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