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