libnvdimm: infrastructure for btt devices
[cascardo/linux.git] / drivers / nvdimm / region_devs.c
1 /*
2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3  *
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.
7  *
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.
12  */
13 #include <linux/scatterlist.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/sort.h>
17 #include <linux/io.h>
18 #include <linux/nd.h>
19 #include "nd-core.h"
20 #include "nd.h"
21
22 static DEFINE_IDA(region_ida);
23
24 static void nd_region_release(struct device *dev)
25 {
26         struct nd_region *nd_region = to_nd_region(dev);
27         u16 i;
28
29         for (i = 0; i < nd_region->ndr_mappings; i++) {
30                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
31                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
32
33                 put_device(&nvdimm->dev);
34         }
35         ida_simple_remove(&region_ida, nd_region->id);
36         kfree(nd_region);
37 }
38
39 static struct device_type nd_blk_device_type = {
40         .name = "nd_blk",
41         .release = nd_region_release,
42 };
43
44 static struct device_type nd_pmem_device_type = {
45         .name = "nd_pmem",
46         .release = nd_region_release,
47 };
48
49 static struct device_type nd_volatile_device_type = {
50         .name = "nd_volatile",
51         .release = nd_region_release,
52 };
53
54 bool is_nd_pmem(struct device *dev)
55 {
56         return dev ? dev->type == &nd_pmem_device_type : false;
57 }
58
59 bool is_nd_blk(struct device *dev)
60 {
61         return dev ? dev->type == &nd_blk_device_type : false;
62 }
63
64 struct nd_region *to_nd_region(struct device *dev)
65 {
66         struct nd_region *nd_region = container_of(dev, struct nd_region, dev);
67
68         WARN_ON(dev->type->release != nd_region_release);
69         return nd_region;
70 }
71 EXPORT_SYMBOL_GPL(to_nd_region);
72
73 /**
74  * nd_region_to_nstype() - region to an integer namespace type
75  * @nd_region: region-device to interrogate
76  *
77  * This is the 'nstype' attribute of a region as well, an input to the
78  * MODALIAS for namespace devices, and bit number for a nvdimm_bus to match
79  * namespace devices with namespace drivers.
80  */
81 int nd_region_to_nstype(struct nd_region *nd_region)
82 {
83         if (is_nd_pmem(&nd_region->dev)) {
84                 u16 i, alias;
85
86                 for (i = 0, alias = 0; i < nd_region->ndr_mappings; i++) {
87                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
88                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
89
90                         if (nvdimm->flags & NDD_ALIASING)
91                                 alias++;
92                 }
93                 if (alias)
94                         return ND_DEVICE_NAMESPACE_PMEM;
95                 else
96                         return ND_DEVICE_NAMESPACE_IO;
97         } else if (is_nd_blk(&nd_region->dev)) {
98                 return ND_DEVICE_NAMESPACE_BLK;
99         }
100
101         return 0;
102 }
103 EXPORT_SYMBOL(nd_region_to_nstype);
104
105 static int is_uuid_busy(struct device *dev, void *data)
106 {
107         struct nd_region *nd_region = to_nd_region(dev->parent);
108         u8 *uuid = data;
109
110         switch (nd_region_to_nstype(nd_region)) {
111         case ND_DEVICE_NAMESPACE_PMEM: {
112                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
113
114                 if (!nspm->uuid)
115                         break;
116                 if (memcmp(uuid, nspm->uuid, NSLABEL_UUID_LEN) == 0)
117                         return -EBUSY;
118                 break;
119         }
120         case ND_DEVICE_NAMESPACE_BLK: {
121                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
122
123                 if (!nsblk->uuid)
124                         break;
125                 if (memcmp(uuid, nsblk->uuid, NSLABEL_UUID_LEN) == 0)
126                         return -EBUSY;
127                 break;
128         }
129         default:
130                 break;
131         }
132
133         return 0;
134 }
135
136 static int is_namespace_uuid_busy(struct device *dev, void *data)
137 {
138         if (is_nd_pmem(dev) || is_nd_blk(dev))
139                 return device_for_each_child(dev, data, is_uuid_busy);
140         return 0;
141 }
142
143 /**
144  * nd_is_uuid_unique - verify that no other namespace has @uuid
145  * @dev: any device on a nvdimm_bus
146  * @uuid: uuid to check
147  */
148 bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
149 {
150         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
151
152         if (!nvdimm_bus)
153                 return false;
154         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
155         if (device_for_each_child(&nvdimm_bus->dev, uuid,
156                                 is_namespace_uuid_busy) != 0)
157                 return false;
158         return true;
159 }
160
161 static ssize_t size_show(struct device *dev,
162                 struct device_attribute *attr, char *buf)
163 {
164         struct nd_region *nd_region = to_nd_region(dev);
165         unsigned long long size = 0;
166
167         if (is_nd_pmem(dev)) {
168                 size = nd_region->ndr_size;
169         } else if (nd_region->ndr_mappings == 1) {
170                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
171
172                 size = nd_mapping->size;
173         }
174
175         return sprintf(buf, "%llu\n", size);
176 }
177 static DEVICE_ATTR_RO(size);
178
179 static ssize_t mappings_show(struct device *dev,
180                 struct device_attribute *attr, char *buf)
181 {
182         struct nd_region *nd_region = to_nd_region(dev);
183
184         return sprintf(buf, "%d\n", nd_region->ndr_mappings);
185 }
186 static DEVICE_ATTR_RO(mappings);
187
188 static ssize_t nstype_show(struct device *dev,
189                 struct device_attribute *attr, char *buf)
190 {
191         struct nd_region *nd_region = to_nd_region(dev);
192
193         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
194 }
195 static DEVICE_ATTR_RO(nstype);
196
197 static ssize_t set_cookie_show(struct device *dev,
198                 struct device_attribute *attr, char *buf)
199 {
200         struct nd_region *nd_region = to_nd_region(dev);
201         struct nd_interleave_set *nd_set = nd_region->nd_set;
202
203         if (is_nd_pmem(dev) && nd_set)
204                 /* pass, should be precluded by region_visible */;
205         else
206                 return -ENXIO;
207
208         return sprintf(buf, "%#llx\n", nd_set->cookie);
209 }
210 static DEVICE_ATTR_RO(set_cookie);
211
212 resource_size_t nd_region_available_dpa(struct nd_region *nd_region)
213 {
214         resource_size_t blk_max_overlap = 0, available, overlap;
215         int i;
216
217         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
218
219  retry:
220         available = 0;
221         overlap = blk_max_overlap;
222         for (i = 0; i < nd_region->ndr_mappings; i++) {
223                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
224                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
225
226                 /* if a dimm is disabled the available capacity is zero */
227                 if (!ndd)
228                         return 0;
229
230                 if (is_nd_pmem(&nd_region->dev)) {
231                         available += nd_pmem_available_dpa(nd_region,
232                                         nd_mapping, &overlap);
233                         if (overlap > blk_max_overlap) {
234                                 blk_max_overlap = overlap;
235                                 goto retry;
236                         }
237                 } else if (is_nd_blk(&nd_region->dev)) {
238                         available += nd_blk_available_dpa(nd_mapping);
239                 }
240         }
241
242         return available;
243 }
244
245 static ssize_t available_size_show(struct device *dev,
246                 struct device_attribute *attr, char *buf)
247 {
248         struct nd_region *nd_region = to_nd_region(dev);
249         unsigned long long available = 0;
250
251         /*
252          * Flush in-flight updates and grab a snapshot of the available
253          * size.  Of course, this value is potentially invalidated the
254          * memory nvdimm_bus_lock() is dropped, but that's userspace's
255          * problem to not race itself.
256          */
257         nvdimm_bus_lock(dev);
258         wait_nvdimm_bus_probe_idle(dev);
259         available = nd_region_available_dpa(nd_region);
260         nvdimm_bus_unlock(dev);
261
262         return sprintf(buf, "%llu\n", available);
263 }
264 static DEVICE_ATTR_RO(available_size);
265
266 static ssize_t init_namespaces_show(struct device *dev,
267                 struct device_attribute *attr, char *buf)
268 {
269         struct nd_region_namespaces *num_ns = dev_get_drvdata(dev);
270         ssize_t rc;
271
272         nvdimm_bus_lock(dev);
273         if (num_ns)
274                 rc = sprintf(buf, "%d/%d\n", num_ns->active, num_ns->count);
275         else
276                 rc = -ENXIO;
277         nvdimm_bus_unlock(dev);
278
279         return rc;
280 }
281 static DEVICE_ATTR_RO(init_namespaces);
282
283 static ssize_t namespace_seed_show(struct device *dev,
284                 struct device_attribute *attr, char *buf)
285 {
286         struct nd_region *nd_region = to_nd_region(dev);
287         ssize_t rc;
288
289         nvdimm_bus_lock(dev);
290         if (nd_region->ns_seed)
291                 rc = sprintf(buf, "%s\n", dev_name(nd_region->ns_seed));
292         else
293                 rc = sprintf(buf, "\n");
294         nvdimm_bus_unlock(dev);
295         return rc;
296 }
297 static DEVICE_ATTR_RO(namespace_seed);
298
299 static ssize_t btt_seed_show(struct device *dev,
300                 struct device_attribute *attr, char *buf)
301 {
302         struct nd_region *nd_region = to_nd_region(dev);
303         ssize_t rc;
304
305         nvdimm_bus_lock(dev);
306         if (nd_region->btt_seed)
307                 rc = sprintf(buf, "%s\n", dev_name(nd_region->btt_seed));
308         else
309                 rc = sprintf(buf, "\n");
310         nvdimm_bus_unlock(dev);
311
312         return rc;
313 }
314 static DEVICE_ATTR_RO(btt_seed);
315
316 static struct attribute *nd_region_attributes[] = {
317         &dev_attr_size.attr,
318         &dev_attr_nstype.attr,
319         &dev_attr_mappings.attr,
320         &dev_attr_btt_seed.attr,
321         &dev_attr_set_cookie.attr,
322         &dev_attr_available_size.attr,
323         &dev_attr_namespace_seed.attr,
324         &dev_attr_init_namespaces.attr,
325         NULL,
326 };
327
328 static umode_t region_visible(struct kobject *kobj, struct attribute *a, int n)
329 {
330         struct device *dev = container_of(kobj, typeof(*dev), kobj);
331         struct nd_region *nd_region = to_nd_region(dev);
332         struct nd_interleave_set *nd_set = nd_region->nd_set;
333         int type = nd_region_to_nstype(nd_region);
334
335         if (a != &dev_attr_set_cookie.attr
336                         && a != &dev_attr_available_size.attr)
337                 return a->mode;
338
339         if ((type == ND_DEVICE_NAMESPACE_PMEM
340                                 || type == ND_DEVICE_NAMESPACE_BLK)
341                         && a == &dev_attr_available_size.attr)
342                 return a->mode;
343         else if (is_nd_pmem(dev) && nd_set)
344                 return a->mode;
345
346         return 0;
347 }
348
349 struct attribute_group nd_region_attribute_group = {
350         .attrs = nd_region_attributes,
351         .is_visible = region_visible,
352 };
353 EXPORT_SYMBOL_GPL(nd_region_attribute_group);
354
355 u64 nd_region_interleave_set_cookie(struct nd_region *nd_region)
356 {
357         struct nd_interleave_set *nd_set = nd_region->nd_set;
358
359         if (nd_set)
360                 return nd_set->cookie;
361         return 0;
362 }
363
364 /*
365  * Upon successful probe/remove, take/release a reference on the
366  * associated interleave set (if present), and plant new btt + namespace
367  * seeds.
368  */
369 static void nd_region_notify_driver_action(struct nvdimm_bus *nvdimm_bus,
370                 struct device *dev, bool probe)
371 {
372         struct nd_region *nd_region;
373
374         if (!probe && (is_nd_pmem(dev) || is_nd_blk(dev))) {
375                 int i;
376
377                 nd_region = to_nd_region(dev);
378                 for (i = 0; i < nd_region->ndr_mappings; i++) {
379                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
380                         struct nvdimm_drvdata *ndd = nd_mapping->ndd;
381                         struct nvdimm *nvdimm = nd_mapping->nvdimm;
382
383                         kfree(nd_mapping->labels);
384                         nd_mapping->labels = NULL;
385                         put_ndd(ndd);
386                         nd_mapping->ndd = NULL;
387                         atomic_dec(&nvdimm->busy);
388                 }
389         }
390         if (dev->parent && is_nd_blk(dev->parent) && probe) {
391                 nd_region = to_nd_region(dev->parent);
392                 nvdimm_bus_lock(dev);
393                 if (nd_region->ns_seed == dev)
394                         nd_region_create_blk_seed(nd_region);
395                 nvdimm_bus_unlock(dev);
396         }
397         if (is_nd_btt(dev) && probe) {
398                 nd_region = to_nd_region(dev->parent);
399                 nvdimm_bus_lock(dev);
400                 if (nd_region->btt_seed == dev)
401                         nd_region_create_btt_seed(nd_region);
402                 nvdimm_bus_unlock(dev);
403         }
404 }
405
406 void nd_region_probe_success(struct nvdimm_bus *nvdimm_bus, struct device *dev)
407 {
408         nd_region_notify_driver_action(nvdimm_bus, dev, true);
409 }
410
411 void nd_region_disable(struct nvdimm_bus *nvdimm_bus, struct device *dev)
412 {
413         nd_region_notify_driver_action(nvdimm_bus, dev, false);
414 }
415
416 static ssize_t mappingN(struct device *dev, char *buf, int n)
417 {
418         struct nd_region *nd_region = to_nd_region(dev);
419         struct nd_mapping *nd_mapping;
420         struct nvdimm *nvdimm;
421
422         if (n >= nd_region->ndr_mappings)
423                 return -ENXIO;
424         nd_mapping = &nd_region->mapping[n];
425         nvdimm = nd_mapping->nvdimm;
426
427         return sprintf(buf, "%s,%llu,%llu\n", dev_name(&nvdimm->dev),
428                         nd_mapping->start, nd_mapping->size);
429 }
430
431 #define REGION_MAPPING(idx) \
432 static ssize_t mapping##idx##_show(struct device *dev,          \
433                 struct device_attribute *attr, char *buf)       \
434 {                                                               \
435         return mappingN(dev, buf, idx);                         \
436 }                                                               \
437 static DEVICE_ATTR_RO(mapping##idx)
438
439 /*
440  * 32 should be enough for a while, even in the presence of socket
441  * interleave a 32-way interleave set is a degenerate case.
442  */
443 REGION_MAPPING(0);
444 REGION_MAPPING(1);
445 REGION_MAPPING(2);
446 REGION_MAPPING(3);
447 REGION_MAPPING(4);
448 REGION_MAPPING(5);
449 REGION_MAPPING(6);
450 REGION_MAPPING(7);
451 REGION_MAPPING(8);
452 REGION_MAPPING(9);
453 REGION_MAPPING(10);
454 REGION_MAPPING(11);
455 REGION_MAPPING(12);
456 REGION_MAPPING(13);
457 REGION_MAPPING(14);
458 REGION_MAPPING(15);
459 REGION_MAPPING(16);
460 REGION_MAPPING(17);
461 REGION_MAPPING(18);
462 REGION_MAPPING(19);
463 REGION_MAPPING(20);
464 REGION_MAPPING(21);
465 REGION_MAPPING(22);
466 REGION_MAPPING(23);
467 REGION_MAPPING(24);
468 REGION_MAPPING(25);
469 REGION_MAPPING(26);
470 REGION_MAPPING(27);
471 REGION_MAPPING(28);
472 REGION_MAPPING(29);
473 REGION_MAPPING(30);
474 REGION_MAPPING(31);
475
476 static umode_t mapping_visible(struct kobject *kobj, struct attribute *a, int n)
477 {
478         struct device *dev = container_of(kobj, struct device, kobj);
479         struct nd_region *nd_region = to_nd_region(dev);
480
481         if (n < nd_region->ndr_mappings)
482                 return a->mode;
483         return 0;
484 }
485
486 static struct attribute *mapping_attributes[] = {
487         &dev_attr_mapping0.attr,
488         &dev_attr_mapping1.attr,
489         &dev_attr_mapping2.attr,
490         &dev_attr_mapping3.attr,
491         &dev_attr_mapping4.attr,
492         &dev_attr_mapping5.attr,
493         &dev_attr_mapping6.attr,
494         &dev_attr_mapping7.attr,
495         &dev_attr_mapping8.attr,
496         &dev_attr_mapping9.attr,
497         &dev_attr_mapping10.attr,
498         &dev_attr_mapping11.attr,
499         &dev_attr_mapping12.attr,
500         &dev_attr_mapping13.attr,
501         &dev_attr_mapping14.attr,
502         &dev_attr_mapping15.attr,
503         &dev_attr_mapping16.attr,
504         &dev_attr_mapping17.attr,
505         &dev_attr_mapping18.attr,
506         &dev_attr_mapping19.attr,
507         &dev_attr_mapping20.attr,
508         &dev_attr_mapping21.attr,
509         &dev_attr_mapping22.attr,
510         &dev_attr_mapping23.attr,
511         &dev_attr_mapping24.attr,
512         &dev_attr_mapping25.attr,
513         &dev_attr_mapping26.attr,
514         &dev_attr_mapping27.attr,
515         &dev_attr_mapping28.attr,
516         &dev_attr_mapping29.attr,
517         &dev_attr_mapping30.attr,
518         &dev_attr_mapping31.attr,
519         NULL,
520 };
521
522 struct attribute_group nd_mapping_attribute_group = {
523         .is_visible = mapping_visible,
524         .attrs = mapping_attributes,
525 };
526 EXPORT_SYMBOL_GPL(nd_mapping_attribute_group);
527
528 void *nd_region_provider_data(struct nd_region *nd_region)
529 {
530         return nd_region->provider_data;
531 }
532 EXPORT_SYMBOL_GPL(nd_region_provider_data);
533
534 static struct nd_region *nd_region_create(struct nvdimm_bus *nvdimm_bus,
535                 struct nd_region_desc *ndr_desc, struct device_type *dev_type,
536                 const char *caller)
537 {
538         struct nd_region *nd_region;
539         struct device *dev;
540         u16 i;
541
542         for (i = 0; i < ndr_desc->num_mappings; i++) {
543                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
544                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
545
546                 if ((nd_mapping->start | nd_mapping->size) % SZ_4K) {
547                         dev_err(&nvdimm_bus->dev, "%s: %s mapping%d is not 4K aligned\n",
548                                         caller, dev_name(&nvdimm->dev), i);
549
550                         return NULL;
551                 }
552         }
553
554         nd_region = kzalloc(sizeof(struct nd_region)
555                         + sizeof(struct nd_mapping) * ndr_desc->num_mappings,
556                         GFP_KERNEL);
557         if (!nd_region)
558                 return NULL;
559         nd_region->id = ida_simple_get(&region_ida, 0, 0, GFP_KERNEL);
560         if (nd_region->id < 0) {
561                 kfree(nd_region);
562                 return NULL;
563         }
564
565         memcpy(nd_region->mapping, ndr_desc->nd_mapping,
566                         sizeof(struct nd_mapping) * ndr_desc->num_mappings);
567         for (i = 0; i < ndr_desc->num_mappings; i++) {
568                 struct nd_mapping *nd_mapping = &ndr_desc->nd_mapping[i];
569                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
570
571                 get_device(&nvdimm->dev);
572         }
573         nd_region->ndr_mappings = ndr_desc->num_mappings;
574         nd_region->provider_data = ndr_desc->provider_data;
575         nd_region->nd_set = ndr_desc->nd_set;
576         ida_init(&nd_region->ns_ida);
577         ida_init(&nd_region->btt_ida);
578         dev = &nd_region->dev;
579         dev_set_name(dev, "region%d", nd_region->id);
580         dev->parent = &nvdimm_bus->dev;
581         dev->type = dev_type;
582         dev->groups = ndr_desc->attr_groups;
583         nd_region->ndr_size = resource_size(ndr_desc->res);
584         nd_region->ndr_start = ndr_desc->res->start;
585         nd_device_register(dev);
586
587         return nd_region;
588 }
589
590 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
591                 struct nd_region_desc *ndr_desc)
592 {
593         return nd_region_create(nvdimm_bus, ndr_desc, &nd_pmem_device_type,
594                         __func__);
595 }
596 EXPORT_SYMBOL_GPL(nvdimm_pmem_region_create);
597
598 struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,
599                 struct nd_region_desc *ndr_desc)
600 {
601         if (ndr_desc->num_mappings > 1)
602                 return NULL;
603         return nd_region_create(nvdimm_bus, ndr_desc, &nd_blk_device_type,
604                         __func__);
605 }
606 EXPORT_SYMBOL_GPL(nvdimm_blk_region_create);
607
608 struct nd_region *nvdimm_volatile_region_create(struct nvdimm_bus *nvdimm_bus,
609                 struct nd_region_desc *ndr_desc)
610 {
611         return nd_region_create(nvdimm_bus, ndr_desc, &nd_volatile_device_type,
612                         __func__);
613 }
614 EXPORT_SYMBOL_GPL(nvdimm_volatile_region_create);