Loading...
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/hdreg.h>
12#include <linux/init.h>
13#include <linux/platform_device.h>
14#include <linux/set_memory.h>
15#include <linux/module.h>
16#include <linux/moduleparam.h>
17#include <linux/badblocks.h>
18#include <linux/memremap.h>
19#include <linux/vmalloc.h>
20#include <linux/blk-mq.h>
21#include <linux/pfn_t.h>
22#include <linux/slab.h>
23#include <linux/uio.h>
24#include <linux/dax.h>
25#include <linux/nd.h>
26#include <linux/backing-dev.h>
27#include <linux/mm.h>
28#include <asm/cacheflush.h>
29#include "pmem.h"
30#include "pfn.h"
31#include "nd.h"
32
33static struct device *to_dev(struct pmem_device *pmem)
34{
35 /*
36 * nvdimm bus services need a 'dev' parameter, and we record the device
37 * at init in bb.dev.
38 */
39 return pmem->bb.dev;
40}
41
42static struct nd_region *to_region(struct pmem_device *pmem)
43{
44 return to_nd_region(to_dev(pmem)->parent);
45}
46
47static void hwpoison_clear(struct pmem_device *pmem,
48 phys_addr_t phys, unsigned int len)
49{
50 unsigned long pfn_start, pfn_end, pfn;
51
52 /* only pmem in the linear map supports HWPoison */
53 if (is_vmalloc_addr(pmem->virt_addr))
54 return;
55
56 pfn_start = PHYS_PFN(phys);
57 pfn_end = pfn_start + PHYS_PFN(len);
58 for (pfn = pfn_start; pfn < pfn_end; pfn++) {
59 struct page *page = pfn_to_page(pfn);
60
61 /*
62 * Note, no need to hold a get_dev_pagemap() reference
63 * here since we're in the driver I/O path and
64 * outstanding I/O requests pin the dev_pagemap.
65 */
66 if (test_and_clear_pmem_poison(page))
67 clear_mce_nospec(pfn);
68 }
69}
70
71static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
72 phys_addr_t offset, unsigned int len)
73{
74 struct device *dev = to_dev(pmem);
75 sector_t sector;
76 long cleared;
77 blk_status_t rc = BLK_STS_OK;
78
79 sector = (offset - pmem->data_offset) / 512;
80
81 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
82 if (cleared < len)
83 rc = BLK_STS_IOERR;
84 if (cleared > 0 && cleared / 512) {
85 hwpoison_clear(pmem, pmem->phys_addr + offset, cleared);
86 cleared /= 512;
87 dev_dbg(dev, "%#llx clear %ld sector%s\n",
88 (unsigned long long) sector, cleared,
89 cleared > 1 ? "s" : "");
90 badblocks_clear(&pmem->bb, sector, cleared);
91 if (pmem->bb_state)
92 sysfs_notify_dirent(pmem->bb_state);
93 }
94
95 arch_invalidate_pmem(pmem->virt_addr + offset, len);
96
97 return rc;
98}
99
100static void write_pmem(void *pmem_addr, struct page *page,
101 unsigned int off, unsigned int len)
102{
103 unsigned int chunk;
104 void *mem;
105
106 while (len) {
107 mem = kmap_atomic(page);
108 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
109 memcpy_flushcache(pmem_addr, mem + off, chunk);
110 kunmap_atomic(mem);
111 len -= chunk;
112 off = 0;
113 page++;
114 pmem_addr += chunk;
115 }
116}
117
118static blk_status_t read_pmem(struct page *page, unsigned int off,
119 void *pmem_addr, unsigned int len)
120{
121 unsigned int chunk;
122 unsigned long rem;
123 void *mem;
124
125 while (len) {
126 mem = kmap_atomic(page);
127 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
128 rem = memcpy_mcsafe(mem + off, pmem_addr, chunk);
129 kunmap_atomic(mem);
130 if (rem)
131 return BLK_STS_IOERR;
132 len -= chunk;
133 off = 0;
134 page++;
135 pmem_addr += chunk;
136 }
137 return BLK_STS_OK;
138}
139
140static blk_status_t pmem_do_read(struct pmem_device *pmem,
141 struct page *page, unsigned int page_off,
142 sector_t sector, unsigned int len)
143{
144 blk_status_t rc;
145 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
146 void *pmem_addr = pmem->virt_addr + pmem_off;
147
148 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
149 return BLK_STS_IOERR;
150
151 rc = read_pmem(page, page_off, pmem_addr, len);
152 flush_dcache_page(page);
153 return rc;
154}
155
156static blk_status_t pmem_do_write(struct pmem_device *pmem,
157 struct page *page, unsigned int page_off,
158 sector_t sector, unsigned int len)
159{
160 blk_status_t rc = BLK_STS_OK;
161 bool bad_pmem = false;
162 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
163 void *pmem_addr = pmem->virt_addr + pmem_off;
164
165 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
166 bad_pmem = true;
167
168 /*
169 * Note that we write the data both before and after
170 * clearing poison. The write before clear poison
171 * handles situations where the latest written data is
172 * preserved and the clear poison operation simply marks
173 * the address range as valid without changing the data.
174 * In this case application software can assume that an
175 * interrupted write will either return the new good
176 * data or an error.
177 *
178 * However, if pmem_clear_poison() leaves the data in an
179 * indeterminate state we need to perform the write
180 * after clear poison.
181 */
182 flush_dcache_page(page);
183 write_pmem(pmem_addr, page, page_off, len);
184 if (unlikely(bad_pmem)) {
185 rc = pmem_clear_poison(pmem, pmem_off, len);
186 write_pmem(pmem_addr, page, page_off, len);
187 }
188
189 return rc;
190}
191
192static blk_qc_t pmem_submit_bio(struct bio *bio)
193{
194 int ret = 0;
195 blk_status_t rc = 0;
196 bool do_acct;
197 unsigned long start;
198 struct bio_vec bvec;
199 struct bvec_iter iter;
200 struct pmem_device *pmem = bio->bi_disk->private_data;
201 struct nd_region *nd_region = to_region(pmem);
202
203 if (bio->bi_opf & REQ_PREFLUSH)
204 ret = nvdimm_flush(nd_region, bio);
205
206 do_acct = blk_queue_io_stat(bio->bi_disk->queue);
207 if (do_acct)
208 start = bio_start_io_acct(bio);
209 bio_for_each_segment(bvec, bio, iter) {
210 if (op_is_write(bio_op(bio)))
211 rc = pmem_do_write(pmem, bvec.bv_page, bvec.bv_offset,
212 iter.bi_sector, bvec.bv_len);
213 else
214 rc = pmem_do_read(pmem, bvec.bv_page, bvec.bv_offset,
215 iter.bi_sector, bvec.bv_len);
216 if (rc) {
217 bio->bi_status = rc;
218 break;
219 }
220 }
221 if (do_acct)
222 bio_end_io_acct(bio, start);
223
224 if (bio->bi_opf & REQ_FUA)
225 ret = nvdimm_flush(nd_region, bio);
226
227 if (ret)
228 bio->bi_status = errno_to_blk_status(ret);
229
230 bio_endio(bio);
231 return BLK_QC_T_NONE;
232}
233
234static int pmem_rw_page(struct block_device *bdev, sector_t sector,
235 struct page *page, unsigned int op)
236{
237 struct pmem_device *pmem = bdev->bd_disk->private_data;
238 blk_status_t rc;
239
240 if (op_is_write(op))
241 rc = pmem_do_write(pmem, page, 0, sector, thp_size(page));
242 else
243 rc = pmem_do_read(pmem, page, 0, sector, thp_size(page));
244 /*
245 * The ->rw_page interface is subtle and tricky. The core
246 * retries on any error, so we can only invoke page_endio() in
247 * the successful completion case. Otherwise, we'll see crashes
248 * caused by double completion.
249 */
250 if (rc == 0)
251 page_endio(page, op_is_write(op), 0);
252
253 return blk_status_to_errno(rc);
254}
255
256/* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
257__weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
258 long nr_pages, void **kaddr, pfn_t *pfn)
259{
260 resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
261
262 if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
263 PFN_PHYS(nr_pages))))
264 return -EIO;
265
266 if (kaddr)
267 *kaddr = pmem->virt_addr + offset;
268 if (pfn)
269 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
270
271 /*
272 * If badblocks are present, limit known good range to the
273 * requested range.
274 */
275 if (unlikely(pmem->bb.count))
276 return nr_pages;
277 return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
278}
279
280static const struct block_device_operations pmem_fops = {
281 .owner = THIS_MODULE,
282 .submit_bio = pmem_submit_bio,
283 .rw_page = pmem_rw_page,
284 .revalidate_disk = nvdimm_revalidate_disk,
285};
286
287static int pmem_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
288 size_t nr_pages)
289{
290 struct pmem_device *pmem = dax_get_private(dax_dev);
291
292 return blk_status_to_errno(pmem_do_write(pmem, ZERO_PAGE(0), 0,
293 PFN_PHYS(pgoff) >> SECTOR_SHIFT,
294 PAGE_SIZE));
295}
296
297static long pmem_dax_direct_access(struct dax_device *dax_dev,
298 pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn)
299{
300 struct pmem_device *pmem = dax_get_private(dax_dev);
301
302 return __pmem_direct_access(pmem, pgoff, nr_pages, kaddr, pfn);
303}
304
305/*
306 * Use the 'no check' versions of copy_from_iter_flushcache() and
307 * copy_to_iter_mcsafe() to bypass HARDENED_USERCOPY overhead. Bounds
308 * checking, both file offset and device offset, is handled by
309 * dax_iomap_actor()
310 */
311static size_t pmem_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
312 void *addr, size_t bytes, struct iov_iter *i)
313{
314 return _copy_from_iter_flushcache(addr, bytes, i);
315}
316
317static size_t pmem_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
318 void *addr, size_t bytes, struct iov_iter *i)
319{
320 return _copy_to_iter_mcsafe(addr, bytes, i);
321}
322
323static const struct dax_operations pmem_dax_ops = {
324 .direct_access = pmem_dax_direct_access,
325 .dax_supported = generic_fsdax_supported,
326 .copy_from_iter = pmem_copy_from_iter,
327 .copy_to_iter = pmem_copy_to_iter,
328 .zero_page_range = pmem_dax_zero_page_range,
329};
330
331static const struct attribute_group *pmem_attribute_groups[] = {
332 &dax_attribute_group,
333 NULL,
334};
335
336static void pmem_pagemap_cleanup(struct dev_pagemap *pgmap)
337{
338 struct request_queue *q =
339 container_of(pgmap->ref, struct request_queue, q_usage_counter);
340
341 blk_cleanup_queue(q);
342}
343
344static void pmem_release_queue(void *pgmap)
345{
346 pmem_pagemap_cleanup(pgmap);
347}
348
349static void pmem_pagemap_kill(struct dev_pagemap *pgmap)
350{
351 struct request_queue *q =
352 container_of(pgmap->ref, struct request_queue, q_usage_counter);
353
354 blk_freeze_queue_start(q);
355}
356
357static void pmem_release_disk(void *__pmem)
358{
359 struct pmem_device *pmem = __pmem;
360
361 kill_dax(pmem->dax_dev);
362 put_dax(pmem->dax_dev);
363 del_gendisk(pmem->disk);
364 put_disk(pmem->disk);
365}
366
367static const struct dev_pagemap_ops fsdax_pagemap_ops = {
368 .kill = pmem_pagemap_kill,
369 .cleanup = pmem_pagemap_cleanup,
370};
371
372static int pmem_attach_disk(struct device *dev,
373 struct nd_namespace_common *ndns)
374{
375 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
376 struct nd_region *nd_region = to_nd_region(dev->parent);
377 int nid = dev_to_node(dev), fua;
378 struct resource *res = &nsio->res;
379 struct resource bb_res;
380 struct nd_pfn *nd_pfn = NULL;
381 struct dax_device *dax_dev;
382 struct nd_pfn_sb *pfn_sb;
383 struct pmem_device *pmem;
384 struct request_queue *q;
385 struct device *gendev;
386 struct gendisk *disk;
387 void *addr;
388 int rc;
389 unsigned long flags = 0UL;
390
391 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
392 if (!pmem)
393 return -ENOMEM;
394
395 rc = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
396 if (rc)
397 return rc;
398
399 /* while nsio_rw_bytes is active, parse a pfn info block if present */
400 if (is_nd_pfn(dev)) {
401 nd_pfn = to_nd_pfn(dev);
402 rc = nvdimm_setup_pfn(nd_pfn, &pmem->pgmap);
403 if (rc)
404 return rc;
405 }
406
407 /* we're attaching a block device, disable raw namespace access */
408 devm_namespace_disable(dev, ndns);
409
410 dev_set_drvdata(dev, pmem);
411 pmem->phys_addr = res->start;
412 pmem->size = resource_size(res);
413 fua = nvdimm_has_flush(nd_region);
414 if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE) || fua < 0) {
415 dev_warn(dev, "unable to guarantee persistence of writes\n");
416 fua = 0;
417 }
418
419 if (!devm_request_mem_region(dev, res->start, resource_size(res),
420 dev_name(&ndns->dev))) {
421 dev_warn(dev, "could not reserve region %pR\n", res);
422 return -EBUSY;
423 }
424
425 q = blk_alloc_queue(dev_to_node(dev));
426 if (!q)
427 return -ENOMEM;
428
429 pmem->pfn_flags = PFN_DEV;
430 pmem->pgmap.ref = &q->q_usage_counter;
431 if (is_nd_pfn(dev)) {
432 pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
433 pmem->pgmap.ops = &fsdax_pagemap_ops;
434 addr = devm_memremap_pages(dev, &pmem->pgmap);
435 pfn_sb = nd_pfn->pfn_sb;
436 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
437 pmem->pfn_pad = resource_size(res) -
438 resource_size(&pmem->pgmap.res);
439 pmem->pfn_flags |= PFN_MAP;
440 memcpy(&bb_res, &pmem->pgmap.res, sizeof(bb_res));
441 bb_res.start += pmem->data_offset;
442 } else if (pmem_should_map_pages(dev)) {
443 memcpy(&pmem->pgmap.res, &nsio->res, sizeof(pmem->pgmap.res));
444 pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
445 pmem->pgmap.ops = &fsdax_pagemap_ops;
446 addr = devm_memremap_pages(dev, &pmem->pgmap);
447 pmem->pfn_flags |= PFN_MAP;
448 memcpy(&bb_res, &pmem->pgmap.res, sizeof(bb_res));
449 } else {
450 if (devm_add_action_or_reset(dev, pmem_release_queue,
451 &pmem->pgmap))
452 return -ENOMEM;
453 addr = devm_memremap(dev, pmem->phys_addr,
454 pmem->size, ARCH_MEMREMAP_PMEM);
455 memcpy(&bb_res, &nsio->res, sizeof(bb_res));
456 }
457
458 if (IS_ERR(addr))
459 return PTR_ERR(addr);
460 pmem->virt_addr = addr;
461
462 blk_queue_write_cache(q, true, fua);
463 blk_queue_physical_block_size(q, PAGE_SIZE);
464 blk_queue_logical_block_size(q, pmem_sector_size(ndns));
465 blk_queue_max_hw_sectors(q, UINT_MAX);
466 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
467 if (pmem->pfn_flags & PFN_MAP)
468 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
469
470 disk = alloc_disk_node(0, nid);
471 if (!disk)
472 return -ENOMEM;
473 pmem->disk = disk;
474
475 disk->fops = &pmem_fops;
476 disk->queue = q;
477 disk->flags = GENHD_FL_EXT_DEVT;
478 disk->private_data = pmem;
479 disk->queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO;
480 nvdimm_namespace_disk_name(ndns, disk->disk_name);
481 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
482 / 512);
483 if (devm_init_badblocks(dev, &pmem->bb))
484 return -ENOMEM;
485 nvdimm_badblocks_populate(nd_region, &pmem->bb, &bb_res);
486 disk->bb = &pmem->bb;
487
488 if (is_nvdimm_sync(nd_region))
489 flags = DAXDEV_F_SYNC;
490 dax_dev = alloc_dax(pmem, disk->disk_name, &pmem_dax_ops, flags);
491 if (IS_ERR(dax_dev)) {
492 put_disk(disk);
493 return PTR_ERR(dax_dev);
494 }
495 dax_write_cache(dax_dev, nvdimm_has_cache(nd_region));
496 pmem->dax_dev = dax_dev;
497 gendev = disk_to_dev(disk);
498 gendev->groups = pmem_attribute_groups;
499
500 device_add_disk(dev, disk, NULL);
501 if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
502 return -ENOMEM;
503
504 revalidate_disk(disk);
505
506 pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
507 "badblocks");
508 if (!pmem->bb_state)
509 dev_warn(dev, "'badblocks' notification disabled\n");
510
511 return 0;
512}
513
514static int nd_pmem_probe(struct device *dev)
515{
516 int ret;
517 struct nd_namespace_common *ndns;
518
519 ndns = nvdimm_namespace_common_probe(dev);
520 if (IS_ERR(ndns))
521 return PTR_ERR(ndns);
522
523 if (is_nd_btt(dev))
524 return nvdimm_namespace_attach_btt(ndns);
525
526 if (is_nd_pfn(dev))
527 return pmem_attach_disk(dev, ndns);
528
529 ret = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
530 if (ret)
531 return ret;
532
533 ret = nd_btt_probe(dev, ndns);
534 if (ret == 0)
535 return -ENXIO;
536
537 /*
538 * We have two failure conditions here, there is no
539 * info reserver block or we found a valid info reserve block
540 * but failed to initialize the pfn superblock.
541 *
542 * For the first case consider namespace as a raw pmem namespace
543 * and attach a disk.
544 *
545 * For the latter, consider this a success and advance the namespace
546 * seed.
547 */
548 ret = nd_pfn_probe(dev, ndns);
549 if (ret == 0)
550 return -ENXIO;
551 else if (ret == -EOPNOTSUPP)
552 return ret;
553
554 ret = nd_dax_probe(dev, ndns);
555 if (ret == 0)
556 return -ENXIO;
557 else if (ret == -EOPNOTSUPP)
558 return ret;
559
560 /* probe complete, attach handles namespace enabling */
561 devm_namespace_disable(dev, ndns);
562
563 return pmem_attach_disk(dev, ndns);
564}
565
566static int nd_pmem_remove(struct device *dev)
567{
568 struct pmem_device *pmem = dev_get_drvdata(dev);
569
570 if (is_nd_btt(dev))
571 nvdimm_namespace_detach_btt(to_nd_btt(dev));
572 else {
573 /*
574 * Note, this assumes nd_device_lock() context to not
575 * race nd_pmem_notify()
576 */
577 sysfs_put(pmem->bb_state);
578 pmem->bb_state = NULL;
579 }
580 nvdimm_flush(to_nd_region(dev->parent), NULL);
581
582 return 0;
583}
584
585static void nd_pmem_shutdown(struct device *dev)
586{
587 nvdimm_flush(to_nd_region(dev->parent), NULL);
588}
589
590static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
591{
592 struct nd_region *nd_region;
593 resource_size_t offset = 0, end_trunc = 0;
594 struct nd_namespace_common *ndns;
595 struct nd_namespace_io *nsio;
596 struct resource res;
597 struct badblocks *bb;
598 struct kernfs_node *bb_state;
599
600 if (event != NVDIMM_REVALIDATE_POISON)
601 return;
602
603 if (is_nd_btt(dev)) {
604 struct nd_btt *nd_btt = to_nd_btt(dev);
605
606 ndns = nd_btt->ndns;
607 nd_region = to_nd_region(ndns->dev.parent);
608 nsio = to_nd_namespace_io(&ndns->dev);
609 bb = &nsio->bb;
610 bb_state = NULL;
611 } else {
612 struct pmem_device *pmem = dev_get_drvdata(dev);
613
614 nd_region = to_region(pmem);
615 bb = &pmem->bb;
616 bb_state = pmem->bb_state;
617
618 if (is_nd_pfn(dev)) {
619 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
620 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
621
622 ndns = nd_pfn->ndns;
623 offset = pmem->data_offset +
624 __le32_to_cpu(pfn_sb->start_pad);
625 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
626 } else {
627 ndns = to_ndns(dev);
628 }
629
630 nsio = to_nd_namespace_io(&ndns->dev);
631 }
632
633 res.start = nsio->res.start + offset;
634 res.end = nsio->res.end - end_trunc;
635 nvdimm_badblocks_populate(nd_region, bb, &res);
636 if (bb_state)
637 sysfs_notify_dirent(bb_state);
638}
639
640MODULE_ALIAS("pmem");
641MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
642MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
643static struct nd_device_driver nd_pmem_driver = {
644 .probe = nd_pmem_probe,
645 .remove = nd_pmem_remove,
646 .notify = nd_pmem_notify,
647 .shutdown = nd_pmem_shutdown,
648 .drv = {
649 .name = "nd_pmem",
650 },
651 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
652};
653
654module_nd_driver(nd_pmem_driver);
655
656MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
657MODULE_LICENSE("GPL v2");
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 "pfn.h"
33#include "nd.h"
34
35struct pmem_device {
36 struct request_queue *pmem_queue;
37 struct gendisk *pmem_disk;
38 struct nd_namespace_common *ndns;
39
40 /* One contiguous memory region per device */
41 phys_addr_t phys_addr;
42 /* when non-zero this device is hosting a 'pfn' instance */
43 phys_addr_t data_offset;
44 u64 pfn_flags;
45 void __pmem *virt_addr;
46 /* immutable base size of the namespace */
47 size_t size;
48 /* trim size when namespace capacity has been section aligned */
49 u32 pfn_pad;
50 struct badblocks bb;
51};
52
53static bool is_bad_pmem(struct badblocks *bb, sector_t sector, unsigned int len)
54{
55 if (bb->count) {
56 sector_t first_bad;
57 int num_bad;
58
59 return !!badblocks_check(bb, sector, len / 512, &first_bad,
60 &num_bad);
61 }
62
63 return false;
64}
65
66static void pmem_clear_poison(struct pmem_device *pmem, phys_addr_t offset,
67 unsigned int len)
68{
69 struct device *dev = disk_to_dev(pmem->pmem_disk);
70 sector_t sector;
71 long cleared;
72
73 sector = (offset - pmem->data_offset) / 512;
74 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
75
76 if (cleared > 0 && cleared / 512) {
77 dev_dbg(dev, "%s: %llx clear %ld sector%s\n",
78 __func__, (unsigned long long) sector,
79 cleared / 512, cleared / 512 > 1 ? "s" : "");
80 badblocks_clear(&pmem->bb, sector, cleared / 512);
81 }
82 invalidate_pmem(pmem->virt_addr + offset, len);
83}
84
85static int pmem_do_bvec(struct pmem_device *pmem, struct page *page,
86 unsigned int len, unsigned int off, int rw,
87 sector_t sector)
88{
89 int rc = 0;
90 bool bad_pmem = false;
91 void *mem = kmap_atomic(page);
92 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
93 void __pmem *pmem_addr = pmem->virt_addr + pmem_off;
94
95 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
96 bad_pmem = true;
97
98 if (rw == READ) {
99 if (unlikely(bad_pmem))
100 rc = -EIO;
101 else {
102 rc = memcpy_from_pmem(mem + off, pmem_addr, len);
103 flush_dcache_page(page);
104 }
105 } else {
106 /*
107 * Note that we write the data both before and after
108 * clearing poison. The write before clear poison
109 * handles situations where the latest written data is
110 * preserved and the clear poison operation simply marks
111 * the address range as valid without changing the data.
112 * In this case application software can assume that an
113 * interrupted write will either return the new good
114 * data or an error.
115 *
116 * However, if pmem_clear_poison() leaves the data in an
117 * indeterminate state we need to perform the write
118 * after clear poison.
119 */
120 flush_dcache_page(page);
121 memcpy_to_pmem(pmem_addr, mem + off, len);
122 if (unlikely(bad_pmem)) {
123 pmem_clear_poison(pmem, pmem_off, len);
124 memcpy_to_pmem(pmem_addr, mem + off, len);
125 }
126 }
127
128 kunmap_atomic(mem);
129 return rc;
130}
131
132static blk_qc_t pmem_make_request(struct request_queue *q, struct bio *bio)
133{
134 int rc = 0;
135 bool do_acct;
136 unsigned long start;
137 struct bio_vec bvec;
138 struct bvec_iter iter;
139 struct block_device *bdev = bio->bi_bdev;
140 struct pmem_device *pmem = bdev->bd_disk->private_data;
141
142 do_acct = nd_iostat_start(bio, &start);
143 bio_for_each_segment(bvec, bio, iter) {
144 rc = pmem_do_bvec(pmem, bvec.bv_page, bvec.bv_len,
145 bvec.bv_offset, bio_data_dir(bio),
146 iter.bi_sector);
147 if (rc) {
148 bio->bi_error = rc;
149 break;
150 }
151 }
152 if (do_acct)
153 nd_iostat_end(bio, start);
154
155 if (bio_data_dir(bio))
156 wmb_pmem();
157
158 bio_endio(bio);
159 return BLK_QC_T_NONE;
160}
161
162static int pmem_rw_page(struct block_device *bdev, sector_t sector,
163 struct page *page, int rw)
164{
165 struct pmem_device *pmem = bdev->bd_disk->private_data;
166 int rc;
167
168 rc = pmem_do_bvec(pmem, page, PAGE_SIZE, 0, rw, sector);
169 if (rw & WRITE)
170 wmb_pmem();
171
172 /*
173 * The ->rw_page interface is subtle and tricky. The core
174 * retries on any error, so we can only invoke page_endio() in
175 * the successful completion case. Otherwise, we'll see crashes
176 * caused by double completion.
177 */
178 if (rc == 0)
179 page_endio(page, rw & WRITE, 0);
180
181 return rc;
182}
183
184static long pmem_direct_access(struct block_device *bdev, sector_t sector,
185 void __pmem **kaddr, pfn_t *pfn)
186{
187 struct pmem_device *pmem = bdev->bd_disk->private_data;
188 resource_size_t offset = sector * 512 + pmem->data_offset;
189
190 *kaddr = pmem->virt_addr + offset;
191 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
192
193 return pmem->size - pmem->pfn_pad - offset;
194}
195
196static const struct block_device_operations pmem_fops = {
197 .owner = THIS_MODULE,
198 .rw_page = pmem_rw_page,
199 .direct_access = pmem_direct_access,
200 .revalidate_disk = nvdimm_revalidate_disk,
201};
202
203static struct pmem_device *pmem_alloc(struct device *dev,
204 struct resource *res, int id)
205{
206 struct pmem_device *pmem;
207 struct request_queue *q;
208
209 pmem = devm_kzalloc(dev, sizeof(*pmem), GFP_KERNEL);
210 if (!pmem)
211 return ERR_PTR(-ENOMEM);
212
213 pmem->phys_addr = res->start;
214 pmem->size = resource_size(res);
215 if (!arch_has_wmb_pmem())
216 dev_warn(dev, "unable to guarantee persistence of writes\n");
217
218 if (!devm_request_mem_region(dev, pmem->phys_addr, pmem->size,
219 dev_name(dev))) {
220 dev_warn(dev, "could not reserve region [0x%pa:0x%zx]\n",
221 &pmem->phys_addr, pmem->size);
222 return ERR_PTR(-EBUSY);
223 }
224
225 q = blk_alloc_queue_node(GFP_KERNEL, dev_to_node(dev));
226 if (!q)
227 return ERR_PTR(-ENOMEM);
228
229 pmem->pfn_flags = PFN_DEV;
230 if (pmem_should_map_pages(dev)) {
231 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, res,
232 &q->q_usage_counter, NULL);
233 pmem->pfn_flags |= PFN_MAP;
234 } else
235 pmem->virt_addr = (void __pmem *) devm_memremap(dev,
236 pmem->phys_addr, pmem->size,
237 ARCH_MEMREMAP_PMEM);
238
239 if (IS_ERR(pmem->virt_addr)) {
240 blk_cleanup_queue(q);
241 return (void __force *) pmem->virt_addr;
242 }
243
244 pmem->pmem_queue = q;
245 return pmem;
246}
247
248static void pmem_detach_disk(struct pmem_device *pmem)
249{
250 if (!pmem->pmem_disk)
251 return;
252
253 del_gendisk(pmem->pmem_disk);
254 put_disk(pmem->pmem_disk);
255 blk_cleanup_queue(pmem->pmem_queue);
256}
257
258static int pmem_attach_disk(struct device *dev,
259 struct nd_namespace_common *ndns, struct pmem_device *pmem)
260{
261 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
262 int nid = dev_to_node(dev);
263 struct resource bb_res;
264 struct gendisk *disk;
265
266 blk_queue_make_request(pmem->pmem_queue, pmem_make_request);
267 blk_queue_physical_block_size(pmem->pmem_queue, PAGE_SIZE);
268 blk_queue_max_hw_sectors(pmem->pmem_queue, UINT_MAX);
269 blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
270 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue);
271
272 disk = alloc_disk_node(0, nid);
273 if (!disk) {
274 blk_cleanup_queue(pmem->pmem_queue);
275 return -ENOMEM;
276 }
277
278 disk->fops = &pmem_fops;
279 disk->private_data = pmem;
280 disk->queue = pmem->pmem_queue;
281 disk->flags = GENHD_FL_EXT_DEVT;
282 nvdimm_namespace_disk_name(ndns, disk->disk_name);
283 disk->driverfs_dev = dev;
284 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
285 / 512);
286 pmem->pmem_disk = disk;
287 devm_exit_badblocks(dev, &pmem->bb);
288 if (devm_init_badblocks(dev, &pmem->bb))
289 return -ENOMEM;
290 bb_res.start = nsio->res.start + pmem->data_offset;
291 bb_res.end = nsio->res.end;
292 if (is_nd_pfn(dev)) {
293 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
294 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
295
296 bb_res.start += __le32_to_cpu(pfn_sb->start_pad);
297 bb_res.end -= __le32_to_cpu(pfn_sb->end_trunc);
298 }
299 nvdimm_badblocks_populate(to_nd_region(dev->parent), &pmem->bb,
300 &bb_res);
301 disk->bb = &pmem->bb;
302 add_disk(disk);
303 revalidate_disk(disk);
304
305 return 0;
306}
307
308static int pmem_rw_bytes(struct nd_namespace_common *ndns,
309 resource_size_t offset, void *buf, size_t size, int rw)
310{
311 struct pmem_device *pmem = dev_get_drvdata(ndns->claim);
312
313 if (unlikely(offset + size > pmem->size)) {
314 dev_WARN_ONCE(&ndns->dev, 1, "request out of range\n");
315 return -EFAULT;
316 }
317
318 if (rw == READ) {
319 unsigned int sz_align = ALIGN(size + (offset & (512 - 1)), 512);
320
321 if (unlikely(is_bad_pmem(&pmem->bb, offset / 512, sz_align)))
322 return -EIO;
323 return memcpy_from_pmem(buf, pmem->virt_addr + offset, size);
324 } else {
325 memcpy_to_pmem(pmem->virt_addr + offset, buf, size);
326 wmb_pmem();
327 }
328
329 return 0;
330}
331
332static int nd_pfn_init(struct nd_pfn *nd_pfn)
333{
334 struct nd_pfn_sb *pfn_sb = kzalloc(sizeof(*pfn_sb), GFP_KERNEL);
335 struct pmem_device *pmem = dev_get_drvdata(&nd_pfn->dev);
336 struct nd_namespace_common *ndns = nd_pfn->ndns;
337 u32 start_pad = 0, end_trunc = 0;
338 resource_size_t start, size;
339 struct nd_namespace_io *nsio;
340 struct nd_region *nd_region;
341 unsigned long npfns;
342 phys_addr_t offset;
343 u64 checksum;
344 int rc;
345
346 if (!pfn_sb)
347 return -ENOMEM;
348
349 nd_pfn->pfn_sb = pfn_sb;
350 rc = nd_pfn_validate(nd_pfn);
351 if (rc == -ENODEV)
352 /* no info block, do init */;
353 else
354 return rc;
355
356 nd_region = to_nd_region(nd_pfn->dev.parent);
357 if (nd_region->ro) {
358 dev_info(&nd_pfn->dev,
359 "%s is read-only, unable to init metadata\n",
360 dev_name(&nd_region->dev));
361 goto err;
362 }
363
364 memset(pfn_sb, 0, sizeof(*pfn_sb));
365
366 /*
367 * Check if pmem collides with 'System RAM' when section aligned and
368 * trim it accordingly
369 */
370 nsio = to_nd_namespace_io(&ndns->dev);
371 start = PHYS_SECTION_ALIGN_DOWN(nsio->res.start);
372 size = resource_size(&nsio->res);
373 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
374 IORES_DESC_NONE) == REGION_MIXED) {
375
376 start = nsio->res.start;
377 start_pad = PHYS_SECTION_ALIGN_UP(start) - start;
378 }
379
380 start = nsio->res.start;
381 size = PHYS_SECTION_ALIGN_UP(start + size) - start;
382 if (region_intersects(start, size, IORESOURCE_SYSTEM_RAM,
383 IORES_DESC_NONE) == REGION_MIXED) {
384 size = resource_size(&nsio->res);
385 end_trunc = start + size - PHYS_SECTION_ALIGN_DOWN(start + size);
386 }
387
388 if (start_pad + end_trunc)
389 dev_info(&nd_pfn->dev, "%s section collision, truncate %d bytes\n",
390 dev_name(&ndns->dev), start_pad + end_trunc);
391
392 /*
393 * Note, we use 64 here for the standard size of struct page,
394 * debugging options may cause it to be larger in which case the
395 * implementation will limit the pfns advertised through
396 * ->direct_access() to those that are included in the memmap.
397 */
398 start += start_pad;
399 npfns = (pmem->size - start_pad - end_trunc - SZ_8K) / SZ_4K;
400 if (nd_pfn->mode == PFN_MODE_PMEM) {
401 unsigned long memmap_size;
402
403 /*
404 * vmemmap_populate_hugepages() allocates the memmap array in
405 * PMD_SIZE chunks.
406 */
407 memmap_size = ALIGN(64 * npfns, PMD_SIZE);
408 offset = ALIGN(start + SZ_8K + memmap_size, nd_pfn->align)
409 - start;
410 } else if (nd_pfn->mode == PFN_MODE_RAM)
411 offset = ALIGN(start + SZ_8K, nd_pfn->align) - start;
412 else
413 goto err;
414
415 if (offset + start_pad + end_trunc >= pmem->size) {
416 dev_err(&nd_pfn->dev, "%s unable to satisfy requested alignment\n",
417 dev_name(&ndns->dev));
418 goto err;
419 }
420
421 npfns = (pmem->size - offset - start_pad - end_trunc) / SZ_4K;
422 pfn_sb->mode = cpu_to_le32(nd_pfn->mode);
423 pfn_sb->dataoff = cpu_to_le64(offset);
424 pfn_sb->npfns = cpu_to_le64(npfns);
425 memcpy(pfn_sb->signature, PFN_SIG, PFN_SIG_LEN);
426 memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
427 memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
428 pfn_sb->version_major = cpu_to_le16(1);
429 pfn_sb->version_minor = cpu_to_le16(1);
430 pfn_sb->start_pad = cpu_to_le32(start_pad);
431 pfn_sb->end_trunc = cpu_to_le32(end_trunc);
432 checksum = nd_sb_checksum((struct nd_gen_sb *) pfn_sb);
433 pfn_sb->checksum = cpu_to_le64(checksum);
434
435 rc = nvdimm_write_bytes(ndns, SZ_4K, pfn_sb, sizeof(*pfn_sb));
436 if (rc)
437 goto err;
438
439 return 0;
440 err:
441 nd_pfn->pfn_sb = NULL;
442 kfree(pfn_sb);
443 return -ENXIO;
444}
445
446static int nvdimm_namespace_detach_pfn(struct nd_namespace_common *ndns)
447{
448 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
449 struct pmem_device *pmem;
450
451 /* free pmem disk */
452 pmem = dev_get_drvdata(&nd_pfn->dev);
453 pmem_detach_disk(pmem);
454
455 /* release nd_pfn resources */
456 kfree(nd_pfn->pfn_sb);
457 nd_pfn->pfn_sb = NULL;
458
459 return 0;
460}
461
462/*
463 * We hotplug memory at section granularity, pad the reserved area from
464 * the previous section base to the namespace base address.
465 */
466static unsigned long init_altmap_base(resource_size_t base)
467{
468 unsigned long base_pfn = PHYS_PFN(base);
469
470 return PFN_SECTION_ALIGN_DOWN(base_pfn);
471}
472
473static unsigned long init_altmap_reserve(resource_size_t base)
474{
475 unsigned long reserve = PHYS_PFN(SZ_8K);
476 unsigned long base_pfn = PHYS_PFN(base);
477
478 reserve += base_pfn - PFN_SECTION_ALIGN_DOWN(base_pfn);
479 return reserve;
480}
481
482static int __nvdimm_namespace_attach_pfn(struct nd_pfn *nd_pfn)
483{
484 int rc;
485 struct resource res;
486 struct request_queue *q;
487 struct pmem_device *pmem;
488 struct vmem_altmap *altmap;
489 struct device *dev = &nd_pfn->dev;
490 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
491 struct nd_namespace_common *ndns = nd_pfn->ndns;
492 u32 start_pad = __le32_to_cpu(pfn_sb->start_pad);
493 u32 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
494 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
495 resource_size_t base = nsio->res.start + start_pad;
496 struct vmem_altmap __altmap = {
497 .base_pfn = init_altmap_base(base),
498 .reserve = init_altmap_reserve(base),
499 };
500
501 pmem = dev_get_drvdata(dev);
502 pmem->data_offset = le64_to_cpu(pfn_sb->dataoff);
503 pmem->pfn_pad = start_pad + end_trunc;
504 nd_pfn->mode = le32_to_cpu(nd_pfn->pfn_sb->mode);
505 if (nd_pfn->mode == PFN_MODE_RAM) {
506 if (pmem->data_offset < SZ_8K)
507 return -EINVAL;
508 nd_pfn->npfns = le64_to_cpu(pfn_sb->npfns);
509 altmap = NULL;
510 } else if (nd_pfn->mode == PFN_MODE_PMEM) {
511 nd_pfn->npfns = (pmem->size - pmem->pfn_pad - pmem->data_offset)
512 / PAGE_SIZE;
513 if (le64_to_cpu(nd_pfn->pfn_sb->npfns) > nd_pfn->npfns)
514 dev_info(&nd_pfn->dev,
515 "number of pfns truncated from %lld to %ld\n",
516 le64_to_cpu(nd_pfn->pfn_sb->npfns),
517 nd_pfn->npfns);
518 altmap = & __altmap;
519 altmap->free = PHYS_PFN(pmem->data_offset - SZ_8K);
520 altmap->alloc = 0;
521 } else {
522 rc = -ENXIO;
523 goto err;
524 }
525
526 /* establish pfn range for lookup, and switch to direct map */
527 q = pmem->pmem_queue;
528 memcpy(&res, &nsio->res, sizeof(res));
529 res.start += start_pad;
530 res.end -= end_trunc;
531 devm_memunmap(dev, (void __force *) pmem->virt_addr);
532 pmem->virt_addr = (void __pmem *) devm_memremap_pages(dev, &res,
533 &q->q_usage_counter, altmap);
534 pmem->pfn_flags |= PFN_MAP;
535 if (IS_ERR(pmem->virt_addr)) {
536 rc = PTR_ERR(pmem->virt_addr);
537 goto err;
538 }
539
540 /* attach pmem disk in "pfn-mode" */
541 rc = pmem_attach_disk(dev, ndns, pmem);
542 if (rc)
543 goto err;
544
545 return rc;
546 err:
547 nvdimm_namespace_detach_pfn(ndns);
548 return rc;
549
550}
551
552static int nvdimm_namespace_attach_pfn(struct nd_namespace_common *ndns)
553{
554 struct nd_pfn *nd_pfn = to_nd_pfn(ndns->claim);
555 int rc;
556
557 if (!nd_pfn->uuid || !nd_pfn->ndns)
558 return -ENODEV;
559
560 rc = nd_pfn_init(nd_pfn);
561 if (rc)
562 return rc;
563 /* we need a valid pfn_sb before we can init a vmem_altmap */
564 return __nvdimm_namespace_attach_pfn(nd_pfn);
565}
566
567static int nd_pmem_probe(struct device *dev)
568{
569 struct nd_region *nd_region = to_nd_region(dev->parent);
570 struct nd_namespace_common *ndns;
571 struct nd_namespace_io *nsio;
572 struct pmem_device *pmem;
573
574 ndns = nvdimm_namespace_common_probe(dev);
575 if (IS_ERR(ndns))
576 return PTR_ERR(ndns);
577
578 nsio = to_nd_namespace_io(&ndns->dev);
579 pmem = pmem_alloc(dev, &nsio->res, nd_region->id);
580 if (IS_ERR(pmem))
581 return PTR_ERR(pmem);
582
583 pmem->ndns = ndns;
584 dev_set_drvdata(dev, pmem);
585 ndns->rw_bytes = pmem_rw_bytes;
586 if (devm_init_badblocks(dev, &pmem->bb))
587 return -ENOMEM;
588 nvdimm_badblocks_populate(nd_region, &pmem->bb, &nsio->res);
589
590 if (is_nd_btt(dev)) {
591 /* btt allocates its own request_queue */
592 blk_cleanup_queue(pmem->pmem_queue);
593 pmem->pmem_queue = NULL;
594 return nvdimm_namespace_attach_btt(ndns);
595 }
596
597 if (is_nd_pfn(dev))
598 return nvdimm_namespace_attach_pfn(ndns);
599
600 if (nd_btt_probe(ndns, pmem) == 0 || nd_pfn_probe(ndns, pmem) == 0) {
601 /*
602 * We'll come back as either btt-pmem, or pfn-pmem, so
603 * drop the queue allocation for now.
604 */
605 blk_cleanup_queue(pmem->pmem_queue);
606 return -ENXIO;
607 }
608
609 return pmem_attach_disk(dev, ndns, pmem);
610}
611
612static int nd_pmem_remove(struct device *dev)
613{
614 struct pmem_device *pmem = dev_get_drvdata(dev);
615
616 if (is_nd_btt(dev))
617 nvdimm_namespace_detach_btt(pmem->ndns);
618 else if (is_nd_pfn(dev))
619 nvdimm_namespace_detach_pfn(pmem->ndns);
620 else
621 pmem_detach_disk(pmem);
622
623 return 0;
624}
625
626static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
627{
628 struct pmem_device *pmem = dev_get_drvdata(dev);
629 struct nd_namespace_common *ndns = pmem->ndns;
630 struct nd_region *nd_region = to_nd_region(dev->parent);
631 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
632 struct resource res = {
633 .start = nsio->res.start + pmem->data_offset,
634 .end = nsio->res.end,
635 };
636
637 if (event != NVDIMM_REVALIDATE_POISON)
638 return;
639
640 if (is_nd_pfn(dev)) {
641 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
642 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
643
644 res.start += __le32_to_cpu(pfn_sb->start_pad);
645 res.end -= __le32_to_cpu(pfn_sb->end_trunc);
646 }
647
648 nvdimm_badblocks_populate(nd_region, &pmem->bb, &res);
649}
650
651MODULE_ALIAS("pmem");
652MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
653MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
654static struct nd_device_driver nd_pmem_driver = {
655 .probe = nd_pmem_probe,
656 .remove = nd_pmem_remove,
657 .notify = nd_pmem_notify,
658 .drv = {
659 .name = "nd_pmem",
660 },
661 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
662};
663
664static int __init pmem_init(void)
665{
666 return nd_driver_register(&nd_pmem_driver);
667}
668module_init(pmem_init);
669
670static void pmem_exit(void)
671{
672 driver_unregister(&nd_pmem_driver.drv);
673}
674module_exit(pmem_exit);
675
676MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
677MODULE_LICENSE("GPL v2");