drivers/nvdimm/region_devs.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/scatterlist.h>
  14 #include <linux/highmem.h>
  15 #include <linux/sched.h>
  16 #include <linux/slab.h>
  17 #include <linux/hash.h>
  18 #include <linux/pmem.h>
  19 #include <linux/sort.h>
  20 #include <linux/io.h>
  21 #include <linux/nd.h>
  22 #include "nd-core.h"
  23 #include "nd.h"
  24
  25 /*
  26  * For readq() and writeq() on 32-bit builds, the hi-lo, lo-hi order is
  27  * irrelevant.
  28  */
  29 #include <linux/io-64-nonatomic-hi-lo.h>
  30
  31 static DEFINE_IDA(region_ida);
  32 static DEFINE_PER_CPU(int, flush_idx);
  33
  34 static int nvdimm_map_flush(struct device *dev, struct nvdimm *nvdimm, int dimm,
  35                 struct nd_region_data *ndrd)
  36 {
  37         int i, j;
  38
  39         dev_dbg(dev, "%s: map %d flush address%s\n", nvdimm_name(nvdimm),
  40                         nvdimm->num_flush, nvdimm->num_flush == 1 ? "" : "es");
  41         for (i = 0; i < (1 << ndrd->hints_shift); i++) {
  42                 struct resource *res = &nvdimm->flush_wpq[i];
  43                 unsigned long pfn = PHYS_PFN(res->start);
  44                 void __iomem *flush_page;
  45
  46                 /* check if flush hints share a page */
  47                 for (j = 0; j < i; j++) {
  48                         struct resource *res_j = &nvdimm->flush_wpq[j];
  49                         unsigned long pfn_j = PHYS_PFN(res_j->start);
  50
  51                         if (pfn == pfn_j)
  52                                 break;
  53                 }
  54
  55                 if (j < i)
  56                         flush_page = (void __iomem *) ((unsigned long)
  57                                         ndrd_get_flush_wpq(ndrd, dimm, j)
  58                                         & PAGE_MASK);
  59                 else
  60                         flush_page = devm_nvdimm_ioremap(dev,
  61                                         PFN_PHYS(pfn), PAGE_SIZE);
  62                 if (!flush_page)
  63                         return -ENXIO;
  64                 ndrd_set_flush_wpq(ndrd, dimm, i, flush_page
  65                                 + (res->start & ~PAGE_MASK));
  66         }
  67
  68         return 0;
  69 }
  70
  71 int nd_region_activate(struct nd_region *nd_region)
  72 {
  73         int i, num_flush = 0;
  74         struct nd_region_data *ndrd;
  75         struct device *dev = &nd_region->dev;
  76         size_t flush_data_size = sizeof(void *);
  77
  78         nvdimm_bus_lock(&nd_region->dev);
  79         for (i = 0; i < nd_region->ndr_mappings; i++) {
  80                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
  81                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
  82
  83                 /* at least one null hint slot per-dimm for the "no-hint" case */
  84                 flush_data_size += sizeof(void *);
  85                 num_flush = min_not_zero(num_flush, nvdimm->num_flush);
  86                 if (!nvdimm->num_flush)
  87                         continue;
  88                 flush_data_size += nvdimm->num_flush * sizeof(void *);
  89         }
  90         nvdimm_bus_unlock(&nd_region->dev);
  91
  92         ndrd = devm_kzalloc(dev, sizeof(*ndrd) + flush_data_size, GFP_KERNEL);
  93         if (!ndrd)
  94                 return -ENOMEM;
  95         dev_set_drvdata(dev, ndrd);
  96
  97         if (!num_flush)
  98                 return 0;
  99
 100         ndrd->hints_shift = ilog2(num_flush);
 101         for (i = 0; i < nd_region->ndr_mappings; i++) {
 102                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 103                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 104                 int rc = nvdimm_map_flush(&nd_region->dev, nvdimm, i, ndrd);
 105
 106                 if (rc)
 107                         return rc;
 108         }
 109
 110         return 0;
 111 }
 112
 113 static void nd_region_release(struct device *dev)
 114 {
 115         struct nd_region *nd_region = to_nd_region(dev);
 116         u16 i;
 117
 118         for (i = 0; i < nd_region->ndr_mappings; i++) {
 119                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 120                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 121
 122                 put_device(&nvdimm->dev);
 123         }
 124         free_percpu(nd_region->lane);
 125         ida_simple_remove(&region_ida, nd_region->id);
 126         if (is_nd_blk(dev))
 127                 kfree(to_nd_blk_region(dev));
 128         else
 129                 kfree(nd_region);
 130 }
 131
 132 static struct device_type nd_blk_device_type = {
 133         .name = "nd_blk",
 134         .release = nd_region_release,
 135 };
 136
 137 static struct device_type nd_pmem_device_type = {
 138         .name = "nd_pmem",
 139         .release = nd_region_release,
 140 };
 141
 142 static struct device_type nd_volatile_device_type = {
 143         .name = "nd_volatile",
 144         .release = nd_region_release,
 145 };
 146
 147 bool is_nd_pmem(struct device *dev)
 148 {
 149         return dev ? dev->type == &nd_pmem_device_type : false;
 150 }
 151
 152 bool is_nd_blk(struct device *dev)
 153 {
 154         return dev ? dev->type == &nd_blk_device_type : false;
 155 }
 156
 157 struct nd_region *to_nd_region(struct device *dev)
 158 {
 159         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
 160
 161         WARN_ON(dev->type->release != nd_region_release);
 162         return nd_region;
 163 }
 164 EXPORT_SYMBOL_GPL(to_nd_region);
 165
 166 struct nd_blk_region *to_nd_blk_region(struct device *dev)
 167 {
 168         struct nd_region *nd_region = to_nd_region(dev);
 169
 170         WARN_ON(!is_nd_blk(dev));
 171         return container_of(nd_region, struct nd_blk_region, nd_region);
 172 }
 173 EXPORT_SYMBOL_GPL(to_nd_blk_region);
 174
 175 void *nd_region_provider_data(struct nd_region *nd_region)
 176 {
 177         return nd_region->provider_data;
 178 }
 179 EXPORT_SYMBOL_GPL(nd_region_provider_data);
 180
 181 void *nd_blk_region_provider_data(struct nd_blk_region *ndbr)
 182 {
 183         return ndbr->blk_provider_data;
 184 }
 185 EXPORT_SYMBOL_GPL(nd_blk_region_provider_data);
 186
 187 void nd_blk_region_set_provider_data(struct nd_blk_region *ndbr, void *data)
 188 {
 189         ndbr->blk_provider_data = data;
 190 }
 191 EXPORT_SYMBOL_GPL(nd_blk_region_set_provider_data);
 192
 193 /**
 194  * nd_region_to_nstype() - region to an integer namespace type
 195  * @nd_region: region-device to interrogate
 196  *
 197  * This is the 'nstype' attribute of a region as well, an input to the
 198  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
 199  * namespace devices with namespace drivers.
 200  */
 201 int nd_region_to_nstype(struct nd_region *nd_region)
 202 {
 203         if (is_nd_pmem(&nd_region->dev)) {
 204                 u16 i, alias;
 205
 206                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
 207                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 208                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 209
 210                         if (nvdimm->flags & NDD_ALIASING)
 211                                 alias++;
 212                 }
 213                 if (alias)
 214                         return ND_DEVICE_NAMESPACE_PMEM;
 215                 else
 216                         return ND_DEVICE_NAMESPACE_IO;
 217         } else if (is_nd_blk(&nd_region->dev)) {
 218                 return ND_DEVICE_NAMESPACE_BLK;
 219         }
 220
 221         return 0;
 222 }
 223 EXPORT_SYMBOL(nd_region_to_nstype);
 224
 225 static ssize_t size_show(struct device *dev,
 226                 struct device_attribute *attr, char *buf)
 227 {
 228         struct nd_region *nd_region = to_nd_region(dev);
 229         unsigned long long size = 0;
 230
 231         if (is_nd_pmem(dev)) {
 232                 size = nd_region->ndr_size;
 233         } else if (nd_region->ndr_mappings == 1) {
 234                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 235
 236                 size = nd_mapping->size;
 237         }
 238
 239         return sprintf(buf, "%llu\n", size);
 240 }
 241 static DEVICE_ATTR_RO(size);
 242
 243 static ssize_t mappings_show(struct device *dev,
 244                 struct device_attribute *attr, char *buf)
 245 {
 246         struct nd_region *nd_region = to_nd_region(dev);
 247
 248         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
 249 }
 250 static DEVICE_ATTR_RO(mappings);
 251
 252 static ssize_t nstype_show(struct device *dev,
 253                 struct device_attribute *attr, char *buf)
 254 {
 255         struct nd_region *nd_region = to_nd_region(dev);
 256
 257         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 258 }
 259 static DEVICE_ATTR_RO(nstype);
 260
 261 static ssize_t set_cookie_show(struct device *dev,
 262                 struct device_attribute *attr, char *buf)
 263 {
 264         struct nd_region *nd_region = to_nd_region(dev);
 265         struct nd_interleave_set *nd_set = nd_region->nd_set;
 266
 267         if (is_nd_pmem(dev) && nd_set)
 268                 /* pass, should be precluded by region_visible */;
 269         else
 270                 return -ENXIO;
 271
 272         return sprintf(buf, "%#llx\n", nd_set->cookie);
 273 }
 274 static DEVICE_ATTR_RO(set_cookie);
 275
 276 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
 277 {
 278         resource_size_t blk_max_overlap = 0, available, overlap;
 279         int i;
 280
 281         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
 282
 283  retry:
 284         available = 0;
 285         overlap = blk_max_overlap;
 286         for (i = 0; i < nd_region->ndr_mappings; i++) {
 287                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 288                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 289
 290                 /* if a dimm is disabled the available capacity is zero */
 291                 if (!ndd)
 292                         return 0;
 293
 294                 if (is_nd_pmem(&nd_region->dev)) {
 295                         available += nd_pmem_available_dpa(nd_region,
 296                                         nd_mapping, &overlap);
 297                         if (overlap > blk_max_overlap) {
 298                                 blk_max_overlap = overlap;
 299                                 goto retry;
 300                         }
 301                 } else if (is_nd_blk(&nd_region->dev)) {
 302                         available += nd_blk_available_dpa(nd_mapping);
 303                 }
 304         }
 305
 306         return available;
 307 }
 308
 309 static ssize_t available_size_show(struct device *dev,
 310                 struct device_attribute *attr, char *buf)
 311 {
 312         struct nd_region *nd_region = to_nd_region(dev);
 313         unsigned long long available = 0;
 314
 315         /*
 316          * Flush in-flight updates and grab a snapshot of the available
 317          * size.  Of course, this value is potentially invalidated the
 318          * memory nvdimm_bus_lock() is dropped, but that's userspace's
 319          * problem to not race itself.
 320          */
 321         nvdimm_bus_lock(dev);
 322         wait_nvdimm_bus_probe_idle(dev);
 323         available = nd_region_available_dpa(nd_region);
 324         nvdimm_bus_unlock(dev);
 325
 326         return sprintf(buf, "%llu\n", available);
 327 }
 328 static DEVICE_ATTR_RO(available_size);
 329
 330 static ssize_t init_namespaces_show(struct device *dev,
 331                 struct device_attribute *attr, char *buf)
 332 {
 333         struct nd_region_data *ndrd = dev_get_drvdata(dev);
 334         ssize_t rc;
 335
 336         nvdimm_bus_lock(dev);
 337         if (ndrd)
 338                 rc = sprintf(buf, "%d/%d\n", ndrd->ns_active, ndrd->ns_count);
 339         else
 340                 rc = -ENXIO;
 341         nvdimm_bus_unlock(dev);
 342
 343         return rc;
 344 }
 345 static DEVICE_ATTR_RO(init_namespaces);
 346
 347 static ssize_t namespace_seed_show(struct device *dev,
 348                 struct device_attribute *attr, char *buf)
 349 {
 350         struct nd_region *nd_region = to_nd_region(dev);
 351         ssize_t rc;
 352
 353         nvdimm_bus_lock(dev);
 354         if (nd_region->ns_seed)
 355                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
 356         else
 357                 rc = sprintf(buf, "\n");
 358         nvdimm_bus_unlock(dev);
 359         return rc;
 360 }
 361 static DEVICE_ATTR_RO(namespace_seed);
 362
 363 static ssize_t btt_seed_show(struct device *dev,
 364                 struct device_attribute *attr, char *buf)
 365 {
 366         struct nd_region *nd_region = to_nd_region(dev);
 367         ssize_t rc;
 368
 369         nvdimm_bus_lock(dev);
 370         if (nd_region->btt_seed)
 371                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
 372         else
 373                 rc = sprintf(buf, "\n");
 374         nvdimm_bus_unlock(dev);
 375
 376         return rc;
 377 }
 378 static DEVICE_ATTR_RO(btt_seed);
 379
 380 static ssize_t pfn_seed_show(struct device *dev,
 381                 struct device_attribute *attr, char *buf)
 382 {
 383         struct nd_region *nd_region = to_nd_region(dev);
 384         ssize_t rc;
 385
 386         nvdimm_bus_lock(dev);
 387         if (nd_region->pfn_seed)
 388                 rc = sprintf(buf, "%s\n", dev_name(nd_region->pfn_seed));
 389         else
 390                 rc = sprintf(buf, "\n");
 391         nvdimm_bus_unlock(dev);
 392
 393         return rc;
 394 }
 395 static DEVICE_ATTR_RO(pfn_seed);
 396
 397 static ssize_t dax_seed_show(struct device *dev,
 398                 struct device_attribute *attr, char *buf)
 399 {
 400         struct nd_region *nd_region = to_nd_region(dev);
 401         ssize_t rc;
 402
 403         nvdimm_bus_lock(dev);
 404         if (nd_region->dax_seed)
 405                 rc = sprintf(buf, "%s\n", dev_name(nd_region->dax_seed));
 406         else
 407                 rc = sprintf(buf, "\n");
 408         nvdimm_bus_unlock(dev);
 409
 410         return rc;
 411 }
 412 static DEVICE_ATTR_RO(dax_seed);
 413
 414 static ssize_t read_only_show(struct device *dev,
 415                 struct device_attribute *attr, char *buf)
 416 {
 417         struct nd_region *nd_region = to_nd_region(dev);
 418
 419         return sprintf(buf, "%d\n", nd_region->ro);
 420 }
 421
 422 static ssize_t read_only_store(struct device *dev,
 423                 struct device_attribute *attr, const char *buf, size_t len)
 424 {
 425         bool ro;
 426         int rc = strtobool(buf, &ro);
 427         struct nd_region *nd_region = to_nd_region(dev);
 428
 429         if (rc)
 430                 return rc;
 431
 432         nd_region->ro = ro;
 433         return len;
 434 }
 435 static DEVICE_ATTR_RW(read_only);
 436
 437 static struct attribute *nd_region_attributes[] = {
 438         &dev_attr_size.attr,
 439         &dev_attr_nstype.attr,
 440         &dev_attr_mappings.attr,
 441         &dev_attr_btt_seed.attr,
 442         &dev_attr_pfn_seed.attr,
 443         &dev_attr_dax_seed.attr,
 444         &dev_attr_read_only.attr,
 445         &dev_attr_set_cookie.attr,
 446         &dev_attr_available_size.attr,
 447         &dev_attr_namespace_seed.attr,
 448         &dev_attr_init_namespaces.attr,
 449         NULL,
 450 };
 451
 452 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
 453 {
 454         struct device *dev = container_of(kobj, typeof(*dev), kobj);
 455         struct nd_region *nd_region = to_nd_region(dev);
 456         struct nd_interleave_set *nd_set = nd_region->nd_set;
 457         int type = nd_region_to_nstype(nd_region);
 458
 459         if (!is_nd_pmem(dev) && a == &dev_attr_pfn_seed.attr)
 460                 return 0;
 461
 462         if (!is_nd_pmem(dev) && a == &dev_attr_dax_seed.attr)
 463                 return 0;
 464
 465         if (a != &dev_attr_set_cookie.attr
 466                         && a != &dev_attr_available_size.attr)
 467                 return a->mode;
 468
 469         if ((type == ND_DEVICE_NAMESPACE_PMEM
 470                                 || type == ND_DEVICE_NAMESPACE_BLK)
 471                         && a == &dev_attr_available_size.attr)
 472                 return a->mode;
 473         else if (is_nd_pmem(dev) && nd_set)
 474                 return a->mode;
 475
 476         return 0;
 477 }
 478
 479 struct attribute_group nd_region_attribute_group = {
 480         .attrs = nd_region_attributes,
 481         .is_visible = region_visible,
 482 };
 483 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
 484
 485 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
 486 {
 487         struct nd_interleave_set *nd_set = nd_region->nd_set;
 488
 489         if (nd_set)
 490                 return nd_set->cookie;
 491         return 0;
 492 }
 493
 494 /*
 495  * Upon successful probe/remove, take/release a reference on the
 496  * associated interleave set (if present), and plant new btt + namespace
 497  * seeds.  Also, on the removal of a BLK region, notify the provider to
 498  * disable the region.
 499  */
 500 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
 501                 struct device *dev, bool probe)
 502 {
 503         struct nd_region *nd_region;
 504
 505         if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
 506                 int i;
 507
 508                 nd_region = to_nd_region(dev);
 509                 for (i = 0; i < nd_region->ndr_mappings; i++) {
 510                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 511                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
 512                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 513
 514                         kfree(nd_mapping->labels);
 515                         nd_mapping->labels = NULL;
 516                         put_ndd(ndd);
 517                         nd_mapping->ndd = NULL;
 518                         if (ndd)
 519                                 atomic_dec(&nvdimm->busy);
 520                 }
 521
 522                 if (is_nd_pmem(dev))
 523                         return;
 524         }
 525         if (dev->parent && is_nd_blk(dev->parent) && probe) {
 526                 nd_region = to_nd_region(dev->parent);
 527                 nvdimm_bus_lock(dev);
 528                 if (nd_region->ns_seed == dev)
 529                         nd_region_create_blk_seed(nd_region);
 530                 nvdimm_bus_unlock(dev);
 531         }
 532         if (is_nd_btt(dev) && probe) {
 533                 struct nd_btt *nd_btt = to_nd_btt(dev);
 534
 535                 nd_region = to_nd_region(dev->parent);
 536                 nvdimm_bus_lock(dev);
 537                 if (nd_region->btt_seed == dev)
 538                         nd_region_create_btt_seed(nd_region);
 539                 if (nd_region->ns_seed == &nd_btt->ndns->dev &&
 540                                 is_nd_blk(dev->parent))
 541                         nd_region_create_blk_seed(nd_region);
 542                 nvdimm_bus_unlock(dev);
 543         }
 544         if (is_nd_pfn(dev) && probe) {
 545                 nd_region = to_nd_region(dev->parent);
 546                 nvdimm_bus_lock(dev);
 547                 if (nd_region->pfn_seed == dev)
 548                         nd_region_create_pfn_seed(nd_region);
 549                 nvdimm_bus_unlock(dev);
 550         }
 551         if (is_nd_dax(dev) && probe) {
 552                 nd_region = to_nd_region(dev->parent);
 553                 nvdimm_bus_lock(dev);
 554                 if (nd_region->dax_seed == dev)
 555                         nd_region_create_dax_seed(nd_region);
 556                 nvdimm_bus_unlock(dev);
 557         }
 558 }
 559
 560 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 561 {
 562         nd_region_notify_driver_action(nvdimm_bus, dev, true);
 563 }
 564
 565 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
 566 {
 567         nd_region_notify_driver_action(nvdimm_bus, dev, false);
 568 }
 569
 570 static ssize_t mappingN(struct device *dev, char *buf, int n)
 571 {
 572         struct nd_region *nd_region = to_nd_region(dev);
 573         struct nd_mapping *nd_mapping;
 574         struct nvdimm *nvdimm;
 575
 576         if (n >= nd_region->ndr_mappings)
 577                 return -ENXIO;
 578         nd_mapping = &nd_region->mapping[n];
 579         nvdimm = nd_mapping->nvdimm;
 580
 581         return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
 582                         nd_mapping->start, nd_mapping->size);
 583 }
 584
 585 #define REGION_MAPPING(idx) \
 586 static ssize_t mapping##idx##_show(struct device *dev,          \
 587                 struct device_attribute *attr, char *buf)       \
 588 {                                                               \
 589         return mappingN(dev, buf, idx);                         \
 590 }                                                               \
 591 static DEVICE_ATTR_RO(mapping##idx)
 592
 593 /*
 594  * 32 should be enough for a while, even in the presence of socket
 595  * interleave a 32-way interleave set is a degenerate case.
 596  */
 597 REGION_MAPPING(0);
 598 REGION_MAPPING(1);
 599 REGION_MAPPING(2);
 600 REGION_MAPPING(3);
 601 REGION_MAPPING(4);
 602 REGION_MAPPING(5);
 603 REGION_MAPPING(6);
 604 REGION_MAPPING(7);
 605 REGION_MAPPING(8);
 606 REGION_MAPPING(9);
 607 REGION_MAPPING(10);
 608 REGION_MAPPING(11);
 609 REGION_MAPPING(12);
 610 REGION_MAPPING(13);
 611 REGION_MAPPING(14);
 612 REGION_MAPPING(15);
 613 REGION_MAPPING(16);
 614 REGION_MAPPING(17);
 615 REGION_MAPPING(18);
 616 REGION_MAPPING(19);
 617 REGION_MAPPING(20);
 618 REGION_MAPPING(21);
 619 REGION_MAPPING(22);
 620 REGION_MAPPING(23);
 621 REGION_MAPPING(24);
 622 REGION_MAPPING(25);
 623 REGION_MAPPING(26);
 624 REGION_MAPPING(27);
 625 REGION_MAPPING(28);
 626 REGION_MAPPING(29);
 627 REGION_MAPPING(30);
 628 REGION_MAPPING(31);
 629
 630 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
 631 {
 632         struct device *dev = container_of(kobj, struct device, kobj);
 633         struct nd_region *nd_region = to_nd_region(dev);
 634
 635         if (n < nd_region->ndr_mappings)
 636                 return a->mode;
 637         return 0;
 638 }
 639
 640 static struct attribute *mapping_attributes[] = {
 641         &dev_attr_mapping0.attr,
 642         &dev_attr_mapping1.attr,
 643         &dev_attr_mapping2.attr,
 644         &dev_attr_mapping3.attr,
 645         &dev_attr_mapping4.attr,
 646         &dev_attr_mapping5.attr,
 647         &dev_attr_mapping6.attr,
 648         &dev_attr_mapping7.attr,
 649         &dev_attr_mapping8.attr,
 650         &dev_attr_mapping9.attr,
 651         &dev_attr_mapping10.attr,
 652         &dev_attr_mapping11.attr,
 653         &dev_attr_mapping12.attr,
 654         &dev_attr_mapping13.attr,
 655         &dev_attr_mapping14.attr,
 656         &dev_attr_mapping15.attr,
 657         &dev_attr_mapping16.attr,
 658         &dev_attr_mapping17.attr,
 659         &dev_attr_mapping18.attr,
 660         &dev_attr_mapping19.attr,
 661         &dev_attr_mapping20.attr,
 662         &dev_attr_mapping21.attr,
 663         &dev_attr_mapping22.attr,
 664         &dev_attr_mapping23.attr,
 665         &dev_attr_mapping24.attr,
 666         &dev_attr_mapping25.attr,
 667         &dev_attr_mapping26.attr,
 668         &dev_attr_mapping27.attr,
 669         &dev_attr_mapping28.attr,
 670         &dev_attr_mapping29.attr,
 671         &dev_attr_mapping30.attr,
 672         &dev_attr_mapping31.attr,
 673         NULL,
 674 };
 675
 676 struct attribute_group nd_mapping_attribute_group = {
 677         .is_visible = mapping_visible,
 678         .attrs = mapping_attributes,
 679 };
 680 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
 681
 682 int nd_blk_region_init(struct nd_region *nd_region)
 683 {
 684         struct device *dev = &nd_region->dev;
 685         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
 686
 687         if (!is_nd_blk(dev))
 688                 return 0;
 689
 690         if (nd_region->ndr_mappings < 1) {
 691                 dev_err(dev, "invalid BLK region\n");
 692                 return -ENXIO;
 693         }
 694
 695         return to_nd_blk_region(dev)->enable(nvdimm_bus, dev);
 696 }
 697
 698 /**
 699  * nd_region_acquire_lane - allocate and lock a lane
 700  * @nd_region: region id and number of lanes possible
 701  *
 702  * A lane correlates to a BLK-data-window and/or a log slot in the BTT.
 703  * We optimize for the common case where there are 256 lanes, one
 704  * per-cpu.  For larger systems we need to lock to share lanes.  For now
 705  * this implementation assumes the cost of maintaining an allocator for
 706  * free lanes is on the order of the lock hold time, so it implements a
 707  * static lane = cpu % num_lanes mapping.
 708  *
 709  * In the case of a BTT instance on top of a BLK namespace a lane may be
 710  * acquired recursively.  We lock on the first instance.
 711  *
 712  * In the case of a BTT instance on top of PMEM, we only acquire a lane
 713  * for the BTT metadata updates.
 714  */
 715 unsigned int nd_region_acquire_lane(struct nd_region *nd_region)
 716 {
 717         unsigned int cpu, lane;
 718
 719         cpu = get_cpu();
 720         if (nd_region->num_lanes < nr_cpu_ids) {
 721                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 722
 723                 lane = cpu % nd_region->num_lanes;
 724                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 725                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 726                 if (ndl_count->count++ == 0)
 727                         spin_lock(&ndl_lock->lock);
 728         } else
 729                 lane = cpu;
 730
 731         return lane;
 732 }
 733 EXPORT_SYMBOL(nd_region_acquire_lane);
 734
 735 void nd_region_release_lane(struct nd_region *nd_region, unsigned int lane)
 736 {
 737         if (nd_region->num_lanes < nr_cpu_ids) {
 738                 unsigned int cpu = get_cpu();
 739                 struct nd_percpu_lane *ndl_lock, *ndl_count;
 740
 741                 ndl_count = per_cpu_ptr(nd_region->lane, cpu);
 742                 ndl_lock = per_cpu_ptr(nd_region->lane, lane);
 743                 if (--ndl_count->count == 0)
 744                         spin_unlock(&ndl_lock->lock);
 745                 put_cpu();
 746         }
 747         put_cpu();
 748 }
 749 EXPORT_SYMBOL(nd_region_release_lane);
 750
 751 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
 752                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
 753                 const char *caller)
 754 {
 755         struct nd_region *nd_region;
 756         struct device *dev;
 757         void *region_buf;
 758         unsigned int i;
 759         int ro = 0;
 760
 761         for (i = 0; i < ndr_desc->num_mappings; i++) {
 762                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
 763                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 764
 765                 if ((nd_mapping->start | nd_mapping->size) % SZ_4K) {
 766                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
 767                                         caller, dev_name(&nvdimm->dev), i);
 768
 769                         return NULL;
 770                 }
 771
 772                 if (nvdimm->flags & NDD_UNARMED)
 773                         ro = 1;
 774         }
 775
 776         if (dev_type == &nd_blk_device_type) {
 777                 struct nd_blk_region_desc *ndbr_desc;
 778                 struct nd_blk_region *ndbr;
 779
 780                 ndbr_desc = to_blk_region_desc(ndr_desc);
 781                 ndbr = kzalloc(sizeof(*ndbr) + sizeof(struct nd_mapping)
 782                                 * ndr_desc->num_mappings,
 783                                 GFP_KERNEL);
 784                 if (ndbr) {
 785                         nd_region = &ndbr->nd_region;
 786                         ndbr->enable = ndbr_desc->enable;
 787                         ndbr->do_io = ndbr_desc->do_io;
 788                 }
 789                 region_buf = ndbr;
 790         } else {
 791                 nd_region = kzalloc(sizeof(struct nd_region)
 792                                 + sizeof(struct nd_mapping)
 793                                 * ndr_desc->num_mappings,
 794                                 GFP_KERNEL);
 795                 region_buf = nd_region;
 796         }
 797
 798         if (!region_buf)
 799                 return NULL;
 800         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
 801         if (nd_region->id < 0)
 802                 goto err_id;
 803
 804         nd_region->lane = alloc_percpu(struct nd_percpu_lane);
 805         if (!nd_region->lane)
 806                 goto err_percpu;
 807
 808         for (i = 0; i < nr_cpu_ids; i++) {
 809                 struct nd_percpu_lane *ndl;
 810
 811                 ndl = per_cpu_ptr(nd_region->lane, i);
 812                 spin_lock_init(&ndl->lock);
 813                 ndl->count = 0;
 814         }
 815
 816         memcpy(nd_region->mapping, ndr_desc->nd_mapping,
 817                         sizeof(struct nd_mapping) * ndr_desc->num_mappings);
 818         for (i = 0; i < ndr_desc->num_mappings; i++) {
 819                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
 820                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 821
 822                 get_device(&nvdimm->dev);
 823         }
 824         nd_region->ndr_mappings = ndr_desc->num_mappings;
 825         nd_region->provider_data = ndr_desc->provider_data;
 826         nd_region->nd_set = ndr_desc->nd_set;
 827         nd_region->num_lanes = ndr_desc->num_lanes;
 828         nd_region->flags = ndr_desc->flags;
 829         nd_region->ro = ro;
 830         nd_region->numa_node = ndr_desc->numa_node;
 831         ida_init(&nd_region->ns_ida);
 832         ida_init(&nd_region->btt_ida);
 833         ida_init(&nd_region->pfn_ida);
 834         ida_init(&nd_region->dax_ida);
 835         dev = &nd_region->dev;
 836         dev_set_name(dev, "region%d", nd_region->id);
 837         dev->parent = &nvdimm_bus->dev;
 838         dev->type = dev_type;
 839         dev->groups = ndr_desc->attr_groups;
 840         nd_region->ndr_size = resource_size(ndr_desc->res);
 841         nd_region->ndr_start = ndr_desc->res->start;
 842         nd_device_register(dev);
 843
 844         return nd_region;
 845
 846  err_percpu:
 847         ida_simple_remove(&region_ida, nd_region->id);
 848  err_id:
 849         kfree(region_buf);
 850         return NULL;
 851 }
 852
 853 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
 854                 struct nd_region_desc *ndr_desc)
 855 {
 856         ndr_desc->num_lanes = ND_MAX_LANES;
 857         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
 858                         __func__);
 859 }
 860 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
 861
 862 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
 863                 struct nd_region_desc *ndr_desc)
 864 {
 865         if (ndr_desc->num_mappings > 1)
 866                 return NULL;
 867         ndr_desc->num_lanes = min(ndr_desc->num_lanes, ND_MAX_LANES);
 868         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
 869                         __func__);
 870 }
 871 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
 872
 873 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
 874                 struct nd_region_desc *ndr_desc)
 875 {
 876         ndr_desc->num_lanes = ND_MAX_LANES;
 877         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
 878                         __func__);
 879 }
 880 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);
 881
 882 /**
 883  * nvdimm_flush - flush any posted write queues between the cpu and pmem media
 884  * @nd_region: blk or interleaved pmem region
 885  */
 886 void nvdimm_flush(struct nd_region *nd_region)
 887 {
 888         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
 889         int i, idx;
 890
 891         /*
 892          * Try to encourage some diversity in flush hint addresses
 893          * across cpus assuming a limited number of flush hints.
 894          */
 895         idx = this_cpu_read(flush_idx);
 896         idx = this_cpu_add_return(flush_idx, hash_32(current->pid + idx, 8));
 897
 898         /*
 899          * The first wmb() is needed to 'sfence' all previous writes
 900          * such that they are architecturally visible for the platform
 901          * buffer flush.  Note that we've already arranged for pmem
 902          * writes to avoid the cache via arch_memcpy_to_pmem().  The
 903          * final wmb() ensures ordering for the NVDIMM flush write.
 904          */
 905         wmb();
 906         for (i = 0; i < nd_region->ndr_mappings; i++)
 907                 if (ndrd_get_flush_wpq(ndrd, i, 0))
 908                         writeq(1, ndrd_get_flush_wpq(ndrd, i, idx));
 909         wmb();
 910 }
 911 EXPORT_SYMBOL_GPL(nvdimm_flush);
 912
 913 /**
 914  * nvdimm_has_flush - determine write flushing requirements
 915  * @nd_region: blk or interleaved pmem region
 916  *
 917  * Returns 1 if writes require flushing
 918  * Returns 0 if writes do not require flushing
 919  * Returns -ENXIO if flushing capability can not be determined
 920  */
 921 int nvdimm_has_flush(struct nd_region *nd_region)
 922 {
 923         struct nd_region_data *ndrd = dev_get_drvdata(&nd_region->dev);
 924         int i;
 925
 926         /* no nvdimm == flushing capability unknown */
 927         if (nd_region->ndr_mappings == 0)
 928                 return -ENXIO;
 929
 930         for (i = 0; i < nd_region->ndr_mappings; i++)
 931                 /* flush hints present, flushing required */
 932                 if (ndrd_get_flush_wpq(ndrd, i, 0))
 933                         return 1;
 934
 935         /*
 936          * The platform defines dimm devices without hints, assume
 937          * platform persistence mechanism like ADR
 938          */
 939         return 0;
 940 }
 941 EXPORT_SYMBOL_GPL(nvdimm_has_flush);
 942
 943 void __exit nd_region_devs_exit(void)
 944 {
 945         ida_destroy(&region_ida);
 946 }