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 "pmem.h"
  33 #include "pfn.h"
  34 #include "nd.h"
  35
  36 static struct device *to_dev(struct pmem_device *pmem)
  37 {
  38         /*
  39          * nvdimm bus services need a 'dev' parameter, and we record the device
  40          * at init in bb.dev.
  41          */
  42         return pmem->bb.dev;
  43 }
  44
  45 static struct nd_region *to_region(struct pmem_device *pmem)
  46 {
  47         return to_nd_region(to_dev(pmem)->parent);
  48 }
  49
  50 static int pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset,
  51                 unsigned int len)
  52 {
  53         struct device *dev = to_dev(pmem);
  54         sector_t sector;
  55         long cleared;
  56
  57         sector = (offset - pmem->data_offset) / 512;
  58         cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
  59
  60         if (cleared > 0 && cleared / 512) {
  61                 dev_dbg(dev, "%s: %#llx clear %ld sector%s\n",
  62                                 __func__, (unsigned long long) sector,
  63                                 cleared / 512, cleared / 512 > 1 ? "s" : "");
  64                 badblocks_clear(&pmem->bb, sector, cleared / 512);
  65         } else {
  66                 return -EIO;
  67         }
  68
  69         invalidate_pmem(pmem->virt_addr + offset, len);
  70         return 0;
  71 }
  72
  73 static void write_pmem(void *pmem_addr, struct page *page,
  74                 unsigned int off, unsigned int len)
  75 {
  76         void *mem = kmap_atomic(page);
  77
  78         memcpy_to_pmem(pmem_addr, mem + off, len);
  79         kunmap_atomic(mem);
  80 }
  81
  82 static int read_pmem(struct page *page, unsigned int off,
  83                 void *pmem_addr, unsigned int len)
  84 {
  85         int rc;
  86         void *mem = kmap_atomic(page);
  87
  88         rc = memcpy_from_pmem(mem + off, pmem_addr, len);
  89         kunmap_atomic(mem);
  90         if (rc)
  91                 return -EIO;
  92         return 0;
  93 }
  94
  95 static int pmem_do_bvec(struct pmem_device *pmem, struct page *page,
  96                         unsigned int len, unsigned int off, bool is_write,
  97                         sector_t sector)
  98 {
  99         int rc = 0;
 100         bool bad_pmem = false;
 101         phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
 102         void *pmem_addr = pmem->virt_addr + pmem_off;
 103
 104         if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
 105                 bad_pmem = true;
 106
 107         if (!is_write) {
 108                 if (unlikely(bad_pmem))
 109                         rc = -EIO;
 110                 else {
 111                         rc = read_pmem(page, off, pmem_addr, len);
 112                         flush_dcache_page(page);
 113                 }
 114         } else {
 115                 /*
 116                  * Note that we write the data both before and after
 117                  * clearing poison.  The write before clear poison
 118                  * handles situations where the latest written data is
 119                  * preserved and the clear poison operation simply marks
 120                  * the address range as valid without changing the data.
 121                  * In this case application software can assume that an
 122                  * interrupted write will either return the new good
 123                  * data or an error.
 124                  *
 125                  * However, if pmem_clear_poison() leaves the data in an
 126                  * indeterminate state we need to perform the write
 127                  * after clear poison.
 128                  */
 129                 flush_dcache_page(page);
 130                 write_pmem(pmem_addr, page, off, len);
 131                 if (unlikely(bad_pmem)) {
 132                         rc = pmem_clear_poison(pmem, pmem_off, len);
 133                         write_pmem(pmem_addr, page, off, len);
 134                 }
 135         }
 136
 137         return rc;
 138 }
 139
 140 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
 141 #ifndef REQ_FLUSH
 142 #define REQ_FLUSH REQ_PREFLUSH
 143 #endif
 144
 145 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
 146 {
 147         int rc = 0;
 148         bool do_acct;
 149         unsigned long start;
 150         struct bio_vec bvec;
 151         struct bvec_iter iter;
 152         struct pmem_device *pmem = q->queuedata;
 153         struct nd_region *nd_region = to_region(pmem);
 154
 155         if (bio->bi_opf & REQ_FLUSH)
 156                 nvdimm_flush(nd_region);
 157
 158         do_acct = nd_iostat_start(bio, &start);
 159         bio_for_each_segment(bvec, bio, iter) {
 160                 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
 161                                 bvec.bv_offset, op_is_write(bio_op(bio)),
 162                                 iter.bi_sector);
 163                 if (rc) {
 164                         bio->bi_error = rc;
 165                         break;
 166                 }
 167         }
 168         if (do_acct)
 169                 nd_iostat_end(bio, start);
 170
 171         if (bio->bi_opf & REQ_FUA)
 172                 nvdimm_flush(nd_region);
 173
 174         bio_endio(bio);
 175         return BLK_QC_T_NONE;
 176 }
 177
 178 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
 179                        struct page *page, bool is_write)
 180 {
 181         struct pmem_device *pmem = bdev->bd_queue->queuedata;
 182         int rc;
 183
 184         rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, is_write, sector);
 185
 186         /*
 187          * The ->rw_page interface is subtle and tricky.  The core
 188          * retries on any error, so we can only invoke page_endio() in
 189          * the successful completion case.  Otherwise, we'll see crashes
 190          * caused by double completion.
 191          */
 192         if (rc == 0)
 193                 page_endio(page, is_write, 0);
 194
 195         return rc;
 196 }
 197
 198 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
 199 __weak long pmem_direct_access(struct block_device *bdev, sector_t sector,
 200                       void **kaddr, pfn_t *pfn, long size)
 201 {
 202         struct pmem_device *pmem = bdev->bd_queue->queuedata;
 203         resource_size_t offset = sector * 512 + pmem->data_offset;
 204
 205         if (unlikely(is_bad_pmem(&pmem->bb, sector, size)))
 206                 return -EIO;
 207         *kaddr = pmem->virt_addr + offset;
 208         *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
 209
 210         /*
 211          * If badblocks are present, limit known good range to the
 212          * requested range.
 213          */
 214         if (unlikely(pmem->bb.count))
 215                 return size;
 216         return pmem->size - pmem->pfn_pad - offset;
 217 }
 218
 219 static const struct block_device_operations pmem_fops = {
 220         .owner =                THIS_MODULE,
 221         .rw_page =              pmem_rw_page,
 222         .direct_access =        pmem_direct_access,
 223         .revalidate_disk =      nvdimm_revalidate_disk,
 224 };
 225
 226 static void pmem_release_queue(void *q)
 227 {
 228         blk_cleanup_queue(q);
 229 }
 230
 231 static void pmem_release_disk(void *disk)
 232 {
 233         del_gendisk(disk);
 234         put_disk(disk);
 235 }
 236
 237 static int pmem_attach_disk(struct device *dev,
 238                 struct nd_namespace_common *ndns)
 239 {
 240         struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
 241         struct nd_region *nd_region = to_nd_region(dev->parent);
 242         struct vmem_altmap __altmap, *altmap = NULL;
 243         struct resource *res = &nsio->res;
 244         struct nd_pfn *nd_pfn = NULL;
 245         int nid = dev_to_node(dev);
 246         struct nd_pfn_sb *pfn_sb;
 247         struct pmem_device *pmem;
 248         struct resource pfn_res;
 249         struct request_queue *q;
 250         struct gendisk *disk;
 251         void *addr;
 252
 253         /* while nsio_rw_bytes is active, parse a pfn info block if present */
 254         if (is_nd_pfn(dev)) {
 255                 nd_pfn = to_nd_pfn(dev);
 256                 altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap);
 257                 if (IS_ERR(altmap))
 258                         return PTR_ERR(altmap);
 259         }
 260
 261         /* we're attaching a block device, disable raw namespace access */
 262         devm_nsio_disable(dev, nsio);
 263
 264         pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
 265         if (!pmem)
 266                 return -ENOMEM;
 267
 268         dev_set_drvdata(dev, pmem);
 269         pmem->phys_addr = res->start;
 270         pmem->size = resource_size(res);
 271         if (nvdimm_has_flush(nd_region) < 0)
 272                 dev_warn(dev, "unable to guarantee persistence of writes\n");
 273
 274         if (!devm_request_mem_region(dev, res->start, resource_size(res),
 275                                 dev_name(dev))) {
 276                 dev_warn(dev, "could not reserve region %pR\n", res);
 277                 return -EBUSY;
 278         }
 279
 280         q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
 281         if (!q)
 282                 return -ENOMEM;
 283
 284         pmem->pfn_flags = PFN_DEV;
 285         if (is_nd_pfn(dev)) {
 286                 addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter,
 287                                 altmap);
 288                 pfn_sb = nd_pfn->pfn_sb;
 289                 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
 290                 pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res);
 291                 pmem->pfn_flags |= PFN_MAP;
 292                 res = &pfn_res; /* for badblocks populate */
 293                 res->start += pmem->data_offset;
 294         } else if (pmem_should_map_pages(dev)) {
 295                 addr = devm_memremap_pages(dev, &nsio->res,
 296                                 &q->q_usage_counter, NULL);
 297                 pmem->pfn_flags |= PFN_MAP;
 298         } else
 299                 addr = devm_memremap(dev, pmem->phys_addr,
 300                                 pmem->size, ARCH_MEMREMAP_PMEM);
 301
 302         /*
 303          * At release time the queue must be dead before
 304          * devm_memremap_pages is unwound
 305          */
 306         if (devm_add_action_or_reset(dev, pmem_release_queue, q))
 307                 return -ENOMEM;
 308
 309         if (IS_ERR(addr))
 310                 return PTR_ERR(addr);
 311         pmem->virt_addr = addr;
 312
 313         blk_queue_write_cache(q, true, true);
 314         blk_queue_make_request(q, pmem_make_request);
 315         blk_queue_physical_block_size(q, PAGE_SIZE);
 316         blk_queue_max_hw_sectors(q, UINT_MAX);
 317         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 318         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
 319         queue_flag_set_unlocked(QUEUE_FLAG_DAX, q);
 320         q->queuedata = pmem;
 321
 322         disk = alloc_disk_node(0, nid);
 323         if (!disk)
 324                 return -ENOMEM;
 325
 326         disk->fops              = &pmem_fops;
 327         disk->queue             = q;
 328         disk->flags             = GENHD_FL_EXT_DEVT;
 329         nvdimm_namespace_disk_name(ndns, disk->disk_name);
 330         set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
 331                         / 512);
 332         if (devm_init_badblocks(dev, &pmem->bb))
 333                 return -ENOMEM;
 334         nvdimm_badblocks_populate(nd_region, &pmem->bb, res);
 335         disk->bb = &pmem->bb;
 336         device_add_disk(dev, disk);
 337
 338         if (devm_add_action_or_reset(dev, pmem_release_disk, disk))
 339                 return -ENOMEM;
 340
 341         revalidate_disk(disk);
 342
 343         return 0;
 344 }
 345
 346 static int nd_pmem_probe(struct device *dev)
 347 {
 348         struct nd_namespace_common *ndns;
 349
 350         ndns = nvdimm_namespace_common_probe(dev);
 351         if (IS_ERR(ndns))
 352                 return PTR_ERR(ndns);
 353
 354         if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev)))
 355                 return -ENXIO;
 356
 357         if (is_nd_btt(dev))
 358                 return nvdimm_namespace_attach_btt(ndns);
 359
 360         if (is_nd_pfn(dev))
 361                 return pmem_attach_disk(dev, ndns);
 362
 363         /* if we find a valid info-block we'll come back as that personality */
 364         if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0
 365                         || nd_dax_probe(dev, ndns) == 0)
 366                 return -ENXIO;
 367
 368         /* ...otherwise we're just a raw pmem device */
 369         return pmem_attach_disk(dev, ndns);
 370 }
 371
 372 static int nd_pmem_remove(struct device *dev)
 373 {
 374         if (is_nd_btt(dev))
 375                 nvdimm_namespace_detach_btt(to_nd_btt(dev));
 376         nvdimm_flush(to_nd_region(dev->parent));
 377
 378         return 0;
 379 }
 380
 381 static void nd_pmem_shutdown(struct device *dev)
 382 {
 383         nvdimm_flush(to_nd_region(dev->parent));
 384 }
 385
 386 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
 387 {
 388         struct nd_region *nd_region;
 389         resource_size_t offset = 0, end_trunc = 0;
 390         struct nd_namespace_common *ndns;
 391         struct nd_namespace_io *nsio;
 392         struct resource res;
 393         struct badblocks *bb;
 394
 395         if (event != NVDIMM_REVALIDATE_POISON)
 396                 return;
 397
 398         if (is_nd_btt(dev)) {
 399                 struct nd_btt *nd_btt = to_nd_btt(dev);
 400
 401                 ndns = nd_btt->ndns;
 402                 nd_region = to_nd_region(ndns->dev.parent);
 403                 nsio = to_nd_namespace_io(&ndns->dev);
 404                 bb = &nsio->bb;
 405         } else {
 406                 struct pmem_device *pmem = dev_get_drvdata(dev);
 407
 408                 nd_region = to_region(pmem);
 409                 bb = &pmem->bb;
 410
 411                 if (is_nd_pfn(dev)) {
 412                         struct nd_pfn *nd_pfn = to_nd_pfn(dev);
 413                         struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
 414
 415                         ndns = nd_pfn->ndns;
 416                         offset = pmem->data_offset +
 417                                         __le32_to_cpu(pfn_sb->start_pad);
 418                         end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
 419                 } else {
 420                         ndns = to_ndns(dev);
 421                 }
 422
 423                 nsio = to_nd_namespace_io(&ndns->dev);
 424         }
 425
 426         res.start = nsio->res.start + offset;
 427         res.end = nsio->res.end - end_trunc;
 428         nvdimm_badblocks_populate(nd_region, bb, &res);
 429 }
 430
 431 MODULE_ALIAS("pmem");
 432 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
 433 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
 434 static struct nd_device_driver nd_pmem_driver = {
 435         .probe = nd_pmem_probe,
 436         .remove = nd_pmem_remove,
 437         .notify = nd_pmem_notify,
 438         .shutdown = nd_pmem_shutdown,
 439         .drv = {
 440                 .name = "nd_pmem",
 441         },
 442         .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
 443 };
 444
 445 static int __init pmem_init(void)
 446 {
 447         return nd_driver_register(&nd_pmem_driver);
 448 }
 449 module_init(pmem_init);
 450
 451 static void pmem_exit(void)
 452 {
 453         driver_unregister(&nd_pmem_driver.drv);
 454 }
 455 module_exit(pmem_exit);
 456
 457 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
 458 MODULE_LICENSE("GPL v2");