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/pmem.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 int pmem_clear_poison(struct pmem_device
*pmem
, phys_addr_t offset
,
55 struct device
*dev
= to_dev(pmem
);
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
);
73 invalidate_pmem(pmem
->virt_addr
+ offset
, len
);
78 static void write_pmem(void *pmem_addr
, struct page
*page
,
79 unsigned int off
, unsigned int len
)
81 void *mem
= kmap_atomic(page
);
83 memcpy_to_pmem(pmem_addr
, mem
+ off
, len
);
87 static int read_pmem(struct page
*page
, unsigned int off
,
88 void *pmem_addr
, unsigned int len
)
91 void *mem
= kmap_atomic(page
);
93 rc
= memcpy_mcsafe(mem
+ off
, pmem_addr
, len
);
100 static int pmem_do_bvec(struct pmem_device
*pmem
, struct page
*page
,
101 unsigned int len
, unsigned int off
, bool is_write
,
105 bool bad_pmem
= false;
106 phys_addr_t pmem_off
= sector
* 512 + pmem
->data_offset
;
107 void *pmem_addr
= pmem
->virt_addr
+ pmem_off
;
109 if (unlikely(is_bad_pmem(&pmem
->bb
, sector
, len
)))
113 if (unlikely(bad_pmem
))
116 rc
= read_pmem(page
, off
, pmem_addr
, len
);
117 flush_dcache_page(page
);
121 * Note that we write the data both before and after
122 * clearing poison. The write before clear poison
123 * handles situations where the latest written data is
124 * preserved and the clear poison operation simply marks
125 * the address range as valid without changing the data.
126 * In this case application software can assume that an
127 * interrupted write will either return the new good
130 * However, if pmem_clear_poison() leaves the data in an
131 * indeterminate state we need to perform the write
132 * after clear poison.
134 flush_dcache_page(page
);
135 write_pmem(pmem_addr
, page
, off
, len
);
136 if (unlikely(bad_pmem
)) {
137 rc
= pmem_clear_poison(pmem
, pmem_off
, len
);
138 write_pmem(pmem_addr
, page
, off
, len
);
145 /* account for REQ_FLUSH rename, replace with REQ_PREFLUSH after v4.8-rc1 */
147 #define REQ_FLUSH REQ_PREFLUSH
150 static blk_qc_t
pmem_make_request(struct request_queue
*q
, struct bio
*bio
)
156 struct bvec_iter iter
;
157 struct pmem_device
*pmem
= q
->queuedata
;
158 struct nd_region
*nd_region
= to_region(pmem
);
160 if (bio
->bi_opf
& REQ_FLUSH
)
161 nvdimm_flush(nd_region
);
163 do_acct
= nd_iostat_start(bio
, &start
);
164 bio_for_each_segment(bvec
, bio
, iter
) {
165 rc
= pmem_do_bvec(pmem
, bvec
.bv_page
, bvec
.bv_len
,
166 bvec
.bv_offset
, op_is_write(bio_op(bio
)),
174 nd_iostat_end(bio
, start
);
176 if (bio
->bi_opf
& REQ_FUA
)
177 nvdimm_flush(nd_region
);
180 return BLK_QC_T_NONE
;
183 static int pmem_rw_page(struct block_device
*bdev
, sector_t sector
,
184 struct page
*page
, bool is_write
)
186 struct pmem_device
*pmem
= bdev
->bd_queue
->queuedata
;
189 rc
= pmem_do_bvec(pmem
, page
, PAGE_SIZE
, 0, is_write
, sector
);
192 * The ->rw_page interface is subtle and tricky. The core
193 * retries on any error, so we can only invoke page_endio() in
194 * the successful completion case. Otherwise, we'll see crashes
195 * caused by double completion.
198 page_endio(page
, is_write
, 0);
203 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
204 __weak
long __pmem_direct_access(struct pmem_device
*pmem
, pgoff_t pgoff
,
205 long nr_pages
, void **kaddr
, pfn_t
*pfn
)
207 resource_size_t offset
= PFN_PHYS(pgoff
) + pmem
->data_offset
;
209 if (unlikely(is_bad_pmem(&pmem
->bb
, PFN_PHYS(pgoff
) / 512,
210 PFN_PHYS(nr_pages
))))
212 *kaddr
= pmem
->virt_addr
+ offset
;
213 *pfn
= phys_to_pfn_t(pmem
->phys_addr
+ offset
, pmem
->pfn_flags
);
216 * If badblocks are present, limit known good range to the
219 if (unlikely(pmem
->bb
.count
))
221 return PHYS_PFN(pmem
->size
- pmem
->pfn_pad
- offset
);
224 static const struct block_device_operations pmem_fops
= {
225 .owner
= THIS_MODULE
,
226 .rw_page
= pmem_rw_page
,
227 .revalidate_disk
= nvdimm_revalidate_disk
,
230 static long pmem_dax_direct_access(struct dax_device
*dax_dev
,
231 pgoff_t pgoff
, long nr_pages
, void **kaddr
, pfn_t
*pfn
)
233 struct pmem_device
*pmem
= dax_get_private(dax_dev
);
235 return __pmem_direct_access(pmem
, pgoff
, nr_pages
, kaddr
, pfn
);
238 static const struct dax_operations pmem_dax_ops
= {
239 .direct_access
= pmem_dax_direct_access
,
242 static void pmem_release_queue(void *q
)
244 blk_cleanup_queue(q
);
247 static void pmem_freeze_queue(void *q
)
249 blk_freeze_queue_start(q
);
252 static void pmem_release_disk(void *__pmem
)
254 struct pmem_device
*pmem
= __pmem
;
256 kill_dax(pmem
->dax_dev
);
257 put_dax(pmem
->dax_dev
);
258 del_gendisk(pmem
->disk
);
259 put_disk(pmem
->disk
);
262 static int pmem_attach_disk(struct device
*dev
,
263 struct nd_namespace_common
*ndns
)
265 struct nd_namespace_io
*nsio
= to_nd_namespace_io(&ndns
->dev
);
266 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
267 struct vmem_altmap __altmap
, *altmap
= NULL
;
268 struct resource
*res
= &nsio
->res
;
269 struct nd_pfn
*nd_pfn
= NULL
;
270 struct dax_device
*dax_dev
;
271 int nid
= dev_to_node(dev
);
272 struct nd_pfn_sb
*pfn_sb
;
273 struct pmem_device
*pmem
;
274 struct resource pfn_res
;
275 struct request_queue
*q
;
276 struct gendisk
*disk
;
279 /* while nsio_rw_bytes is active, parse a pfn info block if present */
280 if (is_nd_pfn(dev
)) {
281 nd_pfn
= to_nd_pfn(dev
);
282 altmap
= nvdimm_setup_pfn(nd_pfn
, &pfn_res
, &__altmap
);
284 return PTR_ERR(altmap
);
287 /* we're attaching a block device, disable raw namespace access */
288 devm_nsio_disable(dev
, nsio
);
290 pmem
= devm_kzalloc(dev
, sizeof(*pmem
), GFP_KERNEL
);
294 dev_set_drvdata(dev
, pmem
);
295 pmem
->phys_addr
= res
->start
;
296 pmem
->size
= resource_size(res
);
297 if (nvdimm_has_flush(nd_region
) < 0)
298 dev_warn(dev
, "unable to guarantee persistence of writes\n");
300 if (!devm_request_mem_region(dev
, res
->start
, resource_size(res
),
301 dev_name(&ndns
->dev
))) {
302 dev_warn(dev
, "could not reserve region %pR\n", res
);
306 q
= blk_alloc_queue_node(GFP_KERNEL
, dev_to_node(dev
));
310 if (devm_add_action_or_reset(dev
, pmem_release_queue
, q
))
313 pmem
->pfn_flags
= PFN_DEV
;
314 if (is_nd_pfn(dev
)) {
315 addr
= devm_memremap_pages(dev
, &pfn_res
, &q
->q_usage_counter
,
317 pfn_sb
= nd_pfn
->pfn_sb
;
318 pmem
->data_offset
= le64_to_cpu(pfn_sb
->dataoff
);
319 pmem
->pfn_pad
= resource_size(res
) - resource_size(&pfn_res
);
320 pmem
->pfn_flags
|= PFN_MAP
;
321 res
= &pfn_res
; /* for badblocks populate */
322 res
->start
+= pmem
->data_offset
;
323 } else if (pmem_should_map_pages(dev
)) {
324 addr
= devm_memremap_pages(dev
, &nsio
->res
,
325 &q
->q_usage_counter
, NULL
);
326 pmem
->pfn_flags
|= PFN_MAP
;
328 addr
= devm_memremap(dev
, pmem
->phys_addr
,
329 pmem
->size
, ARCH_MEMREMAP_PMEM
);
332 * At release time the queue must be frozen before
333 * devm_memremap_pages is unwound
335 if (devm_add_action_or_reset(dev
, pmem_freeze_queue
, q
))
339 return PTR_ERR(addr
);
340 pmem
->virt_addr
= addr
;
342 blk_queue_write_cache(q
, true, true);
343 blk_queue_make_request(q
, pmem_make_request
);
344 blk_queue_physical_block_size(q
, PAGE_SIZE
);
345 blk_queue_max_hw_sectors(q
, UINT_MAX
);
346 blk_queue_bounce_limit(q
, BLK_BOUNCE_ANY
);
347 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
348 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, q
);
351 disk
= alloc_disk_node(0, nid
);
356 disk
->fops
= &pmem_fops
;
358 disk
->flags
= GENHD_FL_EXT_DEVT
;
359 nvdimm_namespace_disk_name(ndns
, disk
->disk_name
);
360 set_capacity(disk
, (pmem
->size
- pmem
->pfn_pad
- pmem
->data_offset
)
362 if (devm_init_badblocks(dev
, &pmem
->bb
))
364 nvdimm_badblocks_populate(nd_region
, &pmem
->bb
, res
);
365 disk
->bb
= &pmem
->bb
;
367 dax_dev
= alloc_dax(pmem
, disk
->disk_name
, &pmem_dax_ops
);
372 pmem
->dax_dev
= dax_dev
;
374 device_add_disk(dev
, disk
);
375 if (devm_add_action_or_reset(dev
, pmem_release_disk
, pmem
))
378 revalidate_disk(disk
);
383 static int nd_pmem_probe(struct device
*dev
)
385 struct nd_namespace_common
*ndns
;
387 ndns
= nvdimm_namespace_common_probe(dev
);
389 return PTR_ERR(ndns
);
391 if (devm_nsio_enable(dev
, to_nd_namespace_io(&ndns
->dev
)))
395 return nvdimm_namespace_attach_btt(ndns
);
398 return pmem_attach_disk(dev
, ndns
);
400 /* if we find a valid info-block we'll come back as that personality */
401 if (nd_btt_probe(dev
, ndns
) == 0 || nd_pfn_probe(dev
, ndns
) == 0
402 || nd_dax_probe(dev
, ndns
) == 0)
405 /* ...otherwise we're just a raw pmem device */
406 return pmem_attach_disk(dev
, ndns
);
409 static int nd_pmem_remove(struct device
*dev
)
412 nvdimm_namespace_detach_btt(to_nd_btt(dev
));
413 nvdimm_flush(to_nd_region(dev
->parent
));
418 static void nd_pmem_shutdown(struct device
*dev
)
420 nvdimm_flush(to_nd_region(dev
->parent
));
423 static void nd_pmem_notify(struct device
*dev
, enum nvdimm_event event
)
425 struct nd_region
*nd_region
;
426 resource_size_t offset
= 0, end_trunc
= 0;
427 struct nd_namespace_common
*ndns
;
428 struct nd_namespace_io
*nsio
;
430 struct badblocks
*bb
;
432 if (event
!= NVDIMM_REVALIDATE_POISON
)
435 if (is_nd_btt(dev
)) {
436 struct nd_btt
*nd_btt
= to_nd_btt(dev
);
439 nd_region
= to_nd_region(ndns
->dev
.parent
);
440 nsio
= to_nd_namespace_io(&ndns
->dev
);
443 struct pmem_device
*pmem
= dev_get_drvdata(dev
);
445 nd_region
= to_region(pmem
);
448 if (is_nd_pfn(dev
)) {
449 struct nd_pfn
*nd_pfn
= to_nd_pfn(dev
);
450 struct nd_pfn_sb
*pfn_sb
= nd_pfn
->pfn_sb
;
453 offset
= pmem
->data_offset
+
454 __le32_to_cpu(pfn_sb
->start_pad
);
455 end_trunc
= __le32_to_cpu(pfn_sb
->end_trunc
);
460 nsio
= to_nd_namespace_io(&ndns
->dev
);
463 res
.start
= nsio
->res
.start
+ offset
;
464 res
.end
= nsio
->res
.end
- end_trunc
;
465 nvdimm_badblocks_populate(nd_region
, bb
, &res
);
468 MODULE_ALIAS("pmem");
469 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO
);
470 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM
);
471 static struct nd_device_driver nd_pmem_driver
= {
472 .probe
= nd_pmem_probe
,
473 .remove
= nd_pmem_remove
,
474 .notify
= nd_pmem_notify
,
475 .shutdown
= nd_pmem_shutdown
,
479 .type
= ND_DRIVER_NAMESPACE_IO
| ND_DRIVER_NAMESPACE_PMEM
,
482 static int __init
pmem_init(void)
484 return nd_driver_register(&nd_pmem_driver
);
486 module_init(pmem_init
);
488 static void pmem_exit(void)
490 driver_unregister(&nd_pmem_driver
.drv
);
492 module_exit(pmem_exit
);
494 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
495 MODULE_LICENSE("GPL v2");