2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/sort.h>
16 #include <linux/slab.h>
17 #include <linux/pmem.h>
18 #include <linux/list.h>
23 static void namespace_io_release(struct device
*dev
)
25 struct nd_namespace_io
*nsio
= to_nd_namespace_io(dev
);
30 static void namespace_pmem_release(struct device
*dev
)
32 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
33 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
36 ida_simple_remove(&nd_region
->ns_ida
, nspm
->id
);
37 kfree(nspm
->alt_name
);
42 static void namespace_blk_release(struct device
*dev
)
44 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
45 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
48 ida_simple_remove(&nd_region
->ns_ida
, nsblk
->id
);
49 kfree(nsblk
->alt_name
);
55 static const struct device_type namespace_io_device_type
= {
56 .name
= "nd_namespace_io",
57 .release
= namespace_io_release
,
60 static const struct device_type namespace_pmem_device_type
= {
61 .name
= "nd_namespace_pmem",
62 .release
= namespace_pmem_release
,
65 static const struct device_type namespace_blk_device_type
= {
66 .name
= "nd_namespace_blk",
67 .release
= namespace_blk_release
,
70 static bool is_namespace_pmem(const struct device
*dev
)
72 return dev
? dev
->type
== &namespace_pmem_device_type
: false;
75 static bool is_namespace_blk(const struct device
*dev
)
77 return dev
? dev
->type
== &namespace_blk_device_type
: false;
80 static bool is_namespace_io(const struct device
*dev
)
82 return dev
? dev
->type
== &namespace_io_device_type
: false;
85 static int is_uuid_busy(struct device
*dev
, void *data
)
87 u8
*uuid1
= data
, *uuid2
= NULL
;
89 if (is_namespace_pmem(dev
)) {
90 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
93 } else if (is_namespace_blk(dev
)) {
94 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
97 } else if (is_nd_btt(dev
)) {
98 struct nd_btt
*nd_btt
= to_nd_btt(dev
);
100 uuid2
= nd_btt
->uuid
;
101 } else if (is_nd_pfn(dev
)) {
102 struct nd_pfn
*nd_pfn
= to_nd_pfn(dev
);
104 uuid2
= nd_pfn
->uuid
;
107 if (uuid2
&& memcmp(uuid1
, uuid2
, NSLABEL_UUID_LEN
) == 0)
113 static int is_namespace_uuid_busy(struct device
*dev
, void *data
)
115 if (is_nd_pmem(dev
) || is_nd_blk(dev
))
116 return device_for_each_child(dev
, data
, is_uuid_busy
);
121 * nd_is_uuid_unique - verify that no other namespace has @uuid
122 * @dev: any device on a nvdimm_bus
123 * @uuid: uuid to check
125 bool nd_is_uuid_unique(struct device
*dev
, u8
*uuid
)
127 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(dev
);
131 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus
->dev
));
132 if (device_for_each_child(&nvdimm_bus
->dev
, uuid
,
133 is_namespace_uuid_busy
) != 0)
138 bool pmem_should_map_pages(struct device
*dev
)
140 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
141 struct nd_namespace_io
*nsio
;
143 if (!IS_ENABLED(CONFIG_ZONE_DEVICE
))
146 if (!test_bit(ND_REGION_PAGEMAP
, &nd_region
->flags
))
149 if (is_nd_pfn(dev
) || is_nd_btt(dev
))
152 nsio
= to_nd_namespace_io(dev
);
153 if (region_intersects(nsio
->res
.start
, resource_size(&nsio
->res
),
154 IORESOURCE_SYSTEM_RAM
,
155 IORES_DESC_NONE
) == REGION_MIXED
)
158 #ifdef ARCH_MEMREMAP_PMEM
159 return ARCH_MEMREMAP_PMEM
== MEMREMAP_WB
;
164 EXPORT_SYMBOL(pmem_should_map_pages
);
166 const char *nvdimm_namespace_disk_name(struct nd_namespace_common
*ndns
,
169 struct nd_region
*nd_region
= to_nd_region(ndns
->dev
.parent
);
170 const char *suffix
= NULL
;
172 if (ndns
->claim
&& is_nd_btt(ndns
->claim
))
175 if (is_namespace_pmem(&ndns
->dev
) || is_namespace_io(&ndns
->dev
)) {
178 if (is_namespace_pmem(&ndns
->dev
)) {
179 struct nd_namespace_pmem
*nspm
;
181 nspm
= to_nd_namespace_pmem(&ndns
->dev
);
186 sprintf(name
, "pmem%d.%d%s", nd_region
->id
, nsidx
,
187 suffix
? suffix
: "");
189 sprintf(name
, "pmem%d%s", nd_region
->id
,
190 suffix
? suffix
: "");
191 } else if (is_namespace_blk(&ndns
->dev
)) {
192 struct nd_namespace_blk
*nsblk
;
194 nsblk
= to_nd_namespace_blk(&ndns
->dev
);
195 sprintf(name
, "ndblk%d.%d%s", nd_region
->id
, nsblk
->id
,
196 suffix
? suffix
: "");
203 EXPORT_SYMBOL(nvdimm_namespace_disk_name
);
205 const u8
*nd_dev_to_uuid(struct device
*dev
)
207 static const u8 null_uuid
[16];
212 if (is_namespace_pmem(dev
)) {
213 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
216 } else if (is_namespace_blk(dev
)) {
217 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
223 EXPORT_SYMBOL(nd_dev_to_uuid
);
225 static ssize_t
nstype_show(struct device
*dev
,
226 struct device_attribute
*attr
, char *buf
)
228 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
230 return sprintf(buf
, "%d\n", nd_region_to_nstype(nd_region
));
232 static DEVICE_ATTR_RO(nstype
);
234 static ssize_t
__alt_name_store(struct device
*dev
, const char *buf
,
237 char *input
, *pos
, *alt_name
, **ns_altname
;
240 if (is_namespace_pmem(dev
)) {
241 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
243 ns_altname
= &nspm
->alt_name
;
244 } else if (is_namespace_blk(dev
)) {
245 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
247 ns_altname
= &nsblk
->alt_name
;
251 if (dev
->driver
|| to_ndns(dev
)->claim
)
254 input
= kmemdup(buf
, len
+ 1, GFP_KERNEL
);
260 if (strlen(pos
) + 1 > NSLABEL_NAME_LEN
) {
265 alt_name
= kzalloc(NSLABEL_NAME_LEN
, GFP_KERNEL
);
271 *ns_altname
= alt_name
;
272 sprintf(*ns_altname
, "%s", pos
);
280 static resource_size_t
nd_namespace_blk_size(struct nd_namespace_blk
*nsblk
)
282 struct nd_region
*nd_region
= to_nd_region(nsblk
->common
.dev
.parent
);
283 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
284 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
285 struct nd_label_id label_id
;
286 resource_size_t size
= 0;
287 struct resource
*res
;
291 nd_label_gen_id(&label_id
, nsblk
->uuid
, NSLABEL_FLAG_LOCAL
);
292 for_each_dpa_resource(ndd
, res
)
293 if (strcmp(res
->name
, label_id
.id
) == 0)
294 size
+= resource_size(res
);
298 static bool __nd_namespace_blk_validate(struct nd_namespace_blk
*nsblk
)
300 struct nd_region
*nd_region
= to_nd_region(nsblk
->common
.dev
.parent
);
301 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
302 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
303 struct nd_label_id label_id
;
304 struct resource
*res
;
307 if (!nsblk
->uuid
|| !nsblk
->lbasize
|| !ndd
)
311 nd_label_gen_id(&label_id
, nsblk
->uuid
, NSLABEL_FLAG_LOCAL
);
312 for_each_dpa_resource(ndd
, res
) {
313 if (strcmp(res
->name
, label_id
.id
) != 0)
316 * Resources with unacknowledged adjustments indicate a
317 * failure to update labels
319 if (res
->flags
& DPA_RESOURCE_ADJUSTED
)
324 /* These values match after a successful label update */
325 if (count
!= nsblk
->num_resources
)
328 for (i
= 0; i
< nsblk
->num_resources
; i
++) {
329 struct resource
*found
= NULL
;
331 for_each_dpa_resource(ndd
, res
)
332 if (res
== nsblk
->res
[i
]) {
344 resource_size_t
nd_namespace_blk_validate(struct nd_namespace_blk
*nsblk
)
346 resource_size_t size
;
348 nvdimm_bus_lock(&nsblk
->common
.dev
);
349 size
= __nd_namespace_blk_validate(nsblk
);
350 nvdimm_bus_unlock(&nsblk
->common
.dev
);
354 EXPORT_SYMBOL(nd_namespace_blk_validate
);
357 static int nd_namespace_label_update(struct nd_region
*nd_region
,
360 dev_WARN_ONCE(dev
, dev
->driver
|| to_ndns(dev
)->claim
,
361 "namespace must be idle during label update\n");
362 if (dev
->driver
|| to_ndns(dev
)->claim
)
366 * Only allow label writes that will result in a valid namespace
367 * or deletion of an existing namespace.
369 if (is_namespace_pmem(dev
)) {
370 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
371 resource_size_t size
= resource_size(&nspm
->nsio
.res
);
373 if (size
== 0 && nspm
->uuid
)
374 /* delete allocation */;
375 else if (!nspm
->uuid
)
378 return nd_pmem_namespace_label_update(nd_region
, nspm
, size
);
379 } else if (is_namespace_blk(dev
)) {
380 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
381 resource_size_t size
= nd_namespace_blk_size(nsblk
);
383 if (size
== 0 && nsblk
->uuid
)
384 /* delete allocation */;
385 else if (!nsblk
->uuid
|| !nsblk
->lbasize
)
388 return nd_blk_namespace_label_update(nd_region
, nsblk
, size
);
393 static ssize_t
alt_name_store(struct device
*dev
,
394 struct device_attribute
*attr
, const char *buf
, size_t len
)
396 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
400 nvdimm_bus_lock(dev
);
401 wait_nvdimm_bus_probe_idle(dev
);
402 rc
= __alt_name_store(dev
, buf
, len
);
404 rc
= nd_namespace_label_update(nd_region
, dev
);
405 dev_dbg(dev
, "%s: %s(%zd)\n", __func__
, rc
< 0 ? "fail " : "", rc
);
406 nvdimm_bus_unlock(dev
);
409 return rc
< 0 ? rc
: len
;
412 static ssize_t
alt_name_show(struct device
*dev
,
413 struct device_attribute
*attr
, char *buf
)
417 if (is_namespace_pmem(dev
)) {
418 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
420 ns_altname
= nspm
->alt_name
;
421 } else if (is_namespace_blk(dev
)) {
422 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
424 ns_altname
= nsblk
->alt_name
;
428 return sprintf(buf
, "%s\n", ns_altname
? ns_altname
: "");
430 static DEVICE_ATTR_RW(alt_name
);
432 static int scan_free(struct nd_region
*nd_region
,
433 struct nd_mapping
*nd_mapping
, struct nd_label_id
*label_id
,
436 bool is_blk
= strncmp(label_id
->id
, "blk", 3) == 0;
437 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
441 struct resource
*res
, *last
;
442 resource_size_t new_start
;
445 for_each_dpa_resource(ndd
, res
)
446 if (strcmp(res
->name
, label_id
->id
) == 0)
452 if (n
>= resource_size(res
)) {
453 n
-= resource_size(res
);
454 nd_dbg_dpa(nd_region
, ndd
, res
, "delete %d\n", rc
);
455 nvdimm_free_dpa(ndd
, res
);
456 /* retry with last resource deleted */
461 * Keep BLK allocations relegated to high DPA as much as
465 new_start
= res
->start
+ n
;
467 new_start
= res
->start
;
469 rc
= adjust_resource(res
, new_start
, resource_size(res
) - n
);
471 res
->flags
|= DPA_RESOURCE_ADJUSTED
;
472 nd_dbg_dpa(nd_region
, ndd
, res
, "shrink %d\n", rc
);
480 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
481 * @nd_region: the set of dimms to reclaim @n bytes from
482 * @label_id: unique identifier for the namespace consuming this dpa range
483 * @n: number of bytes per-dimm to release
485 * Assumes resources are ordered. Starting from the end try to
486 * adjust_resource() the allocation to @n, but if @n is larger than the
487 * allocation delete it and find the 'new' last allocation in the label
490 static int shrink_dpa_allocation(struct nd_region
*nd_region
,
491 struct nd_label_id
*label_id
, resource_size_t n
)
495 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
496 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
499 rc
= scan_free(nd_region
, nd_mapping
, label_id
, n
);
507 static resource_size_t
init_dpa_allocation(struct nd_label_id
*label_id
,
508 struct nd_region
*nd_region
, struct nd_mapping
*nd_mapping
,
511 bool is_blk
= strncmp(label_id
->id
, "blk", 3) == 0;
512 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
513 resource_size_t first_dpa
;
514 struct resource
*res
;
517 /* allocate blk from highest dpa first */
519 first_dpa
= nd_mapping
->start
+ nd_mapping
->size
- n
;
521 first_dpa
= nd_mapping
->start
;
523 /* first resource allocation for this label-id or dimm */
524 res
= nvdimm_allocate_dpa(ndd
, label_id
, first_dpa
, n
);
528 nd_dbg_dpa(nd_region
, ndd
, res
, "init %d\n", rc
);
534 * space_valid() - validate free dpa space against constraints
535 * @nd_region: hosting region of the free space
536 * @ndd: dimm device data for debug
537 * @label_id: namespace id to allocate space
538 * @prev: potential allocation that precedes free space
539 * @next: allocation that follows the given free space range
540 * @exist: first allocation with same id in the mapping
541 * @n: range that must satisfied for pmem allocations
542 * @valid: free space range to validate
544 * BLK-space is valid as long as it does not precede a PMEM
545 * allocation in a given region. PMEM-space must be contiguous
546 * and adjacent to an existing existing allocation (if one
547 * exists). If reserving PMEM any space is valid.
549 static void space_valid(struct nd_region
*nd_region
, struct nvdimm_drvdata
*ndd
,
550 struct nd_label_id
*label_id
, struct resource
*prev
,
551 struct resource
*next
, struct resource
*exist
,
552 resource_size_t n
, struct resource
*valid
)
554 bool is_reserve
= strcmp(label_id
->id
, "pmem-reserve") == 0;
555 bool is_pmem
= strncmp(label_id
->id
, "pmem", 4) == 0;
557 if (valid
->start
>= valid
->end
)
564 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
565 struct nvdimm_bus
*nvdimm_bus
;
566 struct blk_alloc_info info
= {
567 .nd_mapping
= nd_mapping
,
568 .available
= nd_mapping
->size
,
572 WARN_ON(!is_nd_blk(&nd_region
->dev
));
573 nvdimm_bus
= walk_to_nvdimm_bus(&nd_region
->dev
);
574 device_for_each_child(&nvdimm_bus
->dev
, &info
, alias_dpa_busy
);
578 /* allocation needs to be contiguous, so this is all or nothing */
579 if (resource_size(valid
) < n
)
582 /* we've got all the space we need and no existing allocation */
586 /* allocation needs to be contiguous with the existing namespace */
587 if (valid
->start
== exist
->end
+ 1
588 || valid
->end
== exist
->start
- 1)
592 /* truncate @valid size to 0 */
593 valid
->end
= valid
->start
- 1;
597 ALLOC_ERR
= 0, ALLOC_BEFORE
, ALLOC_MID
, ALLOC_AFTER
,
600 static resource_size_t
scan_allocate(struct nd_region
*nd_region
,
601 struct nd_mapping
*nd_mapping
, struct nd_label_id
*label_id
,
604 resource_size_t mapping_end
= nd_mapping
->start
+ nd_mapping
->size
- 1;
605 bool is_pmem
= strncmp(label_id
->id
, "pmem", 4) == 0;
606 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
607 struct resource
*res
, *exist
= NULL
, valid
;
608 const resource_size_t to_allocate
= n
;
611 for_each_dpa_resource(ndd
, res
)
612 if (strcmp(label_id
->id
, res
->name
) == 0)
615 valid
.start
= nd_mapping
->start
;
616 valid
.end
= mapping_end
;
617 valid
.name
= "free space";
620 for_each_dpa_resource(ndd
, res
) {
621 struct resource
*next
= res
->sibling
, *new_res
= NULL
;
622 resource_size_t allocate
, available
= 0;
623 enum alloc_loc loc
= ALLOC_ERR
;
627 /* ignore resources outside this nd_mapping */
628 if (res
->start
> mapping_end
)
630 if (res
->end
< nd_mapping
->start
)
633 /* space at the beginning of the mapping */
634 if (!first
++ && res
->start
> nd_mapping
->start
) {
635 valid
.start
= nd_mapping
->start
;
636 valid
.end
= res
->start
- 1;
637 space_valid(nd_region
, ndd
, label_id
, NULL
, next
, exist
,
638 to_allocate
, &valid
);
639 available
= resource_size(&valid
);
644 /* space between allocations */
646 valid
.start
= res
->start
+ resource_size(res
);
647 valid
.end
= min(mapping_end
, next
->start
- 1);
648 space_valid(nd_region
, ndd
, label_id
, res
, next
, exist
,
649 to_allocate
, &valid
);
650 available
= resource_size(&valid
);
655 /* space at the end of the mapping */
657 valid
.start
= res
->start
+ resource_size(res
);
658 valid
.end
= mapping_end
;
659 space_valid(nd_region
, ndd
, label_id
, res
, next
, exist
,
660 to_allocate
, &valid
);
661 available
= resource_size(&valid
);
666 if (!loc
|| !available
)
668 allocate
= min(available
, n
);
671 if (strcmp(res
->name
, label_id
->id
) == 0) {
672 /* adjust current resource up */
673 rc
= adjust_resource(res
, res
->start
- allocate
,
674 resource_size(res
) + allocate
);
675 action
= "cur grow up";
680 if (strcmp(next
->name
, label_id
->id
) == 0) {
681 /* adjust next resource up */
682 rc
= adjust_resource(next
, next
->start
683 - allocate
, resource_size(next
)
686 action
= "next grow up";
687 } else if (strcmp(res
->name
, label_id
->id
) == 0) {
688 action
= "grow down";
693 if (strcmp(res
->name
, label_id
->id
) == 0)
694 action
= "grow down";
702 if (strcmp(action
, "allocate") == 0) {
703 /* BLK allocate bottom up */
705 valid
.start
+= available
- allocate
;
707 new_res
= nvdimm_allocate_dpa(ndd
, label_id
,
708 valid
.start
, allocate
);
711 } else if (strcmp(action
, "grow down") == 0) {
712 /* adjust current resource down */
713 rc
= adjust_resource(res
, res
->start
, resource_size(res
)
716 res
->flags
|= DPA_RESOURCE_ADJUSTED
;
722 nd_dbg_dpa(nd_region
, ndd
, new_res
, "%s(%d) %d\n",
731 * Retry scan with newly inserted resources.
732 * For example, if we did an ALLOC_BEFORE
733 * insertion there may also have been space
734 * available for an ALLOC_AFTER insertion, so we
735 * need to check this same resource again
743 * If we allocated nothing in the BLK case it may be because we are in
744 * an initial "pmem-reserve pass". Only do an initial BLK allocation
745 * when none of the DPA space is reserved.
747 if ((is_pmem
|| !ndd
->dpa
.child
) && n
== to_allocate
)
748 return init_dpa_allocation(label_id
, nd_region
, nd_mapping
, n
);
752 static int merge_dpa(struct nd_region
*nd_region
,
753 struct nd_mapping
*nd_mapping
, struct nd_label_id
*label_id
)
755 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
756 struct resource
*res
;
758 if (strncmp("pmem", label_id
->id
, 4) == 0)
761 for_each_dpa_resource(ndd
, res
) {
763 struct resource
*next
= res
->sibling
;
764 resource_size_t end
= res
->start
+ resource_size(res
);
766 if (!next
|| strcmp(res
->name
, label_id
->id
) != 0
767 || strcmp(next
->name
, label_id
->id
) != 0
768 || end
!= next
->start
)
770 end
+= resource_size(next
);
771 nvdimm_free_dpa(ndd
, next
);
772 rc
= adjust_resource(res
, res
->start
, end
- res
->start
);
773 nd_dbg_dpa(nd_region
, ndd
, res
, "merge %d\n", rc
);
776 res
->flags
|= DPA_RESOURCE_ADJUSTED
;
783 static int __reserve_free_pmem(struct device
*dev
, void *data
)
785 struct nvdimm
*nvdimm
= data
;
786 struct nd_region
*nd_region
;
787 struct nd_label_id label_id
;
790 if (!is_nd_pmem(dev
))
793 nd_region
= to_nd_region(dev
);
794 if (nd_region
->ndr_mappings
== 0)
797 memset(&label_id
, 0, sizeof(label_id
));
798 strcat(label_id
.id
, "pmem-reserve");
799 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
800 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
801 resource_size_t n
, rem
= 0;
803 if (nd_mapping
->nvdimm
!= nvdimm
)
806 n
= nd_pmem_available_dpa(nd_region
, nd_mapping
, &rem
);
809 rem
= scan_allocate(nd_region
, nd_mapping
, &label_id
, n
);
810 dev_WARN_ONCE(&nd_region
->dev
, rem
,
811 "pmem reserve underrun: %#llx of %#llx bytes\n",
812 (unsigned long long) n
- rem
,
813 (unsigned long long) n
);
814 return rem
? -ENXIO
: 0;
820 static void release_free_pmem(struct nvdimm_bus
*nvdimm_bus
,
821 struct nd_mapping
*nd_mapping
)
823 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
824 struct resource
*res
, *_res
;
826 for_each_dpa_resource_safe(ndd
, res
, _res
)
827 if (strcmp(res
->name
, "pmem-reserve") == 0)
828 nvdimm_free_dpa(ndd
, res
);
831 static int reserve_free_pmem(struct nvdimm_bus
*nvdimm_bus
,
832 struct nd_mapping
*nd_mapping
)
834 struct nvdimm
*nvdimm
= nd_mapping
->nvdimm
;
837 rc
= device_for_each_child(&nvdimm_bus
->dev
, nvdimm
,
838 __reserve_free_pmem
);
840 release_free_pmem(nvdimm_bus
, nd_mapping
);
845 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
846 * @nd_region: the set of dimms to allocate @n more bytes from
847 * @label_id: unique identifier for the namespace consuming this dpa range
848 * @n: number of bytes per-dimm to add to the existing allocation
850 * Assumes resources are ordered. For BLK regions, first consume
851 * BLK-only available DPA free space, then consume PMEM-aliased DPA
852 * space starting at the highest DPA. For PMEM regions start
853 * allocations from the start of an interleave set and end at the first
854 * BLK allocation or the end of the interleave set, whichever comes
857 static int grow_dpa_allocation(struct nd_region
*nd_region
,
858 struct nd_label_id
*label_id
, resource_size_t n
)
860 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(&nd_region
->dev
);
861 bool is_pmem
= strncmp(label_id
->id
, "pmem", 4) == 0;
864 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
865 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
866 resource_size_t rem
= n
;
870 * In the BLK case try once with all unallocated PMEM
871 * reserved, and once without
873 for (j
= is_pmem
; j
< 2; j
++) {
874 bool blk_only
= j
== 0;
877 rc
= reserve_free_pmem(nvdimm_bus
, nd_mapping
);
881 rem
= scan_allocate(nd_region
, nd_mapping
,
884 release_free_pmem(nvdimm_bus
, nd_mapping
);
886 /* try again and allow encroachments into PMEM */
891 dev_WARN_ONCE(&nd_region
->dev
, rem
,
892 "allocation underrun: %#llx of %#llx bytes\n",
893 (unsigned long long) n
- rem
,
894 (unsigned long long) n
);
898 rc
= merge_dpa(nd_region
, nd_mapping
, label_id
);
906 static void nd_namespace_pmem_set_resource(struct nd_region
*nd_region
,
907 struct nd_namespace_pmem
*nspm
, resource_size_t size
)
909 struct resource
*res
= &nspm
->nsio
.res
;
910 resource_size_t offset
= 0;
912 if (size
&& !nspm
->uuid
) {
917 if (size
&& nspm
->uuid
) {
918 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
919 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
920 struct nd_label_id label_id
;
921 struct resource
*res
;
928 nd_label_gen_id(&label_id
, nspm
->uuid
, 0);
930 /* calculate a spa offset from the dpa allocation offset */
931 for_each_dpa_resource(ndd
, res
)
932 if (strcmp(res
->name
, label_id
.id
) == 0) {
933 offset
= (res
->start
- nd_mapping
->start
)
934 * nd_region
->ndr_mappings
;
943 res
->start
= nd_region
->ndr_start
+ offset
;
944 res
->end
= res
->start
+ size
- 1;
947 static bool uuid_not_set(const u8
*uuid
, struct device
*dev
, const char *where
)
950 dev_dbg(dev
, "%s: uuid not set\n", where
);
956 static ssize_t
__size_store(struct device
*dev
, unsigned long long val
)
958 resource_size_t allocated
= 0, available
= 0;
959 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
960 struct nd_namespace_common
*ndns
= to_ndns(dev
);
961 struct nd_mapping
*nd_mapping
;
962 struct nvdimm_drvdata
*ndd
;
963 struct nd_label_id label_id
;
964 u32 flags
= 0, remainder
;
968 if (dev
->driver
|| ndns
->claim
)
971 if (is_namespace_pmem(dev
)) {
972 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
976 } else if (is_namespace_blk(dev
)) {
977 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
980 flags
= NSLABEL_FLAG_LOCAL
;
985 * We need a uuid for the allocation-label and dimm(s) on which
986 * to store the label.
988 if (uuid_not_set(uuid
, dev
, __func__
))
990 if (nd_region
->ndr_mappings
== 0) {
991 dev_dbg(dev
, "%s: not associated with dimm(s)\n", __func__
);
995 div_u64_rem(val
, SZ_4K
* nd_region
->ndr_mappings
, &remainder
);
997 dev_dbg(dev
, "%llu is not %dK aligned\n", val
,
998 (SZ_4K
* nd_region
->ndr_mappings
) / SZ_1K
);
1002 nd_label_gen_id(&label_id
, uuid
, flags
);
1003 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1004 nd_mapping
= &nd_region
->mapping
[i
];
1005 ndd
= to_ndd(nd_mapping
);
1008 * All dimms in an interleave set, or the base dimm for a blk
1009 * region, need to be enabled for the size to be changed.
1014 allocated
+= nvdimm_allocated_dpa(ndd
, &label_id
);
1016 available
= nd_region_available_dpa(nd_region
);
1018 if (val
> available
+ allocated
)
1021 if (val
== allocated
)
1024 val
= div_u64(val
, nd_region
->ndr_mappings
);
1025 allocated
= div_u64(allocated
, nd_region
->ndr_mappings
);
1026 if (val
< allocated
)
1027 rc
= shrink_dpa_allocation(nd_region
, &label_id
,
1030 rc
= grow_dpa_allocation(nd_region
, &label_id
, val
- allocated
);
1035 if (is_namespace_pmem(dev
)) {
1036 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1038 nd_namespace_pmem_set_resource(nd_region
, nspm
,
1039 val
* nd_region
->ndr_mappings
);
1043 * Try to delete the namespace if we deleted all of its
1044 * allocation, this is not the seed or 0th device for the
1045 * region, and it is not actively claimed by a btt, pfn, or dax
1048 if (val
== 0 && id
!= 0 && nd_region
->ns_seed
!= dev
&& !ndns
->claim
)
1049 nd_device_unregister(dev
, ND_ASYNC
);
1054 static ssize_t
size_store(struct device
*dev
,
1055 struct device_attribute
*attr
, const char *buf
, size_t len
)
1057 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
1058 unsigned long long val
;
1062 rc
= kstrtoull(buf
, 0, &val
);
1067 nvdimm_bus_lock(dev
);
1068 wait_nvdimm_bus_probe_idle(dev
);
1069 rc
= __size_store(dev
, val
);
1071 rc
= nd_namespace_label_update(nd_region
, dev
);
1073 if (is_namespace_pmem(dev
)) {
1074 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1077 } else if (is_namespace_blk(dev
)) {
1078 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1080 uuid
= &nsblk
->uuid
;
1083 if (rc
== 0 && val
== 0 && uuid
) {
1084 /* setting size zero == 'delete namespace' */
1089 dev_dbg(dev
, "%s: %llx %s (%d)\n", __func__
, val
, rc
< 0
1090 ? "fail" : "success", rc
);
1092 nvdimm_bus_unlock(dev
);
1095 return rc
< 0 ? rc
: len
;
1098 resource_size_t
__nvdimm_namespace_capacity(struct nd_namespace_common
*ndns
)
1100 struct device
*dev
= &ndns
->dev
;
1102 if (is_namespace_pmem(dev
)) {
1103 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1105 return resource_size(&nspm
->nsio
.res
);
1106 } else if (is_namespace_blk(dev
)) {
1107 return nd_namespace_blk_size(to_nd_namespace_blk(dev
));
1108 } else if (is_namespace_io(dev
)) {
1109 struct nd_namespace_io
*nsio
= to_nd_namespace_io(dev
);
1111 return resource_size(&nsio
->res
);
1113 WARN_ONCE(1, "unknown namespace type\n");
1117 resource_size_t
nvdimm_namespace_capacity(struct nd_namespace_common
*ndns
)
1119 resource_size_t size
;
1121 nvdimm_bus_lock(&ndns
->dev
);
1122 size
= __nvdimm_namespace_capacity(ndns
);
1123 nvdimm_bus_unlock(&ndns
->dev
);
1127 EXPORT_SYMBOL(nvdimm_namespace_capacity
);
1129 static ssize_t
size_show(struct device
*dev
,
1130 struct device_attribute
*attr
, char *buf
)
1132 return sprintf(buf
, "%llu\n", (unsigned long long)
1133 nvdimm_namespace_capacity(to_ndns(dev
)));
1135 static DEVICE_ATTR(size
, 0444, size_show
, size_store
);
1137 static u8
*namespace_to_uuid(struct device
*dev
)
1139 if (is_namespace_pmem(dev
)) {
1140 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1143 } else if (is_namespace_blk(dev
)) {
1144 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1148 return ERR_PTR(-ENXIO
);
1151 static ssize_t
uuid_show(struct device
*dev
,
1152 struct device_attribute
*attr
, char *buf
)
1154 u8
*uuid
= namespace_to_uuid(dev
);
1157 return PTR_ERR(uuid
);
1159 return sprintf(buf
, "%pUb\n", uuid
);
1160 return sprintf(buf
, "\n");
1164 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1165 * @nd_region: parent region so we can updates all dimms in the set
1166 * @dev: namespace type for generating label_id
1167 * @new_uuid: incoming uuid
1168 * @old_uuid: reference to the uuid storage location in the namespace object
1170 static int namespace_update_uuid(struct nd_region
*nd_region
,
1171 struct device
*dev
, u8
*new_uuid
, u8
**old_uuid
)
1173 u32 flags
= is_namespace_blk(dev
) ? NSLABEL_FLAG_LOCAL
: 0;
1174 struct nd_label_id old_label_id
;
1175 struct nd_label_id new_label_id
;
1178 if (!nd_is_uuid_unique(dev
, new_uuid
))
1181 if (*old_uuid
== NULL
)
1185 * If we've already written a label with this uuid, then it's
1186 * too late to rename because we can't reliably update the uuid
1187 * without losing the old namespace. Userspace must delete this
1188 * namespace to abandon the old uuid.
1190 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1191 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
1194 * This check by itself is sufficient because old_uuid
1195 * would be NULL above if this uuid did not exist in the
1196 * currently written set.
1198 * FIXME: can we delete uuid with zero dpa allocated?
1200 if (list_empty(&nd_mapping
->labels
))
1204 nd_label_gen_id(&old_label_id
, *old_uuid
, flags
);
1205 nd_label_gen_id(&new_label_id
, new_uuid
, flags
);
1206 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1207 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
1208 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
1209 struct resource
*res
;
1211 for_each_dpa_resource(ndd
, res
)
1212 if (strcmp(res
->name
, old_label_id
.id
) == 0)
1213 sprintf((void *) res
->name
, "%s",
1218 *old_uuid
= new_uuid
;
1222 static ssize_t
uuid_store(struct device
*dev
,
1223 struct device_attribute
*attr
, const char *buf
, size_t len
)
1225 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
1230 if (is_namespace_pmem(dev
)) {
1231 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1233 ns_uuid
= &nspm
->uuid
;
1234 } else if (is_namespace_blk(dev
)) {
1235 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1237 ns_uuid
= &nsblk
->uuid
;
1242 nvdimm_bus_lock(dev
);
1243 wait_nvdimm_bus_probe_idle(dev
);
1244 if (to_ndns(dev
)->claim
)
1247 rc
= nd_uuid_store(dev
, &uuid
, buf
, len
);
1249 rc
= namespace_update_uuid(nd_region
, dev
, uuid
, ns_uuid
);
1251 rc
= nd_namespace_label_update(nd_region
, dev
);
1254 dev_dbg(dev
, "%s: result: %zd wrote: %s%s", __func__
,
1255 rc
, buf
, buf
[len
- 1] == '\n' ? "" : "\n");
1256 nvdimm_bus_unlock(dev
);
1259 return rc
< 0 ? rc
: len
;
1261 static DEVICE_ATTR_RW(uuid
);
1263 static ssize_t
resource_show(struct device
*dev
,
1264 struct device_attribute
*attr
, char *buf
)
1266 struct resource
*res
;
1268 if (is_namespace_pmem(dev
)) {
1269 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1271 res
= &nspm
->nsio
.res
;
1272 } else if (is_namespace_io(dev
)) {
1273 struct nd_namespace_io
*nsio
= to_nd_namespace_io(dev
);
1279 /* no address to convey if the namespace has no allocation */
1280 if (resource_size(res
) == 0)
1282 return sprintf(buf
, "%#llx\n", (unsigned long long) res
->start
);
1284 static DEVICE_ATTR_RO(resource
);
1286 static const unsigned long ns_lbasize_supported
[] = { 512, 520, 528,
1287 4096, 4104, 4160, 4224, 0 };
1289 static ssize_t
sector_size_show(struct device
*dev
,
1290 struct device_attribute
*attr
, char *buf
)
1292 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1294 if (!is_namespace_blk(dev
))
1297 return nd_sector_size_show(nsblk
->lbasize
, ns_lbasize_supported
, buf
);
1300 static ssize_t
sector_size_store(struct device
*dev
,
1301 struct device_attribute
*attr
, const char *buf
, size_t len
)
1303 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1304 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
1307 if (!is_namespace_blk(dev
))
1311 nvdimm_bus_lock(dev
);
1312 if (to_ndns(dev
)->claim
)
1315 rc
= nd_sector_size_store(dev
, buf
, &nsblk
->lbasize
,
1316 ns_lbasize_supported
);
1318 rc
= nd_namespace_label_update(nd_region
, dev
);
1319 dev_dbg(dev
, "%s: result: %zd %s: %s%s", __func__
,
1320 rc
, rc
< 0 ? "tried" : "wrote", buf
,
1321 buf
[len
- 1] == '\n' ? "" : "\n");
1322 nvdimm_bus_unlock(dev
);
1325 return rc
? rc
: len
;
1327 static DEVICE_ATTR_RW(sector_size
);
1329 static ssize_t
dpa_extents_show(struct device
*dev
,
1330 struct device_attribute
*attr
, char *buf
)
1332 struct nd_region
*nd_region
= to_nd_region(dev
->parent
);
1333 struct nd_label_id label_id
;
1338 nvdimm_bus_lock(dev
);
1339 if (is_namespace_pmem(dev
)) {
1340 struct nd_namespace_pmem
*nspm
= to_nd_namespace_pmem(dev
);
1344 } else if (is_namespace_blk(dev
)) {
1345 struct nd_namespace_blk
*nsblk
= to_nd_namespace_blk(dev
);
1348 flags
= NSLABEL_FLAG_LOCAL
;
1354 nd_label_gen_id(&label_id
, uuid
, flags
);
1355 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1356 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
1357 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
1358 struct resource
*res
;
1360 for_each_dpa_resource(ndd
, res
)
1361 if (strcmp(res
->name
, label_id
.id
) == 0)
1365 nvdimm_bus_unlock(dev
);
1367 return sprintf(buf
, "%d\n", count
);
1369 static DEVICE_ATTR_RO(dpa_extents
);
1371 static ssize_t
holder_show(struct device
*dev
,
1372 struct device_attribute
*attr
, char *buf
)
1374 struct nd_namespace_common
*ndns
= to_ndns(dev
);
1378 rc
= sprintf(buf
, "%s\n", ndns
->claim
? dev_name(ndns
->claim
) : "");
1383 static DEVICE_ATTR_RO(holder
);
1385 static ssize_t
mode_show(struct device
*dev
,
1386 struct device_attribute
*attr
, char *buf
)
1388 struct nd_namespace_common
*ndns
= to_ndns(dev
);
1389 struct device
*claim
;
1394 claim
= ndns
->claim
;
1395 if (claim
&& is_nd_btt(claim
))
1397 else if (claim
&& is_nd_pfn(claim
))
1399 else if (claim
&& is_nd_dax(claim
))
1401 else if (!claim
&& pmem_should_map_pages(dev
))
1405 rc
= sprintf(buf
, "%s\n", mode
);
1410 static DEVICE_ATTR_RO(mode
);
1412 static ssize_t
force_raw_store(struct device
*dev
,
1413 struct device_attribute
*attr
, const char *buf
, size_t len
)
1416 int rc
= strtobool(buf
, &force_raw
);
1421 to_ndns(dev
)->force_raw
= force_raw
;
1425 static ssize_t
force_raw_show(struct device
*dev
,
1426 struct device_attribute
*attr
, char *buf
)
1428 return sprintf(buf
, "%d\n", to_ndns(dev
)->force_raw
);
1430 static DEVICE_ATTR_RW(force_raw
);
1432 static struct attribute
*nd_namespace_attributes
[] = {
1433 &dev_attr_nstype
.attr
,
1434 &dev_attr_size
.attr
,
1435 &dev_attr_mode
.attr
,
1436 &dev_attr_uuid
.attr
,
1437 &dev_attr_holder
.attr
,
1438 &dev_attr_resource
.attr
,
1439 &dev_attr_alt_name
.attr
,
1440 &dev_attr_force_raw
.attr
,
1441 &dev_attr_sector_size
.attr
,
1442 &dev_attr_dpa_extents
.attr
,
1446 static umode_t
namespace_visible(struct kobject
*kobj
,
1447 struct attribute
*a
, int n
)
1449 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
1451 if (a
== &dev_attr_resource
.attr
) {
1452 if (is_namespace_blk(dev
))
1457 if (is_namespace_pmem(dev
) || is_namespace_blk(dev
)) {
1458 if (a
== &dev_attr_size
.attr
)
1461 if (is_namespace_pmem(dev
) && a
== &dev_attr_sector_size
.attr
)
1467 if (a
== &dev_attr_nstype
.attr
|| a
== &dev_attr_size
.attr
1468 || a
== &dev_attr_holder
.attr
1469 || a
== &dev_attr_force_raw
.attr
1470 || a
== &dev_attr_mode
.attr
)
1476 static struct attribute_group nd_namespace_attribute_group
= {
1477 .attrs
= nd_namespace_attributes
,
1478 .is_visible
= namespace_visible
,
1481 static const struct attribute_group
*nd_namespace_attribute_groups
[] = {
1482 &nd_device_attribute_group
,
1483 &nd_namespace_attribute_group
,
1484 &nd_numa_attribute_group
,
1488 struct nd_namespace_common
*nvdimm_namespace_common_probe(struct device
*dev
)
1490 struct nd_btt
*nd_btt
= is_nd_btt(dev
) ? to_nd_btt(dev
) : NULL
;
1491 struct nd_pfn
*nd_pfn
= is_nd_pfn(dev
) ? to_nd_pfn(dev
) : NULL
;
1492 struct nd_dax
*nd_dax
= is_nd_dax(dev
) ? to_nd_dax(dev
) : NULL
;
1493 struct nd_namespace_common
*ndns
= NULL
;
1494 resource_size_t size
;
1496 if (nd_btt
|| nd_pfn
|| nd_dax
) {
1498 ndns
= nd_btt
->ndns
;
1500 ndns
= nd_pfn
->ndns
;
1502 ndns
= nd_dax
->nd_pfn
.ndns
;
1505 return ERR_PTR(-ENODEV
);
1508 * Flush any in-progess probes / removals in the driver
1509 * for the raw personality of this namespace.
1511 device_lock(&ndns
->dev
);
1512 device_unlock(&ndns
->dev
);
1513 if (ndns
->dev
.driver
) {
1514 dev_dbg(&ndns
->dev
, "is active, can't bind %s\n",
1516 return ERR_PTR(-EBUSY
);
1518 if (dev_WARN_ONCE(&ndns
->dev
, ndns
->claim
!= dev
,
1519 "host (%s) vs claim (%s) mismatch\n",
1521 dev_name(ndns
->claim
)))
1522 return ERR_PTR(-ENXIO
);
1524 ndns
= to_ndns(dev
);
1526 dev_dbg(dev
, "claimed by %s, failing probe\n",
1527 dev_name(ndns
->claim
));
1529 return ERR_PTR(-ENXIO
);
1533 size
= nvdimm_namespace_capacity(ndns
);
1534 if (size
< ND_MIN_NAMESPACE_SIZE
) {
1535 dev_dbg(&ndns
->dev
, "%pa, too small must be at least %#x\n",
1536 &size
, ND_MIN_NAMESPACE_SIZE
);
1537 return ERR_PTR(-ENODEV
);
1540 if (is_namespace_pmem(&ndns
->dev
)) {
1541 struct nd_namespace_pmem
*nspm
;
1543 nspm
= to_nd_namespace_pmem(&ndns
->dev
);
1544 if (uuid_not_set(nspm
->uuid
, &ndns
->dev
, __func__
))
1545 return ERR_PTR(-ENODEV
);
1546 } else if (is_namespace_blk(&ndns
->dev
)) {
1547 struct nd_namespace_blk
*nsblk
;
1549 nsblk
= to_nd_namespace_blk(&ndns
->dev
);
1550 if (uuid_not_set(nsblk
->uuid
, &ndns
->dev
, __func__
))
1551 return ERR_PTR(-ENODEV
);
1552 if (!nsblk
->lbasize
) {
1553 dev_dbg(&ndns
->dev
, "%s: sector size not set\n",
1555 return ERR_PTR(-ENODEV
);
1557 if (!nd_namespace_blk_validate(nsblk
))
1558 return ERR_PTR(-ENODEV
);
1563 EXPORT_SYMBOL(nvdimm_namespace_common_probe
);
1565 static struct device
**create_namespace_io(struct nd_region
*nd_region
)
1567 struct nd_namespace_io
*nsio
;
1568 struct device
*dev
, **devs
;
1569 struct resource
*res
;
1571 nsio
= kzalloc(sizeof(*nsio
), GFP_KERNEL
);
1575 devs
= kcalloc(2, sizeof(struct device
*), GFP_KERNEL
);
1581 dev
= &nsio
->common
.dev
;
1582 dev
->type
= &namespace_io_device_type
;
1583 dev
->parent
= &nd_region
->dev
;
1585 res
->name
= dev_name(&nd_region
->dev
);
1586 res
->flags
= IORESOURCE_MEM
;
1587 res
->start
= nd_region
->ndr_start
;
1588 res
->end
= res
->start
+ nd_region
->ndr_size
- 1;
1594 static bool has_uuid_at_pos(struct nd_region
*nd_region
, u8
*uuid
,
1595 u64 cookie
, u16 pos
)
1597 struct nd_namespace_label
*found
= NULL
;
1600 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1601 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
1602 struct nd_label_ent
*label_ent
;
1603 bool found_uuid
= false;
1605 list_for_each_entry(label_ent
, &nd_mapping
->labels
, list
) {
1606 struct nd_namespace_label
*nd_label
= label_ent
->label
;
1607 u16 position
, nlabel
;
1612 isetcookie
= __le64_to_cpu(nd_label
->isetcookie
);
1613 position
= __le16_to_cpu(nd_label
->position
);
1614 nlabel
= __le16_to_cpu(nd_label
->nlabel
);
1616 if (isetcookie
!= cookie
)
1619 if (memcmp(nd_label
->uuid
, uuid
, NSLABEL_UUID_LEN
) != 0)
1623 dev_dbg(to_ndd(nd_mapping
)->dev
,
1624 "%s duplicate entry for uuid\n",
1629 if (nlabel
!= nd_region
->ndr_mappings
)
1631 if (position
!= pos
)
1639 return found
!= NULL
;
1642 static int select_pmem_id(struct nd_region
*nd_region
, u8
*pmem_id
)
1649 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1650 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
1651 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
1652 struct nd_namespace_label
*nd_label
= NULL
;
1653 u64 hw_start
, hw_end
, pmem_start
, pmem_end
;
1654 struct nd_label_ent
*label_ent
;
1656 lockdep_assert_held(&nd_mapping
->lock
);
1657 list_for_each_entry(label_ent
, &nd_mapping
->labels
, list
) {
1658 nd_label
= label_ent
->label
;
1661 if (memcmp(nd_label
->uuid
, pmem_id
, NSLABEL_UUID_LEN
) == 0)
1672 * Check that this label is compliant with the dpa
1673 * range published in NFIT
1675 hw_start
= nd_mapping
->start
;
1676 hw_end
= hw_start
+ nd_mapping
->size
;
1677 pmem_start
= __le64_to_cpu(nd_label
->dpa
);
1678 pmem_end
= pmem_start
+ __le64_to_cpu(nd_label
->rawsize
);
1679 if (pmem_start
>= hw_start
&& pmem_start
< hw_end
1680 && pmem_end
<= hw_end
&& pmem_end
> hw_start
)
1683 dev_dbg(&nd_region
->dev
, "%s invalid label for %pUb\n",
1684 dev_name(ndd
->dev
), nd_label
->uuid
);
1688 /* move recently validated label to the front of the list */
1689 list_move(&label_ent
->list
, &nd_mapping
->labels
);
1695 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1696 * @nd_region: region with mappings to validate
1697 * @nspm: target namespace to create
1698 * @nd_label: target pmem namespace label to evaluate
1700 struct device
*create_namespace_pmem(struct nd_region
*nd_region
,
1701 struct nd_namespace_label
*nd_label
)
1703 u64 altcookie
= nd_region_interleave_set_altcookie(nd_region
);
1704 u64 cookie
= nd_region_interleave_set_cookie(nd_region
);
1705 struct nd_label_ent
*label_ent
;
1706 struct nd_namespace_pmem
*nspm
;
1707 struct nd_mapping
*nd_mapping
;
1708 resource_size_t size
= 0;
1709 struct resource
*res
;
1715 dev_dbg(&nd_region
->dev
, "invalid interleave-set-cookie\n");
1716 return ERR_PTR(-ENXIO
);
1719 if (__le64_to_cpu(nd_label
->isetcookie
) != cookie
) {
1720 dev_dbg(&nd_region
->dev
, "invalid cookie in label: %pUb\n",
1722 if (__le64_to_cpu(nd_label
->isetcookie
) != altcookie
)
1723 return ERR_PTR(-EAGAIN
);
1725 dev_dbg(&nd_region
->dev
, "valid altcookie in label: %pUb\n",
1729 nspm
= kzalloc(sizeof(*nspm
), GFP_KERNEL
);
1731 return ERR_PTR(-ENOMEM
);
1734 dev
= &nspm
->nsio
.common
.dev
;
1735 dev
->type
= &namespace_pmem_device_type
;
1736 dev
->parent
= &nd_region
->dev
;
1737 res
= &nspm
->nsio
.res
;
1738 res
->name
= dev_name(&nd_region
->dev
);
1739 res
->flags
= IORESOURCE_MEM
;
1741 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1742 if (has_uuid_at_pos(nd_region
, nd_label
->uuid
, cookie
, i
))
1744 if (has_uuid_at_pos(nd_region
, nd_label
->uuid
, altcookie
, i
))
1749 if (i
< nd_region
->ndr_mappings
) {
1750 struct nvdimm_drvdata
*ndd
= to_ndd(&nd_region
->mapping
[i
]);
1753 * Give up if we don't find an instance of a uuid at each
1754 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1755 * find a dimm with two instances of the same uuid.
1757 dev_err(&nd_region
->dev
, "%s missing label for %pUb\n",
1758 dev_name(ndd
->dev
), nd_label
->uuid
);
1764 * Fix up each mapping's 'labels' to have the validated pmem label for
1765 * that position at labels[0], and NULL at labels[1]. In the process,
1766 * check that the namespace aligns with interleave-set. We know
1767 * that it does not overlap with any blk namespaces by virtue of
1768 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1771 rc
= select_pmem_id(nd_region
, nd_label
->uuid
);
1775 /* Calculate total size and populate namespace properties from label0 */
1776 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
1777 struct nd_namespace_label
*label0
;
1779 nd_mapping
= &nd_region
->mapping
[i
];
1780 label_ent
= list_first_entry_or_null(&nd_mapping
->labels
,
1781 typeof(*label_ent
), list
);
1782 label0
= label_ent
? label_ent
->label
: 0;
1789 size
+= __le64_to_cpu(label0
->rawsize
);
1790 if (__le16_to_cpu(label0
->position
) != 0)
1792 WARN_ON(nspm
->alt_name
|| nspm
->uuid
);
1793 nspm
->alt_name
= kmemdup((void __force
*) label0
->name
,
1794 NSLABEL_NAME_LEN
, GFP_KERNEL
);
1795 nspm
->uuid
= kmemdup((void __force
*) label0
->uuid
,
1796 NSLABEL_UUID_LEN
, GFP_KERNEL
);
1799 if (!nspm
->alt_name
|| !nspm
->uuid
) {
1804 nd_namespace_pmem_set_resource(nd_region
, nspm
, size
);
1808 namespace_pmem_release(dev
);
1811 dev_dbg(&nd_region
->dev
, "%s: invalid label(s)\n", __func__
);
1814 dev_dbg(&nd_region
->dev
, "%s: label not found\n", __func__
);
1817 dev_dbg(&nd_region
->dev
, "%s: unexpected err: %d\n",
1824 struct resource
*nsblk_add_resource(struct nd_region
*nd_region
,
1825 struct nvdimm_drvdata
*ndd
, struct nd_namespace_blk
*nsblk
,
1826 resource_size_t start
)
1828 struct nd_label_id label_id
;
1829 struct resource
*res
;
1831 nd_label_gen_id(&label_id
, nsblk
->uuid
, NSLABEL_FLAG_LOCAL
);
1832 res
= krealloc(nsblk
->res
,
1833 sizeof(void *) * (nsblk
->num_resources
+ 1),
1837 nsblk
->res
= (struct resource
**) res
;
1838 for_each_dpa_resource(ndd
, res
)
1839 if (strcmp(res
->name
, label_id
.id
) == 0
1840 && res
->start
== start
) {
1841 nsblk
->res
[nsblk
->num_resources
++] = res
;
1847 static struct device
*nd_namespace_blk_create(struct nd_region
*nd_region
)
1849 struct nd_namespace_blk
*nsblk
;
1852 if (!is_nd_blk(&nd_region
->dev
))
1855 nsblk
= kzalloc(sizeof(*nsblk
), GFP_KERNEL
);
1859 dev
= &nsblk
->common
.dev
;
1860 dev
->type
= &namespace_blk_device_type
;
1861 nsblk
->id
= ida_simple_get(&nd_region
->ns_ida
, 0, 0, GFP_KERNEL
);
1862 if (nsblk
->id
< 0) {
1866 dev_set_name(dev
, "namespace%d.%d", nd_region
->id
, nsblk
->id
);
1867 dev
->parent
= &nd_region
->dev
;
1868 dev
->groups
= nd_namespace_attribute_groups
;
1870 return &nsblk
->common
.dev
;
1873 static struct device
*nd_namespace_pmem_create(struct nd_region
*nd_region
)
1875 struct nd_namespace_pmem
*nspm
;
1876 struct resource
*res
;
1879 if (!is_nd_pmem(&nd_region
->dev
))
1882 nspm
= kzalloc(sizeof(*nspm
), GFP_KERNEL
);
1886 dev
= &nspm
->nsio
.common
.dev
;
1887 dev
->type
= &namespace_pmem_device_type
;
1888 dev
->parent
= &nd_region
->dev
;
1889 res
= &nspm
->nsio
.res
;
1890 res
->name
= dev_name(&nd_region
->dev
);
1891 res
->flags
= IORESOURCE_MEM
;
1893 nspm
->id
= ida_simple_get(&nd_region
->ns_ida
, 0, 0, GFP_KERNEL
);
1898 dev_set_name(dev
, "namespace%d.%d", nd_region
->id
, nspm
->id
);
1899 dev
->parent
= &nd_region
->dev
;
1900 dev
->groups
= nd_namespace_attribute_groups
;
1901 nd_namespace_pmem_set_resource(nd_region
, nspm
, 0);
1906 void nd_region_create_ns_seed(struct nd_region
*nd_region
)
1908 WARN_ON(!is_nvdimm_bus_locked(&nd_region
->dev
));
1910 if (nd_region_to_nstype(nd_region
) == ND_DEVICE_NAMESPACE_IO
)
1913 if (is_nd_blk(&nd_region
->dev
))
1914 nd_region
->ns_seed
= nd_namespace_blk_create(nd_region
);
1916 nd_region
->ns_seed
= nd_namespace_pmem_create(nd_region
);
1919 * Seed creation failures are not fatal, provisioning is simply
1920 * disabled until memory becomes available
1922 if (!nd_region
->ns_seed
)
1923 dev_err(&nd_region
->dev
, "failed to create %s namespace\n",
1924 is_nd_blk(&nd_region
->dev
) ? "blk" : "pmem");
1926 nd_device_register(nd_region
->ns_seed
);
1929 void nd_region_create_dax_seed(struct nd_region
*nd_region
)
1931 WARN_ON(!is_nvdimm_bus_locked(&nd_region
->dev
));
1932 nd_region
->dax_seed
= nd_dax_create(nd_region
);
1934 * Seed creation failures are not fatal, provisioning is simply
1935 * disabled until memory becomes available
1937 if (!nd_region
->dax_seed
)
1938 dev_err(&nd_region
->dev
, "failed to create dax namespace\n");
1941 void nd_region_create_pfn_seed(struct nd_region
*nd_region
)
1943 WARN_ON(!is_nvdimm_bus_locked(&nd_region
->dev
));
1944 nd_region
->pfn_seed
= nd_pfn_create(nd_region
);
1946 * Seed creation failures are not fatal, provisioning is simply
1947 * disabled until memory becomes available
1949 if (!nd_region
->pfn_seed
)
1950 dev_err(&nd_region
->dev
, "failed to create pfn namespace\n");
1953 void nd_region_create_btt_seed(struct nd_region
*nd_region
)
1955 WARN_ON(!is_nvdimm_bus_locked(&nd_region
->dev
));
1956 nd_region
->btt_seed
= nd_btt_create(nd_region
);
1958 * Seed creation failures are not fatal, provisioning is simply
1959 * disabled until memory becomes available
1961 if (!nd_region
->btt_seed
)
1962 dev_err(&nd_region
->dev
, "failed to create btt namespace\n");
1965 static int add_namespace_resource(struct nd_region
*nd_region
,
1966 struct nd_namespace_label
*nd_label
, struct device
**devs
,
1969 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
1970 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
1973 for (i
= 0; i
< count
; i
++) {
1974 u8
*uuid
= namespace_to_uuid(devs
[i
]);
1975 struct resource
*res
;
1977 if (IS_ERR_OR_NULL(uuid
)) {
1982 if (memcmp(uuid
, nd_label
->uuid
, NSLABEL_UUID_LEN
) != 0)
1984 if (is_namespace_blk(devs
[i
])) {
1985 res
= nsblk_add_resource(nd_region
, ndd
,
1986 to_nd_namespace_blk(devs
[i
]),
1987 __le64_to_cpu(nd_label
->dpa
));
1990 nd_dbg_dpa(nd_region
, ndd
, res
, "%d assign\n", count
);
1992 dev_err(&nd_region
->dev
,
1993 "error: conflicting extents for uuid: %pUb\n",
2003 struct device
*create_namespace_blk(struct nd_region
*nd_region
,
2004 struct nd_namespace_label
*nd_label
, int count
)
2007 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
2008 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
2009 struct nd_namespace_blk
*nsblk
;
2010 char name
[NSLABEL_NAME_LEN
];
2011 struct device
*dev
= NULL
;
2012 struct resource
*res
;
2014 nsblk
= kzalloc(sizeof(*nsblk
), GFP_KERNEL
);
2016 return ERR_PTR(-ENOMEM
);
2017 dev
= &nsblk
->common
.dev
;
2018 dev
->type
= &namespace_blk_device_type
;
2019 dev
->parent
= &nd_region
->dev
;
2021 nsblk
->lbasize
= __le64_to_cpu(nd_label
->lbasize
);
2022 nsblk
->uuid
= kmemdup(nd_label
->uuid
, NSLABEL_UUID_LEN
,
2026 memcpy(name
, nd_label
->name
, NSLABEL_NAME_LEN
);
2028 nsblk
->alt_name
= kmemdup(name
, NSLABEL_NAME_LEN
,
2030 res
= nsblk_add_resource(nd_region
, ndd
, nsblk
,
2031 __le64_to_cpu(nd_label
->dpa
));
2034 nd_dbg_dpa(nd_region
, ndd
, res
, "%d: assign\n", count
);
2037 namespace_blk_release(dev
);
2038 return ERR_PTR(-ENXIO
);
2041 static int cmp_dpa(const void *a
, const void *b
)
2043 const struct device
*dev_a
= *(const struct device
**) a
;
2044 const struct device
*dev_b
= *(const struct device
**) b
;
2045 struct nd_namespace_blk
*nsblk_a
, *nsblk_b
;
2046 struct nd_namespace_pmem
*nspm_a
, *nspm_b
;
2048 if (is_namespace_io(dev_a
))
2051 if (is_namespace_blk(dev_a
)) {
2052 nsblk_a
= to_nd_namespace_blk(dev_a
);
2053 nsblk_b
= to_nd_namespace_blk(dev_b
);
2055 return memcmp(&nsblk_a
->res
[0]->start
, &nsblk_b
->res
[0]->start
,
2056 sizeof(resource_size_t
));
2059 nspm_a
= to_nd_namespace_pmem(dev_a
);
2060 nspm_b
= to_nd_namespace_pmem(dev_b
);
2062 return memcmp(&nspm_a
->nsio
.res
.start
, &nspm_b
->nsio
.res
.start
,
2063 sizeof(resource_size_t
));
2066 static struct device
**scan_labels(struct nd_region
*nd_region
)
2069 struct device
*dev
, **devs
= NULL
;
2070 struct nd_label_ent
*label_ent
, *e
;
2071 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
2072 resource_size_t map_end
= nd_mapping
->start
+ nd_mapping
->size
- 1;
2074 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2075 list_for_each_entry_safe(label_ent
, e
, &nd_mapping
->labels
, list
) {
2076 struct nd_namespace_label
*nd_label
= label_ent
->label
;
2077 struct device
**__devs
;
2082 flags
= __le32_to_cpu(nd_label
->flags
);
2083 if (is_nd_blk(&nd_region
->dev
)
2084 == !!(flags
& NSLABEL_FLAG_LOCAL
))
2085 /* pass, region matches label type */;
2089 /* skip labels that describe extents outside of the region */
2090 if (nd_label
->dpa
< nd_mapping
->start
|| nd_label
->dpa
> map_end
)
2093 i
= add_namespace_resource(nd_region
, nd_label
, devs
, count
);
2098 __devs
= kcalloc(count
+ 2, sizeof(dev
), GFP_KERNEL
);
2101 memcpy(__devs
, devs
, sizeof(dev
) * count
);
2105 if (is_nd_blk(&nd_region
->dev
)) {
2106 dev
= create_namespace_blk(nd_region
, nd_label
, count
);
2109 devs
[count
++] = dev
;
2111 dev
= create_namespace_pmem(nd_region
, nd_label
);
2113 switch (PTR_ERR(dev
)) {
2115 /* skip invalid labels */
2118 /* fallthrough to seed creation */
2124 devs
[count
++] = dev
;
2128 dev_dbg(&nd_region
->dev
, "%s: discovered %d %s namespace%s\n",
2129 __func__
, count
, is_nd_blk(&nd_region
->dev
)
2130 ? "blk" : "pmem", count
== 1 ? "" : "s");
2133 /* Publish a zero-sized namespace for userspace to configure. */
2134 nd_mapping_free_labels(nd_mapping
);
2136 devs
= kcalloc(2, sizeof(dev
), GFP_KERNEL
);
2139 if (is_nd_blk(&nd_region
->dev
)) {
2140 struct nd_namespace_blk
*nsblk
;
2142 nsblk
= kzalloc(sizeof(*nsblk
), GFP_KERNEL
);
2145 dev
= &nsblk
->common
.dev
;
2146 dev
->type
= &namespace_blk_device_type
;
2148 struct nd_namespace_pmem
*nspm
;
2150 nspm
= kzalloc(sizeof(*nspm
), GFP_KERNEL
);
2153 dev
= &nspm
->nsio
.common
.dev
;
2154 dev
->type
= &namespace_pmem_device_type
;
2155 nd_namespace_pmem_set_resource(nd_region
, nspm
, 0);
2157 dev
->parent
= &nd_region
->dev
;
2158 devs
[count
++] = dev
;
2159 } else if (is_nd_pmem(&nd_region
->dev
)) {
2160 /* clean unselected labels */
2161 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
2162 struct list_head
*l
, *e
;
2166 nd_mapping
= &nd_region
->mapping
[i
];
2167 if (list_empty(&nd_mapping
->labels
)) {
2173 list_for_each_safe(l
, e
, &nd_mapping
->labels
) {
2176 list_move_tail(l
, &list
);
2178 nd_mapping_free_labels(nd_mapping
);
2179 list_splice_init(&list
, &nd_mapping
->labels
);
2184 sort(devs
, count
, sizeof(struct device
*), cmp_dpa
, NULL
);
2190 for (i
= 0; devs
[i
]; i
++)
2191 if (is_nd_blk(&nd_region
->dev
))
2192 namespace_blk_release(devs
[i
]);
2194 namespace_pmem_release(devs
[i
]);
2200 static struct device
**create_namespaces(struct nd_region
*nd_region
)
2202 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
2203 struct device
**devs
;
2206 if (nd_region
->ndr_mappings
== 0)
2209 /* lock down all mappings while we scan labels */
2210 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
2211 nd_mapping
= &nd_region
->mapping
[i
];
2212 mutex_lock_nested(&nd_mapping
->lock
, i
);
2215 devs
= scan_labels(nd_region
);
2217 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
2218 int reverse
= nd_region
->ndr_mappings
- 1 - i
;
2220 nd_mapping
= &nd_region
->mapping
[reverse
];
2221 mutex_unlock(&nd_mapping
->lock
);
2227 static int init_active_labels(struct nd_region
*nd_region
)
2231 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
2232 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[i
];
2233 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
2234 struct nvdimm
*nvdimm
= nd_mapping
->nvdimm
;
2235 struct nd_label_ent
*label_ent
;
2239 * If the dimm is disabled then prevent the region from
2240 * being activated if it aliases DPA.
2243 if ((nvdimm
->flags
& NDD_ALIASING
) == 0)
2245 dev_dbg(&nd_region
->dev
, "%s: is disabled, failing probe\n",
2246 dev_name(&nd_mapping
->nvdimm
->dev
));
2249 nd_mapping
->ndd
= ndd
;
2250 atomic_inc(&nvdimm
->busy
);
2253 count
= nd_label_active_count(ndd
);
2254 dev_dbg(ndd
->dev
, "%s: %d\n", __func__
, count
);
2257 for (j
= 0; j
< count
; j
++) {
2258 struct nd_namespace_label
*label
;
2260 label_ent
= kzalloc(sizeof(*label_ent
), GFP_KERNEL
);
2263 label
= nd_label_active(ndd
, j
);
2264 label_ent
->label
= label
;
2266 mutex_lock(&nd_mapping
->lock
);
2267 list_add_tail(&label_ent
->list
, &nd_mapping
->labels
);
2268 mutex_unlock(&nd_mapping
->lock
);
2274 mutex_lock(&nd_mapping
->lock
);
2275 nd_mapping_free_labels(nd_mapping
);
2276 mutex_unlock(&nd_mapping
->lock
);
2283 int nd_region_register_namespaces(struct nd_region
*nd_region
, int *err
)
2285 struct device
**devs
= NULL
;
2286 int i
, rc
= 0, type
;
2289 nvdimm_bus_lock(&nd_region
->dev
);
2290 rc
= init_active_labels(nd_region
);
2292 nvdimm_bus_unlock(&nd_region
->dev
);
2296 type
= nd_region_to_nstype(nd_region
);
2298 case ND_DEVICE_NAMESPACE_IO
:
2299 devs
= create_namespace_io(nd_region
);
2301 case ND_DEVICE_NAMESPACE_PMEM
:
2302 case ND_DEVICE_NAMESPACE_BLK
:
2303 devs
= create_namespaces(nd_region
);
2308 nvdimm_bus_unlock(&nd_region
->dev
);
2313 for (i
= 0; devs
[i
]; i
++) {
2314 struct device
*dev
= devs
[i
];
2317 if (type
== ND_DEVICE_NAMESPACE_BLK
) {
2318 struct nd_namespace_blk
*nsblk
;
2320 nsblk
= to_nd_namespace_blk(dev
);
2321 id
= ida_simple_get(&nd_region
->ns_ida
, 0, 0,
2324 } else if (type
== ND_DEVICE_NAMESPACE_PMEM
) {
2325 struct nd_namespace_pmem
*nspm
;
2327 nspm
= to_nd_namespace_pmem(dev
);
2328 id
= ida_simple_get(&nd_region
->ns_ida
, 0, 0,
2336 dev_set_name(dev
, "namespace%d.%d", nd_region
->id
, id
);
2337 dev
->groups
= nd_namespace_attribute_groups
;
2338 nd_device_register(dev
);
2341 nd_region
->ns_seed
= devs
[0];
2346 for (j
= i
; devs
[j
]; j
++) {
2347 struct device
*dev
= devs
[j
];
2349 device_initialize(dev
);
2354 * All of the namespaces we tried to register failed, so
2355 * fail region activation.