pmem.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Persistent Memory Driver
  4 *
  5 * Copyright (c) 2014-2015, Intel Corporation.
  6 * Copyright (c) 2015, Christoph Hellwig <hch@lst.de>.
  7 * Copyright (c) 2015, Boaz Harrosh <boaz@plexistor.com>.
  8 */
  9
 10#include <linux/blkdev.h>
 11#include <linux/pagemap.h>
 12#include <linux/hdreg.h>
 13#include <linux/init.h>
 14#include <linux/platform_device.h>
 15#include <linux/set_memory.h>
 16#include <linux/module.h>
 17#include <linux/moduleparam.h>
 18#include <linux/badblocks.h>
 19#include <linux/memremap.h>
 20#include <linux/kstrtox.h>
 21#include <linux/vmalloc.h>
 22#include <linux/blk-mq.h>
 23#include <linux/slab.h>
 24#include <linux/uio.h>
 25#include <linux/dax.h>
 26#include <linux/nd.h>
 27#include <linux/mm.h>
 28#include <asm/cacheflush.h>
 29#include "pmem.h"
 30#include "btt.h"
 31#include "pfn.h"
 32#include "nd.h"
 33
 34static struct device *to_dev(struct pmem_device *pmem)
 35{
 36	/*
 37	 * nvdimm bus services need a 'dev' parameter, and we record the device
 38	 * at init in bb.dev.
 39	 */
 40	return pmem->bb.dev;
 41}
 42
 43static struct nd_region *to_region(struct pmem_device *pmem)
 44{
 45	return to_nd_region(to_dev(pmem)->parent);
 46}
 47
 48static phys_addr_t pmem_to_phys(struct pmem_device *pmem, phys_addr_t offset)
 49{
 50	return pmem->phys_addr + offset;
 51}
 52
 53static sector_t to_sect(struct pmem_device *pmem, phys_addr_t offset)
 54{
 55	return (offset - pmem->data_offset) >> SECTOR_SHIFT;
 56}
 57
 58static phys_addr_t to_offset(struct pmem_device *pmem, sector_t sector)
 59{
 60	return (sector << SECTOR_SHIFT) + pmem->data_offset;
 61}
 62
 63static void pmem_mkpage_present(struct pmem_device *pmem, phys_addr_t offset,
 64		unsigned int len)
 65{
 66	phys_addr_t phys = pmem_to_phys(pmem, offset);
 67	unsigned long pfn_start, pfn_end, pfn;
 68
 69	/* only pmem in the linear map supports HWPoison */
 70	if (is_vmalloc_addr(pmem->virt_addr))
 71		return;
 72
 73	pfn_start = PHYS_PFN(phys);
 74	pfn_end = pfn_start + PHYS_PFN(len);
 75	for (pfn = pfn_start; pfn < pfn_end; pfn++) {
 76		struct page *page = pfn_to_page(pfn);
 77
 78		/*
 79		 * Note, no need to hold a get_dev_pagemap() reference
 80		 * here since we're in the driver I/O path and
 81		 * outstanding I/O requests pin the dev_pagemap.
 82		 */
 83		if (test_and_clear_pmem_poison(page))
 84			clear_mce_nospec(pfn);
 85	}
 86}
 87
 88static void pmem_clear_bb(struct pmem_device *pmem, sector_t sector, long blks)
 89{
 90	if (blks == 0)
 91		return;
 92	badblocks_clear(&pmem->bb, sector, blks);
 93	if (pmem->bb_state)
 94		sysfs_notify_dirent(pmem->bb_state);
 95}
 96
 97static long __pmem_clear_poison(struct pmem_device *pmem,
 98		phys_addr_t offset, unsigned int len)
 99{
100	phys_addr_t phys = pmem_to_phys(pmem, offset);
101	long cleared = nvdimm_clear_poison(to_dev(pmem), phys, len);
102
103	if (cleared > 0) {
104		pmem_mkpage_present(pmem, offset, cleared);
105		arch_invalidate_pmem(pmem->virt_addr + offset, len);
106	}
107	return cleared;
108}
109
110static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
111		phys_addr_t offset, unsigned int len)
112{
113	long cleared = __pmem_clear_poison(pmem, offset, len);
114
115	if (cleared < 0)
116		return BLK_STS_IOERR;
117
118	pmem_clear_bb(pmem, to_sect(pmem, offset), cleared >> SECTOR_SHIFT);
119	if (cleared < len)
120		return BLK_STS_IOERR;
121	return BLK_STS_OK;
122}
123
124static void write_pmem(void *pmem_addr, struct page *page,
125		unsigned int off, unsigned int len)
126{
127	unsigned int chunk;
128	void *mem;
129
130	while (len) {
131		mem = kmap_atomic(page);
132		chunk = min_t(unsigned int, len, PAGE_SIZE - off);
133		memcpy_flushcache(pmem_addr, mem + off, chunk);
134		kunmap_atomic(mem);
135		len -= chunk;
136		off = 0;
137		page++;
138		pmem_addr += chunk;
139	}
140}
141
142static blk_status_t read_pmem(struct page *page, unsigned int off,
143		void *pmem_addr, unsigned int len)
144{
145	unsigned int chunk;
146	unsigned long rem;
147	void *mem;
148
149	while (len) {
150		mem = kmap_atomic(page);
151		chunk = min_t(unsigned int, len, PAGE_SIZE - off);
152		rem = copy_mc_to_kernel(mem + off, pmem_addr, chunk);
153		kunmap_atomic(mem);
154		if (rem)
155			return BLK_STS_IOERR;
156		len -= chunk;
157		off = 0;
158		page++;
159		pmem_addr += chunk;
160	}
161	return BLK_STS_OK;
162}
163
164static blk_status_t pmem_do_read(struct pmem_device *pmem,
165			struct page *page, unsigned int page_off,
166			sector_t sector, unsigned int len)
167{
168	blk_status_t rc;
169	phys_addr_t pmem_off = to_offset(pmem, sector);
170	void *pmem_addr = pmem->virt_addr + pmem_off;
171
172	if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
173		return BLK_STS_IOERR;
174
175	rc = read_pmem(page, page_off, pmem_addr, len);
176	flush_dcache_page(page);
177	return rc;
178}
179
180static blk_status_t pmem_do_write(struct pmem_device *pmem,
181			struct page *page, unsigned int page_off,
182			sector_t sector, unsigned int len)
183{
184	phys_addr_t pmem_off = to_offset(pmem, sector);
185	void *pmem_addr = pmem->virt_addr + pmem_off;
186
187	if (unlikely(is_bad_pmem(&pmem->bb, sector, len))) {
188		blk_status_t rc = pmem_clear_poison(pmem, pmem_off, len);
189
190		if (rc != BLK_STS_OK)
191			return rc;
192	}
193
194	flush_dcache_page(page);
195	write_pmem(pmem_addr, page, page_off, len);
196
197	return BLK_STS_OK;
198}
199
200static void pmem_submit_bio(struct bio *bio)
201{
202	int ret = 0;
203	blk_status_t rc = 0;
204	bool do_acct;
205	unsigned long start;
206	struct bio_vec bvec;
207	struct bvec_iter iter;
208	struct pmem_device *pmem = bio->bi_bdev->bd_disk->private_data;
209	struct nd_region *nd_region = to_region(pmem);
210
211	if (bio->bi_opf & REQ_PREFLUSH)
212		ret = nvdimm_flush(nd_region, bio);
213
214	do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
215	if (do_acct)
216		start = bio_start_io_acct(bio);
217	bio_for_each_segment(bvec, bio, iter) {
218		if (op_is_write(bio_op(bio)))
219			rc = pmem_do_write(pmem, bvec.bv_page, bvec.bv_offset,
220				iter.bi_sector, bvec.bv_len);
221		else
222			rc = pmem_do_read(pmem, bvec.bv_page, bvec.bv_offset,
223				iter.bi_sector, bvec.bv_len);
224		if (rc) {
225			bio->bi_status = rc;
226			break;
227		}
228	}
229	if (do_acct)
230		bio_end_io_acct(bio, start);
231
232	if (bio->bi_opf & REQ_FUA)
233		ret = nvdimm_flush(nd_region, bio);
234
235	if (ret)
236		bio->bi_status = errno_to_blk_status(ret);
237
238	bio_endio(bio);
239}
240
241/* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
242__weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
243		long nr_pages, enum dax_access_mode mode, void **kaddr,
244		unsigned long *pfn)
245{
246	resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
247	sector_t sector = PFN_PHYS(pgoff) >> SECTOR_SHIFT;
248	unsigned int num = PFN_PHYS(nr_pages) >> SECTOR_SHIFT;
249	struct badblocks *bb = &pmem->bb;
250	sector_t first_bad;
251	sector_t num_bad;
252
253	if (kaddr)
254		*kaddr = pmem->virt_addr + offset;
255	if (pfn)
256		*pfn = PHYS_PFN(pmem->phys_addr + offset);
257
258	if (bb->count &&
259	    badblocks_check(bb, sector, num, &first_bad, &num_bad)) {
260		long actual_nr;
261
262		if (mode != DAX_RECOVERY_WRITE)
263			return -EHWPOISON;
264
265		/*
266		 * Set the recovery stride is set to kernel page size because
267		 * the underlying driver and firmware clear poison functions
268		 * don't appear to handle large chunk(such as 2MiB) reliably.
269		 */
270		actual_nr = PHYS_PFN(
271			PAGE_ALIGN((first_bad - sector) << SECTOR_SHIFT));
272		dev_dbg(pmem->bb.dev, "start sector(%llu), nr_pages(%ld), first_bad(%llu), actual_nr(%ld)\n",
273				sector, nr_pages, first_bad, actual_nr);
274		if (actual_nr)
275			return actual_nr;
276		return 1;
277	}
278
279	/*
280	 * If badblocks are present but not in the range, limit known good range
281	 * to the requested range.
282	 */
283	if (bb->count)
284		return nr_pages;
285	return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
286}
287
288static const struct block_device_operations pmem_fops = {
289	.owner =		THIS_MODULE,
290	.submit_bio =		pmem_submit_bio,
291};
292
293static int pmem_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
294				    size_t nr_pages)
295{
296	struct pmem_device *pmem = dax_get_private(dax_dev);
297
298	return blk_status_to_errno(pmem_do_write(pmem, ZERO_PAGE(0), 0,
299				   PFN_PHYS(pgoff) >> SECTOR_SHIFT,
300				   PAGE_SIZE));
301}
302
303static long pmem_dax_direct_access(struct dax_device *dax_dev,
304		pgoff_t pgoff, long nr_pages, enum dax_access_mode mode,
305		void **kaddr, unsigned long *pfn)
306{
307	struct pmem_device *pmem = dax_get_private(dax_dev);
308
309	return __pmem_direct_access(pmem, pgoff, nr_pages, mode, kaddr, pfn);
310}
311
312/*
313 * The recovery write thread started out as a normal pwrite thread and
314 * when the filesystem was told about potential media error in the
315 * range, filesystem turns the normal pwrite to a dax_recovery_write.
316 *
317 * The recovery write consists of clearing media poison, clearing page
318 * HWPoison bit, re-enable page-wide read-write permission, flush the
319 * caches and finally write.  A competing pread thread will be held
320 * off during the recovery process since data read back might not be
321 * valid, and this is achieved by clearing the badblock records after
322 * the recovery write is complete. Competing recovery write threads
323 * are already serialized by writer lock held by dax_iomap_rw().
324 */
325static size_t pmem_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
326		void *addr, size_t bytes, struct iov_iter *i)
327{
328	struct pmem_device *pmem = dax_get_private(dax_dev);
329	size_t olen, len, off;
330	phys_addr_t pmem_off;
331	struct device *dev = pmem->bb.dev;
332	long cleared;
333
334	off = offset_in_page(addr);
335	len = PFN_PHYS(PFN_UP(off + bytes));
336	if (!is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) >> SECTOR_SHIFT, len))
337		return _copy_from_iter_flushcache(addr, bytes, i);
338
339	/*
340	 * Not page-aligned range cannot be recovered. This should not
341	 * happen unless something else went wrong.
342	 */
343	if (off || !PAGE_ALIGNED(bytes)) {
344		dev_dbg(dev, "Found poison, but addr(%p) or bytes(%#zx) not page aligned\n",
345			addr, bytes);
346		return 0;
347	}
348
349	pmem_off = PFN_PHYS(pgoff) + pmem->data_offset;
350	cleared = __pmem_clear_poison(pmem, pmem_off, len);
351	if (cleared > 0 && cleared < len) {
352		dev_dbg(dev, "poison cleared only %ld out of %zu bytes\n",
353			cleared, len);
354		return 0;
355	}
356	if (cleared < 0) {
357		dev_dbg(dev, "poison clear failed: %ld\n", cleared);
358		return 0;
359	}
360
361	olen = _copy_from_iter_flushcache(addr, bytes, i);
362	pmem_clear_bb(pmem, to_sect(pmem, pmem_off), cleared >> SECTOR_SHIFT);
363
364	return olen;
365}
366
367static const struct dax_operations pmem_dax_ops = {
368	.direct_access = pmem_dax_direct_access,
369	.zero_page_range = pmem_dax_zero_page_range,
370	.recovery_write = pmem_recovery_write,
371};
372
373static ssize_t write_cache_show(struct device *dev,
374		struct device_attribute *attr, char *buf)
375{
376	struct pmem_device *pmem = dev_to_disk(dev)->private_data;
377
378	return sprintf(buf, "%d\n", !!dax_write_cache_enabled(pmem->dax_dev));
379}
380
381static ssize_t write_cache_store(struct device *dev,
382		struct device_attribute *attr, const char *buf, size_t len)
383{
384	struct pmem_device *pmem = dev_to_disk(dev)->private_data;
385	bool write_cache;
386	int rc;
387
388	rc = kstrtobool(buf, &write_cache);
389	if (rc)
390		return rc;
391	dax_write_cache(pmem->dax_dev, write_cache);
392	return len;
393}
394static DEVICE_ATTR_RW(write_cache);
395
396static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n)
397{
398#ifndef CONFIG_ARCH_HAS_PMEM_API
399	if (a == &dev_attr_write_cache.attr)
400		return 0;
401#endif
402	return a->mode;
403}
404
405static struct attribute *dax_attributes[] = {
406	&dev_attr_write_cache.attr,
407	NULL,
408};
409
410static const struct attribute_group dax_attribute_group = {
411	.name		= "dax",
412	.attrs		= dax_attributes,
413	.is_visible	= dax_visible,
414};
415
416static const struct attribute_group *pmem_attribute_groups[] = {
417	&dax_attribute_group,
418	NULL,
419};
420
421static void pmem_release_disk(void *__pmem)
422{
423	struct pmem_device *pmem = __pmem;
424
425	dax_remove_host(pmem->disk);
426	kill_dax(pmem->dax_dev);
427	put_dax(pmem->dax_dev);
428	del_gendisk(pmem->disk);
429
430	put_disk(pmem->disk);
431}
432
433static int pmem_pagemap_memory_failure(struct dev_pagemap *pgmap,
434		unsigned long pfn, unsigned long nr_pages, int mf_flags)
435{
436	struct pmem_device *pmem =
437			container_of(pgmap, struct pmem_device, pgmap);
438	u64 offset = PFN_PHYS(pfn) - pmem->phys_addr - pmem->data_offset;
439	u64 len = nr_pages << PAGE_SHIFT;
440
441	return dax_holder_notify_failure(pmem->dax_dev, offset, len, mf_flags);
442}
443
444static const struct dev_pagemap_ops fsdax_pagemap_ops = {
445	.memory_failure		= pmem_pagemap_memory_failure,
446};
447
448static int pmem_attach_disk(struct device *dev,
449		struct nd_namespace_common *ndns)
450{
451	struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
452	struct nd_region *nd_region = to_nd_region(dev->parent);
453	struct queue_limits lim = {
454		.logical_block_size	= pmem_sector_size(ndns),
455		.physical_block_size	= PAGE_SIZE,
456		.max_hw_sectors		= UINT_MAX,
457		.features		= BLK_FEAT_WRITE_CACHE |
458					  BLK_FEAT_SYNCHRONOUS,
459	};
460	int nid = dev_to_node(dev), fua;
461	struct resource *res = &nsio->res;
462	struct range bb_range;
463	struct nd_pfn *nd_pfn = NULL;
464	struct dax_device *dax_dev;
465	struct nd_pfn_sb *pfn_sb;
466	struct pmem_device *pmem;
467	struct gendisk *disk;
468	void *addr;
469	int rc;
470
471	pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
472	if (!pmem)
473		return -ENOMEM;
474
475	rc = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
476	if (rc)
477		return rc;
478
479	/* while nsio_rw_bytes is active, parse a pfn info block if present */
480	if (is_nd_pfn(dev)) {
481		nd_pfn = to_nd_pfn(dev);
482		rc = nvdimm_setup_pfn(nd_pfn, &pmem->pgmap);
483		if (rc)
484			return rc;
485	}
486
487	/* we're attaching a block device, disable raw namespace access */
488	devm_namespace_disable(dev, ndns);
489
490	dev_set_drvdata(dev, pmem);
491	pmem->phys_addr = res->start;
492	pmem->size = resource_size(res);
493	fua = nvdimm_has_flush(nd_region);
494	if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) || fua < 0) {
495		dev_warn(dev, "unable to guarantee persistence of writes\n");
496		fua = 0;
497	}
498	if (fua)
499		lim.features |= BLK_FEAT_FUA;
500	if (is_nd_pfn(dev) || pmem_should_map_pages(dev))
501		lim.features |= BLK_FEAT_DAX;
502
503	if (!devm_request_mem_region(dev, res->start, resource_size(res),
504				dev_name(&ndns->dev))) {
505		dev_warn(dev, "could not reserve region %pR\n", res);
506		return -EBUSY;
507	}
508
509	disk = blk_alloc_disk(&lim, nid);
510	if (IS_ERR(disk))
511		return PTR_ERR(disk);
512
513	pmem->disk = disk;
514	pmem->pgmap.owner = pmem;
515	if (is_nd_pfn(dev)) {
516		pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
517		pmem->pgmap.ops = &fsdax_pagemap_ops;
518		addr = devm_memremap_pages(dev, &pmem->pgmap);
519		pfn_sb = nd_pfn->pfn_sb;
520		pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
521		pmem->pfn_pad = resource_size(res) -
522			range_len(&pmem->pgmap.range);
523		bb_range = pmem->pgmap.range;
524		bb_range.start += pmem->data_offset;
525	} else if (pmem_should_map_pages(dev)) {
526		pmem->pgmap.range.start = res->start;
527		pmem->pgmap.range.end = res->end;
528		pmem->pgmap.nr_range = 1;
529		pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
530		pmem->pgmap.ops = &fsdax_pagemap_ops;
531		addr = devm_memremap_pages(dev, &pmem->pgmap);
532		bb_range = pmem->pgmap.range;
533	} else {
534		addr = devm_memremap(dev, pmem->phys_addr,
535				pmem->size, ARCH_MEMREMAP_PMEM);
536		bb_range.start =  res->start;
537		bb_range.end = res->end;
538	}
539
540	if (IS_ERR(addr)) {
541		rc = PTR_ERR(addr);
542		goto out;
543	}
544	pmem->virt_addr = addr;
545
546	disk->fops		= &pmem_fops;
547	disk->private_data	= pmem;
548	nvdimm_namespace_disk_name(ndns, disk->disk_name);
549	set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
550			/ 512);
551	if (devm_init_badblocks(dev, &pmem->bb))
552		return -ENOMEM;
553	nvdimm_badblocks_populate(nd_region, &pmem->bb, &bb_range);
554	disk->bb = &pmem->bb;
555
556	dax_dev = alloc_dax(pmem, &pmem_dax_ops);
557	if (IS_ERR(dax_dev)) {
558		rc = PTR_ERR(dax_dev);
559		if (rc != -EOPNOTSUPP)
560			goto out;
561	} else {
562		set_dax_nocache(dax_dev);
563		set_dax_nomc(dax_dev);
564		if (is_nvdimm_sync(nd_region))
565			set_dax_synchronous(dax_dev);
566		pmem->dax_dev = dax_dev;
567		rc = dax_add_host(dax_dev, disk);
568		if (rc)
569			goto out_cleanup_dax;
570		dax_write_cache(dax_dev, nvdimm_has_cache(nd_region));
571	}
572	rc = device_add_disk(dev, disk, pmem_attribute_groups);
573	if (rc)
574		goto out_remove_host;
575	if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
576		return -ENOMEM;
577
578	nvdimm_check_and_set_ro(disk);
579
580	pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
581					  "badblocks");
582	if (!pmem->bb_state)
583		dev_warn(dev, "'badblocks' notification disabled\n");
584	return 0;
585
586out_remove_host:
587	dax_remove_host(pmem->disk);
588out_cleanup_dax:
589	kill_dax(pmem->dax_dev);
590	put_dax(pmem->dax_dev);
591out:
592	put_disk(pmem->disk);
593	return rc;
594}
595
596static int nd_pmem_probe(struct device *dev)
597{
598	int ret;
599	struct nd_namespace_common *ndns;
600
601	ndns = nvdimm_namespace_common_probe(dev);
602	if (IS_ERR(ndns))
603		return PTR_ERR(ndns);
604
605	if (is_nd_btt(dev))
606		return nvdimm_namespace_attach_btt(ndns);
607
608	if (is_nd_pfn(dev))
609		return pmem_attach_disk(dev, ndns);
610
611	ret = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
612	if (ret)
613		return ret;
614
615	ret = nd_btt_probe(dev, ndns);
616	if (ret == 0)
617		return -ENXIO;
618
619	/*
620	 * We have two failure conditions here, there is no
621	 * info reserver block or we found a valid info reserve block
622	 * but failed to initialize the pfn superblock.
623	 *
624	 * For the first case consider namespace as a raw pmem namespace
625	 * and attach a disk.
626	 *
627	 * For the latter, consider this a success and advance the namespace
628	 * seed.
629	 */
630	ret = nd_pfn_probe(dev, ndns);
631	if (ret == 0)
632		return -ENXIO;
633	else if (ret == -EOPNOTSUPP)
634		return ret;
635
636	ret = nd_dax_probe(dev, ndns);
637	if (ret == 0)
638		return -ENXIO;
639	else if (ret == -EOPNOTSUPP)
640		return ret;
641
642	/* probe complete, attach handles namespace enabling */
643	devm_namespace_disable(dev, ndns);
644
645	return pmem_attach_disk(dev, ndns);
646}
647
648static void nd_pmem_remove(struct device *dev)
649{
650	struct pmem_device *pmem = dev_get_drvdata(dev);
651
652	if (is_nd_btt(dev))
653		nvdimm_namespace_detach_btt(to_nd_btt(dev));
654	else {
655		/*
656		 * Note, this assumes device_lock() context to not
657		 * race nd_pmem_notify()
658		 */
659		sysfs_put(pmem->bb_state);
660		pmem->bb_state = NULL;
661	}
662	nvdimm_flush(to_nd_region(dev->parent), NULL);
663}
664
665static void nd_pmem_shutdown(struct device *dev)
666{
667	nvdimm_flush(to_nd_region(dev->parent), NULL);
668}
669
670static void pmem_revalidate_poison(struct device *dev)
671{
672	struct nd_region *nd_region;
673	resource_size_t offset = 0, end_trunc = 0;
674	struct nd_namespace_common *ndns;
675	struct nd_namespace_io *nsio;
676	struct badblocks *bb;
677	struct range range;
678	struct kernfs_node *bb_state;
679
680	if (is_nd_btt(dev)) {
681		struct nd_btt *nd_btt = to_nd_btt(dev);
682
683		ndns = nd_btt->ndns;
684		nd_region = to_nd_region(ndns->dev.parent);
685		nsio = to_nd_namespace_io(&ndns->dev);
686		bb = &nsio->bb;
687		bb_state = NULL;
688	} else {
689		struct pmem_device *pmem = dev_get_drvdata(dev);
690
691		nd_region = to_region(pmem);
692		bb = &pmem->bb;
693		bb_state = pmem->bb_state;
694
695		if (is_nd_pfn(dev)) {
696			struct nd_pfn *nd_pfn = to_nd_pfn(dev);
697			struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
698
699			ndns = nd_pfn->ndns;
700			offset = pmem->data_offset +
701					__le32_to_cpu(pfn_sb->start_pad);
702			end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
703		} else {
704			ndns = to_ndns(dev);
705		}
706
707		nsio = to_nd_namespace_io(&ndns->dev);
708	}
709
710	range.start = nsio->res.start + offset;
711	range.end = nsio->res.end - end_trunc;
712	nvdimm_badblocks_populate(nd_region, bb, &range);
713	if (bb_state)
714		sysfs_notify_dirent(bb_state);
715}
716
717static void pmem_revalidate_region(struct device *dev)
718{
719	struct pmem_device *pmem;
720
721	if (is_nd_btt(dev)) {
722		struct nd_btt *nd_btt = to_nd_btt(dev);
723		struct btt *btt = nd_btt->btt;
724
725		nvdimm_check_and_set_ro(btt->btt_disk);
726		return;
727	}
728
729	pmem = dev_get_drvdata(dev);
730	nvdimm_check_and_set_ro(pmem->disk);
731}
732
733static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
734{
735	switch (event) {
736	case NVDIMM_REVALIDATE_POISON:
737		pmem_revalidate_poison(dev);
738		break;
739	case NVDIMM_REVALIDATE_REGION:
740		pmem_revalidate_region(dev);
741		break;
742	default:
743		dev_WARN_ONCE(dev, 1, "notify: unknown event: %d\n", event);
744		break;
745	}
746}
747
748MODULE_ALIAS("pmem");
749MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
750MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
751static struct nd_device_driver nd_pmem_driver = {
752	.probe = nd_pmem_probe,
753	.remove = nd_pmem_remove,
754	.notify = nd_pmem_notify,
755	.shutdown = nd_pmem_shutdown,
756	.drv = {
757		.name = "nd_pmem",
758	},
759	.type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
760};
761
762module_nd_driver(nd_pmem_driver);
763
764MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
765MODULE_DESCRIPTION("NVDIMM Persistent Memory Driver");
766MODULE_LICENSE("GPL v2");