drivers/nvdimm/pmem.c

   1 /*
   2  * Persistent Memory Driver
   3  *
   4  * Copyright (c) 2014-2015, Intel Corporation.
   5  * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>.
   6  * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>.
   7  *
   8  * This program is free software; you can redistribute it and/or modify it
   9  * under the terms and conditions of the GNU General Public License,
  10  * version 2, as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope it will be useful, but WITHOUT
  13  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  15  * more details.
  16  */
  17
  18 #include <asm/cacheflush.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/hdreg.h>
  21 #include <linux/init.h>
  22 #include <linux/platform_device.h>
  23 #include <linux/module.h>
  24 #include <linux/moduleparam.h>
  25 #include <linux/badblocks.h>
  26 #include <linux/memremap.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/pfn_t.h>
  29 #include <linux/slab.h>
  30 #include <linux/pmem.h>
  31 #include <linux/nd.h>
  32 #include "pfn.h"
  33 #include "nd.h"
  34
  35 struct pmem_device {
  36         struct request_queue    *pmem_queue;
  37         struct gendisk          *pmem_disk;
  38
  39         /* One contiguous memory region per device */
  40         phys_addr_t             phys_addr;
  41         /* when non-zero this device is hosting a 'pfn' instance */
  42         phys_addr_t             data_offset;
  43         u64                     pfn_flags;
  44         void __pmem             *virt_addr;
  45         /* immutable base size of the namespace */
  46         size_t                  size;
  47         /* trim size when namespace capacity has been section aligned */
  48         u32                     pfn_pad;
  49         struct badblocks        bb;
  50 };
  51
  52 static bool is_bad_pmem(struct badblocks *bb, sector_t sector, unsigned int len)
  53 {
  54         if (bb->count) {
  55                 sector_t first_bad;
  56                 int num_bad;
  57
  58                 return !!badblocks_check(bb, sector, len / 512, &first_bad,
  59                                 &num_bad);
  60         }
  61
  62         return false;
  63 }
  64
  65 static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset,
  66                 unsigned int len)
  67 {
  68         struct device *dev = disk_to_dev(pmem->pmem_disk);
  69         sector_t sector;
  70         long cleared;
  71
  72         sector = (offset - pmem->data_offset) / 512;
  73         cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
  74
  75         if (cleared > 0 && cleared / 512) {
  76                 dev_dbg(dev, "%s: %llx clear %ld sector%s\n",
  77                                 __func__, (unsigned long long) sector,
  78                                 cleared / 512, cleared / 512 > 1 ? "s" : "");
  79                 badblocks_clear(&pmem->bb, sector, cleared / 512);
  80         }
  81         invalidate_pmem(pmem->virt_addr + offset, len);
  82 }
  83
  84 static int pmem_do_bvec(struct pmem_device *pmem, struct page *page,
  85                         unsigned int len, unsigned int off, int rw,
  86                         sector_t sector)
  87 {
  88         int rc = 0;
  89         bool bad_pmem = false;
  90         void *mem = kmap_atomic(page);
  91         phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
  92         void __pmem *pmem_addr = pmem->virt_addr + pmem_off;
  93
  94         if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
  95                 bad_pmem = true;
  96
  97         if (rw == READ) {
  98                 if (unlikely(bad_pmem))
  99                         rc = -EIO;
 100                 else {
 101                         rc = memcpy_from_pmem(mem + off, pmem_addr, len);
 102                         flush_dcache_page(page);
 103                 }
 104         } else {
 105                 /*
 106                  * Note that we write the data both before and after
 107                  * clearing poison.  The write before clear poison
 108                  * handles situations where the latest written data is
 109                  * preserved and the clear poison operation simply marks
 110                  * the address range as valid without changing the data.
 111                  * In this case application software can assume that an
 112                  * interrupted write will either return the new good
 113                  * data or an error.
 114                  *
 115                  * However, if pmem_clear_poison() leaves the data in an
 116                  * indeterminate state we need to perform the write
 117                  * after clear poison.
 118                  */
 119                 flush_dcache_page(page);
 120                 memcpy_to_pmem(pmem_addr, mem + off, len);
 121                 if (unlikely(bad_pmem)) {
 122                         pmem_clear_poison(pmem, pmem_off, len);
 123                         memcpy_to_pmem(pmem_addr, mem + off, len);
 124                 }
 125         }
 126
 127         kunmap_atomic(mem);
 128         return rc;
 129 }
 130
 131 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
 132 {
 133         int rc = 0;
 134         bool do_acct;
 135         unsigned long start;
 136         struct bio_vec bvec;
 137         struct bvec_iter iter;
 138         struct pmem_device *pmem = q->queuedata;
 139
 140         do_acct = nd_iostat_start(bio, &start);
 141         bio_for_each_segment(bvec, bio, iter) {
 142                 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
 143                                 bvec.bv_offset, bio_data_dir(bio),
 144                                 iter.bi_sector);
 145                 if (rc) {
 146                         bio->bi_error = rc;
 147                         break;
 148                 }
 149         }
 150         if (do_acct)
 151                 nd_iostat_end(bio, start);
 152
 153         if (bio_data_dir(bio))
 154                 wmb_pmem();
 155
 156         bio_endio(bio);
 157         return BLK_QC_T_NONE;
 158 }
 159
 160 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
 161                        struct page *page, int rw)
 162 {
 163         struct pmem_device *pmem = bdev->bd_queue->queuedata;
 164         int rc;
 165
 166         rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector);
 167         if (rw & WRITE)
 168                 wmb_pmem();
 169
 170         /*
 171          * The ->rw_page interface is subtle and tricky.  The core
 172          * retries on any error, so we can only invoke page_endio() in
 173          * the successful completion case.  Otherwise, we'll see crashes
 174          * caused by double completion.
 175          */
 176         if (rc == 0)
 177                 page_endio(page, rw & WRITE, 0);
 178
 179         return rc;
 180 }
 181
 182 static long pmem_direct_access(struct block_device *bdev, sector_t sector,
 183                       void __pmem **kaddr, pfn_t *pfn)
 184 {
 185         struct pmem_device *pmem = bdev->bd_queue->queuedata;
 186         resource_size_t offset = sector * 512 + pmem->data_offset;
 187
 188         *kaddr = pmem->virt_addr + offset;
 189         *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
 190
 191         return pmem->size - pmem->pfn_pad - offset;
 192 }
 193
 194 static const struct block_device_operations pmem_fops = {
 195         .owner =                THIS_MODULE,
 196         .rw_page =              pmem_rw_page,
 197         .direct_access =        pmem_direct_access,
 198         .revalidate_disk =      nvdimm_revalidate_disk,
 199 };
 200
 201 static struct pmem_device *pmem_alloc(struct device *dev,
 202                 struct resource *res, int id)
 203 {
 204         struct pmem_device *pmem;
 205         struct request_queue *q;
 206
 207         pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
 208         if (!pmem)
 209                 return ERR_PTR(-ENOMEM);
 210
 211         pmem->phys_addr = res->start;
 212         pmem->size = resource_size(res);
 213         if (!arch_has_wmb_pmem())
 214                 dev_warn(dev, "unable to guarantee persistence of writes\n");
 215
 216         if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size,
 217                         dev_name(dev))) {
 218                 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n",
 219                                 &pmem->phys_addr, pmem->size);
 220                 return ERR_PTR(-EBUSY);
 221         }
 222
 223         q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
 224         if (!q)
 225                 return ERR_PTR(-ENOMEM);
 226
 227         pmem->pfn_flags = PFN_DEV;
 228         if (pmem_should_map_pages(dev)) {
 229                 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res,
 230                                 &q->q_usage_counter, NULL);
 231                 pmem->pfn_flags |= PFN_MAP;
 232         } else
 233                 pmem->virt_addr = (void __pmem *) devm_memremap(dev,
 234                                 pmem->phys_addr, pmem->size,
 235                                 ARCH_MEMREMAP_PMEM);
 236
 237         if (IS_ERR(pmem->virt_addr)) {
 238                 blk_cleanup_queue(q);
 239                 return (void __force *) pmem->virt_addr;
 240         }
 241
 242         pmem->pmem_queue = q;
 243         return pmem;
 244 }
 245
 246 static void pmem_detach_disk(struct pmem_device *pmem)
 247 {
 248         if (!pmem->pmem_disk)
 249                 return;
 250
 251         del_gendisk(pmem->pmem_disk);
 252         put_disk(pmem->pmem_disk);
 253         blk_cleanup_queue(pmem->pmem_queue);
 254 }
 255
 256 static int pmem_attach_disk(struct device *dev,
 257                 struct nd_namespace_common *ndns, struct pmem_device *pmem)
 258 {
 259         struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
 260         int nid = dev_to_node(dev);
 261         struct resource bb_res;
 262         struct gendisk *disk;
 263
 264         blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
 265         blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE);
 266         blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX);
 267         blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
 268         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue);
 269         pmem->pmem_queue->queuedata = pmem;
 270
 271         disk = alloc_disk_node(0, nid);
 272         if (!disk) {
 273                 blk_cleanup_queue(pmem->pmem_queue);
 274                 return -ENOMEM;
 275         }
 276
 277         disk->fops              = &pmem_fops;
 278         disk->queue             = pmem->pmem_queue;
 279         disk->flags             = GENHD_FL_EXT_DEVT;
 280         nvdimm_namespace_disk_name(ndns, disk->disk_name);
 281         disk->driverfs_dev = dev;
 282         set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
 283                         / 512);
 284         pmem->pmem_disk = disk;
 285         devm_exit_badblocks(dev, &pmem->bb);
 286         if (devm_init_badblocks(dev, &pmem->bb))
 287                 return -ENOMEM;
 288         bb_res.start = nsio->res.start + pmem->data_offset;
 289         bb_res.end = nsio->res.end;
 290         if (is_nd_pfn(dev)) {
 291                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 292                 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 293
 294                 bb_res.start += __le32_to_cpu(pfn_sb->start_pad);
 295                 bb_res.end -= __le32_to_cpu(pfn_sb->end_trunc);
 296         }
 297         nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb,
 298                         &bb_res);
 299         disk->bb = &pmem->bb;
 300         add_disk(disk);
 301         revalidate_disk(disk);
 302
 303         return 0;
 304 }
 305
 306 static int pmem_rw_bytes(struct nd_namespace_common *ndns,
 307                 resource_size_t offset, void *buf, size_t size, int rw)
 308 {
 309         struct pmem_device *pmem = dev_get_drvdata(ndns->claim);
 310
 311         if (unlikely(offset + size > pmem->size)) {
 312                 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
 313                 return -EFAULT;
 314         }
 315
 316         if (rw == READ) {
 317                 unsigned int sz_align = ALIGN(size + (offset & (512 - 1)), 512);
 318
 319                 if (unlikely(is_bad_pmem(&pmem->bb, offset / 512, sz_align)))
 320                         return -EIO;
 321                 return memcpy_from_pmem(buf, pmem->virt_addr + offset, size);
 322         } else {
 323                 memcpy_to_pmem(pmem->virt_addr + offset, buf, size);
 324                 wmb_pmem();
 325         }
 326
 327         return 0;
 328 }
 329
 330 static int nd_pfn_init(struct nd_pfn *nd_pfn)
 331 {
 332         struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
 333         struct nd_namespace_common *ndns = nd_pfn->ndns;
 334         u32 start_pad = 0, end_trunc = 0;
 335         resource_size_t start, size;
 336         struct nd_namespace_io *nsio;
 337         struct nd_region *nd_region;
 338         struct nd_pfn_sb *pfn_sb;
 339         unsigned long npfns;
 340         phys_addr_t offset;
 341         u64 checksum;
 342         int rc;
 343
 344         pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
 345         if (!pfn_sb)
 346                 return -ENOMEM;
 347
 348         nd_pfn->pfn_sb = pfn_sb;
 349         rc = nd_pfn_validate(nd_pfn);
 350         if (rc == -ENODEV)
 351                 /* no info block, do init */;
 352         else
 353                 return rc;
 354
 355         nd_region = to_nd_region(nd_pfn->dev.parent);
 356         if (nd_region->ro) {
 357                 dev_info(&nd_pfn->dev,
 358                                 "%s is read-only, unable to init metadata\n",
 359                                 dev_name(&nd_region->dev));
 360                 return -ENXIO;
 361         }
 362
 363         memset(pfn_sb, 0, sizeof(*pfn_sb));
 364
 365         /*
 366          * Check if pmem collides with 'System RAM' when section aligned and
 367          * trim it accordingly
 368          */
 369         nsio = to_nd_namespace_io(&ndns->dev);
 370         start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
 371         size = resource_size(&nsio->res);
 372         if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
 373                                 IORES_DESC_NONE) == REGION_MIXED) {
 374
 375                 start = nsio->res.start;
 376                 start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
 377         }
 378
 379         start = nsio->res.start;
 380         size = PHYS_SECTION_ALIGN_UP(start + size) - start;
 381         if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
 382                                 IORES_DESC_NONE) == REGION_MIXED) {
 383                 size = resource_size(&nsio->res);
 384                 end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
 385         }
 386
 387         if (start_pad + end_trunc)
 388                 dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
 389                                 dev_name(&ndns->dev), start_pad + end_trunc);
 390
 391         /*
 392          * Note, we use 64 here for the standard size of struct page,
 393          * debugging options may cause it to be larger in which case the
 394          * implementation will limit the pfns advertised through
 395          * ->direct_access() to those that are included in the memmap.
 396          */
 397         start += start_pad;
 398         npfns = (pmem->size - start_pad - end_trunc - SZ_8K) / SZ_4K;
 399         if (nd_pfn->mode == PFN_MODE_PMEM)
 400                 offset = ALIGN(start + SZ_8K + 64 * npfns, nd_pfn->align)
 401                         - start;
 402         else if (nd_pfn->mode == PFN_MODE_RAM)
 403                 offset = ALIGN(start + SZ_8K, nd_pfn->align) - start;
 404         else
 405                 return -ENXIO;
 406
 407         if (offset + start_pad + end_trunc >= pmem->size) {
 408                 dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
 409                                 dev_name(&ndns->dev));
 410                 return -ENXIO;
 411         }
 412
 413         npfns = (pmem->size - offset - start_pad - end_trunc) / SZ_4K;
 414         pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
 415         pfn_sb->dataoff = cpu_to_le64(offset);
 416         pfn_sb->npfns = cpu_to_le64(npfns);
 417         memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN);
 418         memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
 419         memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
 420         pfn_sb->version_major = cpu_to_le16(1);
 421         pfn_sb->version_minor = cpu_to_le16(1);
 422         pfn_sb->start_pad = cpu_to_le32(start_pad);
 423         pfn_sb->end_trunc = cpu_to_le32(end_trunc);
 424         checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
 425         pfn_sb->checksum = cpu_to_le64(checksum);
 426
 427         return nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb));
 428 }
 429
 430 static void nvdimm_namespace_detach_pfn(struct nd_pfn *nd_pfn)
 431 {
 432         struct pmem_device *pmem;
 433
 434         /* free pmem disk */
 435         pmem = dev_get_drvdata(&nd_pfn->dev);
 436         pmem_detach_disk(pmem);
 437 }
 438
 439 /*
 440  * We hotplug memory at section granularity, pad the reserved area from
 441  * the previous section base to the namespace base address.
 442  */
 443 static unsigned long init_altmap_base(resource_size_t base)
 444 {
 445         unsigned long base_pfn = PHYS_PFN(base);
 446
 447         return PFN_SECTION_ALIGN_DOWN(base_pfn);
 448 }
 449
 450 static unsigned long init_altmap_reserve(resource_size_t base)
 451 {
 452         unsigned long reserve = PHYS_PFN(SZ_8K);
 453         unsigned long base_pfn = PHYS_PFN(base);
 454
 455         reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
 456         return reserve;
 457 }
 458
 459 static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn)
 460 {
 461         int rc;
 462         struct resource res;
 463         struct request_queue *q;
 464         struct pmem_device *pmem;
 465         struct vmem_altmap *altmap;
 466         struct device *dev = &nd_pfn->dev;
 467         struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 468         struct nd_namespace_common *ndns = nd_pfn->ndns;
 469         u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
 470         u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
 471         struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
 472         resource_size_t base = nsio->res.start + start_pad;
 473         struct vmem_altmap __altmap = {
 474                 .base_pfn = init_altmap_base(base),
 475                 .reserve = init_altmap_reserve(base),
 476         };
 477
 478         pmem = dev_get_drvdata(dev);
 479         pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
 480         pmem->pfn_pad = start_pad + end_trunc;
 481         nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
 482         if (nd_pfn->mode == PFN_MODE_RAM) {
 483                 if (pmem->data_offset < SZ_8K)
 484                         return -EINVAL;
 485                 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
 486                 altmap = NULL;
 487         } else if (nd_pfn->mode == PFN_MODE_PMEM) {
 488                 nd_pfn->npfns = (pmem->size - pmem->pfn_pad - pmem->data_offset)
 489                         / PAGE_SIZE;
 490                 if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
 491                         dev_info(&nd_pfn->dev,
 492                                         "number of pfns truncated from %lld to %ld\n",
 493                                         le64_to_cpu(nd_pfn->pfn_sb->npfns),
 494                                         nd_pfn->npfns);
 495                 altmap = & __altmap;
 496                 altmap->free = PHYS_PFN(pmem->data_offset - SZ_8K);
 497                 altmap->alloc = 0;
 498         } else {
 499                 rc = -ENXIO;
 500                 goto err;
 501         }
 502
 503         /* establish pfn range for lookup, and switch to direct map */
 504         q = pmem->pmem_queue;
 505         memcpy(&res, &nsio->res, sizeof(res));
 506         res.start += start_pad;
 507         res.end -= end_trunc;
 508         devm_memunmap(dev, (void __force *) pmem->virt_addr);
 509         pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &res,
 510                         &q->q_usage_counter, altmap);
 511         pmem->pfn_flags |= PFN_MAP;
 512         if (IS_ERR(pmem->virt_addr)) {
 513                 rc = PTR_ERR(pmem->virt_addr);
 514                 goto err;
 515         }
 516
 517         /* attach pmem disk in "pfn-mode" */
 518         rc = pmem_attach_disk(dev, ndns, pmem);
 519         if (rc)
 520                 goto err;
 521
 522         return rc;
 523  err:
 524         nvdimm_namespace_detach_pfn(nd_pfn);
 525         return rc;
 526
 527 }
 528
 529 static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
 530 {
 531         struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
 532         int rc;
 533
 534         if (!nd_pfn->uuid || !nd_pfn->ndns)
 535                 return -ENODEV;
 536
 537         rc = nd_pfn_init(nd_pfn);
 538         if (rc)
 539                 return rc;
 540         /* we need a valid pfn_sb before we can init a vmem_altmap */
 541         return __nvdimm_namespace_attach_pfn(nd_pfn);
 542 }
 543
 544 static int nd_pmem_probe(struct device *dev)
 545 {
 546         struct nd_region *nd_region = to_nd_region(dev->parent);
 547         struct nd_namespace_common *ndns;
 548         struct nd_namespace_io *nsio;
 549         struct pmem_device *pmem;
 550
 551         ndns = nvdimm_namespace_common_probe(dev);
 552         if (IS_ERR(ndns))
 553                 return PTR_ERR(ndns);
 554
 555         nsio = to_nd_namespace_io(&ndns->dev);
 556         pmem = pmem_alloc(dev, &nsio->res, nd_region->id);
 557         if (IS_ERR(pmem))
 558                 return PTR_ERR(pmem);
 559
 560         dev_set_drvdata(dev, pmem);
 561         ndns->rw_bytes = pmem_rw_bytes;
 562         if (devm_init_badblocks(dev, &pmem->bb))
 563                 return -ENOMEM;
 564         nvdimm_badblocks_populate(nd_region, &pmem->bb, &nsio->res);
 565
 566         if (is_nd_btt(dev)) {
 567                 /* btt allocates its own request_queue */
 568                 blk_cleanup_queue(pmem->pmem_queue);
 569                 pmem->pmem_queue = NULL;
 570                 return nvdimm_namespace_attach_btt(ndns);
 571         }
 572
 573         if (is_nd_pfn(dev))
 574                 return nvdimm_namespace_attach_pfn(ndns);
 575
 576         if (nd_btt_probe(dev, ndns, pmem) == 0
 577                         || nd_pfn_probe(dev, ndns, pmem) == 0) {
 578                 /*
 579                  * We'll come back as either btt-pmem, or pfn-pmem, so
 580                  * drop the queue allocation for now.
 581                  */
 582                 blk_cleanup_queue(pmem->pmem_queue);
 583                 return -ENXIO;
 584         }
 585
 586         return pmem_attach_disk(dev, ndns, pmem);
 587 }
 588
 589 static int nd_pmem_remove(struct device *dev)
 590 {
 591         struct pmem_device *pmem = dev_get_drvdata(dev);
 592
 593         if (is_nd_btt(dev))
 594                 nvdimm_namespace_detach_btt(to_nd_btt(dev));
 595         else if (is_nd_pfn(dev))
 596                 nvdimm_namespace_detach_pfn(to_nd_pfn(dev));
 597         else
 598                 pmem_detach_disk(pmem);
 599
 600         return 0;
 601 }
 602
 603 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
 604 {
 605         struct nd_region *nd_region = to_nd_region(dev->parent);
 606         struct pmem_device *pmem = dev_get_drvdata(dev);
 607         resource_size_t offset = 0, end_trunc = 0;
 608         struct nd_namespace_common *ndns;
 609         struct nd_namespace_io *nsio;
 610         struct resource res;
 611
 612         if (event != NVDIMM_REVALIDATE_POISON)
 613                 return;
 614
 615         if (is_nd_btt(dev)) {
 616                 struct nd_btt *nd_btt = to_nd_btt(dev);
 617
 618                 ndns = nd_btt->ndns;
 619         } else if (is_nd_pfn(dev)) {
 620                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 621                 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 622
 623                 ndns = nd_pfn->ndns;
 624                 offset = pmem->data_offset + __le32_to_cpu(pfn_sb->start_pad);
 625                 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
 626         } else
 627                 ndns = to_ndns(dev);
 628
 629         nsio = to_nd_namespace_io(&ndns->dev);
 630         res.start = nsio->res.start + offset;
 631         res.end = nsio->res.end - end_trunc;
 632         nvdimm_badblocks_populate(nd_region, &pmem->bb, &res);
 633 }
 634
 635 MODULE_ALIAS("pmem");
 636 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
 637 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
 638 static struct nd_device_driver nd_pmem_driver = {
 639         .probe = nd_pmem_probe,
 640         .remove = nd_pmem_remove,
 641         .notify = nd_pmem_notify,
 642         .drv = {
 643                 .name = "nd_pmem",
 644         },
 645         .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
 646 };
 647
 648 static int __init pmem_init(void)
 649 {
 650         return nd_driver_register(&nd_pmem_driver);
 651 }
 652 module_init(pmem_init);
 653
 654 static void pmem_exit(void)
 655 {
 656         driver_unregister(&nd_pmem_driver.drv);
 657 }
 658 module_exit(pmem_exit);
 659
 660 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
 661 MODULE_LICENSE("GPL v2");