drivers/nvdimm/pmem.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Persistent Memory Driver
   4  *
   5  * Copyright (c) 2014-2015, Intel Corporation.
   6  * Copyright (c) 2015, Christoph Hellwig <[email protected]>.
   7  * Copyright (c) 2015, Boaz Harrosh <[email protected]>.
   8  */
   9
  10 #include <linux/blkdev.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/init.h>
  14 #include <linux/platform_device.h>
  15 #include <linux/set_memory.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/badblocks.h>
  19 #include <linux/memremap.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/blk-mq.h>
  22 #include <linux/pfn_t.h>
  23 #include <linux/slab.h>
  24 #include <linux/uio.h>
  25 #include <linux/dax.h>
  26 #include <linux/nd.h>
  27 #include <linux/mm.h>
  28 #include <asm/cacheflush.h>
  29 #include "pmem.h"
  30 #include "btt.h"
  31 #include "pfn.h"
  32 #include "nd.h"
  33
  34 static struct device *to_dev(struct pmem_device *pmem)
  35 {
  36         /*
  37          * nvdimm bus services need a 'dev' parameter, and we record the device
  38          * at init in bb.dev.
  39          */
  40         return pmem->bb.dev;
  41 }
  42
  43 static struct nd_region *to_region(struct pmem_device *pmem)
  44 {
  45         return to_nd_region(to_dev(pmem)->parent);
  46 }
  47
  48 static phys_addr_t pmem_to_phys(struct pmem_device *pmem, phys_addr_t offset)
  49 {
  50         return pmem->phys_addr + offset;
  51 }
  52
  53 static sector_t to_sect(struct pmem_device *pmem, phys_addr_t offset)
  54 {
  55         return (offset - pmem->data_offset) >> SECTOR_SHIFT;
  56 }
  57
  58 static phys_addr_t to_offset(struct pmem_device *pmem, sector_t sector)
  59 {
  60         return (sector << SECTOR_SHIFT) + pmem->data_offset;
  61 }
  62
  63 static void pmem_mkpage_present(struct pmem_device *pmem, phys_addr_t offset,
  64                 unsigned int len)
  65 {
  66         phys_addr_t phys = pmem_to_phys(pmem, offset);
  67         unsigned long pfn_start, pfn_end, pfn;
  68
  69         /* only pmem in the linear map supports HWPoison */
  70         if (is_vmalloc_addr(pmem->virt_addr))
  71                 return;
  72
  73         pfn_start = PHYS_PFN(phys);
  74         pfn_end = pfn_start + PHYS_PFN(len);
  75         for (pfn = pfn_start; pfn < pfn_end; pfn++) {
  76                 struct page *page = pfn_to_page(pfn);
  77
  78                 /*
  79                  * Note, no need to hold a get_dev_pagemap() reference
  80                  * here since we're in the driver I/O path and
  81                  * outstanding I/O requests pin the dev_pagemap.
  82                  */
  83                 if (test_and_clear_pmem_poison(page))
  84                         clear_mce_nospec(pfn);
  85         }
  86 }
  87
  88 static void pmem_clear_bb(struct pmem_device *pmem, sector_t sector, long blks)
  89 {
  90         if (blks == 0)
  91                 return;
  92         badblocks_clear(&pmem->bb, sector, blks);
  93         if (pmem->bb_state)
  94                 sysfs_notify_dirent(pmem->bb_state);
  95 }
  96
  97 static long __pmem_clear_poison(struct pmem_device *pmem,
  98                 phys_addr_t offset, unsigned int len)
  99 {
 100         phys_addr_t phys = pmem_to_phys(pmem, offset);
 101         long cleared = nvdimm_clear_poison(to_dev(pmem), phys, len);
 102
 103         if (cleared > 0) {
 104                 pmem_mkpage_present(pmem, offset, cleared);
 105                 arch_invalidate_pmem(pmem->virt_addr + offset, len);
 106         }
 107         return cleared;
 108 }
 109
 110 static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
 111                 phys_addr_t offset, unsigned int len)
 112 {
 113         long cleared = __pmem_clear_poison(pmem, offset, len);
 114
 115         if (cleared < 0)
 116                 return BLK_STS_IOERR;
 117
 118         pmem_clear_bb(pmem, to_sect(pmem, offset), cleared >> SECTOR_SHIFT);
 119         if (cleared < len)
 120                 return BLK_STS_IOERR;
 121         return BLK_STS_OK;
 122 }
 123
 124 static void write_pmem(void *pmem_addr, struct page *page,
 125                 unsigned int off, unsigned int len)
 126 {
 127         unsigned int chunk;
 128         void *mem;
 129
 130         while (len) {
 131                 mem = kmap_atomic(page);
 132                 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
 133                 memcpy_flushcache(pmem_addr, mem + off, chunk);
 134                 kunmap_atomic(mem);
 135                 len -= chunk;
 136                 off = 0;
 137                 page++;
 138                 pmem_addr += chunk;
 139         }
 140 }
 141
 142 static blk_status_t read_pmem(struct page *page, unsigned int off,
 143                 void *pmem_addr, unsigned int len)
 144 {
 145         unsigned int chunk;
 146         unsigned long rem;
 147         void *mem;
 148
 149         while (len) {
 150                 mem = kmap_atomic(page);
 151                 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
 152                 rem = copy_mc_to_kernel(mem + off, pmem_addr, chunk);
 153                 kunmap_atomic(mem);
 154                 if (rem)
 155                         return BLK_STS_IOERR;
 156                 len -= chunk;
 157                 off = 0;
 158                 page++;
 159                 pmem_addr += chunk;
 160         }
 161         return BLK_STS_OK;
 162 }
 163
 164 static blk_status_t pmem_do_read(struct pmem_device *pmem,
 165                         struct page *page, unsigned int page_off,
 166                         sector_t sector, unsigned int len)
 167 {
 168         blk_status_t rc;
 169         phys_addr_t pmem_off = to_offset(pmem, sector);
 170         void *pmem_addr = pmem->virt_addr + pmem_off;
 171
 172         if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
 173                 return BLK_STS_IOERR;
 174
 175         rc = read_pmem(page, page_off, pmem_addr, len);
 176         flush_dcache_page(page);
 177         return rc;
 178 }
 179
 180 static blk_status_t pmem_do_write(struct pmem_device *pmem,
 181                         struct page *page, unsigned int page_off,
 182                         sector_t sector, unsigned int len)
 183 {
 184         phys_addr_t pmem_off = to_offset(pmem, sector);
 185         void *pmem_addr = pmem->virt_addr + pmem_off;
 186
 187         if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) {
 188                 blk_status_t rc = pmem_clear_poison(pmem, pmem_off, len);
 189
 190                 if (rc != BLK_STS_OK)
 191                         return rc;
 192         }
 193
 194         flush_dcache_page(page);
 195         write_pmem(pmem_addr, page, page_off, len);
 196
 197         return BLK_STS_OK;
 198 }
 199
 200 static void pmem_submit_bio(struct bio *bio)
 201 {
 202         int ret = 0;
 203         blk_status_t rc = 0;
 204         bool do_acct;
 205         unsigned long start;
 206         struct bio_vec bvec;
 207         struct bvec_iter iter;
 208         struct pmem_device *pmem = bio->bi_bdev->bd_disk->private_data;
 209         struct nd_region *nd_region = to_region(pmem);
 210
 211         if (bio->bi_opf & REQ_PREFLUSH)
 212                 ret = nvdimm_flush(nd_region, bio);
 213
 214         do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
 215         if (do_acct)
 216                 start = bio_start_io_acct(bio);
 217         bio_for_each_segment(bvec, bio, iter) {
 218                 if (op_is_write(bio_op(bio)))
 219                         rc = pmem_do_write(pmem, bvec.bv_page, bvec.bv_offset,
 220                                 iter.bi_sector, bvec.bv_len);
 221                 else
 222                         rc = pmem_do_read(pmem, bvec.bv_page, bvec.bv_offset,
 223                                 iter.bi_sector, bvec.bv_len);
 224                 if (rc) {
 225                         bio->bi_status = rc;
 226                         break;
 227                 }
 228         }
 229         if (do_acct)
 230                 bio_end_io_acct(bio, start);
 231
 232         if (bio->bi_opf & REQ_FUA)
 233                 ret = nvdimm_flush(nd_region, bio);
 234
 235         if (ret)
 236                 bio->bi_status = errno_to_blk_status(ret);
 237
 238         bio_endio(bio);
 239 }
 240
 241 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
 242 __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
 243                 long nr_pages, enum dax_access_mode mode, void **kaddr,
 244                 pfn_t *pfn)
 245 {
 246         resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
 247         sector_t sector = PFN_PHYS(pgoff) >> SECTOR_SHIFT;
 248         unsigned int num = PFN_PHYS(nr_pages) >> SECTOR_SHIFT;
 249         struct badblocks *bb = &pmem->bb;
 250         sector_t first_bad;
 251         int num_bad;
 252
 253         if (kaddr)
 254                 *kaddr = pmem->virt_addr + offset;
 255         if (pfn)
 256                 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
 257
 258         if (bb->count &&
 259             badblocks_check(bb, sector, num, &first_bad, &num_bad)) {
 260                 long actual_nr;
 261
 262                 if (mode != DAX_RECOVERY_WRITE)
 263                         return -EHWPOISON;
 264
 265                 /*
 266                  * Set the recovery stride is set to kernel page size because
 267                  * the underlying driver and firmware clear poison functions
 268                  * don't appear to handle large chunk(such as 2MiB) reliably.
 269                  */
 270                 actual_nr = PHYS_PFN(
 271                         PAGE_ALIGN((first_bad - sector) << SECTOR_SHIFT));
 272                 dev_dbg(pmem->bb.dev, "start sector(%llu), nr_pages(%ld), first_bad(%llu), actual_nr(%ld)\n",
 273                                 sector, nr_pages, first_bad, actual_nr);
 274                 if (actual_nr)
 275                         return actual_nr;
 276                 return 1;
 277         }
 278
 279         /*
 280          * If badblocks are present but not in the range, limit known good range
 281          * to the requested range.
 282          */
 283         if (bb->count)
 284                 return nr_pages;
 285         return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
 286 }
 287
 288 static const struct block_device_operations pmem_fops = {
 289         .owner =                THIS_MODULE,
 290         .submit_bio =           pmem_submit_bio,
 291 };
 292
 293 static int pmem_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
 294                                     size_t nr_pages)
 295 {
 296         struct pmem_device *pmem = dax_get_private(dax_dev);
 297
 298         return blk_status_to_errno(pmem_do_write(pmem, ZERO_PAGE(0), 0,
 299                                    PFN_PHYS(pgoff) >> SECTOR_SHIFT,
 300                                    PAGE_SIZE));
 301 }
 302
 303 static long pmem_dax_direct_access(struct dax_device *dax_dev,
 304                 pgoff_t pgoff, long nr_pages, enum dax_access_mode mode,
 305                 void **kaddr, pfn_t *pfn)
 306 {
 307         struct pmem_device *pmem = dax_get_private(dax_dev);
 308
 309         return __pmem_direct_access(pmem, pgoff, nr_pages, mode, kaddr, pfn);
 310 }
 311
 312 /*
 313  * The recovery write thread started out as a normal pwrite thread and
 314  * when the filesystem was told about potential media error in the
 315  * range, filesystem turns the normal pwrite to a dax_recovery_write.
 316  *
 317  * The recovery write consists of clearing media poison, clearing page
 318  * HWPoison bit, reenable page-wide read-write permission, flush the
 319  * caches and finally write.  A competing pread thread will be held
 320  * off during the recovery process since data read back might not be
 321  * valid, and this is achieved by clearing the badblock records after
 322  * the recovery write is complete. Competing recovery write threads
 323  * are already serialized by writer lock held by dax_iomap_rw().
 324  */
 325 static size_t pmem_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
 326                 void *addr, size_t bytes, struct iov_iter *i)
 327 {
 328         struct pmem_device *pmem = dax_get_private(dax_dev);
 329         size_t olen, len, off;
 330         phys_addr_t pmem_off;
 331         struct device *dev = pmem->bb.dev;
 332         long cleared;
 333
 334         off = offset_in_page(addr);
 335         len = PFN_PHYS(PFN_UP(off + bytes));
 336         if (!is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) >> SECTOR_SHIFT, len))
 337                 return _copy_from_iter_flushcache(addr, bytes, i);
 338
 339         /*
 340          * Not page-aligned range cannot be recovered. This should not
 341          * happen unless something else went wrong.
 342          */
 343         if (off || !PAGE_ALIGNED(bytes)) {
 344                 dev_dbg(dev, "Found poison, but addr(%p) or bytes(%#zx) not page aligned\n",
 345                         addr, bytes);
 346                 return 0;
 347         }
 348
 349         pmem_off = PFN_PHYS(pgoff) + pmem->data_offset;
 350         cleared = __pmem_clear_poison(pmem, pmem_off, len);
 351         if (cleared > 0 && cleared < len) {
 352                 dev_dbg(dev, "poison cleared only %ld out of %zu bytes\n",
 353                         cleared, len);
 354                 return 0;
 355         }
 356         if (cleared < 0) {
 357                 dev_dbg(dev, "poison clear failed: %ld\n", cleared);
 358                 return 0;
 359         }
 360
 361         olen = _copy_from_iter_flushcache(addr, bytes, i);
 362         pmem_clear_bb(pmem, to_sect(pmem, pmem_off), cleared >> SECTOR_SHIFT);
 363
 364         return olen;
 365 }
 366
 367 static const struct dax_operations pmem_dax_ops = {
 368         .direct_access = pmem_dax_direct_access,
 369         .zero_page_range = pmem_dax_zero_page_range,
 370         .recovery_write = pmem_recovery_write,
 371 };
 372
 373 static ssize_t write_cache_show(struct device *dev,
 374                 struct device_attribute *attr, char *buf)
 375 {
 376         struct pmem_device *pmem = dev_to_disk(dev)->private_data;
 377
 378         return sprintf(buf, "%d\n", !!dax_write_cache_enabled(pmem->dax_dev));
 379 }
 380
 381 static ssize_t write_cache_store(struct device *dev,
 382                 struct device_attribute *attr, const char *buf, size_t len)
 383 {
 384         struct pmem_device *pmem = dev_to_disk(dev)->private_data;
 385         bool write_cache;
 386         int rc;
 387
 388         rc = strtobool(buf, &write_cache);
 389         if (rc)
 390                 return rc;
 391         dax_write_cache(pmem->dax_dev, write_cache);
 392         return len;
 393 }
 394 static DEVICE_ATTR_RW(write_cache);
 395
 396 static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
 397 {
 398 #ifndef CONFIG_ARCH_HAS_PMEM_API
 399         if (a == &dev_attr_write_cache.attr)
 400                 return 0;
 401 #endif
 402         return a->mode;
 403 }
 404
 405 static struct attribute *dax_attributes[] = {
 406         &dev_attr_write_cache.attr,
 407         NULL,
 408 };
 409
 410 static const struct attribute_group dax_attribute_group = {
 411         .name           = "dax",
 412         .attrs          = dax_attributes,
 413         .is_visible     = dax_visible,
 414 };
 415
 416 static const struct attribute_group *pmem_attribute_groups[] = {
 417         &dax_attribute_group,
 418         NULL,
 419 };
 420
 421 static void pmem_release_disk(void *__pmem)
 422 {
 423         struct pmem_device *pmem = __pmem;
 424
 425         dax_remove_host(pmem->disk);
 426         kill_dax(pmem->dax_dev);
 427         put_dax(pmem->dax_dev);
 428         del_gendisk(pmem->disk);
 429
 430         put_disk(pmem->disk);
 431 }
 432
 433 static int pmem_pagemap_memory_failure(struct dev_pagemap *pgmap,
 434                 unsigned long pfn, unsigned long nr_pages, int mf_flags)
 435 {
 436         struct pmem_device *pmem =
 437                         container_of(pgmap, struct pmem_device, pgmap);
 438         u64 offset = PFN_PHYS(pfn) - pmem->phys_addr - pmem->data_offset;
 439         u64 len = nr_pages << PAGE_SHIFT;
 440
 441         return dax_holder_notify_failure(pmem->dax_dev, offset, len, mf_flags);
 442 }
 443
 444 static const struct dev_pagemap_ops fsdax_pagemap_ops = {
 445         .memory_failure         = pmem_pagemap_memory_failure,
 446 };
 447
 448 static int pmem_attach_disk(struct device *dev,
 449                 struct nd_namespace_common *ndns)
 450 {
 451         struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
 452         struct nd_region *nd_region = to_nd_region(dev->parent);
 453         int nid = dev_to_node(dev), fua;
 454         struct resource *res = &nsio->res;
 455         struct range bb_range;
 456         struct nd_pfn *nd_pfn = NULL;
 457         struct dax_device *dax_dev;
 458         struct nd_pfn_sb *pfn_sb;
 459         struct pmem_device *pmem;
 460         struct request_queue *q;
 461         struct gendisk *disk;
 462         void *addr;
 463         int rc;
 464
 465         pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
 466         if (!pmem)
 467                 return -ENOMEM;
 468
 469         rc = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
 470         if (rc)
 471                 return rc;
 472
 473         /* while nsio_rw_bytes is active, parse a pfn info block if present */
 474         if (is_nd_pfn(dev)) {
 475                 nd_pfn = to_nd_pfn(dev);
 476                 rc = nvdimm_setup_pfn(nd_pfn, &pmem->pgmap);
 477                 if (rc)
 478                         return rc;
 479         }
 480
 481         /* we're attaching a block device, disable raw namespace access */
 482         devm_namespace_disable(dev, ndns);
 483
 484         dev_set_drvdata(dev, pmem);
 485         pmem->phys_addr = res->start;
 486         pmem->size = resource_size(res);
 487         fua = nvdimm_has_flush(nd_region);
 488         if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) || fua < 0) {
 489                 dev_warn(dev, "unable to guarantee persistence of writes\n");
 490                 fua = 0;
 491         }
 492
 493         if (!devm_request_mem_region(dev, res->start, resource_size(res),
 494                                 dev_name(&ndns->dev))) {
 495                 dev_warn(dev, "could not reserve region %pR\n", res);
 496                 return -EBUSY;
 497         }
 498
 499         disk = blk_alloc_disk(nid);
 500         if (!disk)
 501                 return -ENOMEM;
 502         q = disk->queue;
 503
 504         pmem->disk = disk;
 505         pmem->pgmap.owner = pmem;
 506         pmem->pfn_flags = PFN_DEV;
 507         if (is_nd_pfn(dev)) {
 508                 pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
 509                 pmem->pgmap.ops = &fsdax_pagemap_ops;
 510                 addr = devm_memremap_pages(dev, &pmem->pgmap);
 511                 pfn_sb = nd_pfn->pfn_sb;
 512                 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
 513                 pmem->pfn_pad = resource_size(res) -
 514                         range_len(&pmem->pgmap.range);
 515                 pmem->pfn_flags |= PFN_MAP;
 516                 bb_range = pmem->pgmap.range;
 517                 bb_range.start += pmem->data_offset;
 518         } else if (pmem_should_map_pages(dev)) {
 519                 pmem->pgmap.range.start = res->start;
 520                 pmem->pgmap.range.end = res->end;
 521                 pmem->pgmap.nr_range = 1;
 522                 pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
 523                 pmem->pgmap.ops = &fsdax_pagemap_ops;
 524                 addr = devm_memremap_pages(dev, &pmem->pgmap);
 525                 pmem->pfn_flags |= PFN_MAP;
 526                 bb_range = pmem->pgmap.range;
 527         } else {
 528                 addr = devm_memremap(dev, pmem->phys_addr,
 529                                 pmem->size, ARCH_MEMREMAP_PMEM);
 530                 bb_range.start =  res->start;
 531                 bb_range.end = res->end;
 532         }
 533
 534         if (IS_ERR(addr)) {
 535                 rc = PTR_ERR(addr);
 536                 goto out;
 537         }
 538         pmem->virt_addr = addr;
 539
 540         blk_queue_write_cache(q, true, fua);
 541         blk_queue_physical_block_size(q, PAGE_SIZE);
 542         blk_queue_logical_block_size(q, pmem_sector_size(ndns));
 543         blk_queue_max_hw_sectors(q, UINT_MAX);
 544         blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
 545         blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, q);
 546         if (pmem->pfn_flags & PFN_MAP)
 547                 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
 548
 549         disk->fops              = &pmem_fops;
 550         disk->private_data      = pmem;
 551         nvdimm_namespace_disk_name(ndns, disk->disk_name);
 552         set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
 553                         / 512);
 554         if (devm_init_badblocks(dev, &pmem->bb))
 555                 return -ENOMEM;
 556         nvdimm_badblocks_populate(nd_region, &pmem->bb, &bb_range);
 557         disk->bb = &pmem->bb;
 558
 559         dax_dev = alloc_dax(pmem, &pmem_dax_ops);
 560         if (IS_ERR(dax_dev)) {
 561                 rc = PTR_ERR(dax_dev);
 562                 goto out;
 563         }
 564         set_dax_nocache(dax_dev);
 565         set_dax_nomc(dax_dev);
 566         if (is_nvdimm_sync(nd_region))
 567                 set_dax_synchronous(dax_dev);
 568         rc = dax_add_host(dax_dev, disk);
 569         if (rc)
 570                 goto out_cleanup_dax;
 571         dax_write_cache(dax_dev, nvdimm_has_cache(nd_region));
 572         pmem->dax_dev = dax_dev;
 573
 574         rc = device_add_disk(dev, disk, pmem_attribute_groups);
 575         if (rc)
 576                 goto out_remove_host;
 577         if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
 578                 return -ENOMEM;
 579
 580         nvdimm_check_and_set_ro(disk);
 581
 582         pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
 583                                           "badblocks");
 584         if (!pmem->bb_state)
 585                 dev_warn(dev, "'badblocks' notification disabled\n");
 586         return 0;
 587
 588 out_remove_host:
 589         dax_remove_host(pmem->disk);
 590 out_cleanup_dax:
 591         kill_dax(pmem->dax_dev);
 592         put_dax(pmem->dax_dev);
 593 out:
 594         put_disk(pmem->disk);
 595         return rc;
 596 }
 597
 598 static int nd_pmem_probe(struct device *dev)
 599 {
 600         int ret;
 601         struct nd_namespace_common *ndns;
 602
 603         ndns = nvdimm_namespace_common_probe(dev);
 604         if (IS_ERR(ndns))
 605                 return PTR_ERR(ndns);
 606
 607         if (is_nd_btt(dev))
 608                 return nvdimm_namespace_attach_btt(ndns);
 609
 610         if (is_nd_pfn(dev))
 611                 return pmem_attach_disk(dev, ndns);
 612
 613         ret = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
 614         if (ret)
 615                 return ret;
 616
 617         ret = nd_btt_probe(dev, ndns);
 618         if (ret == 0)
 619                 return -ENXIO;
 620
 621         /*
 622          * We have two failure conditions here, there is no
 623          * info reserver block or we found a valid info reserve block
 624          * but failed to initialize the pfn superblock.
 625          *
 626          * For the first case consider namespace as a raw pmem namespace
 627          * and attach a disk.
 628          *
 629          * For the latter, consider this a success and advance the namespace
 630          * seed.
 631          */
 632         ret = nd_pfn_probe(dev, ndns);
 633         if (ret == 0)
 634                 return -ENXIO;
 635         else if (ret == -EOPNOTSUPP)
 636                 return ret;
 637
 638         ret = nd_dax_probe(dev, ndns);
 639         if (ret == 0)
 640                 return -ENXIO;
 641         else if (ret == -EOPNOTSUPP)
 642                 return ret;
 643
 644         /* probe complete, attach handles namespace enabling */
 645         devm_namespace_disable(dev, ndns);
 646
 647         return pmem_attach_disk(dev, ndns);
 648 }
 649
 650 static void nd_pmem_remove(struct device *dev)
 651 {
 652         struct pmem_device *pmem = dev_get_drvdata(dev);
 653
 654         if (is_nd_btt(dev))
 655                 nvdimm_namespace_detach_btt(to_nd_btt(dev));
 656         else {
 657                 /*
 658                  * Note, this assumes device_lock() context to not
 659                  * race nd_pmem_notify()
 660                  */
 661                 sysfs_put(pmem->bb_state);
 662                 pmem->bb_state = NULL;
 663         }
 664         nvdimm_flush(to_nd_region(dev->parent), NULL);
 665 }
 666
 667 static void nd_pmem_shutdown(struct device *dev)
 668 {
 669         nvdimm_flush(to_nd_region(dev->parent), NULL);
 670 }
 671
 672 static void pmem_revalidate_poison(struct device *dev)
 673 {
 674         struct nd_region *nd_region;
 675         resource_size_t offset = 0, end_trunc = 0;
 676         struct nd_namespace_common *ndns;
 677         struct nd_namespace_io *nsio;
 678         struct badblocks *bb;
 679         struct range range;
 680         struct kernfs_node *bb_state;
 681
 682         if (is_nd_btt(dev)) {
 683                 struct nd_btt *nd_btt = to_nd_btt(dev);
 684
 685                 ndns = nd_btt->ndns;
 686                 nd_region = to_nd_region(ndns->dev.parent);
 687                 nsio = to_nd_namespace_io(&ndns->dev);
 688                 bb = &nsio->bb;
 689                 bb_state = NULL;
 690         } else {
 691                 struct pmem_device *pmem = dev_get_drvdata(dev);
 692
 693                 nd_region = to_region(pmem);
 694                 bb = &pmem->bb;
 695                 bb_state = pmem->bb_state;
 696
 697                 if (is_nd_pfn(dev)) {
 698                         struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 699                         struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 700
 701                         ndns = nd_pfn->ndns;
 702                         offset = pmem->data_offset +
 703                                         __le32_to_cpu(pfn_sb->start_pad);
 704                         end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
 705                 } else {
 706                         ndns = to_ndns(dev);
 707                 }
 708
 709                 nsio = to_nd_namespace_io(&ndns->dev);
 710         }
 711
 712         range.start = nsio->res.start + offset;
 713         range.end = nsio->res.end - end_trunc;
 714         nvdimm_badblocks_populate(nd_region, bb, &range);
 715         if (bb_state)
 716                 sysfs_notify_dirent(bb_state);
 717 }
 718
 719 static void pmem_revalidate_region(struct device *dev)
 720 {
 721         struct pmem_device *pmem;
 722
 723         if (is_nd_btt(dev)) {
 724                 struct nd_btt *nd_btt = to_nd_btt(dev);
 725                 struct btt *btt = nd_btt->btt;
 726
 727                 nvdimm_check_and_set_ro(btt->btt_disk);
 728                 return;
 729         }
 730
 731         pmem = dev_get_drvdata(dev);
 732         nvdimm_check_and_set_ro(pmem->disk);
 733 }
 734
 735 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
 736 {
 737         switch (event) {
 738         case NVDIMM_REVALIDATE_POISON:
 739                 pmem_revalidate_poison(dev);
 740                 break;
 741         case NVDIMM_REVALIDATE_REGION:
 742                 pmem_revalidate_region(dev);
 743                 break;
 744         default:
 745                 dev_WARN_ONCE(dev, 1, "notify: unknown event: %d\n", event);
 746                 break;
 747         }
 748 }
 749
 750 MODULE_ALIAS("pmem");
 751 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
 752 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
 753 static struct nd_device_driver nd_pmem_driver = {
 754         .probe = nd_pmem_probe,
 755         .remove = nd_pmem_remove,
 756         .notify = nd_pmem_notify,
 757         .shutdown = nd_pmem_shutdown,
 758         .drv = {
 759                 .name = "nd_pmem",
 760         },
 761         .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
 762 };
 763
 764 module_nd_driver(nd_pmem_driver);
 765
 766 MODULE_AUTHOR("Ross Zwisler <[email protected]>");
 767 MODULE_LICENSE("GPL v2");