]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - drivers/nvdimm/dimm_devs.c
Merge tag 'devprop-fix-4.13-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[karo-tx-linux.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 static void nvdimm_release(struct device *dev)
204 {
205         struct nvdimm *nvdimm = to_nvdimm(dev);
206
207         ida_simple_remove(&dimm_ida, nvdimm->id);
208         kfree(nvdimm);
209 }
210
211 static struct device_type nvdimm_device_type = {
212         .name = "nvdimm",
213         .release = nvdimm_release,
214 };
215
216 bool is_nvdimm(struct device *dev)
217 {
218         return dev->type == &nvdimm_device_type;
219 }
220
221 struct nvdimm *to_nvdimm(struct device *dev)
222 {
223         struct nvdimm *nvdimm = container_of(dev, struct nvdimm, dev);
224
225         WARN_ON(!is_nvdimm(dev));
226         return nvdimm;
227 }
228 EXPORT_SYMBOL_GPL(to_nvdimm);
229
230 struct nvdimm *nd_blk_region_to_dimm(struct nd_blk_region *ndbr)
231 {
232         struct nd_region *nd_region = &ndbr->nd_region;
233         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
234
235         return nd_mapping->nvdimm;
236 }
237 EXPORT_SYMBOL_GPL(nd_blk_region_to_dimm);
238
239 unsigned long nd_blk_memremap_flags(struct nd_blk_region *ndbr)
240 {
241         /* pmem mapping properties are private to libnvdimm */
242         return ARCH_MEMREMAP_PMEM;
243 }
244 EXPORT_SYMBOL_GPL(nd_blk_memremap_flags);
245
246 struct nvdimm_drvdata *to_ndd(struct nd_mapping *nd_mapping)
247 {
248         struct nvdimm *nvdimm = nd_mapping->nvdimm;
249
250         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm->dev));
251
252         return dev_get_drvdata(&nvdimm->dev);
253 }
254 EXPORT_SYMBOL(to_ndd);
255
256 void nvdimm_drvdata_release(struct kref *kref)
257 {
258         struct nvdimm_drvdata *ndd = container_of(kref, typeof(*ndd), kref);
259         struct device *dev = ndd->dev;
260         struct resource *res, *_r;
261
262         dev_dbg(dev, "%s\n", __func__);
263
264         nvdimm_bus_lock(dev);
265         for_each_dpa_resource_safe(ndd, res, _r)
266                 nvdimm_free_dpa(ndd, res);
267         nvdimm_bus_unlock(dev);
268
269         kvfree(ndd->data);
270         kfree(ndd);
271         put_device(dev);
272 }
273
274 void get_ndd(struct nvdimm_drvdata *ndd)
275 {
276         kref_get(&ndd->kref);
277 }
278
279 void put_ndd(struct nvdimm_drvdata *ndd)
280 {
281         if (ndd)
282                 kref_put(&ndd->kref, nvdimm_drvdata_release);
283 }
284
285 const char *nvdimm_name(struct nvdimm *nvdimm)
286 {
287         return dev_name(&nvdimm->dev);
288 }
289 EXPORT_SYMBOL_GPL(nvdimm_name);
290
291 struct kobject *nvdimm_kobj(struct nvdimm *nvdimm)
292 {
293         return &nvdimm->dev.kobj;
294 }
295 EXPORT_SYMBOL_GPL(nvdimm_kobj);
296
297 unsigned long nvdimm_cmd_mask(struct nvdimm *nvdimm)
298 {
299         return nvdimm->cmd_mask;
300 }
301 EXPORT_SYMBOL_GPL(nvdimm_cmd_mask);
302
303 void *nvdimm_provider_data(struct nvdimm *nvdimm)
304 {
305         if (nvdimm)
306                 return nvdimm->provider_data;
307         return NULL;
308 }
309 EXPORT_SYMBOL_GPL(nvdimm_provider_data);
310
311 static ssize_t commands_show(struct device *dev,
312                 struct device_attribute *attr, char *buf)
313 {
314         struct nvdimm *nvdimm = to_nvdimm(dev);
315         int cmd, len = 0;
316
317         if (!nvdimm->cmd_mask)
318                 return sprintf(buf, "\n");
319
320         for_each_set_bit(cmd, &nvdimm->cmd_mask, BITS_PER_LONG)
321                 len += sprintf(buf + len, "%s ", nvdimm_cmd_name(cmd));
322         len += sprintf(buf + len, "\n");
323         return len;
324 }
325 static DEVICE_ATTR_RO(commands);
326
327 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
328                 char *buf)
329 {
330         struct nvdimm *nvdimm = to_nvdimm(dev);
331
332         /*
333          * The state may be in the process of changing, userspace should
334          * quiesce probing if it wants a static answer
335          */
336         nvdimm_bus_lock(dev);
337         nvdimm_bus_unlock(dev);
338         return sprintf(buf, "%s\n", atomic_read(&nvdimm->busy)
339                         ? "active" : "idle");
340 }
341 static DEVICE_ATTR_RO(state);
342
343 static ssize_t available_slots_show(struct device *dev,
344                 struct device_attribute *attr, char *buf)
345 {
346         struct nvdimm_drvdata *ndd = dev_get_drvdata(dev);
347         ssize_t rc;
348         u32 nfree;
349
350         if (!ndd)
351                 return -ENXIO;
352
353         nvdimm_bus_lock(dev);
354         nfree = nd_label_nfree(ndd);
355         if (nfree - 1 > nfree) {
356                 dev_WARN_ONCE(dev, 1, "we ate our last label?\n");
357                 nfree = 0;
358         } else
359                 nfree--;
360         rc = sprintf(buf, "%d\n", nfree);
361         nvdimm_bus_unlock(dev);
362         return rc;
363 }
364 static DEVICE_ATTR_RO(available_slots);
365
366 static struct attribute *nvdimm_attributes[] = {
367         &dev_attr_state.attr,
368         &dev_attr_commands.attr,
369         &dev_attr_available_slots.attr,
370         NULL,
371 };
372
373 struct attribute_group nvdimm_attribute_group = {
374         .attrs = nvdimm_attributes,
375 };
376 EXPORT_SYMBOL_GPL(nvdimm_attribute_group);
377
378 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
379                 const struct attribute_group **groups, unsigned long flags,
380                 unsigned long cmd_mask, int num_flush,
381                 struct resource *flush_wpq)
382 {
383         struct nvdimm *nvdimm = kzalloc(sizeof(*nvdimm), GFP_KERNEL);
384         struct device *dev;
385
386         if (!nvdimm)
387                 return NULL;
388
389         nvdimm->id = ida_simple_get(&dimm_ida, 0, 0, GFP_KERNEL);
390         if (nvdimm->id < 0) {
391                 kfree(nvdimm);
392                 return NULL;
393         }
394         nvdimm->provider_data = provider_data;
395         nvdimm->flags = flags;
396         nvdimm->cmd_mask = cmd_mask;
397         nvdimm->num_flush = num_flush;
398         nvdimm->flush_wpq = flush_wpq;
399         atomic_set(&nvdimm->busy, 0);
400         dev = &nvdimm->dev;
401         dev_set_name(dev, "nmem%d", nvdimm->id);
402         dev->parent = &nvdimm_bus->dev;
403         dev->type = &nvdimm_device_type;
404         dev->devt = MKDEV(nvdimm_major, nvdimm->id);
405         dev->groups = groups;
406         nd_device_register(dev);
407
408         return nvdimm;
409 }
410 EXPORT_SYMBOL_GPL(nvdimm_create);
411
412 int alias_dpa_busy(struct device *dev, void *data)
413 {
414         resource_size_t map_end, blk_start, new;
415         struct blk_alloc_info *info = data;
416         struct nd_mapping *nd_mapping;
417         struct nd_region *nd_region;
418         struct nvdimm_drvdata *ndd;
419         struct resource *res;
420         int i;
421
422         if (!is_memory(dev))
423                 return 0;
424
425         nd_region = to_nd_region(dev);
426         for (i = 0; i < nd_region->ndr_mappings; i++) {
427                 nd_mapping  = &nd_region->mapping[i];
428                 if (nd_mapping->nvdimm == info->nd_mapping->nvdimm)
429                         break;
430         }
431
432         if (i >= nd_region->ndr_mappings)
433                 return 0;
434
435         ndd = to_ndd(nd_mapping);
436         map_end = nd_mapping->start + nd_mapping->size - 1;
437         blk_start = nd_mapping->start;
438
439         /*
440          * In the allocation case ->res is set to free space that we are
441          * looking to validate against PMEM aliasing collision rules
442          * (i.e. BLK is allocated after all aliased PMEM).
443          */
444         if (info->res) {
445                 if (info->res->start >= nd_mapping->start
446                                 && info->res->start < map_end)
447                         /* pass */;
448                 else
449                         return 0;
450         }
451
452  retry:
453         /*
454          * Find the free dpa from the end of the last pmem allocation to
455          * the end of the interleave-set mapping.
456          */
457         for_each_dpa_resource(ndd, res) {
458                 if (strncmp(res->name, "pmem", 4) != 0)
459                         continue;
460                 if ((res->start >= blk_start && res->start < map_end)
461                                 || (res->end >= blk_start
462                                         && res->end <= map_end)) {
463                         new = max(blk_start, min(map_end + 1, res->end + 1));
464                         if (new != blk_start) {
465                                 blk_start = new;
466                                 goto retry;
467                         }
468                 }
469         }
470
471         /* update the free space range with the probed blk_start */
472         if (info->res && blk_start > info->res->start) {
473                 info->res->start = max(info->res->start, blk_start);
474                 if (info->res->start > info->res->end)
475                         info->res->end = info->res->start - 1;
476                 return 1;
477         }
478
479         info->available -= blk_start - nd_mapping->start;
480
481         return 0;
482 }
483
484 /**
485  * nd_blk_available_dpa - account the unused dpa of BLK region
486  * @nd_mapping: container of dpa-resource-root + labels
487  *
488  * Unlike PMEM, BLK namespaces can occupy discontiguous DPA ranges, but
489  * we arrange for them to never start at an lower dpa than the last
490  * PMEM allocation in an aliased region.
491  */
492 resource_size_t nd_blk_available_dpa(struct nd_region *nd_region)
493 {
494         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
495         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
496         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
497         struct blk_alloc_info info = {
498                 .nd_mapping = nd_mapping,
499                 .available = nd_mapping->size,
500                 .res = NULL,
501         };
502         struct resource *res;
503
504         if (!ndd)
505                 return 0;
506
507         device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
508
509         /* now account for busy blk allocations in unaliased dpa */
510         for_each_dpa_resource(ndd, res) {
511                 if (strncmp(res->name, "blk", 3) != 0)
512                         continue;
513                 info.available -= resource_size(res);
514         }
515
516         return info.available;
517 }
518
519 /**
520  * nd_pmem_available_dpa - for the given dimm+region account unallocated dpa
521  * @nd_mapping: container of dpa-resource-root + labels
522  * @nd_region: constrain available space check to this reference region
523  * @overlap: calculate available space assuming this level of overlap
524  *
525  * Validate that a PMEM label, if present, aligns with the start of an
526  * interleave set and truncate the available size at the lowest BLK
527  * overlap point.
528  *
529  * The expectation is that this routine is called multiple times as it
530  * probes for the largest BLK encroachment for any single member DIMM of
531  * the interleave set.  Once that value is determined the PMEM-limit for
532  * the set can be established.
533  */
534 resource_size_t nd_pmem_available_dpa(struct nd_region *nd_region,
535                 struct nd_mapping *nd_mapping, resource_size_t *overlap)
536 {
537         resource_size_t map_start, map_end, busy = 0, available, blk_start;
538         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
539         struct resource *res;
540         const char *reason;
541
542         if (!ndd)
543                 return 0;
544
545         map_start = nd_mapping->start;
546         map_end = map_start + nd_mapping->size - 1;
547         blk_start = max(map_start, map_end + 1 - *overlap);
548         for_each_dpa_resource(ndd, res) {
549                 if (res->start >= map_start && res->start < map_end) {
550                         if (strncmp(res->name, "blk", 3) == 0)
551                                 blk_start = min(blk_start,
552                                                 max(map_start, res->start));
553                         else if (res->end > map_end) {
554                                 reason = "misaligned to iset";
555                                 goto err;
556                         } else
557                                 busy += resource_size(res);
558                 } else if (res->end >= map_start && res->end <= map_end) {
559                         if (strncmp(res->name, "blk", 3) == 0) {
560                                 /*
561                                  * If a BLK allocation overlaps the start of
562                                  * PMEM the entire interleave set may now only
563                                  * be used for BLK.
564                                  */
565                                 blk_start = map_start;
566                         } else
567                                 busy += resource_size(res);
568                 } else if (map_start > res->start && map_start < res->end) {
569                         /* total eclipse of the mapping */
570                         busy += nd_mapping->size;
571                         blk_start = map_start;
572                 }
573         }
574
575         *overlap = map_end + 1 - blk_start;
576         available = blk_start - map_start;
577         if (busy < available)
578                 return available - busy;
579         return 0;
580
581  err:
582         nd_dbg_dpa(nd_region, ndd, res, "%s\n", reason);
583         return 0;
584 }
585
586 void nvdimm_free_dpa(struct nvdimm_drvdata *ndd, struct resource *res)
587 {
588         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
589         kfree(res->name);
590         __release_region(&ndd->dpa, res->start, resource_size(res));
591 }
592
593 struct resource *nvdimm_allocate_dpa(struct nvdimm_drvdata *ndd,
594                 struct nd_label_id *label_id, resource_size_t start,
595                 resource_size_t n)
596 {
597         char *name = kmemdup(label_id, sizeof(*label_id), GFP_KERNEL);
598         struct resource *res;
599
600         if (!name)
601                 return NULL;
602
603         WARN_ON_ONCE(!is_nvdimm_bus_locked(ndd->dev));
604         res = __request_region(&ndd->dpa, start, n, name, 0);
605         if (!res)
606                 kfree(name);
607         return res;
608 }
609
610 /**
611  * nvdimm_allocated_dpa - sum up the dpa currently allocated to this label_id
612  * @nvdimm: container of dpa-resource-root + labels
613  * @label_id: dpa resource name of the form {pmem|blk}-<human readable uuid>
614  */
615 resource_size_t nvdimm_allocated_dpa(struct nvdimm_drvdata *ndd,
616                 struct nd_label_id *label_id)
617 {
618         resource_size_t allocated = 0;
619         struct resource *res;
620
621         for_each_dpa_resource(ndd, res)
622                 if (strcmp(res->name, label_id->id) == 0)
623                         allocated += resource_size(res);
624
625         return allocated;
626 }
627
628 static int count_dimms(struct device *dev, void *c)
629 {
630         int *count = c;
631
632         if (is_nvdimm(dev))
633                 (*count)++;
634         return 0;
635 }
636
637 int nvdimm_bus_check_dimm_count(struct nvdimm_bus *nvdimm_bus, int dimm_count)
638 {
639         int count = 0;
640         /* Flush any possible dimm registration failures */
641         nd_synchronize();
642
643         device_for_each_child(&nvdimm_bus->dev, &count, count_dimms);
644         dev_dbg(&nvdimm_bus->dev, "%s: count: %d\n", __func__, count);
645         if (count != dimm_count)
646                 return -ENXIO;
647         return 0;
648 }
649 EXPORT_SYMBOL_GPL(nvdimm_bus_check_dimm_count);
650
651 void __exit nvdimm_devs_exit(void)
652 {
653         ida_destroy(&dimm_ida);
654 }