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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/vmalloc.h>
15 #include <linux/device.h>
16 #include <linux/ndctl.h>
17 #include <linux/slab.h>
25 static DEFINE_IDA(dimm_ida
);
28 * Retrieve bus and dimm handle and return if this bus supports
29 * get_config_data commands
31 int nvdimm_check_config_data(struct device
*dev
)
33 struct nvdimm
*nvdimm
= to_nvdimm(dev
);
35 if (!nvdimm
->cmd_mask
||
36 !test_bit(ND_CMD_GET_CONFIG_DATA
, &nvdimm
->cmd_mask
)) {
37 if (nvdimm
->flags
& NDD_ALIASING
)
46 static int validate_dimm(struct nvdimm_drvdata
*ndd
)
53 rc
= nvdimm_check_config_data(ndd
->dev
);
55 dev_dbg(ndd
->dev
, "%pf: %s error: %d\n",
56 __builtin_return_address(0), __func__
, rc
);
61 * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
62 * @nvdimm: dimm to initialize
64 int nvdimm_init_nsarea(struct nvdimm_drvdata
*ndd
)
66 struct nd_cmd_get_config_size
*cmd
= &ndd
->nsarea
;
67 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(ndd
->dev
);
68 struct nvdimm_bus_descriptor
*nd_desc
;
69 int rc
= validate_dimm(ndd
);
75 return 0; /* already valid */
77 memset(cmd
, 0, sizeof(*cmd
));
78 nd_desc
= nvdimm_bus
->nd_desc
;
79 return nd_desc
->ndctl(nd_desc
, to_nvdimm(ndd
->dev
),
80 ND_CMD_GET_CONFIG_SIZE
, cmd
, sizeof(*cmd
), NULL
);
83 int nvdimm_init_config_data(struct nvdimm_drvdata
*ndd
)
85 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(ndd
->dev
);
86 struct nd_cmd_get_config_data_hdr
*cmd
;
87 struct nvdimm_bus_descriptor
*nd_desc
;
88 int rc
= validate_dimm(ndd
);
89 u32 max_cmd_size
, config_size
;
98 if (ndd
->nsarea
.status
|| ndd
->nsarea
.max_xfer
== 0
99 || ndd
->nsarea
.config_size
< ND_LABEL_MIN_SIZE
) {
100 dev_dbg(ndd
->dev
, "failed to init config data area: (%d:%d)\n",
101 ndd
->nsarea
.max_xfer
, ndd
->nsarea
.config_size
);
105 ndd
->data
= kmalloc(ndd
->nsarea
.config_size
, GFP_KERNEL
);
107 ndd
->data
= vmalloc(ndd
->nsarea
.config_size
);
112 max_cmd_size
= min_t(u32
, PAGE_SIZE
, ndd
->nsarea
.max_xfer
);
113 cmd
= kzalloc(max_cmd_size
+ sizeof(*cmd
), GFP_KERNEL
);
117 nd_desc
= nvdimm_bus
->nd_desc
;
118 for (config_size
= ndd
->nsarea
.config_size
, offset
= 0;
119 config_size
; config_size
-= cmd
->in_length
,
120 offset
+= cmd
->in_length
) {
121 cmd
->in_length
= min(config_size
, max_cmd_size
);
122 cmd
->in_offset
= offset
;
123 rc
= nd_desc
->ndctl(nd_desc
, to_nvdimm(ndd
->dev
),
124 ND_CMD_GET_CONFIG_DATA
, cmd
,
125 cmd
->in_length
+ sizeof(*cmd
), NULL
);
126 if (rc
|| cmd
->status
) {
130 memcpy(ndd
->data
+ offset
, cmd
->out_buf
, cmd
->in_length
);
132 dev_dbg(ndd
->dev
, "%s: len: %zu rc: %d\n", __func__
, offset
, rc
);
138 int nvdimm_set_config_data(struct nvdimm_drvdata
*ndd
, size_t offset
,
139 void *buf
, size_t len
)
141 int rc
= validate_dimm(ndd
);
142 size_t max_cmd_size
, buf_offset
;
143 struct nd_cmd_set_config_hdr
*cmd
;
144 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(ndd
->dev
);
145 struct nvdimm_bus_descriptor
*nd_desc
= nvdimm_bus
->nd_desc
;
153 if (offset
+ len
> ndd
->nsarea
.config_size
)
156 max_cmd_size
= min_t(u32
, PAGE_SIZE
, len
);
157 max_cmd_size
= min_t(u32
, max_cmd_size
, ndd
->nsarea
.max_xfer
);
158 cmd
= kzalloc(max_cmd_size
+ sizeof(*cmd
) + sizeof(u32
), GFP_KERNEL
);
162 for (buf_offset
= 0; len
; len
-= cmd
->in_length
,
163 buf_offset
+= cmd
->in_length
) {
167 cmd
->in_offset
= offset
+ buf_offset
;
168 cmd
->in_length
= min(max_cmd_size
, len
);
169 memcpy(cmd
->in_buf
, buf
+ buf_offset
, cmd
->in_length
);
171 /* status is output in the last 4-bytes of the command buffer */
172 cmd_size
= sizeof(*cmd
) + cmd
->in_length
+ sizeof(u32
);
173 status
= ((void *) cmd
) + cmd_size
- sizeof(u32
);
175 rc
= nd_desc
->ndctl(nd_desc
, to_nvdimm(ndd
->dev
),
176 ND_CMD_SET_CONFIG_DATA
, cmd
, cmd_size
, NULL
);
178 rc
= rc
? rc
: -ENXIO
;
187 void nvdimm_set_aliasing(struct device
*dev
)
189 struct nvdimm
*nvdimm
= to_nvdimm(dev
);
191 nvdimm
->flags
|= NDD_ALIASING
;
194 static void nvdimm_release(struct device
*dev
)
196 struct nvdimm
*nvdimm
= to_nvdimm(dev
);
198 ida_simple_remove(&dimm_ida
, nvdimm
->id
);
202 static struct device_type nvdimm_device_type
= {
204 .release
= nvdimm_release
,
207 bool is_nvdimm(struct device
*dev
)
209 return dev
->type
== &nvdimm_device_type
;
212 struct nvdimm
*to_nvdimm(struct device
*dev
)
214 struct nvdimm
*nvdimm
= container_of(dev
, struct nvdimm
, dev
);
216 WARN_ON(!is_nvdimm(dev
));
219 EXPORT_SYMBOL_GPL(to_nvdimm
);
221 struct nvdimm
*nd_blk_region_to_dimm(struct nd_blk_region
*ndbr
)
223 struct nd_region
*nd_region
= &ndbr
->nd_region
;
224 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
226 return nd_mapping
->nvdimm
;
228 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm
);
230 struct nvdimm_drvdata
*to_ndd(struct nd_mapping
*nd_mapping
)
232 struct nvdimm
*nvdimm
= nd_mapping
->nvdimm
;
234 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm
->dev
));
236 return dev_get_drvdata(&nvdimm
->dev
);
238 EXPORT_SYMBOL(to_ndd
);
240 void nvdimm_drvdata_release(struct kref
*kref
)
242 struct nvdimm_drvdata
*ndd
= container_of(kref
, typeof(*ndd
), kref
);
243 struct device
*dev
= ndd
->dev
;
244 struct resource
*res
, *_r
;
246 dev_dbg(dev
, "%s\n", __func__
);
248 nvdimm_bus_lock(dev
);
249 for_each_dpa_resource_safe(ndd
, res
, _r
)
250 nvdimm_free_dpa(ndd
, res
);
251 nvdimm_bus_unlock(dev
);
258 void get_ndd(struct nvdimm_drvdata
*ndd
)
260 kref_get(&ndd
->kref
);
263 void put_ndd(struct nvdimm_drvdata
*ndd
)
266 kref_put(&ndd
->kref
, nvdimm_drvdata_release
);
269 const char *nvdimm_name(struct nvdimm
*nvdimm
)
271 return dev_name(&nvdimm
->dev
);
273 EXPORT_SYMBOL_GPL(nvdimm_name
);
275 struct kobject
*nvdimm_kobj(struct nvdimm
*nvdimm
)
277 return &nvdimm
->dev
.kobj
;
279 EXPORT_SYMBOL_GPL(nvdimm_kobj
);
281 unsigned long nvdimm_cmd_mask(struct nvdimm
*nvdimm
)
283 return nvdimm
->cmd_mask
;
285 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask
);
287 void *nvdimm_provider_data(struct nvdimm
*nvdimm
)
290 return nvdimm
->provider_data
;
293 EXPORT_SYMBOL_GPL(nvdimm_provider_data
);
295 static ssize_t
commands_show(struct device
*dev
,
296 struct device_attribute
*attr
, char *buf
)
298 struct nvdimm
*nvdimm
= to_nvdimm(dev
);
301 if (!nvdimm
->cmd_mask
)
302 return sprintf(buf
, "\n");
304 for_each_set_bit(cmd
, &nvdimm
->cmd_mask
, BITS_PER_LONG
)
305 len
+= sprintf(buf
+ len
, "%s ", nvdimm_cmd_name(cmd
));
306 len
+= sprintf(buf
+ len
, "\n");
309 static DEVICE_ATTR_RO(commands
);
311 static ssize_t
state_show(struct device
*dev
, struct device_attribute
*attr
,
314 struct nvdimm
*nvdimm
= to_nvdimm(dev
);
317 * The state may be in the process of changing, userspace should
318 * quiesce probing if it wants a static answer
320 nvdimm_bus_lock(dev
);
321 nvdimm_bus_unlock(dev
);
322 return sprintf(buf
, "%s\n", atomic_read(&nvdimm
->busy
)
323 ? "active" : "idle");
325 static DEVICE_ATTR_RO(state
);
327 static ssize_t
available_slots_show(struct device
*dev
,
328 struct device_attribute
*attr
, char *buf
)
330 struct nvdimm_drvdata
*ndd
= dev_get_drvdata(dev
);
337 nvdimm_bus_lock(dev
);
338 nfree
= nd_label_nfree(ndd
);
339 if (nfree
- 1 > nfree
) {
340 dev_WARN_ONCE(dev
, 1, "we ate our last label?\n");
344 rc
= sprintf(buf
, "%d\n", nfree
);
345 nvdimm_bus_unlock(dev
);
348 static DEVICE_ATTR_RO(available_slots
);
350 static struct attribute
*nvdimm_attributes
[] = {
351 &dev_attr_state
.attr
,
352 &dev_attr_commands
.attr
,
353 &dev_attr_available_slots
.attr
,
357 struct attribute_group nvdimm_attribute_group
= {
358 .attrs
= nvdimm_attributes
,
360 EXPORT_SYMBOL_GPL(nvdimm_attribute_group
);
362 struct nvdimm
*nvdimm_create(struct nvdimm_bus
*nvdimm_bus
, void *provider_data
,
363 const struct attribute_group
**groups
, unsigned long flags
,
364 unsigned long cmd_mask
, int num_flush
,
365 struct resource
*flush_wpq
)
367 struct nvdimm
*nvdimm
= kzalloc(sizeof(*nvdimm
), GFP_KERNEL
);
373 nvdimm
->id
= ida_simple_get(&dimm_ida
, 0, 0, GFP_KERNEL
);
374 if (nvdimm
->id
< 0) {
378 nvdimm
->provider_data
= provider_data
;
379 nvdimm
->flags
= flags
;
380 nvdimm
->cmd_mask
= cmd_mask
;
381 nvdimm
->num_flush
= num_flush
;
382 nvdimm
->flush_wpq
= flush_wpq
;
383 atomic_set(&nvdimm
->busy
, 0);
385 dev_set_name(dev
, "nmem%d", nvdimm
->id
);
386 dev
->parent
= &nvdimm_bus
->dev
;
387 dev
->type
= &nvdimm_device_type
;
388 dev
->devt
= MKDEV(nvdimm_major
, nvdimm
->id
);
389 dev
->groups
= groups
;
390 nd_device_register(dev
);
394 EXPORT_SYMBOL_GPL(nvdimm_create
);
396 int alias_dpa_busy(struct device
*dev
, void *data
)
398 resource_size_t map_end
, blk_start
, new, busy
;
399 struct blk_alloc_info
*info
= data
;
400 struct nd_mapping
*nd_mapping
;
401 struct nd_region
*nd_region
;
402 struct nvdimm_drvdata
*ndd
;
403 struct resource
*res
;
406 if (!is_nd_pmem(dev
))
409 nd_region
= to_nd_region(dev
);
410 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
411 nd_mapping
= &nd_region
->mapping
[i
];
412 if (nd_mapping
->nvdimm
== info
->nd_mapping
->nvdimm
)
416 if (i
>= nd_region
->ndr_mappings
)
419 ndd
= to_ndd(nd_mapping
);
420 map_end
= nd_mapping
->start
+ nd_mapping
->size
- 1;
421 blk_start
= nd_mapping
->start
;
424 * In the allocation case ->res is set to free space that we are
425 * looking to validate against PMEM aliasing collision rules
426 * (i.e. BLK is allocated after all aliased PMEM).
429 if (info
->res
->start
>= nd_mapping
->start
430 && info
->res
->start
< map_end
)
438 * Find the free dpa from the end of the last pmem allocation to
439 * the end of the interleave-set mapping that is not already
440 * covered by a blk allocation.
443 for_each_dpa_resource(ndd
, res
) {
444 if ((res
->start
>= blk_start
&& res
->start
< map_end
)
445 || (res
->end
>= blk_start
446 && res
->end
<= map_end
)) {
447 if (strncmp(res
->name
, "pmem", 4) == 0) {
448 new = max(blk_start
, min(map_end
+ 1,
450 if (new != blk_start
) {
455 busy
+= min(map_end
, res
->end
)
456 - max(nd_mapping
->start
, res
->start
) + 1;
457 } else if (nd_mapping
->start
> res
->start
458 && map_end
< res
->end
) {
459 /* total eclipse of the PMEM region mapping */
460 busy
+= nd_mapping
->size
;
465 /* update the free space range with the probed blk_start */
466 if (info
->res
&& blk_start
> info
->res
->start
) {
467 info
->res
->start
= max(info
->res
->start
, blk_start
);
468 if (info
->res
->start
> info
->res
->end
)
469 info
->res
->end
= info
->res
->start
- 1;
473 info
->available
-= blk_start
- nd_mapping
->start
+ busy
;
478 static int blk_dpa_busy(struct device
*dev
, void *data
)
480 struct blk_alloc_info
*info
= data
;
481 struct nd_mapping
*nd_mapping
;
482 struct nd_region
*nd_region
;
483 resource_size_t map_end
;
486 if (!is_nd_pmem(dev
))
489 nd_region
= to_nd_region(dev
);
490 for (i
= 0; i
< nd_region
->ndr_mappings
; i
++) {
491 nd_mapping
= &nd_region
->mapping
[i
];
492 if (nd_mapping
->nvdimm
== info
->nd_mapping
->nvdimm
)
496 if (i
>= nd_region
->ndr_mappings
)
499 map_end
= nd_mapping
->start
+ nd_mapping
->size
- 1;
500 if (info
->res
->start
>= nd_mapping
->start
501 && info
->res
->start
< map_end
) {
502 if (info
->res
->end
<= map_end
) {
506 info
->busy
-= info
->res
->end
- map_end
;
509 } else if (info
->res
->end
>= nd_mapping
->start
510 && info
->res
->end
<= map_end
) {
511 info
->busy
-= nd_mapping
->start
- info
->res
->start
;
514 info
->busy
-= nd_mapping
->size
;
520 * nd_blk_available_dpa - account the unused dpa of BLK region
521 * @nd_mapping: container of dpa-resource-root + labels
523 * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
524 * we arrange for them to never start at an lower dpa than the last
525 * PMEM allocation in an aliased region.
527 resource_size_t
nd_blk_available_dpa(struct nd_region
*nd_region
)
529 struct nvdimm_bus
*nvdimm_bus
= walk_to_nvdimm_bus(&nd_region
->dev
);
530 struct nd_mapping
*nd_mapping
= &nd_region
->mapping
[0];
531 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
532 struct blk_alloc_info info
= {
533 .nd_mapping
= nd_mapping
,
534 .available
= nd_mapping
->size
,
537 struct resource
*res
;
542 device_for_each_child(&nvdimm_bus
->dev
, &info
, alias_dpa_busy
);
544 /* now account for busy blk allocations in unaliased dpa */
545 for_each_dpa_resource(ndd
, res
) {
546 if (strncmp(res
->name
, "blk", 3) != 0)
550 info
.busy
= resource_size(res
);
551 device_for_each_child(&nvdimm_bus
->dev
, &info
, blk_dpa_busy
);
552 info
.available
-= info
.busy
;
555 return info
.available
;
559 * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
560 * @nd_mapping: container of dpa-resource-root + labels
561 * @nd_region: constrain available space check to this reference region
562 * @overlap: calculate available space assuming this level of overlap
564 * Validate that a PMEM label, if present, aligns with the start of an
565 * interleave set and truncate the available size at the lowest BLK
568 * The expectation is that this routine is called multiple times as it
569 * probes for the largest BLK encroachment for any single member DIMM of
570 * the interleave set. Once that value is determined the PMEM-limit for
571 * the set can be established.
573 resource_size_t
nd_pmem_available_dpa(struct nd_region
*nd_region
,
574 struct nd_mapping
*nd_mapping
, resource_size_t
*overlap
)
576 resource_size_t map_start
, map_end
, busy
= 0, available
, blk_start
;
577 struct nvdimm_drvdata
*ndd
= to_ndd(nd_mapping
);
578 struct resource
*res
;
584 map_start
= nd_mapping
->start
;
585 map_end
= map_start
+ nd_mapping
->size
- 1;
586 blk_start
= max(map_start
, map_end
+ 1 - *overlap
);
587 for_each_dpa_resource(ndd
, res
) {
588 if (res
->start
>= map_start
&& res
->start
< map_end
) {
589 if (strncmp(res
->name
, "blk", 3) == 0)
590 blk_start
= min(blk_start
,
591 max(map_start
, res
->start
));
592 else if (res
->end
> map_end
) {
593 reason
= "misaligned to iset";
596 busy
+= resource_size(res
);
597 } else if (res
->end
>= map_start
&& res
->end
<= map_end
) {
598 if (strncmp(res
->name
, "blk", 3) == 0) {
600 * If a BLK allocation overlaps the start of
601 * PMEM the entire interleave set may now only
604 blk_start
= map_start
;
606 busy
+= resource_size(res
);
607 } else if (map_start
> res
->start
&& map_start
< res
->end
) {
608 /* total eclipse of the mapping */
609 busy
+= nd_mapping
->size
;
610 blk_start
= map_start
;
614 *overlap
= map_end
+ 1 - blk_start
;
615 available
= blk_start
- map_start
;
616 if (busy
< available
)
617 return available
- busy
;
621 nd_dbg_dpa(nd_region
, ndd
, res
, "%s\n", reason
);
625 void nvdimm_free_dpa(struct nvdimm_drvdata
*ndd
, struct resource
*res
)
627 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd
->dev
));
629 __release_region(&ndd
->dpa
, res
->start
, resource_size(res
));
632 struct resource
*nvdimm_allocate_dpa(struct nvdimm_drvdata
*ndd
,
633 struct nd_label_id
*label_id
, resource_size_t start
,
636 char *name
= kmemdup(label_id
, sizeof(*label_id
), GFP_KERNEL
);
637 struct resource
*res
;
642 WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd
->dev
));
643 res
= __request_region(&ndd
->dpa
, start
, n
, name
, 0);
650 * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
651 * @nvdimm: container of dpa-resource-root + labels
652 * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
654 resource_size_t
nvdimm_allocated_dpa(struct nvdimm_drvdata
*ndd
,
655 struct nd_label_id
*label_id
)
657 resource_size_t allocated
= 0;
658 struct resource
*res
;
660 for_each_dpa_resource(ndd
, res
)
661 if (strcmp(res
->name
, label_id
->id
) == 0)
662 allocated
+= resource_size(res
);
667 static int count_dimms(struct device
*dev
, void *c
)
676 int nvdimm_bus_check_dimm_count(struct nvdimm_bus
*nvdimm_bus
, int dimm_count
)
679 /* Flush any possible dimm registration failures */
682 device_for_each_child(&nvdimm_bus
->dev
, &count
, count_dimms
);
683 dev_dbg(&nvdimm_bus
->dev
, "%s: count: %d\n", __func__
, count
);
684 if (count
!= dimm_count
)
688 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count
);
690 void __exit
nvdimm_devs_exit(void)
692 ida_destroy(&dimm_ida
);