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/pfn_t.h>
29 #include <linux/slab.h>
30 #include <linux/pmem.h>
36 static void pmem_clear_poison(struct pmem_device
*pmem
, phys_addr_t offset
,
39 struct device
*dev
= pmem
->bb
.dev
;
43 sector
= (offset
- pmem
->data_offset
) / 512;
44 cleared
= nvdimm_clear_poison(dev
, pmem
->phys_addr
+ offset
, len
);
46 if (cleared
> 0 && cleared
/ 512) {
47 dev_dbg(dev
, "%s: %llx clear %ld sector%s\n",
48 __func__
, (unsigned long long) sector
,
49 cleared
/ 512, cleared
/ 512 > 1 ? "s" : "");
50 badblocks_clear(&pmem
->bb
, sector
, cleared
/ 512);
52 invalidate_pmem(pmem
->virt_addr
+ offset
, len
);
55 static int pmem_do_bvec(struct pmem_device
*pmem
, struct page
*page
,
56 unsigned int len
, unsigned int off
, int rw
,
60 bool bad_pmem
= false;
61 void *mem
= kmap_atomic(page
);
62 phys_addr_t pmem_off
= sector
* 512 + pmem
->data_offset
;
63 void __pmem
*pmem_addr
= pmem
->virt_addr
+ pmem_off
;
65 if (unlikely(is_bad_pmem(&pmem
->bb
, sector
, len
)))
69 if (unlikely(bad_pmem
))
72 rc
= memcpy_from_pmem(mem
+ off
, pmem_addr
, len
);
73 flush_dcache_page(page
);
77 * Note that we write the data both before and after
78 * clearing poison. The write before clear poison
79 * handles situations where the latest written data is
80 * preserved and the clear poison operation simply marks
81 * the address range as valid without changing the data.
82 * In this case application software can assume that an
83 * interrupted write will either return the new good
86 * However, if pmem_clear_poison() leaves the data in an
87 * indeterminate state we need to perform the write
90 flush_dcache_page(page
);
91 memcpy_to_pmem(pmem_addr
, mem
+ off
, len
);
92 if (unlikely(bad_pmem
)) {
93 pmem_clear_poison(pmem
, pmem_off
, len
);
94 memcpy_to_pmem(pmem_addr
, mem
+ off
, len
);
102 static blk_qc_t
pmem_make_request(struct request_queue
*q
, struct bio
*bio
)
108 struct bvec_iter iter
;
109 struct pmem_device
*pmem
= q
->queuedata
;
111 do_acct
= nd_iostat_start(bio
, &start
);
112 bio_for_each_segment(bvec
, bio
, iter
) {
113 rc
= pmem_do_bvec(pmem
, bvec
.bv_page
, bvec
.bv_len
,
114 bvec
.bv_offset
, bio_data_dir(bio
),
122 nd_iostat_end(bio
, start
);
124 if (bio_data_dir(bio
))
128 return BLK_QC_T_NONE
;
131 static int pmem_rw_page(struct block_device
*bdev
, sector_t sector
,
132 struct page
*page
, int rw
)
134 struct pmem_device
*pmem
= bdev
->bd_queue
->queuedata
;
137 rc
= pmem_do_bvec(pmem
, page
, PAGE_SIZE
, 0, rw
, sector
);
142 * The ->rw_page interface is subtle and tricky. The core
143 * retries on any error, so we can only invoke page_endio() in
144 * the successful completion case. Otherwise, we'll see crashes
145 * caused by double completion.
148 page_endio(page
, rw
& WRITE
, 0);
153 /* see "strong" declaration in tools/testing/nvdimm/pmem-dax.c */
154 __weak
long pmem_direct_access(struct block_device
*bdev
, sector_t sector
,
155 void __pmem
**kaddr
, pfn_t
*pfn
, long size
)
157 struct pmem_device
*pmem
= bdev
->bd_queue
->queuedata
;
158 resource_size_t offset
= sector
* 512 + pmem
->data_offset
;
160 if (unlikely(is_bad_pmem(&pmem
->bb
, sector
, size
)))
162 *kaddr
= pmem
->virt_addr
+ offset
;
163 *pfn
= phys_to_pfn_t(pmem
->phys_addr
+ offset
, pmem
->pfn_flags
);
166 * If badblocks are present, limit known good range to the
169 if (unlikely(pmem
->bb
.count
))
171 return pmem
->size
- pmem
->pfn_pad
- offset
;
174 static const struct block_device_operations pmem_fops
= {
175 .owner
= THIS_MODULE
,
176 .rw_page
= pmem_rw_page
,
177 .direct_access
= pmem_direct_access
,
178 .revalidate_disk
= nvdimm_revalidate_disk
,
181 static void pmem_release_queue(void *q
)
183 blk_cleanup_queue(q
);
186 static void pmem_release_disk(void *disk
)
192 static int pmem_attach_disk(struct device
*dev
,
193 struct nd_namespace_common
*ndns
)
195 struct nd_namespace_io
*nsio
= to_nd_namespace_io(&ndns
->dev
);
196 struct vmem_altmap __altmap
, *altmap
= NULL
;
197 struct resource
*res
= &nsio
->res
;
198 struct nd_pfn
*nd_pfn
= NULL
;
199 int nid
= dev_to_node(dev
);
200 struct nd_pfn_sb
*pfn_sb
;
201 struct pmem_device
*pmem
;
202 struct resource pfn_res
;
203 struct request_queue
*q
;
204 struct gendisk
*disk
;
207 /* while nsio_rw_bytes is active, parse a pfn info block if present */
208 if (is_nd_pfn(dev
)) {
209 nd_pfn
= to_nd_pfn(dev
);
210 altmap
= nvdimm_setup_pfn(nd_pfn
, &pfn_res
, &__altmap
);
212 return PTR_ERR(altmap
);
215 /* we're attaching a block device, disable raw namespace access */
216 devm_nsio_disable(dev
, nsio
);
218 pmem
= devm_kzalloc(dev
, sizeof(*pmem
), GFP_KERNEL
);
222 dev_set_drvdata(dev
, pmem
);
223 pmem
->phys_addr
= res
->start
;
224 pmem
->size
= resource_size(res
);
225 if (!arch_has_wmb_pmem())
226 dev_warn(dev
, "unable to guarantee persistence of writes\n");
228 if (!devm_request_mem_region(dev
, res
->start
, resource_size(res
),
230 dev_warn(dev
, "could not reserve region %pR\n", res
);
234 q
= blk_alloc_queue_node(GFP_KERNEL
, dev_to_node(dev
));
238 pmem
->pfn_flags
= PFN_DEV
;
239 if (is_nd_pfn(dev
)) {
240 addr
= devm_memremap_pages(dev
, &pfn_res
, &q
->q_usage_counter
,
242 pfn_sb
= nd_pfn
->pfn_sb
;
243 pmem
->data_offset
= le64_to_cpu(pfn_sb
->dataoff
);
244 pmem
->pfn_pad
= resource_size(res
) - resource_size(&pfn_res
);
245 pmem
->pfn_flags
|= PFN_MAP
;
246 res
= &pfn_res
; /* for badblocks populate */
247 res
->start
+= pmem
->data_offset
;
248 } else if (pmem_should_map_pages(dev
)) {
249 addr
= devm_memremap_pages(dev
, &nsio
->res
,
250 &q
->q_usage_counter
, NULL
);
251 pmem
->pfn_flags
|= PFN_MAP
;
253 addr
= devm_memremap(dev
, pmem
->phys_addr
,
254 pmem
->size
, ARCH_MEMREMAP_PMEM
);
257 * At release time the queue must be dead before
258 * devm_memremap_pages is unwound
260 if (devm_add_action_or_reset(dev
, pmem_release_queue
, q
))
264 return PTR_ERR(addr
);
265 pmem
->virt_addr
= (void __pmem
*) addr
;
267 blk_queue_make_request(q
, pmem_make_request
);
268 blk_queue_physical_block_size(q
, PAGE_SIZE
);
269 blk_queue_max_hw_sectors(q
, UINT_MAX
);
270 blk_queue_bounce_limit(q
, BLK_BOUNCE_ANY
);
271 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, q
);
274 disk
= alloc_disk_node(0, nid
);
278 disk
->fops
= &pmem_fops
;
280 disk
->flags
= GENHD_FL_EXT_DEVT
;
281 nvdimm_namespace_disk_name(ndns
, disk
->disk_name
);
282 disk
->driverfs_dev
= dev
;
283 set_capacity(disk
, (pmem
->size
- pmem
->pfn_pad
- pmem
->data_offset
)
285 if (devm_init_badblocks(dev
, &pmem
->bb
))
287 nvdimm_badblocks_populate(to_nd_region(dev
->parent
), &pmem
->bb
, res
);
288 disk
->bb
= &pmem
->bb
;
291 if (devm_add_action_or_reset(dev
, pmem_release_disk
, disk
))
294 revalidate_disk(disk
);
299 static int nd_pmem_probe(struct device
*dev
)
301 struct nd_namespace_common
*ndns
;
303 ndns
= nvdimm_namespace_common_probe(dev
);
305 return PTR_ERR(ndns
);
307 if (devm_nsio_enable(dev
, to_nd_namespace_io(&ndns
->dev
)))
311 return nvdimm_namespace_attach_btt(ndns
);
314 return pmem_attach_disk(dev
, ndns
);
316 /* if we find a valid info-block we'll come back as that personality */
317 if (nd_btt_probe(dev
, ndns
) == 0 || nd_pfn_probe(dev
, ndns
) == 0
318 || nd_dax_probe(dev
, ndns
) == 0)
321 /* ...otherwise we're just a raw pmem device */
322 return pmem_attach_disk(dev
, ndns
);
325 static int nd_pmem_remove(struct device
*dev
)
328 nvdimm_namespace_detach_btt(to_nd_btt(dev
));
332 static void nd_pmem_notify(struct device
*dev
, enum nvdimm_event event
)
334 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
335 struct pmem_device
*pmem
= dev_get_drvdata(dev
);
336 resource_size_t offset
= 0, end_trunc
= 0;
337 struct nd_namespace_common
*ndns
;
338 struct nd_namespace_io
*nsio
;
341 if (event
!= NVDIMM_REVALIDATE_POISON
)
344 if (is_nd_btt(dev
)) {
345 struct nd_btt
*nd_btt
= to_nd_btt(dev
);
348 } else if (is_nd_pfn(dev
)) {
349 struct nd_pfn
*nd_pfn
= to_nd_pfn(dev
);
350 struct nd_pfn_sb
*pfn_sb
= nd_pfn
->pfn_sb
;
353 offset
= pmem
->data_offset
+ __le32_to_cpu(pfn_sb
->start_pad
);
354 end_trunc
= __le32_to_cpu(pfn_sb
->end_trunc
);
358 nsio
= to_nd_namespace_io(&ndns
->dev
);
359 res
.start
= nsio
->res
.start
+ offset
;
360 res
.end
= nsio
->res
.end
- end_trunc
;
361 nvdimm_badblocks_populate(nd_region
, &pmem
->bb
, &res
);
364 MODULE_ALIAS("pmem");
365 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_IO
);
366 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_NAMESPACE_PMEM
);
367 static struct nd_device_driver nd_pmem_driver
= {
368 .probe
= nd_pmem_probe
,
369 .remove
= nd_pmem_remove
,
370 .notify
= nd_pmem_notify
,
374 .type
= ND_DRIVER_NAMESPACE_IO
| ND_DRIVER_NAMESPACE_PMEM
,
377 static int __init
pmem_init(void)
379 return nd_driver_register(&nd_pmem_driver
);
381 module_init(pmem_init
);
383 static void pmem_exit(void)
385 driver_unregister(&nd_pmem_driver
.drv
);
387 module_exit(pmem_exit
);
389 MODULE_AUTHOR("Ross Zwisler <ross.zwisler@linux.intel.com>");
390 MODULE_LICENSE("GPL v2");