2 * Persistent Memory Driver
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>.
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.
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
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/blk-mq.h>
29 #include <linux/pfn_t.h>
30 #include <linux/slab.h>
31 #include <linux/uio.h>
32 #include <linux/dax.h>
34 #include <linux/backing-dev.h>
40 static struct device *to_dev(struct pmem_device *pmem)
43 * nvdimm bus services need a 'dev' parameter, and we record the device
49 static struct nd_region *to_region(struct pmem_device *pmem)
51 return to_nd_region(to_dev(pmem)->parent);
54 static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
55 phys_addr_t offset, unsigned int len)
57 struct device *dev = to_dev(pmem);
60 blk_status_t rc = BLK_STS_OK;
62 sector = (offset - pmem->data_offset) / 512;
64 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
67 if (cleared > 0 && cleared / 512) {
69 dev_dbg(dev, "%s: %#llx clear %ld sector%s\n", __func__,
70 (unsigned long long) sector, cleared,
71 cleared > 1 ? "s" : "");
72 badblocks_clear(&pmem->bb, sector, cleared);
74 sysfs_notify_dirent(pmem->bb_state);
77 arch_invalidate_pmem(pmem->virt_addr + offset, len);
82 static void write_pmem(void *pmem_addr, struct page *page,
83 unsigned int off, unsigned int len)
89 mem = kmap_atomic(page);
90 chunk = min_t(unsigned int, len, PAGE_SIZE);
91 memcpy_flushcache(pmem_addr, mem + off, chunk);
96 pmem_addr += PAGE_SIZE;
100 static blk_status_t read_pmem(struct page *page, unsigned int off,
101 void *pmem_addr, unsigned int len)
108 mem = kmap_atomic(page);
109 chunk = min_t(unsigned int, len, PAGE_SIZE);
110 rc = memcpy_mcsafe(mem + off, pmem_addr, chunk);
113 return BLK_STS_IOERR;
117 pmem_addr += PAGE_SIZE;
122 static blk_status_t pmem_do_bvec(struct pmem_device *pmem, struct page *page,
123 unsigned int len, unsigned int off, bool is_write,
126 blk_status_t rc = BLK_STS_OK;
127 bool bad_pmem = false;
128 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
129 void *pmem_addr = pmem->virt_addr + pmem_off;
131 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
135 if (unlikely(bad_pmem))
138 rc = read_pmem(page, off, pmem_addr, len);
139 flush_dcache_page(page);
143 * Note that we write the data both before and after
144 * clearing poison. The write before clear poison
145 * handles situations where the latest written data is
146 * preserved and the clear poison operation simply marks
147 * the address range as valid without changing the data.
148 * In this case application software can assume that an
149 * interrupted write will either return the new good
152 * However, if pmem_clear_poison() leaves the data in an
153 * indeterminate state we need to perform the write
154 * after clear poison.
156 flush_dcache_page(page);
157 write_pmem(pmem_addr, page, off, len);
158 if (unlikely(bad_pmem)) {
159 rc = pmem_clear_poison(pmem, pmem_off, len);
160 write_pmem(pmem_addr, page, off, len);
167 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
169 #define REQ_FLUSH REQ_PREFLUSH
172 static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
178 struct bvec_iter iter;
179 struct pmem_device *pmem = q->queuedata;
180 struct nd_region *nd_region = to_region(pmem);
182 if (bio->bi_opf & REQ_FLUSH)
183 nvdimm_flush(nd_region);
185 do_acct = nd_iostat_start(bio, &start);
186 bio_for_each_segment(bvec, bio, iter) {
187 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
188 bvec.bv_offset, op_is_write(bio_op(bio)),
196 nd_iostat_end(bio, start);
198 if (bio->bi_opf & REQ_FUA)
199 nvdimm_flush(nd_region);
202 return BLK_QC_T_NONE;
205 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
206 struct page *page, bool is_write)
208 struct pmem_device *pmem = bdev->bd_queue->queuedata;
211 rc = pmem_do_bvec(pmem, page, hpage_nr_pages(page) * PAGE_SIZE,
212 0, is_write, sector);
215 * The ->rw_page interface is subtle and tricky. The core
216 * retries on any error, so we can only invoke page_endio() in
217 * the successful completion case. Otherwise, we'll see crashes
218 * caused by double completion.
221 page_endio(page, is_write, 0);
223 return blk_status_to_errno(rc);
226 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
227 __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
228 long nr_pages, void **kaddr, pfn_t *pfn)
230 resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
232 if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
233 PFN_PHYS(nr_pages))))
235 *kaddr = pmem->virt_addr + offset;
236 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
239 * If badblocks are present, limit known good range to the
242 if (unlikely(pmem->bb.count))
244 return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
247 static const struct block_device_operations pmem_fops = {
248 .owner = THIS_MODULE,
249 .rw_page = pmem_rw_page,
250 .revalidate_disk = nvdimm_revalidate_disk,
253 static long pmem_dax_direct_access(struct dax_device *dax_dev,
254 pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
256 struct pmem_device *pmem = dax_get_private(dax_dev);
258 return __pmem_direct_access(pmem, pgoff, nr_pages, kaddr, pfn);
261 static size_t pmem_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
262 void *addr, size_t bytes, struct iov_iter *i)
264 return copy_from_iter_flushcache(addr, bytes, i);
267 static const struct dax_operations pmem_dax_ops = {
268 .direct_access = pmem_dax_direct_access,
269 .copy_from_iter = pmem_copy_from_iter,
272 static const struct attribute_group *pmem_attribute_groups[] = {
273 &dax_attribute_group,
277 static void pmem_release_queue(void *q)
279 blk_cleanup_queue(q);
282 static void pmem_freeze_queue(void *q)
284 blk_freeze_queue_start(q);
287 static void pmem_release_disk(void *__pmem)
289 struct pmem_device *pmem = __pmem;
291 kill_dax(pmem->dax_dev);
292 put_dax(pmem->dax_dev);
293 del_gendisk(pmem->disk);
294 put_disk(pmem->disk);
297 static int pmem_attach_disk(struct device *dev,
298 struct nd_namespace_common *ndns)
300 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
301 struct nd_region *nd_region = to_nd_region(dev->parent);
302 struct vmem_altmap __altmap, *altmap = NULL;
303 int nid = dev_to_node(dev), fua, wbc;
304 struct resource *res = &nsio->res;
305 struct nd_pfn *nd_pfn = NULL;
306 struct dax_device *dax_dev;
307 struct nd_pfn_sb *pfn_sb;
308 struct pmem_device *pmem;
309 struct resource pfn_res;
310 struct request_queue *q;
311 struct device *gendev;
312 struct gendisk *disk;
315 /* while nsio_rw_bytes is active, parse a pfn info block if present */
316 if (is_nd_pfn(dev)) {
317 nd_pfn = to_nd_pfn(dev);
318 altmap = nvdimm_setup_pfn(nd_pfn, &pfn_res, &__altmap);
320 return PTR_ERR(altmap);
323 /* we're attaching a block device, disable raw namespace access */
324 devm_nsio_disable(dev, nsio);
326 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
330 dev_set_drvdata(dev, pmem);
331 pmem->phys_addr = res->start;
332 pmem->size = resource_size(res);
333 fua = nvdimm_has_flush(nd_region);
334 if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) || fua < 0) {
335 dev_warn(dev, "unable to guarantee persistence of writes\n");
338 wbc = nvdimm_has_cache(nd_region) &&
339 !test_bit(ND_REGION_PERSIST_CACHE, &nd_region->flags);
341 if (!devm_request_mem_region(dev, res->start, resource_size(res),
342 dev_name(&ndns->dev))) {
343 dev_warn(dev, "could not reserve region %pR\n", res);
347 q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
351 if (devm_add_action_or_reset(dev, pmem_release_queue, q))
354 pmem->pfn_flags = PFN_DEV;
355 if (is_nd_pfn(dev)) {
356 addr = devm_memremap_pages(dev, &pfn_res, &q->q_usage_counter,
358 pfn_sb = nd_pfn->pfn_sb;
359 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
360 pmem->pfn_pad = resource_size(res) - resource_size(&pfn_res);
361 pmem->pfn_flags |= PFN_MAP;
362 res = &pfn_res; /* for badblocks populate */
363 res->start += pmem->data_offset;
364 } else if (pmem_should_map_pages(dev)) {
365 addr = devm_memremap_pages(dev, &nsio->res,
366 &q->q_usage_counter, NULL);
367 pmem->pfn_flags |= PFN_MAP;
369 addr = devm_memremap(dev, pmem->phys_addr,
370 pmem->size, ARCH_MEMREMAP_PMEM);
373 * At release time the queue must be frozen before
374 * devm_memremap_pages is unwound
376 if (devm_add_action_or_reset(dev, pmem_freeze_queue, q))
380 return PTR_ERR(addr);
381 pmem->virt_addr = addr;
383 blk_queue_write_cache(q, wbc, fua);
384 blk_queue_make_request(q, pmem_make_request);
385 blk_queue_physical_block_size(q, PAGE_SIZE);
386 blk_queue_logical_block_size(q, pmem_sector_size(ndns));
387 blk_queue_max_hw_sectors(q, UINT_MAX);
388 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
389 queue_flag_set_unlocked(QUEUE_FLAG_DAX, q);
392 disk = alloc_disk_node(0, nid);
397 disk->fops = &pmem_fops;
399 disk->flags = GENHD_FL_EXT_DEVT;
400 disk->queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO;
401 nvdimm_namespace_disk_name(ndns, disk->disk_name);
402 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
404 if (devm_init_badblocks(dev, &pmem->bb))
406 nvdimm_badblocks_populate(nd_region, &pmem->bb, res);
407 disk->bb = &pmem->bb;
409 dax_dev = alloc_dax(pmem, disk->disk_name, &pmem_dax_ops);
414 dax_write_cache(dax_dev, wbc);
415 pmem->dax_dev = dax_dev;
417 gendev = disk_to_dev(disk);
418 gendev->groups = pmem_attribute_groups;
420 device_add_disk(dev, disk);
421 if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
424 revalidate_disk(disk);
426 pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
429 dev_warn(dev, "'badblocks' notification disabled\n");
434 static int nd_pmem_probe(struct device *dev)
436 struct nd_namespace_common *ndns;
438 ndns = nvdimm_namespace_common_probe(dev);
440 return PTR_ERR(ndns);
442 if (devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev)))
446 return nvdimm_namespace_attach_btt(ndns);
449 return pmem_attach_disk(dev, ndns);
451 /* if we find a valid info-block we'll come back as that personality */
452 if (nd_btt_probe(dev, ndns) == 0 || nd_pfn_probe(dev, ndns) == 0
453 || nd_dax_probe(dev, ndns) == 0)
456 /* ...otherwise we're just a raw pmem device */
457 return pmem_attach_disk(dev, ndns);
460 static int nd_pmem_remove(struct device *dev)
462 struct pmem_device *pmem = dev_get_drvdata(dev);
465 nvdimm_namespace_detach_btt(to_nd_btt(dev));
468 * Note, this assumes device_lock() context to not race
471 sysfs_put(pmem->bb_state);
472 pmem->bb_state = NULL;
474 nvdimm_flush(to_nd_region(dev->parent));
479 static void nd_pmem_shutdown(struct device *dev)
481 nvdimm_flush(to_nd_region(dev->parent));
484 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
486 struct nd_region *nd_region;
487 resource_size_t offset = 0, end_trunc = 0;
488 struct nd_namespace_common *ndns;
489 struct nd_namespace_io *nsio;
491 struct badblocks *bb;
492 struct kernfs_node *bb_state;
494 if (event != NVDIMM_REVALIDATE_POISON)
497 if (is_nd_btt(dev)) {
498 struct nd_btt *nd_btt = to_nd_btt(dev);
501 nd_region = to_nd_region(ndns->dev.parent);
502 nsio = to_nd_namespace_io(&ndns->dev);
506 struct pmem_device *pmem = dev_get_drvdata(dev);
508 nd_region = to_region(pmem);
510 bb_state = pmem->bb_state;
512 if (is_nd_pfn(dev)) {
513 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
514 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
517 offset = pmem->data_offset +
518 __le32_to_cpu(pfn_sb->start_pad);
519 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
524 nsio = to_nd_namespace_io(&ndns->dev);
527 res.start = nsio->res.start + offset;
528 res.end = nsio->res.end - end_trunc;
529 nvdimm_badblocks_populate(nd_region, bb, &res);
531 sysfs_notify_dirent(bb_state);
534 MODULE_ALIAS("pmem");
535 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
536 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
537 static struct nd_device_driver nd_pmem_driver = {
538 .probe = nd_pmem_probe,
539 .remove = nd_pmem_remove,
540 .notify = nd_pmem_notify,
541 .shutdown = nd_pmem_shutdown,
545 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
548 static int __init pmem_init(void)
550 return nd_driver_register(&nd_pmem_driver);
552 module_init(pmem_init);
554 static void pmem_exit(void)
556 driver_unregister(&nd_pmem_driver.drv);
558 module_exit(pmem_exit);
560 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
561 MODULE_LICENSE("GPL v2");