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/memory_hotplug.h>
  25 #include <linux/moduleparam.h>
  26 #include <linux/badblocks.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/slab.h>
  29 #include <linux/pmem.h>
  30 #include <linux/nd.h>
  31 #include "pfn.h"
  32 #include "nd.h"
  33
  34 struct pmem_device {
  35         struct request_queue    *pmem_queue;
  36         struct gendisk          *pmem_disk;
  37         struct nd_namespace_common *ndns;
  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         void __pmem             *virt_addr;
  44         size_t                  size;
  45         struct badblocks        bb;
  46 };
  47
  48 static int pmem_major;
  49
  50 static bool is_bad_pmem(struct badblocks *bb, sector_t sector, unsigned int len)
  51 {
  52         if (bb->count) {
  53                 sector_t first_bad;
  54                 int num_bad;
  55
  56                 return !!badblocks_check(bb, sector, len / 512, &first_bad,
  57                                 &num_bad);
  58         }
  59
  60         return false;
  61 }
  62
  63 static int pmem_do_bvec(struct pmem_device *pmem, struct page *page,
  64                         unsigned int len, unsigned int off, int rw,
  65                         sector_t sector)
  66 {
  67         void *mem = kmap_atomic(page);
  68         phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
  69         void __pmem *pmem_addr = pmem->virt_addr + pmem_off;
  70
  71         if (rw == READ) {
  72                 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
  73                         return -EIO;
  74                 memcpy_from_pmem(mem + off, pmem_addr, len);
  75                 flush_dcache_page(page);
  76         } else {
  77                 flush_dcache_page(page);
  78                 memcpy_to_pmem(pmem_addr, mem + off, len);
  79         }
  80
  81         kunmap_atomic(mem);
  82         return 0;
  83 }
  84
  85 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
  86 {
  87         int rc = 0;
  88         bool do_acct;
  89         unsigned long start;
  90         struct bio_vec bvec;
  91         struct bvec_iter iter;
  92         struct block_device *bdev = bio->bi_bdev;
  93         struct pmem_device *pmem = bdev->bd_disk->private_data;
  94
  95         do_acct = nd_iostat_start(bio, &start);
  96         bio_for_each_segment(bvec, bio, iter) {
  97                 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
  98                                 bvec.bv_offset, bio_data_dir(bio),
  99                                 iter.bi_sector);
 100                 if (rc) {
 101                         bio->bi_error = rc;
 102                         break;
 103                 }
 104         }
 105         if (do_acct)
 106                 nd_iostat_end(bio, start);
 107
 108         if (bio_data_dir(bio))
 109                 wmb_pmem();
 110
 111         bio_endio(bio);
 112         return BLK_QC_T_NONE;
 113 }
 114
 115 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
 116                        struct page *page, int rw)
 117 {
 118         struct pmem_device *pmem = bdev->bd_disk->private_data;
 119         int rc;
 120
 121         rc = pmem_do_bvec(pmem, page, PAGE_CACHE_SIZE, 0, rw, sector);
 122         if (rw & WRITE)
 123                 wmb_pmem();
 124
 125         /*
 126          * The ->rw_page interface is subtle and tricky.  The core
 127          * retries on any error, so we can only invoke page_endio() in
 128          * the successful completion case.  Otherwise, we'll see crashes
 129          * caused by double completion.
 130          */
 131         if (rc == 0)
 132                 page_endio(page, rw & WRITE, 0);
 133
 134         return rc;
 135 }
 136
 137 static long pmem_direct_access(struct block_device *bdev, sector_t sector,
 138                       void __pmem **kaddr, unsigned long *pfn)
 139 {
 140         struct pmem_device *pmem = bdev->bd_disk->private_data;
 141         resource_size_t offset = sector * 512 + pmem->data_offset;
 142
 143         *kaddr = pmem->virt_addr + offset;
 144         *pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT;
 145
 146         return pmem->size - offset;
 147 }
 148
 149 static const struct block_device_operations pmem_fops = {
 150         .owner =                THIS_MODULE,
 151         .rw_page =              pmem_rw_page,
 152         .direct_access =        pmem_direct_access,
 153         .revalidate_disk =      nvdimm_revalidate_disk,
 154 };
 155
 156 static struct pmem_device *pmem_alloc(struct device *dev,
 157                 struct resource *res, int id)
 158 {
 159         struct pmem_device *pmem;
 160
 161         pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
 162         if (!pmem)
 163                 return ERR_PTR(-ENOMEM);
 164
 165         pmem->phys_addr = res->start;
 166         pmem->size = resource_size(res);
 167         if (!arch_has_wmb_pmem())
 168                 dev_warn(dev, "unable to guarantee persistence of writes\n");
 169
 170         if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size,
 171                         dev_name(dev))) {
 172                 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n",
 173                                 &pmem->phys_addr, pmem->size);
 174                 return ERR_PTR(-EBUSY);
 175         }
 176
 177         if (pmem_should_map_pages(dev))
 178                 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res);
 179         else
 180                 pmem->virt_addr = (void __pmem *) devm_memremap(dev,
 181                                 pmem->phys_addr, pmem->size,
 182                                 ARCH_MEMREMAP_PMEM);
 183
 184         if (IS_ERR(pmem->virt_addr))
 185                 return (void __force *) pmem->virt_addr;
 186
 187         return pmem;
 188 }
 189
 190 static void pmem_detach_disk(struct pmem_device *pmem)
 191 {
 192         if (!pmem->pmem_disk)
 193                 return;
 194
 195         del_gendisk(pmem->pmem_disk);
 196         put_disk(pmem->pmem_disk);
 197         blk_cleanup_queue(pmem->pmem_queue);
 198 }
 199
 200 static int pmem_attach_disk(struct device *dev,
 201                 struct nd_namespace_common *ndns, struct pmem_device *pmem)
 202 {
 203         int nid = dev_to_node(dev);
 204         struct gendisk *disk;
 205
 206         pmem->pmem_queue = blk_alloc_queue_node(GFP_KERNEL, nid);
 207         if (!pmem->pmem_queue)
 208                 return -ENOMEM;
 209
 210         blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
 211         blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE);
 212         blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX);
 213         blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
 214         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue);
 215
 216         disk = alloc_disk_node(0, nid);
 217         if (!disk) {
 218                 blk_cleanup_queue(pmem->pmem_queue);
 219                 return -ENOMEM;
 220         }
 221
 222         disk->major             = pmem_major;
 223         disk->first_minor       = 0;
 224         disk->fops              = &pmem_fops;
 225         disk->private_data      = pmem;
 226         disk->queue             = pmem->pmem_queue;
 227         disk->flags             = GENHD_FL_EXT_DEVT;
 228         nvdimm_namespace_disk_name(ndns, disk->disk_name);
 229         disk->driverfs_dev = dev;
 230         set_capacity(disk, (pmem->size - pmem->data_offset) / 512);
 231         pmem->pmem_disk = disk;
 232         devm_exit_badblocks(dev, &pmem->bb);
 233         if (devm_init_badblocks(dev, &pmem->bb))
 234                 return -ENOMEM;
 235         nvdimm_namespace_add_poison(ndns, &pmem->bb, pmem->data_offset);
 236
 237         disk->bb = &pmem->bb;
 238         add_disk(disk);
 239         revalidate_disk(disk);
 240
 241         return 0;
 242 }
 243
 244 static int pmem_rw_bytes(struct nd_namespace_common *ndns,
 245                 resource_size_t offset, void *buf, size_t size, int rw)
 246 {
 247         struct pmem_device *pmem = dev_get_drvdata(ndns->claim);
 248
 249         if (unlikely(offset + size > pmem->size)) {
 250                 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
 251                 return -EFAULT;
 252         }
 253
 254         if (rw == READ) {
 255                 unsigned int sz_align = ALIGN(size + (offset & (512 - 1)), 512);
 256
 257                 if (unlikely(is_bad_pmem(&pmem->bb, offset / 512, sz_align)))
 258                         return -EIO;
 259                 memcpy_from_pmem(buf, pmem->virt_addr + offset, size);
 260         } else {
 261                 memcpy_to_pmem(pmem->virt_addr + offset, buf, size);
 262                 wmb_pmem();
 263         }
 264
 265         return 0;
 266 }
 267
 268 static int nd_pfn_init(struct nd_pfn *nd_pfn)
 269 {
 270         struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL);
 271         struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
 272         struct nd_namespace_common *ndns = nd_pfn->ndns;
 273         struct nd_region *nd_region;
 274         unsigned long npfns;
 275         phys_addr_t offset;
 276         u64 checksum;
 277         int rc;
 278
 279         if (!pfn_sb)
 280                 return -ENOMEM;
 281
 282         nd_pfn->pfn_sb = pfn_sb;
 283         rc = nd_pfn_validate(nd_pfn);
 284         if (rc == 0 || rc == -EBUSY)
 285                 return rc;
 286
 287         /* section alignment for simple hotplug */
 288         if (nvdimm_namespace_capacity(ndns) < ND_PFN_ALIGN
 289                         || pmem->phys_addr & ND_PFN_MASK)
 290                 return -ENODEV;
 291
 292         nd_region = to_nd_region(nd_pfn->dev.parent);
 293         if (nd_region->ro) {
 294                 dev_info(&nd_pfn->dev,
 295                                 "%s is read-only, unable to init metadata\n",
 296                                 dev_name(&nd_region->dev));
 297                 goto err;
 298         }
 299
 300         memset(pfn_sb, 0, sizeof(*pfn_sb));
 301         npfns = (pmem->size - SZ_8K) / SZ_4K;
 302         /*
 303          * Note, we use 64 here for the standard size of struct page,
 304          * debugging options may cause it to be larger in which case the
 305          * implementation will limit the pfns advertised through
 306          * ->direct_access() to those that are included in the memmap.
 307          */
 308         if (nd_pfn->mode == PFN_MODE_PMEM)
 309                 offset = ALIGN(SZ_8K + 64 * npfns, PMD_SIZE);
 310         else if (nd_pfn->mode == PFN_MODE_RAM)
 311                 offset = SZ_8K;
 312         else
 313                 goto err;
 314
 315         npfns = (pmem->size - offset) / SZ_4K;
 316         pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
 317         pfn_sb->dataoff = cpu_to_le64(offset);
 318         pfn_sb->npfns = cpu_to_le64(npfns);
 319         memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN);
 320         memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
 321         pfn_sb->version_major = cpu_to_le16(1);
 322         checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
 323         pfn_sb->checksum = cpu_to_le64(checksum);
 324
 325         rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb));
 326         if (rc)
 327                 goto err;
 328
 329         return 0;
 330  err:
 331         nd_pfn->pfn_sb = NULL;
 332         kfree(pfn_sb);
 333         return -ENXIO;
 334 }
 335
 336 static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns)
 337 {
 338         struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
 339         struct pmem_device *pmem;
 340
 341         /* free pmem disk */
 342         pmem = dev_get_drvdata(&nd_pfn->dev);
 343         pmem_detach_disk(pmem);
 344
 345         /* release nd_pfn resources */
 346         kfree(nd_pfn->pfn_sb);
 347         nd_pfn->pfn_sb = NULL;
 348
 349         return 0;
 350 }
 351
 352 static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
 353 {
 354         struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
 355         struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
 356         struct device *dev = &nd_pfn->dev;
 357         struct vmem_altmap *altmap;
 358         struct nd_region *nd_region;
 359         struct nd_pfn_sb *pfn_sb;
 360         struct pmem_device *pmem;
 361         phys_addr_t offset;
 362         int rc;
 363
 364         if (!nd_pfn->uuid || !nd_pfn->ndns)
 365                 return -ENODEV;
 366
 367         nd_region = to_nd_region(dev->parent);
 368         rc = nd_pfn_init(nd_pfn);
 369         if (rc)
 370                 return rc;
 371
 372         if (PAGE_SIZE != SZ_4K) {
 373                 dev_err(dev, "only supported on systems with 4K PAGE_SIZE\n");
 374                 return -ENXIO;
 375         }
 376         if (nsio->res.start & ND_PFN_MASK) {
 377                 dev_err(dev, "%s not memory hotplug section aligned\n",
 378                                 dev_name(&ndns->dev));
 379                 return -ENXIO;
 380         }
 381
 382         pfn_sb = nd_pfn->pfn_sb;
 383         offset = le64_to_cpu(pfn_sb->dataoff);
 384         nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
 385         if (nd_pfn->mode == PFN_MODE_RAM) {
 386                 if (offset != SZ_8K)
 387                         return -EINVAL;
 388                 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
 389                 altmap = NULL;
 390         } else {
 391                 rc = -ENXIO;
 392                 goto err;
 393         }
 394
 395         /* establish pfn range for lookup, and switch to direct map */
 396         pmem = dev_get_drvdata(dev);
 397         devm_memunmap(dev, (void __force *) pmem->virt_addr);
 398         pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &nsio->res);
 399         if (IS_ERR(pmem->virt_addr)) {
 400                 rc = PTR_ERR(pmem->virt_addr);
 401                 goto err;
 402         }
 403
 404         /* attach pmem disk in "pfn-mode" */
 405         pmem->data_offset = offset;
 406         rc = pmem_attach_disk(dev, ndns, pmem);
 407         if (rc)
 408                 goto err;
 409
 410         return rc;
 411  err:
 412         nvdimm_namespace_detach_pfn(ndns);
 413         return rc;
 414 }
 415
 416 static int nd_pmem_probe(struct device *dev)
 417 {
 418         struct nd_region *nd_region = to_nd_region(dev->parent);
 419         struct nd_namespace_common *ndns;
 420         struct nd_namespace_io *nsio;
 421         struct pmem_device *pmem;
 422
 423         ndns = nvdimm_namespace_common_probe(dev);
 424         if (IS_ERR(ndns))
 425                 return PTR_ERR(ndns);
 426
 427         nsio = to_nd_namespace_io(&ndns->dev);
 428         pmem = pmem_alloc(dev, &nsio->res, nd_region->id);
 429         if (IS_ERR(pmem))
 430                 return PTR_ERR(pmem);
 431
 432         pmem->ndns = ndns;
 433         dev_set_drvdata(dev, pmem);
 434         ndns->rw_bytes = pmem_rw_bytes;
 435         if (devm_init_badblocks(dev, &pmem->bb))
 436                 return -ENOMEM;
 437         nvdimm_namespace_add_poison(ndns, &pmem->bb, 0);
 438
 439         if (is_nd_btt(dev))
 440                 return nvdimm_namespace_attach_btt(ndns);
 441
 442         if (is_nd_pfn(dev))
 443                 return nvdimm_namespace_attach_pfn(ndns);
 444
 445         if (nd_btt_probe(ndns, pmem) == 0) {
 446                 /* we'll come back as btt-pmem */
 447                 return -ENXIO;
 448         }
 449
 450         if (nd_pfn_probe(ndns, pmem) == 0) {
 451                 /* we'll come back as pfn-pmem */
 452                 return -ENXIO;
 453         }
 454
 455         return pmem_attach_disk(dev, ndns, pmem);
 456 }
 457
 458 static int nd_pmem_remove(struct device *dev)
 459 {
 460         struct pmem_device *pmem = dev_get_drvdata(dev);
 461
 462         if (is_nd_btt(dev))
 463                 nvdimm_namespace_detach_btt(pmem->ndns);
 464         else if (is_nd_pfn(dev))
 465                 nvdimm_namespace_detach_pfn(pmem->ndns);
 466         else
 467                 pmem_detach_disk(pmem);
 468
 469         return 0;
 470 }
 471
 472 MODULE_ALIAS("pmem");
 473 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
 474 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
 475 static struct nd_device_driver nd_pmem_driver = {
 476         .probe = nd_pmem_probe,
 477         .remove = nd_pmem_remove,
 478         .drv = {
 479                 .name = "nd_pmem",
 480         },
 481         .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
 482 };
 483
 484 static int __init pmem_init(void)
 485 {
 486         int error;
 487
 488         pmem_major = register_blkdev(0, "pmem");
 489         if (pmem_major < 0)
 490                 return pmem_major;
 491
 492         error = nd_driver_register(&nd_pmem_driver);
 493         if (error) {
 494                 unregister_blkdev(pmem_major, "pmem");
 495                 return error;
 496         }
 497
 498         return 0;
 499 }
 500 module_init(pmem_init);
 501
 502 static void pmem_exit(void)
 503 {
 504         driver_unregister(&nd_pmem_driver.drv);
 505         unregister_blkdev(pmem_major, "pmem");
 506 }
 507 module_exit(pmem_exit);
 508
 509 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
 510 MODULE_LICENSE("GPL v2");