GNU Linux-libre 4.14.290-gnu1
[releases.git] / drivers / nvdimm / dimm_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 #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>
18 #include <linux/io.h>
19 #include <linux/fs.h>
20 #include <linux/mm.h>
21 #include "nd-core.h"
22 #include "label.h"
23 #include "pmem.h"
24 #include "nd.h"
25
26 static DEFINE_IDA(dimm_ida);
27
28 /*
29  * Retrieve bus and dimm handle and return if this bus supports
30  * get_config_data commands
31  */
32 int nvdimm_check_config_data(struct device *dev)
33 {
34         struct nvdimm *nvdimm = to_nvdimm(dev);
35
36         if (!nvdimm->cmd_mask ||
37             !test_bit(ND_CMD_GET_CONFIG_DATA, &nvdimm->cmd_mask)) {
38                 if (test_bit(NDD_ALIASING, &nvdimm->flags))
39                         return -ENXIO;
40                 else
41                         return -ENOTTY;
42         }
43
44         return 0;
45 }
46
47 static int validate_dimm(struct nvdimm_drvdata *ndd)
48 {
49         int rc;
50
51         if (!ndd)
52                 return -EINVAL;
53
54         rc = nvdimm_check_config_data(ndd->dev);
55         if (rc)
56                 dev_dbg(ndd->dev, "%pf: %s error: %d\n",
57                                 __builtin_return_address(0), __func__, rc);
58         return rc;
59 }
60
61 /**
62  * nvdimm_init_nsarea - determine the geometry of a dimm's namespace area
63  * @nvdimm: dimm to initialize
64  */
65 int nvdimm_init_nsarea(struct nvdimm_drvdata *ndd)
66 {
67         struct nd_cmd_get_config_size *cmd = &ndd->nsarea;
68         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
69         struct nvdimm_bus_descriptor *nd_desc;
70         int rc = validate_dimm(ndd);
71         int cmd_rc = 0;
72
73         if (rc)
74                 return rc;
75
76         if (cmd->config_size)
77                 return 0; /* already valid */
78
79         memset(cmd, 0, sizeof(*cmd));
80         nd_desc = nvdimm_bus->nd_desc;
81         rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
82                         ND_CMD_GET_CONFIG_SIZE, cmd, sizeof(*cmd), &cmd_rc);
83         if (rc < 0)
84                 return rc;
85         return cmd_rc;
86 }
87
88 int nvdimm_init_config_data(struct nvdimm_drvdata *ndd)
89 {
90         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
91         struct nd_cmd_get_config_data_hdr *cmd;
92         struct nvdimm_bus_descriptor *nd_desc;
93         int rc = validate_dimm(ndd);
94         u32 max_cmd_size, config_size;
95         size_t offset;
96
97         if (rc)
98                 return rc;
99
100         if (ndd->data)
101                 return 0;
102
103         if (ndd->nsarea.status || ndd->nsarea.max_xfer == 0
104                         || ndd->nsarea.config_size < ND_LABEL_MIN_SIZE) {
105                 dev_dbg(ndd->dev, "failed to init config data area: (%d:%d)\n",
106                                 ndd->nsarea.max_xfer, ndd->nsarea.config_size);
107                 return -ENXIO;
108         }
109
110         ndd->data = kvmalloc(ndd->nsarea.config_size, GFP_KERNEL);
111         if (!ndd->data)
112                 return -ENOMEM;
113
114         max_cmd_size = min_t(u32, PAGE_SIZE, ndd->nsarea.max_xfer);
115         cmd = kzalloc(max_cmd_size + sizeof(*cmd), GFP_KERNEL);
116         if (!cmd)
117                 return -ENOMEM;
118
119         nd_desc = nvdimm_bus->nd_desc;
120         for (config_size = ndd->nsarea.config_size, offset = 0;
121                         config_size; config_size -= cmd->in_length,
122                         offset += cmd->in_length) {
123                 cmd->in_length = min(config_size, max_cmd_size);
124                 cmd->in_offset = offset;
125                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
126                                 ND_CMD_GET_CONFIG_DATA, cmd,
127                                 cmd->in_length + sizeof(*cmd), NULL);
128                 if (rc || cmd->status) {
129                         rc = -ENXIO;
130                         break;
131                 }
132                 memcpy(ndd->data + offset, cmd->out_buf, cmd->in_length);
133         }
134         dev_dbg(ndd->dev, "%s: len: %zu rc: %d\n", __func__, offset, rc);
135         kfree(cmd);
136
137         return rc;
138 }
139
140 int nvdimm_set_config_data(struct nvdimm_drvdata *ndd, size_t offset,
141                 void *buf, size_t len)
142 {
143         int rc = validate_dimm(ndd);
144         size_t max_cmd_size, buf_offset;
145         struct nd_cmd_set_config_hdr *cmd;
146         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(ndd->dev);
147         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
148
149         if (rc)
150                 return rc;
151
152         if (!ndd->data)
153                 return -ENXIO;
154
155         if (offset + len > ndd->nsarea.config_size)
156                 return -ENXIO;
157
158         max_cmd_size = min_t(u32, PAGE_SIZE, len);
159         max_cmd_size = min_t(u32, max_cmd_size, ndd->nsarea.max_xfer);
160         cmd = kzalloc(max_cmd_size + sizeof(*cmd) + sizeof(u32), GFP_KERNEL);
161         if (!cmd)
162                 return -ENOMEM;
163
164         for (buf_offset = 0; len; len -= cmd->in_length,
165                         buf_offset += cmd->in_length) {
166                 size_t cmd_size;
167                 u32 *status;
168
169                 cmd->in_offset = offset + buf_offset;
170                 cmd->in_length = min(max_cmd_size, len);
171                 memcpy(cmd->in_buf, buf + buf_offset, cmd->in_length);
172
173                 /* status is output in the last 4-bytes of the command buffer */
174                 cmd_size = sizeof(*cmd) + cmd->in_length + sizeof(u32);
175                 status = ((void *) cmd) + cmd_size - sizeof(u32);
176
177                 rc = nd_desc->ndctl(nd_desc, to_nvdimm(ndd->dev),
178                                 ND_CMD_SET_CONFIG_DATA, cmd, cmd_size, NULL);
179                 if (rc || *status) {
180                         rc = rc ? rc : -ENXIO;
181                         break;
182                 }
183         }
184         kfree(cmd);
185
186         return rc;
187 }
188
189 void nvdimm_set_aliasing(struct device *dev)
190 {
191         struct nvdimm *nvdimm = to_nvdimm(dev);
192
193         set_bit(NDD_ALIASING, &nvdimm->flags);
194 }
195
196 void nvdimm_set_locked(struct device *dev)
197 {
198         struct nvdimm *nvdimm = to_nvdimm(dev);
199
200         set_bit(NDD_LOCKED, &nvdimm->flags);
201 }
202
203 void nvdimm_clear_locked(struct device *dev)
204 {
205         struct nvdimm *nvdimm = to_nvdimm(dev);
206
207         clear_bit(NDD_LOCKED, &nvdimm->flags);
208 }
209
210 static void nvdimm_release(struct device *dev)
211 {
212         struct nvdimm *nvdimm = to_nvdimm(dev);
213
214         ida_simple_remove(&dimm_ida, nvdimm->id);
215         kfree(nvdimm);
216 }
217
218 static struct device_type nvdimm_device_type = {
219         .name = "nvdimm",
220         .release = nvdimm_release,
221 };
222
223 bool is_nvdimm(struct device *dev)
224 {
225         return dev->type == &nvdimm_device_type;
226 }
227
228 struct nvdimm *to_nvdimm(struct device *dev)
229 {
230         struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
231
232         WARN_ON(!is_nvdimm(dev));
233         return nvdimm;
234 }
235 EXPORT_SYMBOL_GPL(to_nvdimm);
236
237 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
238 {
239         struct nd_region *nd_region = &ndbr->nd_region;
240         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
241
242         return nd_mapping->nvdimm;
243 }
244 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
245
246 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
247 {
248         /* pmem mapping properties are private to libnvdimm */
249         return ARCH_MEMREMAP_PMEM;
250 }
251 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
252
253 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
254 {
255         struct nvdimm *nvdimm = nd_mapping->nvdimm;
256
257         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
258
259         return dev_get_drvdata(&nvdimm->dev);
260 }
261 EXPORT_SYMBOL(to_ndd);
262
263 void nvdimm_drvdata_release(struct kref *kref)
264 {
265         struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
266         struct device *dev = ndd->dev;
267         struct resource *res, *_r;
268
269         dev_dbg(dev, "%s\n", __func__);
270
271         nvdimm_bus_lock(dev);
272         for_each_dpa_resource_safe(ndd, res, _r)
273                 nvdimm_free_dpa(ndd, res);
274         nvdimm_bus_unlock(dev);
275
276         kvfree(ndd->data);
277         kfree(ndd);
278         put_device(dev);
279 }
280
281 void get_ndd(struct nvdimm_drvdata *ndd)
282 {
283         kref_get(&ndd->kref);
284 }
285
286 void put_ndd(struct nvdimm_drvdata *ndd)
287 {
288         if (ndd)
289                 kref_put(&ndd->kref, nvdimm_drvdata_release);
290 }
291
292 const char *nvdimm_name(struct nvdimm *nvdimm)
293 {
294         return dev_name(&nvdimm->dev);
295 }
296 EXPORT_SYMBOL_GPL(nvdimm_name);
297
298 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
299 {
300         return &nvdimm->dev.kobj;
301 }
302 EXPORT_SYMBOL_GPL(nvdimm_kobj);
303
304 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
305 {
306         return nvdimm->cmd_mask;
307 }
308 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
309
310 void *nvdimm_provider_data(struct nvdimm *nvdimm)
311 {
312         if (nvdimm)
313                 return nvdimm->provider_data;
314         return NULL;
315 }
316 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
317
318 static ssize_t commands_show(struct device *dev,
319                 struct device_attribute *attr, char *buf)
320 {
321         struct nvdimm *nvdimm = to_nvdimm(dev);
322         int cmd, len = 0;
323
324         if (!nvdimm->cmd_mask)
325                 return sprintf(buf, "\n");
326
327         for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
328                 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
329         len += sprintf(buf + len, "\n");
330         return len;
331 }
332 static DEVICE_ATTR_RO(commands);
333
334 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
335                 char *buf)
336 {
337         struct nvdimm *nvdimm = to_nvdimm(dev);
338
339         /*
340          * The state may be in the process of changing, userspace should
341          * quiesce probing if it wants a static answer
342          */
343         nvdimm_bus_lock(dev);
344         nvdimm_bus_unlock(dev);
345         return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
346                         ? "active" : "idle");
347 }
348 static DEVICE_ATTR_RO(state);
349
350 static ssize_t __available_slots_show(struct nvdimm_drvdata *ndd, char *buf)
351 {
352         struct device *dev;
353         ssize_t rc;
354         u32 nfree;
355
356         if (!ndd)
357                 return -ENXIO;
358
359         dev = ndd->dev;
360         nvdimm_bus_lock(dev);
361         nfree = nd_label_nfree(ndd);
362         if (nfree - 1 > nfree) {
363                 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
364                 nfree = 0;
365         } else
366                 nfree--;
367         rc = sprintf(buf, "%d\n", nfree);
368         nvdimm_bus_unlock(dev);
369         return rc;
370 }
371
372 static ssize_t available_slots_show(struct device *dev,
373                                     struct device_attribute *attr, char *buf)
374 {
375         ssize_t rc;
376
377         device_lock(dev);
378         rc = __available_slots_show(dev_get_drvdata(dev), buf);
379         device_unlock(dev);
380
381         return rc;
382 }
383 static DEVICE_ATTR_RO(available_slots);
384
385 static struct attribute *nvdimm_attributes[] = {
386         &dev_attr_state.attr,
387         &dev_attr_commands.attr,
388         &dev_attr_available_slots.attr,
389         NULL,
390 };
391
392 struct attribute_group nvdimm_attribute_group = {
393         .attrs = nvdimm_attributes,
394 };
395 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
396
397 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
398                 const struct attribute_group **groups, unsigned long flags,
399                 unsigned long cmd_mask, int num_flush,
400                 struct resource *flush_wpq)
401 {
402         struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
403         struct device *dev;
404
405         if (!nvdimm)
406                 return NULL;
407
408         nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
409         if (nvdimm->id < 0) {
410                 kfree(nvdimm);
411                 return NULL;
412         }
413         nvdimm->provider_data = provider_data;
414         nvdimm->flags = flags;
415         nvdimm->cmd_mask = cmd_mask;
416         nvdimm->num_flush = num_flush;
417         nvdimm->flush_wpq = flush_wpq;
418         atomic_set(&nvdimm->busy, 0);
419         dev = &nvdimm->dev;
420         dev_set_name(dev, "nmem%d", nvdimm->id);
421         dev->parent = &nvdimm_bus->dev;
422         dev->type = &nvdimm_device_type;
423         dev->devt = MKDEV(nvdimm_major, nvdimm->id);
424         dev->groups = groups;
425         nd_device_register(dev);
426
427         return nvdimm;
428 }
429 EXPORT_SYMBOL_GPL(nvdimm_create);
430
431 int alias_dpa_busy(struct device *dev, void *data)
432 {
433         resource_size_t map_end, blk_start, new;
434         struct blk_alloc_info *info = data;
435         struct nd_mapping *nd_mapping;
436         struct nd_region *nd_region;
437         struct nvdimm_drvdata *ndd;
438         struct resource *res;
439         int i;
440
441         if (!is_memory(dev))
442                 return 0;
443
444         nd_region = to_nd_region(dev);
445         for (i = 0; i < nd_region->ndr_mappings; i++) {
446                 nd_mapping  = &nd_region->mapping[i];
447                 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
448                         break;
449         }
450
451         if (i >= nd_region->ndr_mappings)
452                 return 0;
453
454         ndd = to_ndd(nd_mapping);
455         map_end = nd_mapping->start + nd_mapping->size - 1;
456         blk_start = nd_mapping->start;
457
458         /*
459          * In the allocation case ->res is set to free space that we are
460          * looking to validate against PMEM aliasing collision rules
461          * (i.e. BLK is allocated after all aliased PMEM).
462          */
463         if (info->res) {
464                 if (info->res->start >= nd_mapping->start
465                                 && info->res->start < map_end)
466                         /* pass */;
467                 else
468                         return 0;
469         }
470
471  retry:
472         /*
473          * Find the free dpa from the end of the last pmem allocation to
474          * the end of the interleave-set mapping.
475          */
476         for_each_dpa_resource(ndd, res) {
477                 if (strncmp(res->name, "pmem", 4) != 0)
478                         continue;
479                 if ((res->start >= blk_start && res->start < map_end)
480                                 || (res->end >= blk_start
481                                         && res->end <= map_end)) {
482                         new = max(blk_start, min(map_end + 1, res->end + 1));
483                         if (new != blk_start) {
484                                 blk_start = new;
485                                 goto retry;
486                         }
487                 }
488         }
489
490         /* update the free space range with the probed blk_start */
491         if (info->res && blk_start > info->res->start) {
492                 info->res->start = max(info->res->start, blk_start);
493                 if (info->res->start > info->res->end)
494                         info->res->end = info->res->start - 1;
495                 return 1;
496         }
497
498         info->available -= blk_start - nd_mapping->start;
499
500         return 0;
501 }
502
503 /**
504  * nd_blk_available_dpa - account the unused dpa of BLK region
505  * @nd_mapping: container of dpa-resource-root + labels
506  *
507  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
508  * we arrange for them to never start at an lower dpa than the last
509  * PMEM allocation in an aliased region.
510  */
511 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
512 {
513         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
514         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
515         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
516         struct blk_alloc_info info = {
517                 .nd_mapping = nd_mapping,
518                 .available = nd_mapping->size,
519                 .res = NULL,
520         };
521         struct resource *res;
522
523         if (!ndd)
524                 return 0;
525
526         device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
527
528         /* now account for busy blk allocations in unaliased dpa */
529         for_each_dpa_resource(ndd, res) {
530                 if (strncmp(res->name, "blk", 3) != 0)
531                         continue;
532                 info.available -= resource_size(res);
533         }
534
535         return info.available;
536 }
537
538 /**
539  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
540  * @nd_mapping: container of dpa-resource-root + labels
541  * @nd_region: constrain available space check to this reference region
542  * @overlap: calculate available space assuming this level of overlap
543  *
544  * Validate that a PMEM label, if present, aligns with the start of an
545  * interleave set and truncate the available size at the lowest BLK
546  * overlap point.
547  *
548  * The expectation is that this routine is called multiple times as it
549  * probes for the largest BLK encroachment for any single member DIMM of
550  * the interleave set.  Once that value is determined the PMEM-limit for
551  * the set can be established.
552  */
553 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
554                 struct nd_mapping *nd_mapping, resource_size_t *overlap)
555 {
556         resource_size_t map_start, map_end, busy = 0, available, blk_start;
557         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
558         struct resource *res;
559         const char *reason;
560
561         if (!ndd)
562                 return 0;
563
564         map_start = nd_mapping->start;
565         map_end = map_start + nd_mapping->size - 1;
566         blk_start = max(map_start, map_end + 1 - *overlap);
567         for_each_dpa_resource(ndd, res) {
568                 if (res->start >= map_start && res->start < map_end) {
569                         if (strncmp(res->name, "blk", 3) == 0)
570                                 blk_start = min(blk_start,
571                                                 max(map_start, res->start));
572                         else if (res->end > map_end) {
573                                 reason = "misaligned to iset";
574                                 goto err;
575                         } else
576                                 busy += resource_size(res);
577                 } else if (res->end >= map_start && res->end <= map_end) {
578                         if (strncmp(res->name, "blk", 3) == 0) {
579                                 /*
580                                  * If a BLK allocation overlaps the start of
581                                  * PMEM the entire interleave set may now only
582                                  * be used for BLK.
583                                  */
584                                 blk_start = map_start;
585                         } else
586                                 busy += resource_size(res);
587                 } else if (map_start > res->start && map_start < res->end) {
588                         /* total eclipse of the mapping */
589                         busy += nd_mapping->size;
590                         blk_start = map_start;
591                 }
592         }
593
594         *overlap = map_end + 1 - blk_start;
595         available = blk_start - map_start;
596         if (busy < available)
597                 return available - busy;
598         return 0;
599
600  err:
601         nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
602         return 0;
603 }
604
605 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
606 {
607         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
608         kfree(res->name);
609         __release_region(&ndd->dpa, res->start, resource_size(res));
610 }
611
612 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
613                 struct nd_label_id *label_id, resource_size_t start,
614                 resource_size_t n)
615 {
616         char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
617         struct resource *res;
618
619         if (!name)
620                 return NULL;
621
622         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
623         res = __request_region(&ndd->dpa, start, n, name, 0);
624         if (!res)
625                 kfree(name);
626         return res;
627 }
628
629 /**
630  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
631  * @nvdimm: container of dpa-resource-root + labels
632  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
633  */
634 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
635                 struct nd_label_id *label_id)
636 {
637         resource_size_t allocated = 0;
638         struct resource *res;
639
640         for_each_dpa_resource(ndd, res)
641                 if (strcmp(res->name, label_id->id) == 0)
642                         allocated += resource_size(res);
643
644         return allocated;
645 }
646
647 static int count_dimms(struct device *dev, void *c)
648 {
649         int *count = c;
650
651         if (is_nvdimm(dev))
652                 (*count)++;
653         return 0;
654 }
655
656 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
657 {
658         int count = 0;
659         /* Flush any possible dimm registration failures */
660         nd_synchronize();
661
662         device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
663         dev_dbg(&nvdimm_bus->dev, "%s: count: %d\n", __func__, count);
664         if (count != dimm_count)
665                 return -ENXIO;
666         return 0;
667 }
668 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
669
670 void __exit nvdimm_devs_exit(void)
671 {
672         ida_destroy(&dimm_ida);
673 }