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>
38 static struct device
*to_dev(struct pmem_device
*pmem
)
41 * nvdimm bus services need a 'dev' parameter, and we record the device
47 static struct nd_region
*to_region(struct pmem_device
*pmem
)
49 return to_nd_region(to_dev(pmem
)->parent
);
52 static blk_status_t
pmem_clear_poison(struct pmem_device
*pmem
,
53 phys_addr_t offset
, unsigned int len
)
55 struct device
*dev
= to_dev(pmem
);
58 blk_status_t rc
= BLK_STS_OK
;
60 sector
= (offset
- pmem
->data_offset
) / 512;
62 cleared
= nvdimm_clear_poison(dev
, pmem
->phys_addr
+ offset
, len
);
65 if (cleared
> 0 && cleared
/ 512) {
67 dev_dbg(dev
, "%s: %#llx clear %ld sector%s\n", __func__
,
68 (unsigned long long) sector
, cleared
,
69 cleared
> 1 ? "s" : "");
70 badblocks_clear(&pmem
->bb
, sector
, cleared
);
72 sysfs_notify_dirent(pmem
->bb_state
);
75 arch_invalidate_pmem(pmem
->virt_addr
+ offset
, len
);
80 static void write_pmem(void *pmem_addr
, struct page
*page
,
81 unsigned int off
, unsigned int len
)
83 void *mem
= kmap_atomic(page
);
85 memcpy_flushcache(pmem_addr
, mem
+ off
, len
);
89 static blk_status_t
read_pmem(struct page
*page
, unsigned int off
,
90 void *pmem_addr
, unsigned int len
)
93 void *mem
= kmap_atomic(page
);
95 rc
= memcpy_mcsafe(mem
+ off
, pmem_addr
, len
);
102 static blk_status_t
pmem_do_bvec(struct pmem_device
*pmem
, struct page
*page
,
103 unsigned int len
, unsigned int off
, bool is_write
,
106 blk_status_t rc
= BLK_STS_OK
;
107 bool bad_pmem
= false;
108 phys_addr_t pmem_off
= sector
* 512 + pmem
->data_offset
;
109 void *pmem_addr
= pmem
->virt_addr
+ pmem_off
;
111 if (unlikely(is_bad_pmem(&pmem
->bb
, sector
, len
)))
115 if (unlikely(bad_pmem
))
118 rc
= read_pmem(page
, off
, pmem_addr
, len
);
119 flush_dcache_page(page
);
123 * Note that we write the data both before and after
124 * clearing poison. The write before clear poison
125 * handles situations where the latest written data is
126 * preserved and the clear poison operation simply marks
127 * the address range as valid without changing the data.
128 * In this case application software can assume that an
129 * interrupted write will either return the new good
132 * However, if pmem_clear_poison() leaves the data in an
133 * indeterminate state we need to perform the write
134 * after clear poison.
136 flush_dcache_page(page
);
137 write_pmem(pmem_addr
, page
, off
, len
);
138 if (unlikely(bad_pmem
)) {
139 rc
= pmem_clear_poison(pmem
, pmem_off
, len
);
140 write_pmem(pmem_addr
, page
, off
, len
);
147 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
149 #define REQ_FLUSH REQ_PREFLUSH
152 static blk_qc_t
pmem_make_request(struct request_queue
*q
, struct bio
*bio
)
158 struct bvec_iter iter
;
159 struct pmem_device
*pmem
= q
->queuedata
;
160 struct nd_region
*nd_region
= to_region(pmem
);
162 if (bio
->bi_opf
& REQ_FLUSH
)
163 nvdimm_flush(nd_region
);
165 do_acct
= nd_iostat_start(bio
, &start
);
166 bio_for_each_segment(bvec
, bio
, iter
) {
167 rc
= pmem_do_bvec(pmem
, bvec
.bv_page
, bvec
.bv_len
,
168 bvec
.bv_offset
, op_is_write(bio_op(bio
)),
176 nd_iostat_end(bio
, start
);
178 if (bio
->bi_opf
& REQ_FUA
)
179 nvdimm_flush(nd_region
);
182 return BLK_QC_T_NONE
;
185 static int pmem_rw_page(struct block_device
*bdev
, sector_t sector
,
186 struct page
*page
, bool is_write
)
188 struct pmem_device
*pmem
= bdev
->bd_queue
->queuedata
;
191 rc
= pmem_do_bvec(pmem
, page
, PAGE_SIZE
, 0, is_write
, sector
);
194 * The ->rw_page interface is subtle and tricky. The core
195 * retries on any error, so we can only invoke page_endio() in
196 * the successful completion case. Otherwise, we'll see crashes
197 * caused by double completion.
200 page_endio(page
, is_write
, 0);
202 return blk_status_to_errno(rc
);
205 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
206 __weak
long __pmem_direct_access(struct pmem_device
*pmem
, pgoff_t pgoff
,
207 long nr_pages
, void **kaddr
, pfn_t
*pfn
)
209 resource_size_t offset
= PFN_PHYS(pgoff
) + pmem
->data_offset
;
211 if (unlikely(is_bad_pmem(&pmem
->bb
, PFN_PHYS(pgoff
) / 512,
212 PFN_PHYS(nr_pages
))))
214 *kaddr
= pmem
->virt_addr
+ offset
;
215 *pfn
= phys_to_pfn_t(pmem
->phys_addr
+ offset
, pmem
->pfn_flags
);
218 * If badblocks are present, limit known good range to the
221 if (unlikely(pmem
->bb
.count
))
223 return PHYS_PFN(pmem
->size
- pmem
->pfn_pad
- offset
);
226 static const struct block_device_operations pmem_fops
= {
227 .owner
= THIS_MODULE
,
228 .rw_page
= pmem_rw_page
,
229 .revalidate_disk
= nvdimm_revalidate_disk
,
232 static long pmem_dax_direct_access(struct dax_device
*dax_dev
,
233 pgoff_t pgoff
, long nr_pages
, void **kaddr
, pfn_t
*pfn
)
235 struct pmem_device
*pmem
= dax_get_private(dax_dev
);
237 return __pmem_direct_access(pmem
, pgoff
, nr_pages
, kaddr
, pfn
);
240 static size_t pmem_copy_from_iter(struct dax_device
*dax_dev
, pgoff_t pgoff
,
241 void *addr
, size_t bytes
, struct iov_iter
*i
)
243 return copy_from_iter_flushcache(addr
, bytes
, i
);
246 static const struct dax_operations pmem_dax_ops
= {
247 .direct_access
= pmem_dax_direct_access
,
248 .copy_from_iter
= pmem_copy_from_iter
,
251 static const struct attribute_group
*pmem_attribute_groups
[] = {
252 &dax_attribute_group
,
256 static void pmem_release_queue(void *q
)
258 blk_cleanup_queue(q
);
261 static void pmem_freeze_queue(void *q
)
263 blk_freeze_queue_start(q
);
266 static void pmem_release_disk(void *__pmem
)
268 struct pmem_device
*pmem
= __pmem
;
270 kill_dax(pmem
->dax_dev
);
271 put_dax(pmem
->dax_dev
);
272 del_gendisk(pmem
->disk
);
273 put_disk(pmem
->disk
);
276 static int pmem_attach_disk(struct device
*dev
,
277 struct nd_namespace_common
*ndns
)
279 struct nd_namespace_io
*nsio
= to_nd_namespace_io(&ndns
->dev
);
280 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
281 struct vmem_altmap __altmap
, *altmap
= NULL
;
282 int nid
= dev_to_node(dev
), fua
, wbc
;
283 struct resource
*res
= &nsio
->res
;
284 struct nd_pfn
*nd_pfn
= NULL
;
285 struct dax_device
*dax_dev
;
286 struct nd_pfn_sb
*pfn_sb
;
287 struct pmem_device
*pmem
;
288 struct resource pfn_res
;
289 struct request_queue
*q
;
290 struct device
*gendev
;
291 struct gendisk
*disk
;
294 /* while nsio_rw_bytes is active, parse a pfn info block if present */
295 if (is_nd_pfn(dev
)) {
296 nd_pfn
= to_nd_pfn(dev
);
297 altmap
= nvdimm_setup_pfn(nd_pfn
, &pfn_res
, &__altmap
);
299 return PTR_ERR(altmap
);
302 /* we're attaching a block device, disable raw namespace access */
303 devm_nsio_disable(dev
, nsio
);
305 pmem
= devm_kzalloc(dev
, sizeof(*pmem
), GFP_KERNEL
);
309 dev_set_drvdata(dev
, pmem
);
310 pmem
->phys_addr
= res
->start
;
311 pmem
->size
= resource_size(res
);
312 fua
= nvdimm_has_flush(nd_region
);
313 if (!IS_ENABLED(CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
) || fua
< 0) {
314 dev_warn(dev
, "unable to guarantee persistence of writes\n");
317 wbc
= nvdimm_has_cache(nd_region
);
319 if (!devm_request_mem_region(dev
, res
->start
, resource_size(res
),
320 dev_name(&ndns
->dev
))) {
321 dev_warn(dev
, "could not reserve region %pR\n", res
);
325 q
= blk_alloc_queue_node(GFP_KERNEL
, dev_to_node(dev
));
329 if (devm_add_action_or_reset(dev
, pmem_release_queue
, q
))
332 pmem
->pfn_flags
= PFN_DEV
;
333 if (is_nd_pfn(dev
)) {
334 addr
= devm_memremap_pages(dev
, &pfn_res
, &q
->q_usage_counter
,
336 pfn_sb
= nd_pfn
->pfn_sb
;
337 pmem
->data_offset
= le64_to_cpu(pfn_sb
->dataoff
);
338 pmem
->pfn_pad
= resource_size(res
) - resource_size(&pfn_res
);
339 pmem
->pfn_flags
|= PFN_MAP
;
340 res
= &pfn_res
; /* for badblocks populate */
341 res
->start
+= pmem
->data_offset
;
342 } else if (pmem_should_map_pages(dev
)) {
343 addr
= devm_memremap_pages(dev
, &nsio
->res
,
344 &q
->q_usage_counter
, NULL
);
345 pmem
->pfn_flags
|= PFN_MAP
;
347 addr
= devm_memremap(dev
, pmem
->phys_addr
,
348 pmem
->size
, ARCH_MEMREMAP_PMEM
);
351 * At release time the queue must be frozen before
352 * devm_memremap_pages is unwound
354 if (devm_add_action_or_reset(dev
, pmem_freeze_queue
, q
))
358 return PTR_ERR(addr
);
359 pmem
->virt_addr
= addr
;
361 blk_queue_write_cache(q
, wbc
, fua
);
362 blk_queue_make_request(q
, pmem_make_request
);
363 blk_queue_physical_block_size(q
, PAGE_SIZE
);
364 blk_queue_logical_block_size(q
, pmem_sector_size(ndns
));
365 blk_queue_max_hw_sectors(q
, UINT_MAX
);
366 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
367 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, q
);
370 disk
= alloc_disk_node(0, nid
);
375 disk
->fops
= &pmem_fops
;
377 disk
->flags
= GENHD_FL_EXT_DEVT
;
378 nvdimm_namespace_disk_name(ndns
, disk
->disk_name
);
379 set_capacity(disk
, (pmem
->size
- pmem
->pfn_pad
- pmem
->data_offset
)
381 if (devm_init_badblocks(dev
, &pmem
->bb
))
383 nvdimm_badblocks_populate(nd_region
, &pmem
->bb
, res
);
384 disk
->bb
= &pmem
->bb
;
386 dax_dev
= alloc_dax(pmem
, disk
->disk_name
, &pmem_dax_ops
);
391 dax_write_cache(dax_dev
, wbc
);
392 pmem
->dax_dev
= dax_dev
;
394 gendev
= disk_to_dev(disk
);
395 gendev
->groups
= pmem_attribute_groups
;
397 device_add_disk(dev
, disk
);
398 if (devm_add_action_or_reset(dev
, pmem_release_disk
, pmem
))
401 revalidate_disk(disk
);
403 pmem
->bb_state
= sysfs_get_dirent(disk_to_dev(disk
)->kobj
.sd
,
406 dev_warn(dev
, "'badblocks' notification disabled\n");
411 static int nd_pmem_probe(struct device
*dev
)
413 struct nd_namespace_common
*ndns
;
415 ndns
= nvdimm_namespace_common_probe(dev
);
417 return PTR_ERR(ndns
);
419 if (devm_nsio_enable(dev
, to_nd_namespace_io(&ndns
->dev
)))
423 return nvdimm_namespace_attach_btt(ndns
);
426 return pmem_attach_disk(dev
, ndns
);
428 /* if we find a valid info-block we'll come back as that personality */
429 if (nd_btt_probe(dev
, ndns
) == 0 || nd_pfn_probe(dev
, ndns
) == 0
430 || nd_dax_probe(dev
, ndns
) == 0)
433 /* ...otherwise we're just a raw pmem device */
434 return pmem_attach_disk(dev
, ndns
);
437 static int nd_pmem_remove(struct device
*dev
)
439 struct pmem_device
*pmem
= dev_get_drvdata(dev
);
442 nvdimm_namespace_detach_btt(to_nd_btt(dev
));
445 * Note, this assumes device_lock() context to not race
448 sysfs_put(pmem
->bb_state
);
449 pmem
->bb_state
= NULL
;
451 nvdimm_flush(to_nd_region(dev
->parent
));
456 static void nd_pmem_shutdown(struct device
*dev
)
458 nvdimm_flush(to_nd_region(dev
->parent
));
461 static void nd_pmem_notify(struct device
*dev
, enum nvdimm_event event
)
463 struct nd_region
*nd_region
;
464 resource_size_t offset
= 0, end_trunc
= 0;
465 struct nd_namespace_common
*ndns
;
466 struct nd_namespace_io
*nsio
;
468 struct badblocks
*bb
;
469 struct kernfs_node
*bb_state
;
471 if (event
!= NVDIMM_REVALIDATE_POISON
)
474 if (is_nd_btt(dev
)) {
475 struct nd_btt
*nd_btt
= to_nd_btt(dev
);
478 nd_region
= to_nd_region(ndns
->dev
.parent
);
479 nsio
= to_nd_namespace_io(&ndns
->dev
);
483 struct pmem_device
*pmem
= dev_get_drvdata(dev
);
485 nd_region
= to_region(pmem
);
487 bb_state
= pmem
->bb_state
;
489 if (is_nd_pfn(dev
)) {
490 struct nd_pfn
*nd_pfn
= to_nd_pfn(dev
);
491 struct nd_pfn_sb
*pfn_sb
= nd_pfn
->pfn_sb
;
494 offset
= pmem
->data_offset
+
495 __le32_to_cpu(pfn_sb
->start_pad
);
496 end_trunc
= __le32_to_cpu(pfn_sb
->end_trunc
);
501 nsio
= to_nd_namespace_io(&ndns
->dev
);
504 res
.start
= nsio
->res
.start
+ offset
;
505 res
.end
= nsio
->res
.end
- end_trunc
;
506 nvdimm_badblocks_populate(nd_region
, bb
, &res
);
508 sysfs_notify_dirent(bb_state
);
511 MODULE_ALIAS("pmem");
512 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO
);
513 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM
);
514 static struct nd_device_driver nd_pmem_driver
= {
515 .probe
= nd_pmem_probe
,
516 .remove
= nd_pmem_remove
,
517 .notify
= nd_pmem_notify
,
518 .shutdown
= nd_pmem_shutdown
,
522 .type
= ND_DRIVER_NAMESPACE_IO
| ND_DRIVER_NAMESPACE_PMEM
,
525 static int __init
pmem_init(void)
527 return nd_driver_register(&nd_pmem_driver
);
529 module_init(pmem_init
);
531 static void pmem_exit(void)
533 driver_unregister(&nd_pmem_driver
.drv
);
535 module_exit(pmem_exit
);
537 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
538 MODULE_LICENSE("GPL v2");