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/vmalloc.h>
21 #include <linux/blk-mq.h>
22 #include <linux/pfn_t.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
to_dev(struct pmem_device * pmem)34 static 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
to_region(struct pmem_device * pmem)43 static struct nd_region *to_region(struct pmem_device *pmem)
44 {
45 return to_nd_region(to_dev(pmem)->parent);
46 }
47
hwpoison_clear(struct pmem_device * pmem,phys_addr_t phys,unsigned int len)48 static void hwpoison_clear(struct pmem_device *pmem,
49 phys_addr_t phys, unsigned int len)
50 {
51 unsigned long pfn_start, pfn_end, pfn;
52
53 /* only pmem in the linear map supports HWPoison */
54 if (is_vmalloc_addr(pmem->virt_addr))
55 return;
56
57 pfn_start = PHYS_PFN(phys);
58 pfn_end = pfn_start + PHYS_PFN(len);
59 for (pfn = pfn_start; pfn < pfn_end; pfn++) {
60 struct page *page = pfn_to_page(pfn);
61
62 /*
63 * Note, no need to hold a get_dev_pagemap() reference
64 * here since we're in the driver I/O path and
65 * outstanding I/O requests pin the dev_pagemap.
66 */
67 if (test_and_clear_pmem_poison(page))
68 clear_mce_nospec(pfn);
69 }
70 }
71
pmem_clear_poison(struct pmem_device * pmem,phys_addr_t offset,unsigned int len)72 static blk_status_t pmem_clear_poison(struct pmem_device *pmem,
73 phys_addr_t offset, unsigned int len)
74 {
75 struct device *dev = to_dev(pmem);
76 sector_t sector;
77 long cleared;
78 blk_status_t rc = BLK_STS_OK;
79
80 sector = (offset - pmem->data_offset) / 512;
81
82 cleared = nvdimm_clear_poison(dev, pmem->phys_addr + offset, len);
83 if (cleared < len)
84 rc = BLK_STS_IOERR;
85 if (cleared > 0 && cleared / 512) {
86 hwpoison_clear(pmem, pmem->phys_addr + offset, cleared);
87 cleared /= 512;
88 dev_dbg(dev, "%#llx clear %ld sector%s\n",
89 (unsigned long long) sector, cleared,
90 cleared > 1 ? "s" : "");
91 badblocks_clear(&pmem->bb, sector, cleared);
92 if (pmem->bb_state)
93 sysfs_notify_dirent(pmem->bb_state);
94 }
95
96 arch_invalidate_pmem(pmem->virt_addr + offset, len);
97
98 return rc;
99 }
100
write_pmem(void * pmem_addr,struct page * page,unsigned int off,unsigned int len)101 static void write_pmem(void *pmem_addr, struct page *page,
102 unsigned int off, unsigned int len)
103 {
104 unsigned int chunk;
105 void *mem;
106
107 while (len) {
108 mem = kmap_atomic(page);
109 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
110 memcpy_flushcache(pmem_addr, mem + off, chunk);
111 kunmap_atomic(mem);
112 len -= chunk;
113 off = 0;
114 page++;
115 pmem_addr += chunk;
116 }
117 }
118
read_pmem(struct page * page,unsigned int off,void * pmem_addr,unsigned int len)119 static blk_status_t read_pmem(struct page *page, unsigned int off,
120 void *pmem_addr, unsigned int len)
121 {
122 unsigned int chunk;
123 unsigned long rem;
124 void *mem;
125
126 while (len) {
127 mem = kmap_atomic(page);
128 chunk = min_t(unsigned int, len, PAGE_SIZE - off);
129 rem = copy_mc_to_kernel(mem + off, pmem_addr, chunk);
130 kunmap_atomic(mem);
131 if (rem)
132 return BLK_STS_IOERR;
133 len -= chunk;
134 off = 0;
135 page++;
136 pmem_addr += chunk;
137 }
138 return BLK_STS_OK;
139 }
140
pmem_do_read(struct pmem_device * pmem,struct page * page,unsigned int page_off,sector_t sector,unsigned int len)141 static blk_status_t pmem_do_read(struct pmem_device *pmem,
142 struct page *page, unsigned int page_off,
143 sector_t sector, unsigned int len)
144 {
145 blk_status_t rc;
146 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
147 void *pmem_addr = pmem->virt_addr + pmem_off;
148
149 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
150 return BLK_STS_IOERR;
151
152 rc = read_pmem(page, page_off, pmem_addr, len);
153 flush_dcache_page(page);
154 return rc;
155 }
156
pmem_do_write(struct pmem_device * pmem,struct page * page,unsigned int page_off,sector_t sector,unsigned int len)157 static blk_status_t pmem_do_write(struct pmem_device *pmem,
158 struct page *page, unsigned int page_off,
159 sector_t sector, unsigned int len)
160 {
161 blk_status_t rc = BLK_STS_OK;
162 bool bad_pmem = false;
163 phys_addr_t pmem_off = sector * 512 + pmem->data_offset;
164 void *pmem_addr = pmem->virt_addr + pmem_off;
165
166 if (unlikely(is_bad_pmem(&pmem->bb, sector, len)))
167 bad_pmem = true;
168
169 /*
170 * Note that we write the data both before and after
171 * clearing poison. The write before clear poison
172 * handles situations where the latest written data is
173 * preserved and the clear poison operation simply marks
174 * the address range as valid without changing the data.
175 * In this case application software can assume that an
176 * interrupted write will either return the new good
177 * data or an error.
178 *
179 * However, if pmem_clear_poison() leaves the data in an
180 * indeterminate state we need to perform the write
181 * after clear poison.
182 */
183 flush_dcache_page(page);
184 write_pmem(pmem_addr, page, page_off, len);
185 if (unlikely(bad_pmem)) {
186 rc = pmem_clear_poison(pmem, pmem_off, len);
187 write_pmem(pmem_addr, page, page_off, len);
188 }
189
190 return rc;
191 }
192
pmem_submit_bio(struct bio * bio)193 static blk_qc_t pmem_submit_bio(struct bio *bio)
194 {
195 int ret = 0;
196 blk_status_t rc = 0;
197 bool do_acct;
198 unsigned long start;
199 struct bio_vec bvec;
200 struct bvec_iter iter;
201 struct pmem_device *pmem = bio->bi_bdev->bd_disk->private_data;
202 struct nd_region *nd_region = to_region(pmem);
203
204 if (bio->bi_opf & REQ_PREFLUSH)
205 ret = nvdimm_flush(nd_region, bio);
206
207 do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
208 if (do_acct)
209 start = bio_start_io_acct(bio);
210 bio_for_each_segment(bvec, bio, iter) {
211 if (op_is_write(bio_op(bio)))
212 rc = pmem_do_write(pmem, bvec.bv_page, bvec.bv_offset,
213 iter.bi_sector, bvec.bv_len);
214 else
215 rc = pmem_do_read(pmem, bvec.bv_page, bvec.bv_offset,
216 iter.bi_sector, bvec.bv_len);
217 if (rc) {
218 bio->bi_status = rc;
219 break;
220 }
221 }
222 if (do_acct)
223 bio_end_io_acct(bio, start);
224
225 if (bio->bi_opf & REQ_FUA)
226 ret = nvdimm_flush(nd_region, bio);
227
228 if (ret)
229 bio->bi_status = errno_to_blk_status(ret);
230
231 bio_endio(bio);
232 return BLK_QC_T_NONE;
233 }
234
pmem_rw_page(struct block_device * bdev,sector_t sector,struct page * page,unsigned int op)235 static int pmem_rw_page(struct block_device *bdev, sector_t sector,
236 struct page *page, unsigned int op)
237 {
238 struct pmem_device *pmem = bdev->bd_disk->private_data;
239 blk_status_t rc;
240
241 if (op_is_write(op))
242 rc = pmem_do_write(pmem, page, 0, sector, thp_size(page));
243 else
244 rc = pmem_do_read(pmem, page, 0, sector, thp_size(page));
245 /*
246 * The ->rw_page interface is subtle and tricky. The core
247 * retries on any error, so we can only invoke page_endio() in
248 * the successful completion case. Otherwise, we'll see crashes
249 * caused by double completion.
250 */
251 if (rc == 0)
252 page_endio(page, op_is_write(op), 0);
253
254 return blk_status_to_errno(rc);
255 }
256
257 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
__pmem_direct_access(struct pmem_device * pmem,pgoff_t pgoff,long nr_pages,void ** kaddr,pfn_t * pfn)258 __weak long __pmem_direct_access(struct pmem_device *pmem, pgoff_t pgoff,
259 long nr_pages, void **kaddr, pfn_t *pfn)
260 {
261 resource_size_t offset = PFN_PHYS(pgoff) + pmem->data_offset;
262
263 if (unlikely(is_bad_pmem(&pmem->bb, PFN_PHYS(pgoff) / 512,
264 PFN_PHYS(nr_pages))))
265 return -EIO;
266
267 if (kaddr)
268 *kaddr = pmem->virt_addr + offset;
269 if (pfn)
270 *pfn = phys_to_pfn_t(pmem->phys_addr + offset, pmem->pfn_flags);
271
272 /*
273 * If badblocks are present, limit known good range to the
274 * requested range.
275 */
276 if (unlikely(pmem->bb.count))
277 return nr_pages;
278 return PHYS_PFN(pmem->size - pmem->pfn_pad - offset);
279 }
280
281 static const struct block_device_operations pmem_fops = {
282 .owner = THIS_MODULE,
283 .submit_bio = pmem_submit_bio,
284 .rw_page = pmem_rw_page,
285 };
286
pmem_dax_zero_page_range(struct dax_device * dax_dev,pgoff_t pgoff,size_t nr_pages)287 static 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
pmem_dax_direct_access(struct dax_device * dax_dev,pgoff_t pgoff,long nr_pages,void ** kaddr,pfn_t * pfn)297 static 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_mc_to_iter() to bypass HARDENED_USERCOPY overhead. Bounds
308 * checking, both file offset and device offset, is handled by
309 * dax_iomap_actor()
310 */
pmem_copy_from_iter(struct dax_device * dax_dev,pgoff_t pgoff,void * addr,size_t bytes,struct iov_iter * i)311 static 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
pmem_copy_to_iter(struct dax_device * dax_dev,pgoff_t pgoff,void * addr,size_t bytes,struct iov_iter * i)317 static 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_mc_to_iter(addr, bytes, i);
321 }
322
323 static 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
331 static const struct attribute_group *pmem_attribute_groups[] = {
332 &dax_attribute_group,
333 NULL,
334 };
335
pmem_pagemap_cleanup(struct dev_pagemap * pgmap)336 static 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
pmem_release_queue(void * pgmap)344 static void pmem_release_queue(void *pgmap)
345 {
346 pmem_pagemap_cleanup(pgmap);
347 }
348
pmem_pagemap_kill(struct dev_pagemap * pgmap)349 static 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
pmem_release_disk(void * __pmem)357 static 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
367 static const struct dev_pagemap_ops fsdax_pagemap_ops = {
368 .kill = pmem_pagemap_kill,
369 .cleanup = pmem_pagemap_cleanup,
370 };
371
pmem_attach_disk(struct device * dev,struct nd_namespace_common * ndns)372 static 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 range bb_range;
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 range_len(&pmem->pgmap.range);
439 pmem->pfn_flags |= PFN_MAP;
440 bb_range = pmem->pgmap.range;
441 bb_range.start += pmem->data_offset;
442 } else if (pmem_should_map_pages(dev)) {
443 pmem->pgmap.range.start = res->start;
444 pmem->pgmap.range.end = res->end;
445 pmem->pgmap.nr_range = 1;
446 pmem->pgmap.type = MEMORY_DEVICE_FS_DAX;
447 pmem->pgmap.ops = &fsdax_pagemap_ops;
448 addr = devm_memremap_pages(dev, &pmem->pgmap);
449 pmem->pfn_flags |= PFN_MAP;
450 bb_range = pmem->pgmap.range;
451 } else {
452 if (devm_add_action_or_reset(dev, pmem_release_queue,
453 &pmem->pgmap))
454 return -ENOMEM;
455 addr = devm_memremap(dev, pmem->phys_addr,
456 pmem->size, ARCH_MEMREMAP_PMEM);
457 bb_range.start = res->start;
458 bb_range.end = res->end;
459 }
460
461 if (IS_ERR(addr))
462 return PTR_ERR(addr);
463 pmem->virt_addr = addr;
464
465 blk_queue_write_cache(q, true, fua);
466 blk_queue_physical_block_size(q, PAGE_SIZE);
467 blk_queue_logical_block_size(q, pmem_sector_size(ndns));
468 blk_queue_max_hw_sectors(q, UINT_MAX);
469 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
470 if (pmem->pfn_flags & PFN_MAP)
471 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
472
473 disk = alloc_disk_node(0, nid);
474 if (!disk)
475 return -ENOMEM;
476 pmem->disk = disk;
477
478 disk->fops = &pmem_fops;
479 disk->queue = q;
480 disk->flags = GENHD_FL_EXT_DEVT;
481 disk->private_data = pmem;
482 nvdimm_namespace_disk_name(ndns, disk->disk_name);
483 set_capacity(disk, (pmem->size - pmem->pfn_pad - pmem->data_offset)
484 / 512);
485 if (devm_init_badblocks(dev, &pmem->bb))
486 return -ENOMEM;
487 nvdimm_badblocks_populate(nd_region, &pmem->bb, &bb_range);
488 disk->bb = &pmem->bb;
489
490 if (is_nvdimm_sync(nd_region))
491 flags = DAXDEV_F_SYNC;
492 dax_dev = alloc_dax(pmem, disk->disk_name, &pmem_dax_ops, flags);
493 if (IS_ERR(dax_dev)) {
494 put_disk(disk);
495 return PTR_ERR(dax_dev);
496 }
497 dax_write_cache(dax_dev, nvdimm_has_cache(nd_region));
498 pmem->dax_dev = dax_dev;
499 gendev = disk_to_dev(disk);
500 gendev->groups = pmem_attribute_groups;
501
502 device_add_disk(dev, disk, NULL);
503 if (devm_add_action_or_reset(dev, pmem_release_disk, pmem))
504 return -ENOMEM;
505
506 nvdimm_check_and_set_ro(disk);
507
508 pmem->bb_state = sysfs_get_dirent(disk_to_dev(disk)->kobj.sd,
509 "badblocks");
510 if (!pmem->bb_state)
511 dev_warn(dev, "'badblocks' notification disabled\n");
512
513 return 0;
514 }
515
nd_pmem_probe(struct device * dev)516 static int nd_pmem_probe(struct device *dev)
517 {
518 int ret;
519 struct nd_namespace_common *ndns;
520
521 ndns = nvdimm_namespace_common_probe(dev);
522 if (IS_ERR(ndns))
523 return PTR_ERR(ndns);
524
525 if (is_nd_btt(dev))
526 return nvdimm_namespace_attach_btt(ndns);
527
528 if (is_nd_pfn(dev))
529 return pmem_attach_disk(dev, ndns);
530
531 ret = devm_namespace_enable(dev, ndns, nd_info_block_reserve());
532 if (ret)
533 return ret;
534
535 ret = nd_btt_probe(dev, ndns);
536 if (ret == 0)
537 return -ENXIO;
538
539 /*
540 * We have two failure conditions here, there is no
541 * info reserver block or we found a valid info reserve block
542 * but failed to initialize the pfn superblock.
543 *
544 * For the first case consider namespace as a raw pmem namespace
545 * and attach a disk.
546 *
547 * For the latter, consider this a success and advance the namespace
548 * seed.
549 */
550 ret = nd_pfn_probe(dev, ndns);
551 if (ret == 0)
552 return -ENXIO;
553 else if (ret == -EOPNOTSUPP)
554 return ret;
555
556 ret = nd_dax_probe(dev, ndns);
557 if (ret == 0)
558 return -ENXIO;
559 else if (ret == -EOPNOTSUPP)
560 return ret;
561
562 /* probe complete, attach handles namespace enabling */
563 devm_namespace_disable(dev, ndns);
564
565 return pmem_attach_disk(dev, ndns);
566 }
567
nd_pmem_remove(struct device * dev)568 static void nd_pmem_remove(struct device *dev)
569 {
570 struct pmem_device *pmem = dev_get_drvdata(dev);
571
572 if (is_nd_btt(dev))
573 nvdimm_namespace_detach_btt(to_nd_btt(dev));
574 else {
575 /*
576 * Note, this assumes nd_device_lock() context to not
577 * race nd_pmem_notify()
578 */
579 sysfs_put(pmem->bb_state);
580 pmem->bb_state = NULL;
581 }
582 nvdimm_flush(to_nd_region(dev->parent), NULL);
583 }
584
nd_pmem_shutdown(struct device * dev)585 static void nd_pmem_shutdown(struct device *dev)
586 {
587 nvdimm_flush(to_nd_region(dev->parent), NULL);
588 }
589
pmem_revalidate_poison(struct device * dev)590 static void pmem_revalidate_poison(struct device *dev)
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 badblocks *bb;
597 struct range range;
598 struct kernfs_node *bb_state;
599
600 if (is_nd_btt(dev)) {
601 struct nd_btt *nd_btt = to_nd_btt(dev);
602
603 ndns = nd_btt->ndns;
604 nd_region = to_nd_region(ndns->dev.parent);
605 nsio = to_nd_namespace_io(&ndns->dev);
606 bb = &nsio->bb;
607 bb_state = NULL;
608 } else {
609 struct pmem_device *pmem = dev_get_drvdata(dev);
610
611 nd_region = to_region(pmem);
612 bb = &pmem->bb;
613 bb_state = pmem->bb_state;
614
615 if (is_nd_pfn(dev)) {
616 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
617 struct nd_pfn_sb *pfn_sb = nd_pfn->pfn_sb;
618
619 ndns = nd_pfn->ndns;
620 offset = pmem->data_offset +
621 __le32_to_cpu(pfn_sb->start_pad);
622 end_trunc = __le32_to_cpu(pfn_sb->end_trunc);
623 } else {
624 ndns = to_ndns(dev);
625 }
626
627 nsio = to_nd_namespace_io(&ndns->dev);
628 }
629
630 range.start = nsio->res.start + offset;
631 range.end = nsio->res.end - end_trunc;
632 nvdimm_badblocks_populate(nd_region, bb, &range);
633 if (bb_state)
634 sysfs_notify_dirent(bb_state);
635 }
636
pmem_revalidate_region(struct device * dev)637 static void pmem_revalidate_region(struct device *dev)
638 {
639 struct pmem_device *pmem;
640
641 if (is_nd_btt(dev)) {
642 struct nd_btt *nd_btt = to_nd_btt(dev);
643 struct btt *btt = nd_btt->btt;
644
645 nvdimm_check_and_set_ro(btt->btt_disk);
646 return;
647 }
648
649 pmem = dev_get_drvdata(dev);
650 nvdimm_check_and_set_ro(pmem->disk);
651 }
652
nd_pmem_notify(struct device * dev,enum nvdimm_event event)653 static void nd_pmem_notify(struct device *dev, enum nvdimm_event event)
654 {
655 switch (event) {
656 case NVDIMM_REVALIDATE_POISON:
657 pmem_revalidate_poison(dev);
658 break;
659 case NVDIMM_REVALIDATE_REGION:
660 pmem_revalidate_region(dev);
661 break;
662 default:
663 dev_WARN_ONCE(dev, 1, "notify: unknown event: %d\n", event);
664 break;
665 }
666 }
667
668 MODULE_ALIAS("pmem");
669 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO);
670 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM);
671 static struct nd_device_driver nd_pmem_driver = {
672 .probe = nd_pmem_probe,
673 .remove = nd_pmem_remove,
674 .notify = nd_pmem_notify,
675 .shutdown = nd_pmem_shutdown,
676 .drv = {
677 .name = "nd_pmem",
678 },
679 .type = ND_DRIVER_NAMESPACE_IO | ND_DRIVER_NAMESPACE_PMEM,
680 };
681
682 module_nd_driver(nd_pmem_driver);
683
684 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
685 MODULE_LICENSE("GPL v2");
686