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