {
 952                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 953
 954                 lane = cpu % nd_region->num_lanes;
 955                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 956                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 957                 if (ndl_count->count++ == 0)
 958                         spin_lock(&ndl_lock->lock);
 959         } else
 960                 lane = cpu;
 961
 962         return lane;
 963 }
 964 EXPORT_SYMBOL(nd_region_acquire_lane);
 965
 966 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 967 {
 968         if (nd_region->num_lanes < nr_cpu_ids) {
 969                 unsigned int cpu = get_cpu();
 970                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 971
 972                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 973                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 974                 if (--ndl_count->count == 0)
 975                         spin_unlock(&ndl_lock->lock);
 976                 put_cpu();
 977         }
 978         put_cpu();
 979 }
 980 EXPORT_SYMBOL(nd_region_release_lane);
 981
 982 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 983                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 984                 const char *caller)
 985 {
 986         struct nd_region *nd_region;
 987         struct device *dev;
 988         void *region_buf;
 989         unsigned int i;
 990         int ro = 0;
 991
 992         for (i = 0; i < ndr_desc->num_mappings; i++) {
 993                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
 994                 struct nvdimm *nvdimm = mapping->nvdimm;
 995
 996                 if ((mapping->start | mapping->size) % SZ_4K) {
 997                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 998                                         caller, dev_name(&nvdimm->dev), i);
 999
1000                         return NULL;
1001                 }
1002
1003                 if (test_bit(NDD_UNARMED, &nvdimm->flags))
1004                         ro = 1;
1005         }
1006
1007         if (dev_type == &nd_blk_device_type) {
1008                 struct nd_blk_region_desc *ndbr_desc;
1009                 struct nd_blk_region *ndbr;
1010
1011                 ndbr_desc = to_blk_region_desc(ndr_desc);
1012                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
1013                                 * ndr_desc->num_mappings,
1014                                 GFP_KERNEL);
1015                 if (ndbr) {
1016                         nd_region = &ndbr->nd_region;
1017                         ndbr->enable = ndbr_desc->enable;
1018                         ndbr->do_io = ndbr_desc->do_io;
1019                 }
1020                 region_buf = ndbr;
1021         } else {
1022                 nd_region = kzalloc(sizeof(struct nd_region)
1023                                 + sizeof(struct nd_mapping)
1024                                 * ndr_desc->num_mappings,
1025                                 GFP_KERNEL);
1026                 region_buf = nd_region;
1027         }
1028
1029         if (!region_buf)
1030                 return NULL;
1031         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
1032         if (nd_region->id < 0)
1033                 goto err_id;
1034
1035         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
1036         if (!nd_region->lane)
1037                 goto err_percpu;
1038
1039         for (i = 0; i < nr_cpu_ids; i++) {
1040                 struct nd_percpu_lane *ndl;
1041
1042                 ndl = per_cpu_ptr(nd_region->lane, i);
1043                 spin_lock_init(&ndl->lock);
1044                 ndl->count = 0;
1045         }
1046
1047         for (i = 0; i < ndr_desc->num_mappings; i++) {
1048                 struct nd_mapping_desc *mapping = &ndr_desc->mapping[i];
1049                 struct nvdimm *nvdimm = mapping->nvdimm;
1050
1051                 nd_region->mapping[i].nvdimm = nvdimm;
1052                 nd_region->mapping[i].start = mapping->start;
1053                 nd_region->mapping[i].size = mapping->size;
1054                 nd_region->mapping[i].position = mapping->position;
1055                 INIT_LIST_HEAD(&nd_region->mapping[i].labels);
1056                 mutex_init(&nd_region->mapping[i].lock);
1057
1058                 get_device(&nvdimm->dev);
1059         }
1060         nd_region->ndr_mappings = ndr_desc->num_mappings;
1061         nd_region->provider_data = ndr_desc->provider_data;
1062         nd_region->nd_set = ndr_desc->nd_set;
1063         nd_region->num_lanes = ndr_desc->num_lanes;
1064         nd_region->flags = ndr_desc->flags;
1065         nd_region->ro = ro;
1066         nd_region->numa_node = ndr_desc->numa_node;
1067         ida_init(&nd_region->ns_ida);
1068         ida_init(&nd_region->btt_ida);
1069         ida_init(&nd_region->pfn_ida);
1070         ida_init(&nd_region->dax_ida);
1071         dev = &nd_region->dev;
1072         dev_set_name(dev, "region%d", nd_region->id);
1073         dev->parent = &nvdimm_bus->dev;
1074         dev->type = dev_type;
1075         dev->groups = ndr_desc->attr_groups;
1076         dev->of_node = ndr_desc->of_node;
1077         nd_region->ndr_size = resource_size(ndr_desc->res);
1078         nd_region->ndr_start = ndr_desc->res->start;
1079         nd_device_register(dev);
1080
1081         return nd_region;
1082
1083  err_percpu:
1084         ida_simple_remove(&region_ida, nd_region->id);
1085  err_id:
1086         kfree(region_buf);
1087         return NULL;
1088 }
1089
1090 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
1091                 struct nd_region_desc *ndr_desc)
1092 {
1093         ndr_desc->num_lanes = ND_MAX_LANES;
1094         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
1095                         __func__);
1096 }
1097 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
1098
1099 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
1100                 struct nd_region_desc *ndr_desc)
1101 {
1102         if (ndr_desc->num_mappings > 1)
1103                 return NULL;
1104         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
1105         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
1106                         __func__);
1107 }
1108 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
1109
1110 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
1111                 struct nd_region_desc *ndr_desc)
1112 {
1113         ndr_desc->num_lanes = ND_MAX_LANES;
1114         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
1115                         __func__);
1116 }
1117 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
1118
1119 /**
1120  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
1121  * @nd_region: blk or interleaved pmem region
1122  */
1123 void nvdimm_flush(struct nd_region *nd_region)
1124 {
1125         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
1126         int i, idx;
1127
1128         /*
1129          * Try to encourage some diversity in flush hint addresses
1130          * across cpus assuming a limited number of flush hints.
1131          */
1132         idx = this_cpu_read(flush_idx);
1133         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
1134
1135         /*
1136          * The first wmb() is needed to 'sfence' all previous writes
1137          * such that they are architecturally visible for the platform
1138          * buffer flush.  Note that we've already arranged for pmem
1139          * writes to avoid the cache via memcpy_flushcache().  The final
1140          * wmb() ensures ordering for the NVDIMM flush write.
1141          */
1142         wmb();
1143         for (i = 0; i < nd_region->ndr_mappings; i++)
1144                 if (ndrd_get_flush_wpq(ndrd, i, 0))
1145                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
1146         wmb();
1147 }
1148 EXPORT_SYMBOL_GPL(nvdimm_flush);
1149
1150 /**
1151  * nvdimm_has_flush - determine write flushing requirements
1152  * @nd_region: blk or interleaved pmem region
1153  *
1154  * Returns 1 if writes require flushing
1155  * Returns 0 if writes do not require flushing
1156  * Returns -ENXIO if flushing capability can not be determined
1157  */
1158 int nvdimm_has_flush(struct nd_region *nd_region)
1159 {
1160         int i;
1161
1162         /* no nvdimm or pmem api == flushing capability unknown */
1163         if (nd_region->ndr_mappings == 0
1164                         || !IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
1165                 return -ENXIO;
1166
1167         for (i = 0; i < nd_region->ndr_mappings; i++) {
1168                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1169                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1170
1171                 /* flush hints present / available */
1172                 if (nvdimm->num_flush)
1173                         return 1;
1174         }
1175
1176         /*
1177          * The platform defines dimm devices without hints, assume
1178          * platform persistence mechanism like ADR
1179          */
1180         return 0;
1181 }
1182 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
1183
1184 int nvdimm_has_cache(struct nd_region *nd_region)
1185 {
1186         return is_nd_pmem(&nd_region->dev) &&
1187                 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
1188 }
1189 EXPORT_SYMBOL_GPL(nvdimm_has_cache);
1190
1191 struct conflict_context {
1192         struct nd_region *nd_region;
1193         resource_size_t start, size;
1194 };
1195
1196 static int region_conflict(struct device *dev, void *data)
1197 {
1198         struct nd_region *nd_region;
1199         struct conflict_context *ctx = data;
1200         resource_size_t res_end, region_end, region_start;
1201
1202         if (!is_memory(dev))
1203                 return 0;
1204
1205         nd_region = to_nd_region(dev);
1206         if (nd_region == ctx->nd_region)
1207                 return 0;
1208
1209         res_end = ctx->start + ctx->size;
1210         region_start = nd_region->ndr_start;
1211         region_end = region_start + nd_region->ndr_size;
1212         if (ctx->start >= region_start && ctx->start < region_end)
1213                 return -EBUSY;
1214         if (res_end > region_start && res_end <= region_end)
1215                 return -EBUSY;
1216         return 0;
1217 }
1218
1219 int nd_region_conflict(struct nd_region *nd_region, resource_size_t start,
1220                 resource_size_t size)
1221 {
1222         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
1223         struct conflict_context ctx = {
1224                 .nd_region = nd_region,
1225                 .start = start,
1226                 .size = size,
1227         };
1228
1229         return device_for_each_child(&nvdimm_bus->dev, &ctx, region_conflict);
1230 }
1231
1232 void __exit nd_region_devs_exit(void)
1233 {
1234         ida_destroy(&region_ida);
1235 }