?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.1.15
[linux/fpc-iii.git] / drivers / nvdimm / namespace_devs.c
blobd0214644e3341d90d38f41589d3d58921dc1b8fd
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/module.h>
14 #include <linux/device.h>
15 #include <linux/sort.h>
16 #include <linux/slab.h>
17 #include <linux/list.h>
18 #include <linux/nd.h>
19 #include "nd-core.h"
20 #include "pmem.h"
21 #include "nd.h"
22
23 static void namespace_io_release(struct device *dev)
24 {
25 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
26
27 kfree(nsio);
28 }
29
30 static void namespace_pmem_release(struct device *dev)
31 {
32 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
33 struct nd_region *nd_region = to_nd_region(dev->parent);
34
35 if (nspm->id >= 0)
36 ida_simple_remove(&nd_region->ns_ida, nspm->id);
37 kfree(nspm->alt_name);
38 kfree(nspm->uuid);
39 kfree(nspm);
40 }
41
42 static void namespace_blk_release(struct device *dev)
43 {
44 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
45 struct nd_region *nd_region = to_nd_region(dev->parent);
46
47 if (nsblk->id >= 0)
48 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
49 kfree(nsblk->alt_name);
50 kfree(nsblk->uuid);
51 kfree(nsblk->res);
52 kfree(nsblk);
53 }
54
55 static const struct device_type namespace_io_device_type = {
56 .name = "nd_namespace_io",
57 .release = namespace_io_release,
58 };
59
60 static const struct device_type namespace_pmem_device_type = {
61 .name = "nd_namespace_pmem",
62 .release = namespace_pmem_release,
63 };
64
65 static const struct device_type namespace_blk_device_type = {
66 .name = "nd_namespace_blk",
67 .release = namespace_blk_release,
68 };
69
70 static bool is_namespace_pmem(const struct device *dev)
71 {
72 return dev ? dev->type == &namespace_pmem_device_type : false;
73 }
74
75 static bool is_namespace_blk(const struct device *dev)
76 {
77 return dev ? dev->type == &namespace_blk_device_type : false;
78 }
79
80 static bool is_namespace_io(const struct device *dev)
81 {
82 return dev ? dev->type == &namespace_io_device_type : false;
83 }
84
85 static int is_uuid_busy(struct device *dev, void *data)
86 {
87 u8 *uuid1 = data, *uuid2 = NULL;
88
89 if (is_namespace_pmem(dev)) {
90 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
91
92 uuid2 = nspm->uuid;
93 } else if (is_namespace_blk(dev)) {
94 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
95
96 uuid2 = nsblk->uuid;
97 } else if (is_nd_btt(dev)) {
98 struct nd_btt *nd_btt = to_nd_btt(dev);
99
100 uuid2 = nd_btt->uuid;
101 } else if (is_nd_pfn(dev)) {
102 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
103
104 uuid2 = nd_pfn->uuid;
105 }
106
107 if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
108 return -EBUSY;
109
110 return 0;
111 }
112
113 static int is_namespace_uuid_busy(struct device *dev, void *data)
114 {
115 if (is_nd_region(dev))
116 return device_for_each_child(dev, data, is_uuid_busy);
117 return 0;
118 }
119
120 /**
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
124 */
125 bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
126 {
127 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
128
129 if (!nvdimm_bus)
130 return false;
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)
134 return false;
135 return true;
136 }
137
138 bool pmem_should_map_pages(struct device *dev)
139 {
140 struct nd_region *nd_region = to_nd_region(dev->parent);
141 struct nd_namespace_common *ndns = to_ndns(dev);
142 struct nd_namespace_io *nsio;
143
144 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
145 return false;
146
147 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
148 return false;
149
150 if (is_nd_pfn(dev) || is_nd_btt(dev))
151 return false;
152
153 if (ndns->force_raw)
154 return false;
155
156 nsio = to_nd_namespace_io(dev);
157 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
158 IORESOURCE_SYSTEM_RAM,
159 IORES_DESC_NONE) == REGION_MIXED)
160 return false;
161
162 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
163 }
164 EXPORT_SYMBOL(pmem_should_map_pages);
165
166 unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
167 {
168 if (is_namespace_pmem(&ndns->dev)) {
169 struct nd_namespace_pmem *nspm;
170
171 nspm = to_nd_namespace_pmem(&ndns->dev);
172 if (nspm->lbasize == 0 || nspm->lbasize == 512)
173 /* default */;
174 else if (nspm->lbasize == 4096)
175 return 4096;
176 else
177 dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
178 nspm->lbasize);
179 }
180
181 /*
182 * There is no namespace label (is_namespace_io()), or the label
183 * indicates the default sector size.
184 */
185 return 512;
186 }
187 EXPORT_SYMBOL(pmem_sector_size);
188
189 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
190 char *name)
191 {
192 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
193 const char *suffix = NULL;
194
195 if (ndns->claim && is_nd_btt(ndns->claim))
196 suffix = "s";
197
198 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
199 int nsidx = 0;
200
201 if (is_namespace_pmem(&ndns->dev)) {
202 struct nd_namespace_pmem *nspm;
203
204 nspm = to_nd_namespace_pmem(&ndns->dev);
205 nsidx = nspm->id;
206 }
207
208 if (nsidx)
209 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
210 suffix ? suffix : "");
211 else
212 sprintf(name, "pmem%d%s", nd_region->id,
213 suffix ? suffix : "");
214 } else if (is_namespace_blk(&ndns->dev)) {
215 struct nd_namespace_blk *nsblk;
216
217 nsblk = to_nd_namespace_blk(&ndns->dev);
218 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
219 suffix ? suffix : "");
220 } else {
221 return NULL;
222 }
223
224 return name;
225 }
226 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
227
228 const u8 *nd_dev_to_uuid(struct device *dev)
229 {
230 static const u8 null_uuid[16];
231
232 if (!dev)
233 return null_uuid;
234
235 if (is_namespace_pmem(dev)) {
236 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
237
238 return nspm->uuid;
239 } else if (is_namespace_blk(dev)) {
240 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
241
242 return nsblk->uuid;
243 } else
244 return null_uuid;
245 }
246 EXPORT_SYMBOL(nd_dev_to_uuid);
247
248 static ssize_t nstype_show(struct device *dev,
249 struct device_attribute *attr, char *buf)
250 {
251 struct nd_region *nd_region = to_nd_region(dev->parent);
252
253 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
254 }
255 static DEVICE_ATTR_RO(nstype);
256
257 static ssize_t __alt_name_store(struct device *dev, const char *buf,
258 const size_t len)
259 {
260 char *input, *pos, *alt_name, **ns_altname;
261 ssize_t rc;
262
263 if (is_namespace_pmem(dev)) {
264 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
265
266 ns_altname = &nspm->alt_name;
267 } else if (is_namespace_blk(dev)) {
268 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
269
270 ns_altname = &nsblk->alt_name;
271 } else
272 return -ENXIO;
273
274 if (dev->driver || to_ndns(dev)->claim)
275 return -EBUSY;
276
277 input = kstrndup(buf, len, GFP_KERNEL);
278 if (!input)
279 return -ENOMEM;
280
281 pos = strim(input);
282 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
283 rc = -EINVAL;
284 goto out;
285 }
286
287 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
288 if (!alt_name) {
289 rc = -ENOMEM;
290 goto out;
291 }
292 kfree(*ns_altname);
293 *ns_altname = alt_name;
294 sprintf(*ns_altname, "%s", pos);
295 rc = len;
296
297 out:
298 kfree(input);
299 return rc;
300 }
301
302 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
303 {
304 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
305 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
306 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
307 struct nd_label_id label_id;
308 resource_size_t size = 0;
309 struct resource *res;
310
311 if (!nsblk->uuid)
312 return 0;
313 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
314 for_each_dpa_resource(ndd, res)
315 if (strcmp(res->name, label_id.id) == 0)
316 size += resource_size(res);
317 return size;
318 }
319
320 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
321 {
322 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
323 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
324 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
325 struct nd_label_id label_id;
326 struct resource *res;
327 int count, i;
328
329 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
330 return false;
331
332 count = 0;
333 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
334 for_each_dpa_resource(ndd, res) {
335 if (strcmp(res->name, label_id.id) != 0)
336 continue;
337 /*
338 * Resources with unacknowledged adjustments indicate a
339 * failure to update labels
340 */
341 if (res->flags & DPA_RESOURCE_ADJUSTED)
342 return false;
343 count++;
344 }
345
346 /* These values match after a successful label update */
347 if (count != nsblk->num_resources)
348 return false;
349
350 for (i = 0; i < nsblk->num_resources; i++) {
351 struct resource *found = NULL;
352
353 for_each_dpa_resource(ndd, res)
354 if (res == nsblk->res[i]) {
355 found = res;
356 break;
357 }
358 /* stale resource */
359 if (!found)
360 return false;
361 }
362
363 return true;
364 }
365
366 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
367 {
368 resource_size_t size;
369
370 nvdimm_bus_lock(&nsblk->common.dev);
371 size = __nd_namespace_blk_validate(nsblk);
372 nvdimm_bus_unlock(&nsblk->common.dev);
373
374 return size;
375 }
376 EXPORT_SYMBOL(nd_namespace_blk_validate);
377
378
379 static int nd_namespace_label_update(struct nd_region *nd_region,
380 struct device *dev)
381 {
382 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
383 "namespace must be idle during label update\n");
384 if (dev->driver || to_ndns(dev)->claim)
385 return 0;
386
387 /*
388 * Only allow label writes that will result in a valid namespace
389 * or deletion of an existing namespace.
390 */
391 if (is_namespace_pmem(dev)) {
392 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
393 resource_size_t size = resource_size(&nspm->nsio.res);
394
395 if (size == 0 && nspm->uuid)
396 /* delete allocation */;
397 else if (!nspm->uuid)
398 return 0;
399
400 return nd_pmem_namespace_label_update(nd_region, nspm, size);
401 } else if (is_namespace_blk(dev)) {
402 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
403 resource_size_t size = nd_namespace_blk_size(nsblk);
404
405 if (size == 0 && nsblk->uuid)
406 /* delete allocation */;
407 else if (!nsblk->uuid || !nsblk->lbasize)
408 return 0;
409
410 return nd_blk_namespace_label_update(nd_region, nsblk, size);
411 } else
412 return -ENXIO;
413 }
414
415 static ssize_t alt_name_store(struct device *dev,
416 struct device_attribute *attr, const char *buf, size_t len)
417 {
418 struct nd_region *nd_region = to_nd_region(dev->parent);
419 ssize_t rc;
420
421 device_lock(dev);
422 nvdimm_bus_lock(dev);
423 wait_nvdimm_bus_probe_idle(dev);
424 rc = __alt_name_store(dev, buf, len);
425 if (rc >= 0)
426 rc = nd_namespace_label_update(nd_region, dev);
427 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
428 nvdimm_bus_unlock(dev);
429 device_unlock(dev);
430
431 return rc < 0 ? rc : len;
432 }
433
434 static ssize_t alt_name_show(struct device *dev,
435 struct device_attribute *attr, char *buf)
436 {
437 char *ns_altname;
438
439 if (is_namespace_pmem(dev)) {
440 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
441
442 ns_altname = nspm->alt_name;
443 } else if (is_namespace_blk(dev)) {
444 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
445
446 ns_altname = nsblk->alt_name;
447 } else
448 return -ENXIO;
449
450 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
451 }
452 static DEVICE_ATTR_RW(alt_name);
453
454 static int scan_free(struct nd_region *nd_region,
455 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
456 resource_size_t n)
457 {
458 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
459 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
460 int rc = 0;
461
462 while (n) {
463 struct resource *res, *last;
464 resource_size_t new_start;
465
466 last = NULL;
467 for_each_dpa_resource(ndd, res)
468 if (strcmp(res->name, label_id->id) == 0)
469 last = res;
470 res = last;
471 if (!res)
472 return 0;
473
474 if (n >= resource_size(res)) {
475 n -= resource_size(res);
476 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
477 nvdimm_free_dpa(ndd, res);
478 /* retry with last resource deleted */
479 continue;
480 }
481
482 /*
483 * Keep BLK allocations relegated to high DPA as much as
484 * possible
485 */
486 if (is_blk)
487 new_start = res->start + n;
488 else
489 new_start = res->start;
490
491 rc = adjust_resource(res, new_start, resource_size(res) - n);
492 if (rc == 0)
493 res->flags |= DPA_RESOURCE_ADJUSTED;
494 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
495 break;
496 }
497
498 return rc;
499 }
500
501 /**
502 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
503 * @nd_region: the set of dimms to reclaim @n bytes from
504 * @label_id: unique identifier for the namespace consuming this dpa range
505 * @n: number of bytes per-dimm to release
506 *
507 * Assumes resources are ordered. Starting from the end try to
508 * adjust_resource() the allocation to @n, but if @n is larger than the
509 * allocation delete it and find the 'new' last allocation in the label
510 * set.
511 */
512 static int shrink_dpa_allocation(struct nd_region *nd_region,
513 struct nd_label_id *label_id, resource_size_t n)
514 {
515 int i;
516
517 for (i = 0; i < nd_region->ndr_mappings; i++) {
518 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
519 int rc;
520
521 rc = scan_free(nd_region, nd_mapping, label_id, n);
522 if (rc)
523 return rc;
524 }
525
526 return 0;
527 }
528
529 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
530 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
531 resource_size_t n)
532 {
533 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
534 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
535 resource_size_t first_dpa;
536 struct resource *res;
537 int rc = 0;
538
539 /* allocate blk from highest dpa first */
540 if (is_blk)
541 first_dpa = nd_mapping->start + nd_mapping->size - n;
542 else
543 first_dpa = nd_mapping->start;
544
545 /* first resource allocation for this label-id or dimm */
546 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
547 if (!res)
548 rc = -EBUSY;
549
550 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
551 return rc ? n : 0;
552 }
553
554
555 /**
556 * space_valid() - validate free dpa space against constraints
557 * @nd_region: hosting region of the free space
558 * @ndd: dimm device data for debug
559 * @label_id: namespace id to allocate space
560 * @prev: potential allocation that precedes free space
561 * @next: allocation that follows the given free space range
562 * @exist: first allocation with same id in the mapping
563 * @n: range that must satisfied for pmem allocations
564 * @valid: free space range to validate
565 *
566 * BLK-space is valid as long as it does not precede a PMEM
567 * allocation in a given region. PMEM-space must be contiguous
568 * and adjacent to an existing existing allocation (if one
569 * exists). If reserving PMEM any space is valid.
570 */
571 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
572 struct nd_label_id *label_id, struct resource *prev,
573 struct resource *next, struct resource *exist,
574 resource_size_t n, struct resource *valid)
575 {
576 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
577 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
578
579 if (valid->start >= valid->end)
580 goto invalid;
581
582 if (is_reserve)
583 return;
584
585 if (!is_pmem) {
586 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
587 struct nvdimm_bus *nvdimm_bus;
588 struct blk_alloc_info info = {
589 .nd_mapping = nd_mapping,
590 .available = nd_mapping->size,
591 .res = valid,
592 };
593
594 WARN_ON(!is_nd_blk(&nd_region->dev));
595 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
596 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
597 return;
598 }
599
600 /* allocation needs to be contiguous, so this is all or nothing */
601 if (resource_size(valid) < n)
602 goto invalid;
603
604 /* we've got all the space we need and no existing allocation */
605 if (!exist)
606 return;
607
608 /* allocation needs to be contiguous with the existing namespace */
609 if (valid->start == exist->end + 1
610 || valid->end == exist->start - 1)
611 return;
612
613 invalid:
614 /* truncate @valid size to 0 */
615 valid->end = valid->start - 1;
616 }
617
618 enum alloc_loc {
619 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
620 };
621
622 static resource_size_t scan_allocate(struct nd_region *nd_region,
623 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
624 resource_size_t n)
625 {
626 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
627 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
628 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
629 struct resource *res, *exist = NULL, valid;
630 const resource_size_t to_allocate = n;
631 int first;
632
633 for_each_dpa_resource(ndd, res)
634 if (strcmp(label_id->id, res->name) == 0)
635 exist = res;
636
637 valid.start = nd_mapping->start;
638 valid.end = mapping_end;
639 valid.name = "free space";
640 retry:
641 first = 0;
642 for_each_dpa_resource(ndd, res) {
643 struct resource *next = res->sibling, *new_res = NULL;
644 resource_size_t allocate, available = 0;
645 enum alloc_loc loc = ALLOC_ERR;
646 const char *action;
647 int rc = 0;
648
649 /* ignore resources outside this nd_mapping */
650 if (res->start > mapping_end)
651 continue;
652 if (res->end < nd_mapping->start)
653 continue;
654
655 /* space at the beginning of the mapping */
656 if (!first++ && res->start > nd_mapping->start) {
657 valid.start = nd_mapping->start;
658 valid.end = res->start - 1;
659 space_valid(nd_region, ndd, label_id, NULL, next, exist,
660 to_allocate, &valid);
661 available = resource_size(&valid);
662 if (available)
663 loc = ALLOC_BEFORE;
664 }
665
666 /* space between allocations */
667 if (!loc && next) {
668 valid.start = res->start + resource_size(res);
669 valid.end = min(mapping_end, next->start - 1);
670 space_valid(nd_region, ndd, label_id, res, next, exist,
671 to_allocate, &valid);
672 available = resource_size(&valid);
673 if (available)
674 loc = ALLOC_MID;
675 }
676
677 /* space at the end of the mapping */
678 if (!loc && !next) {
679 valid.start = res->start + resource_size(res);
680 valid.end = mapping_end;
681 space_valid(nd_region, ndd, label_id, res, next, exist,
682 to_allocate, &valid);
683 available = resource_size(&valid);
684 if (available)
685 loc = ALLOC_AFTER;
686 }
687
688 if (!loc || !available)
689 continue;
690 allocate = min(available, n);
691 switch (loc) {
692 case ALLOC_BEFORE:
693 if (strcmp(res->name, label_id->id) == 0) {
694 /* adjust current resource up */
695 rc = adjust_resource(res, res->start - allocate,
696 resource_size(res) + allocate);
697 action = "cur grow up";
698 } else
699 action = "allocate";
700 break;
701 case ALLOC_MID:
702 if (strcmp(next->name, label_id->id) == 0) {
703 /* adjust next resource up */
704 rc = adjust_resource(next, next->start
705 - allocate, resource_size(next)
706 + allocate);
707 new_res = next;
708 action = "next grow up";
709 } else if (strcmp(res->name, label_id->id) == 0) {
710 action = "grow down";
711 } else
712 action = "allocate";
713 break;
714 case ALLOC_AFTER:
715 if (strcmp(res->name, label_id->id) == 0)
716 action = "grow down";
717 else
718 action = "allocate";
719 break;
720 default:
721 return n;
722 }
723
724 if (strcmp(action, "allocate") == 0) {
725 /* BLK allocate bottom up */
726 if (!is_pmem)
727 valid.start += available - allocate;
728
729 new_res = nvdimm_allocate_dpa(ndd, label_id,
730 valid.start, allocate);
731 if (!new_res)
732 rc = -EBUSY;
733 } else if (strcmp(action, "grow down") == 0) {
734 /* adjust current resource down */
735 rc = adjust_resource(res, res->start, resource_size(res)
736 + allocate);
737 if (rc == 0)
738 res->flags |= DPA_RESOURCE_ADJUSTED;
739 }
740
741 if (!new_res)
742 new_res = res;
743
744 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
745 action, loc, rc);
746
747 if (rc)
748 return n;
749
750 n -= allocate;
751 if (n) {
752 /*
753 * Retry scan with newly inserted resources.
754 * For example, if we did an ALLOC_BEFORE
755 * insertion there may also have been space
756 * available for an ALLOC_AFTER insertion, so we
757 * need to check this same resource again
758 */
759 goto retry;
760 } else
761 return 0;
762 }
763
764 /*
765 * If we allocated nothing in the BLK case it may be because we are in
766 * an initial "pmem-reserve pass". Only do an initial BLK allocation
767 * when none of the DPA space is reserved.
768 */
769 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
770 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
771 return n;
772 }
773
774 static int merge_dpa(struct nd_region *nd_region,
775 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
776 {
777 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
778 struct resource *res;
779
780 if (strncmp("pmem", label_id->id, 4) == 0)
781 return 0;
782 retry:
783 for_each_dpa_resource(ndd, res) {
784 int rc;
785 struct resource *next = res->sibling;
786 resource_size_t end = res->start + resource_size(res);
787
788 if (!next || strcmp(res->name, label_id->id) != 0
789 || strcmp(next->name, label_id->id) != 0
790 || end != next->start)
791 continue;
792 end += resource_size(next);
793 nvdimm_free_dpa(ndd, next);
794 rc = adjust_resource(res, res->start, end - res->start);
795 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
796 if (rc)
797 return rc;
798 res->flags |= DPA_RESOURCE_ADJUSTED;
799 goto retry;
800 }
801
802 return 0;
803 }
804
805 int __reserve_free_pmem(struct device *dev, void *data)
806 {
807 struct nvdimm *nvdimm = data;
808 struct nd_region *nd_region;
809 struct nd_label_id label_id;
810 int i;
811
812 if (!is_memory(dev))
813 return 0;
814
815 nd_region = to_nd_region(dev);
816 if (nd_region->ndr_mappings == 0)
817 return 0;
818
819 memset(&label_id, 0, sizeof(label_id));
820 strcat(label_id.id, "pmem-reserve");
821 for (i = 0; i < nd_region->ndr_mappings; i++) {
822 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
823 resource_size_t n, rem = 0;
824
825 if (nd_mapping->nvdimm != nvdimm)
826 continue;
827
828 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
829 if (n == 0)
830 return 0;
831 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
832 dev_WARN_ONCE(&nd_region->dev, rem,
833 "pmem reserve underrun: %#llx of %#llx bytes\n",
834 (unsigned long long) n - rem,
835 (unsigned long long) n);
836 return rem ? -ENXIO : 0;
837 }
838
839 return 0;
840 }
841
842 void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
843 struct nd_mapping *nd_mapping)
844 {
845 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
846 struct resource *res, *_res;
847
848 for_each_dpa_resource_safe(ndd, res, _res)
849 if (strcmp(res->name, "pmem-reserve") == 0)
850 nvdimm_free_dpa(ndd, res);
851 }
852
853 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
854 struct nd_mapping *nd_mapping)
855 {
856 struct nvdimm *nvdimm = nd_mapping->nvdimm;
857 int rc;
858
859 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
860 __reserve_free_pmem);
861 if (rc)
862 release_free_pmem(nvdimm_bus, nd_mapping);
863 return rc;
864 }
865
866 /**
867 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
868 * @nd_region: the set of dimms to allocate @n more bytes from
869 * @label_id: unique identifier for the namespace consuming this dpa range
870 * @n: number of bytes per-dimm to add to the existing allocation
871 *
872 * Assumes resources are ordered. For BLK regions, first consume
873 * BLK-only available DPA free space, then consume PMEM-aliased DPA
874 * space starting at the highest DPA. For PMEM regions start
875 * allocations from the start of an interleave set and end at the first
876 * BLK allocation or the end of the interleave set, whichever comes
877 * first.
878 */
879 static int grow_dpa_allocation(struct nd_region *nd_region,
880 struct nd_label_id *label_id, resource_size_t n)
881 {
882 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
883 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
884 int i;
885
886 for (i = 0; i < nd_region->ndr_mappings; i++) {
887 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
888 resource_size_t rem = n;
889 int rc, j;
890
891 /*
892 * In the BLK case try once with all unallocated PMEM
893 * reserved, and once without
894 */
895 for (j = is_pmem; j < 2; j++) {
896 bool blk_only = j == 0;
897
898 if (blk_only) {
899 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
900 if (rc)
901 return rc;
902 }
903 rem = scan_allocate(nd_region, nd_mapping,
904 label_id, rem);
905 if (blk_only)
906 release_free_pmem(nvdimm_bus, nd_mapping);
907
908 /* try again and allow encroachments into PMEM */
909 if (rem == 0)
910 break;
911 }
912
913 dev_WARN_ONCE(&nd_region->dev, rem,
914 "allocation underrun: %#llx of %#llx bytes\n",
915 (unsigned long long) n - rem,
916 (unsigned long long) n);
917 if (rem)
918 return -ENXIO;
919
920 rc = merge_dpa(nd_region, nd_mapping, label_id);
921 if (rc)
922 return rc;
923 }
924
925 return 0;
926 }
927
928 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
929 struct nd_namespace_pmem *nspm, resource_size_t size)
930 {
931 struct resource *res = &nspm->nsio.res;
932 resource_size_t offset = 0;
933
934 if (size && !nspm->uuid) {
935 WARN_ON_ONCE(1);
936 size = 0;
937 }
938
939 if (size && nspm->uuid) {
940 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
941 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
942 struct nd_label_id label_id;
943 struct resource *res;
944
945 if (!ndd) {
946 size = 0;
947 goto out;
948 }
949
950 nd_label_gen_id(&label_id, nspm->uuid, 0);
951
952 /* calculate a spa offset from the dpa allocation offset */
953 for_each_dpa_resource(ndd, res)
954 if (strcmp(res->name, label_id.id) == 0) {
955 offset = (res->start - nd_mapping->start)
956 * nd_region->ndr_mappings;
957 goto out;
958 }
959
960 WARN_ON_ONCE(1);
961 size = 0;
962 }
963
964 out:
965 res->start = nd_region->ndr_start + offset;
966 res->end = res->start + size - 1;
967 }
968
969 static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
970 {
971 if (!uuid) {
972 dev_dbg(dev, "%s: uuid not set\n", where);
973 return true;
974 }
975 return false;
976 }
977
978 static ssize_t __size_store(struct device *dev, unsigned long long val)
979 {
980 resource_size_t allocated = 0, available = 0;
981 struct nd_region *nd_region = to_nd_region(dev->parent);
982 struct nd_namespace_common *ndns = to_ndns(dev);
983 struct nd_mapping *nd_mapping;
984 struct nvdimm_drvdata *ndd;
985 struct nd_label_id label_id;
986 u32 flags = 0, remainder;
987 int rc, i, id = -1;
988 u8 *uuid = NULL;
989
990 if (dev->driver || ndns->claim)
991 return -EBUSY;
992
993 if (is_namespace_pmem(dev)) {
994 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
995
996 uuid = nspm->uuid;
997 id = nspm->id;
998 } else if (is_namespace_blk(dev)) {
999 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1000
1001 uuid = nsblk->uuid;
1002 flags = NSLABEL_FLAG_LOCAL;
1003 id = nsblk->id;
1004 }
1005
1006 /*
1007 * We need a uuid for the allocation-label and dimm(s) on which
1008 * to store the label.
1009 */
1010 if (uuid_not_set(uuid, dev, __func__))
1011 return -ENXIO;
1012 if (nd_region->ndr_mappings == 0) {
1013 dev_dbg(dev, "not associated with dimm(s)\n");
1014 return -ENXIO;
1015 }
1016
1017 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
1018 if (remainder) {
1019 dev_dbg(dev, "%llu is not %dK aligned\n", val,
1020 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
1021 return -EINVAL;
1022 }
1023
1024 nd_label_gen_id(&label_id, uuid, flags);
1025 for (i = 0; i < nd_region->ndr_mappings; i++) {
1026 nd_mapping = &nd_region->mapping[i];
1027 ndd = to_ndd(nd_mapping);
1028
1029 /*
1030 * All dimms in an interleave set, or the base dimm for a blk
1031 * region, need to be enabled for the size to be changed.
1032 */
1033 if (!ndd)
1034 return -ENXIO;
1035
1036 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1037 }
1038 available = nd_region_allocatable_dpa(nd_region);
1039
1040 if (val > available + allocated)
1041 return -ENOSPC;
1042
1043 if (val == allocated)
1044 return 0;
1045
1046 val = div_u64(val, nd_region->ndr_mappings);
1047 allocated = div_u64(allocated, nd_region->ndr_mappings);
1048 if (val < allocated)
1049 rc = shrink_dpa_allocation(nd_region, &label_id,
1050 allocated - val);
1051 else
1052 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1053
1054 if (rc)
1055 return rc;
1056
1057 if (is_namespace_pmem(dev)) {
1058 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1059
1060 nd_namespace_pmem_set_resource(nd_region, nspm,
1061 val * nd_region->ndr_mappings);
1062 }
1063
1064 /*
1065 * Try to delete the namespace if we deleted all of its
1066 * allocation, this is not the seed or 0th device for the
1067 * region, and it is not actively claimed by a btt, pfn, or dax
1068 * instance.
1069 */
1070 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1071 nd_device_unregister(dev, ND_ASYNC);
1072
1073 return rc;
1074 }
1075
1076 static ssize_t size_store(struct device *dev,
1077 struct device_attribute *attr, const char *buf, size_t len)
1078 {
1079 struct nd_region *nd_region = to_nd_region(dev->parent);
1080 unsigned long long val;
1081 u8 **uuid = NULL;
1082 int rc;
1083
1084 rc = kstrtoull(buf, 0, &val);
1085 if (rc)
1086 return rc;
1087
1088 device_lock(dev);
1089 nvdimm_bus_lock(dev);
1090 wait_nvdimm_bus_probe_idle(dev);
1091 rc = __size_store(dev, val);
1092 if (rc >= 0)
1093 rc = nd_namespace_label_update(nd_region, dev);
1094
1095 if (is_namespace_pmem(dev)) {
1096 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1097
1098 uuid = &nspm->uuid;
1099 } else if (is_namespace_blk(dev)) {
1100 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1101
1102 uuid = &nsblk->uuid;
1103 }
1104
1105 if (rc == 0 && val == 0 && uuid) {
1106 /* setting size zero == 'delete namespace' */
1107 kfree(*uuid);
1108 *uuid = NULL;
1109 }
1110
1111 dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
1112
1113 nvdimm_bus_unlock(dev);
1114 device_unlock(dev);
1115
1116 return rc < 0 ? rc : len;
1117 }
1118
1119 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1120 {
1121 struct device *dev = &ndns->dev;
1122
1123 if (is_namespace_pmem(dev)) {
1124 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1125
1126 return resource_size(&nspm->nsio.res);
1127 } else if (is_namespace_blk(dev)) {
1128 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1129 } else if (is_namespace_io(dev)) {
1130 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1131
1132 return resource_size(&nsio->res);
1133 } else
1134 WARN_ONCE(1, "unknown namespace type\n");
1135 return 0;
1136 }
1137
1138 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1139 {
1140 resource_size_t size;
1141
1142 nvdimm_bus_lock(&ndns->dev);
1143 size = __nvdimm_namespace_capacity(ndns);
1144 nvdimm_bus_unlock(&ndns->dev);
1145
1146 return size;
1147 }
1148 EXPORT_SYMBOL(nvdimm_namespace_capacity);
1149
1150 bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
1151 {
1152 int i;
1153 bool locked = false;
1154 struct device *dev = &ndns->dev;
1155 struct nd_region *nd_region = to_nd_region(dev->parent);
1156
1157 for (i = 0; i < nd_region->ndr_mappings; i++) {
1158 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1159 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1160
1161 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
1162 dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
1163 locked = true;
1164 }
1165 }
1166 return locked;
1167 }
1168 EXPORT_SYMBOL(nvdimm_namespace_locked);
1169
1170 static ssize_t size_show(struct device *dev,
1171 struct device_attribute *attr, char *buf)
1172 {
1173 return sprintf(buf, "%llu\n", (unsigned long long)
1174 nvdimm_namespace_capacity(to_ndns(dev)));
1175 }
1176 static DEVICE_ATTR(size, 0444, size_show, size_store);
1177
1178 static u8 *namespace_to_uuid(struct device *dev)
1179 {
1180 if (is_namespace_pmem(dev)) {
1181 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1182
1183 return nspm->uuid;
1184 } else if (is_namespace_blk(dev)) {
1185 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1186
1187 return nsblk->uuid;
1188 } else
1189 return ERR_PTR(-ENXIO);
1190 }
1191
1192 static ssize_t uuid_show(struct device *dev,
1193 struct device_attribute *attr, char *buf)
1194 {
1195 u8 *uuid = namespace_to_uuid(dev);
1196
1197 if (IS_ERR(uuid))
1198 return PTR_ERR(uuid);
1199 if (uuid)
1200 return sprintf(buf, "%pUb\n", uuid);
1201 return sprintf(buf, "\n");
1202 }
1203
1204 /**
1205 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1206 * @nd_region: parent region so we can updates all dimms in the set
1207 * @dev: namespace type for generating label_id
1208 * @new_uuid: incoming uuid
1209 * @old_uuid: reference to the uuid storage location in the namespace object
1210 */
1211 static int namespace_update_uuid(struct nd_region *nd_region,
1212 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1213 {
1214 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1215 struct nd_label_id old_label_id;
1216 struct nd_label_id new_label_id;
1217 int i;
1218
1219 if (!nd_is_uuid_unique(dev, new_uuid))
1220 return -EINVAL;
1221
1222 if (*old_uuid == NULL)
1223 goto out;
1224
1225 /*
1226 * If we've already written a label with this uuid, then it's
1227 * too late to rename because we can't reliably update the uuid
1228 * without losing the old namespace. Userspace must delete this
1229 * namespace to abandon the old uuid.
1230 */
1231 for (i = 0; i < nd_region->ndr_mappings; i++) {
1232 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1233
1234 /*
1235 * This check by itself is sufficient because old_uuid
1236 * would be NULL above if this uuid did not exist in the
1237 * currently written set.
1238 *
1239 * FIXME: can we delete uuid with zero dpa allocated?
1240 */
1241 if (list_empty(&nd_mapping->labels))
1242 return -EBUSY;
1243 }
1244
1245 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1246 nd_label_gen_id(&new_label_id, new_uuid, flags);
1247 for (i = 0; i < nd_region->ndr_mappings; i++) {
1248 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1249 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1250 struct nd_label_ent *label_ent;
1251 struct resource *res;
1252
1253 for_each_dpa_resource(ndd, res)
1254 if (strcmp(res->name, old_label_id.id) == 0)
1255 sprintf((void *) res->name, "%s",
1256 new_label_id.id);
1257
1258 mutex_lock(&nd_mapping->lock);
1259 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1260 struct nd_namespace_label *nd_label = label_ent->label;
1261 struct nd_label_id label_id;
1262
1263 if (!nd_label)
1264 continue;
1265 nd_label_gen_id(&label_id, nd_label->uuid,
1266 __le32_to_cpu(nd_label->flags));
1267 if (strcmp(old_label_id.id, label_id.id) == 0)
1268 set_bit(ND_LABEL_REAP, &label_ent->flags);
1269 }
1270 mutex_unlock(&nd_mapping->lock);
1271 }
1272 kfree(*old_uuid);
1273 out:
1274 *old_uuid = new_uuid;
1275 return 0;
1276 }
1277
1278 static ssize_t uuid_store(struct device *dev,
1279 struct device_attribute *attr, const char *buf, size_t len)
1280 {
1281 struct nd_region *nd_region = to_nd_region(dev->parent);
1282 u8 *uuid = NULL;
1283 ssize_t rc = 0;
1284 u8 **ns_uuid;
1285
1286 if (is_namespace_pmem(dev)) {
1287 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1288
1289 ns_uuid = &nspm->uuid;
1290 } else if (is_namespace_blk(dev)) {
1291 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1292
1293 ns_uuid = &nsblk->uuid;
1294 } else
1295 return -ENXIO;
1296
1297 device_lock(dev);
1298 nvdimm_bus_lock(dev);
1299 wait_nvdimm_bus_probe_idle(dev);
1300 if (to_ndns(dev)->claim)
1301 rc = -EBUSY;
1302 if (rc >= 0)
1303 rc = nd_uuid_store(dev, &uuid, buf, len);
1304 if (rc >= 0)
1305 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1306 if (rc >= 0)
1307 rc = nd_namespace_label_update(nd_region, dev);
1308 else
1309 kfree(uuid);
1310 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1311 buf[len - 1] == '\n' ? "" : "\n");
1312 nvdimm_bus_unlock(dev);
1313 device_unlock(dev);
1314
1315 return rc < 0 ? rc : len;
1316 }
1317 static DEVICE_ATTR_RW(uuid);
1318
1319 static ssize_t resource_show(struct device *dev,
1320 struct device_attribute *attr, char *buf)
1321 {
1322 struct resource *res;
1323
1324 if (is_namespace_pmem(dev)) {
1325 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1326
1327 res = &nspm->nsio.res;
1328 } else if (is_namespace_io(dev)) {
1329 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1330
1331 res = &nsio->res;
1332 } else
1333 return -ENXIO;
1334
1335 /* no address to convey if the namespace has no allocation */
1336 if (resource_size(res) == 0)
1337 return -ENXIO;
1338 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1339 }
1340 static DEVICE_ATTR_RO(resource);
1341
1342 static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1343 4096, 4104, 4160, 4224, 0 };
1344
1345 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1346
1347 static ssize_t sector_size_show(struct device *dev,
1348 struct device_attribute *attr, char *buf)
1349 {
1350 if (is_namespace_blk(dev)) {
1351 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1352
1353 return nd_size_select_show(nsblk->lbasize,
1354 blk_lbasize_supported, buf);
1355 }
1356
1357 if (is_namespace_pmem(dev)) {
1358 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1359
1360 return nd_size_select_show(nspm->lbasize,
1361 pmem_lbasize_supported, buf);
1362 }
1363 return -ENXIO;
1364 }
1365
1366 static ssize_t sector_size_store(struct device *dev,
1367 struct device_attribute *attr, const char *buf, size_t len)
1368 {
1369 struct nd_region *nd_region = to_nd_region(dev->parent);
1370 const unsigned long *supported;
1371 unsigned long *lbasize;
1372 ssize_t rc = 0;
1373
1374 if (is_namespace_blk(dev)) {
1375 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1376
1377 lbasize = &nsblk->lbasize;
1378 supported = blk_lbasize_supported;
1379 } else if (is_namespace_pmem(dev)) {
1380 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1381
1382 lbasize = &nspm->lbasize;
1383 supported = pmem_lbasize_supported;
1384 } else
1385 return -ENXIO;
1386
1387 device_lock(dev);
1388 nvdimm_bus_lock(dev);
1389 if (to_ndns(dev)->claim)
1390 rc = -EBUSY;
1391 if (rc >= 0)
1392 rc = nd_size_select_store(dev, buf, lbasize, supported);
1393 if (rc >= 0)
1394 rc = nd_namespace_label_update(nd_region, dev);
1395 dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1396 buf, buf[len - 1] == '\n' ? "" : "\n");
1397 nvdimm_bus_unlock(dev);
1398 device_unlock(dev);
1399
1400 return rc ? rc : len;
1401 }
1402 static DEVICE_ATTR_RW(sector_size);
1403
1404 static ssize_t dpa_extents_show(struct device *dev,
1405 struct device_attribute *attr, char *buf)
1406 {
1407 struct nd_region *nd_region = to_nd_region(dev->parent);
1408 struct nd_label_id label_id;
1409 int count = 0, i;
1410 u8 *uuid = NULL;
1411 u32 flags = 0;
1412
1413 nvdimm_bus_lock(dev);
1414 if (is_namespace_pmem(dev)) {
1415 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1416
1417 uuid = nspm->uuid;
1418 flags = 0;
1419 } else if (is_namespace_blk(dev)) {
1420 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1421
1422 uuid = nsblk->uuid;
1423 flags = NSLABEL_FLAG_LOCAL;
1424 }
1425
1426 if (!uuid)
1427 goto out;
1428
1429 nd_label_gen_id(&label_id, uuid, flags);
1430 for (i = 0; i < nd_region->ndr_mappings; i++) {
1431 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1432 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1433 struct resource *res;
1434
1435 for_each_dpa_resource(ndd, res)
1436 if (strcmp(res->name, label_id.id) == 0)
1437 count++;
1438 }
1439 out:
1440 nvdimm_bus_unlock(dev);
1441
1442 return sprintf(buf, "%d\n", count);
1443 }
1444 static DEVICE_ATTR_RO(dpa_extents);
1445
1446 static int btt_claim_class(struct device *dev)
1447 {
1448 struct nd_region *nd_region = to_nd_region(dev->parent);
1449 int i, loop_bitmask = 0;
1450
1451 for (i = 0; i < nd_region->ndr_mappings; i++) {
1452 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1453 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1454 struct nd_namespace_index *nsindex;
1455
1456 /*
1457 * If any of the DIMMs do not support labels the only
1458 * possible BTT format is v1.
1459 */
1460 if (!ndd) {
1461 loop_bitmask = 0;
1462 break;
1463 }
1464
1465 nsindex = to_namespace_index(ndd, ndd->ns_current);
1466 if (nsindex == NULL)
1467 loop_bitmask |= 1;
1468 else {
1469 /* check whether existing labels are v1.1 or v1.2 */
1470 if (__le16_to_cpu(nsindex->major) == 1
1471 && __le16_to_cpu(nsindex->minor) == 1)
1472 loop_bitmask |= 2;
1473 else
1474 loop_bitmask |= 4;
1475 }
1476 }
1477 /*
1478 * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1479 * block is found, a v1.1 label for any mapping will set bit 1, and a
1480 * v1.2 label will set bit 2.
1481 *
1482 * At the end of the loop, at most one of the three bits must be set.
1483 * If multiple bits were set, it means the different mappings disagree
1484 * about their labels, and this must be cleaned up first.
1485 *
1486 * If all the label index blocks are found to agree, nsindex of NULL
1487 * implies labels haven't been initialized yet, and when they will,
1488 * they will be of the 1.2 format, so we can assume BTT2.0
1489 *
1490 * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1491 * found, we enforce BTT2.0
1492 *
1493 * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1494 */
1495 switch (loop_bitmask) {
1496 case 0:
1497 case 2:
1498 return NVDIMM_CCLASS_BTT;
1499 case 1:
1500 case 4:
1501 return NVDIMM_CCLASS_BTT2;
1502 default:
1503 return -ENXIO;
1504 }
1505 }
1506
1507 static ssize_t holder_show(struct device *dev,
1508 struct device_attribute *attr, char *buf)
1509 {
1510 struct nd_namespace_common *ndns = to_ndns(dev);
1511 ssize_t rc;
1512
1513 device_lock(dev);
1514 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1515 device_unlock(dev);
1516
1517 return rc;
1518 }
1519 static DEVICE_ATTR_RO(holder);
1520
1521 static ssize_t __holder_class_store(struct device *dev, const char *buf)
1522 {
1523 struct nd_namespace_common *ndns = to_ndns(dev);
1524
1525 if (dev->driver || ndns->claim)
1526 return -EBUSY;
1527
1528 if (sysfs_streq(buf, "btt"))
1529 ndns->claim_class = btt_claim_class(dev);
1530 else if (sysfs_streq(buf, "pfn"))
1531 ndns->claim_class = NVDIMM_CCLASS_PFN;
1532 else if (sysfs_streq(buf, "dax"))
1533 ndns->claim_class = NVDIMM_CCLASS_DAX;
1534 else if (sysfs_streq(buf, ""))
1535 ndns->claim_class = NVDIMM_CCLASS_NONE;
1536 else
1537 return -EINVAL;
1538
1539 /* btt_claim_class() could've returned an error */
1540 if (ndns->claim_class < 0)
1541 return ndns->claim_class;
1542
1543 return 0;
1544 }
1545
1546 static ssize_t holder_class_store(struct device *dev,
1547 struct device_attribute *attr, const char *buf, size_t len)
1548 {
1549 struct nd_region *nd_region = to_nd_region(dev->parent);
1550 ssize_t rc;
1551
1552 device_lock(dev);
1553 nvdimm_bus_lock(dev);
1554 wait_nvdimm_bus_probe_idle(dev);
1555 rc = __holder_class_store(dev, buf);
1556 if (rc >= 0)
1557 rc = nd_namespace_label_update(nd_region, dev);
1558 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
1559 nvdimm_bus_unlock(dev);
1560 device_unlock(dev);
1561
1562 return rc < 0 ? rc : len;
1563 }
1564
1565 static ssize_t holder_class_show(struct device *dev,
1566 struct device_attribute *attr, char *buf)
1567 {
1568 struct nd_namespace_common *ndns = to_ndns(dev);
1569 ssize_t rc;
1570
1571 device_lock(dev);
1572 if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1573 rc = sprintf(buf, "\n");
1574 else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1575 (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1576 rc = sprintf(buf, "btt\n");
1577 else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1578 rc = sprintf(buf, "pfn\n");
1579 else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1580 rc = sprintf(buf, "dax\n");
1581 else
1582 rc = sprintf(buf, "<unknown>\n");
1583 device_unlock(dev);
1584
1585 return rc;
1586 }
1587 static DEVICE_ATTR_RW(holder_class);
1588
1589 static ssize_t mode_show(struct device *dev,
1590 struct device_attribute *attr, char *buf)
1591 {
1592 struct nd_namespace_common *ndns = to_ndns(dev);
1593 struct device *claim;
1594 char *mode;
1595 ssize_t rc;
1596
1597 device_lock(dev);
1598 claim = ndns->claim;
1599 if (claim && is_nd_btt(claim))
1600 mode = "safe";
1601 else if (claim && is_nd_pfn(claim))
1602 mode = "memory";
1603 else if (claim && is_nd_dax(claim))
1604 mode = "dax";
1605 else if (!claim && pmem_should_map_pages(dev))
1606 mode = "memory";
1607 else
1608 mode = "raw";
1609 rc = sprintf(buf, "%s\n", mode);
1610 device_unlock(dev);
1611
1612 return rc;
1613 }
1614 static DEVICE_ATTR_RO(mode);
1615
1616 static ssize_t force_raw_store(struct device *dev,
1617 struct device_attribute *attr, const char *buf, size_t len)
1618 {
1619 bool force_raw;
1620 int rc = strtobool(buf, &force_raw);
1621
1622 if (rc)
1623 return rc;
1624
1625 to_ndns(dev)->force_raw = force_raw;
1626 return len;
1627 }
1628
1629 static ssize_t force_raw_show(struct device *dev,
1630 struct device_attribute *attr, char *buf)
1631 {
1632 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1633 }
1634 static DEVICE_ATTR_RW(force_raw);
1635
1636 static struct attribute *nd_namespace_attributes[] = {
1637 &dev_attr_nstype.attr,
1638 &dev_attr_size.attr,
1639 &dev_attr_mode.attr,
1640 &dev_attr_uuid.attr,
1641 &dev_attr_holder.attr,
1642 &dev_attr_resource.attr,
1643 &dev_attr_alt_name.attr,
1644 &dev_attr_force_raw.attr,
1645 &dev_attr_sector_size.attr,
1646 &dev_attr_dpa_extents.attr,
1647 &dev_attr_holder_class.attr,
1648 NULL,
1649 };
1650
1651 static umode_t namespace_visible(struct kobject *kobj,
1652 struct attribute *a, int n)
1653 {
1654 struct device *dev = container_of(kobj, struct device, kobj);
1655
1656 if (a == &dev_attr_resource.attr) {
1657 if (is_namespace_blk(dev))
1658 return 0;
1659 return 0400;
1660 }
1661
1662 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1663 if (a == &dev_attr_size.attr)
1664 return 0644;
1665
1666 return a->mode;
1667 }
1668
1669 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1670 || a == &dev_attr_holder.attr
1671 || a == &dev_attr_holder_class.attr
1672 || a == &dev_attr_force_raw.attr
1673 || a == &dev_attr_mode.attr)
1674 return a->mode;
1675
1676 return 0;
1677 }
1678
1679 static struct attribute_group nd_namespace_attribute_group = {
1680 .attrs = nd_namespace_attributes,
1681 .is_visible = namespace_visible,
1682 };
1683
1684 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1685 &nd_device_attribute_group,
1686 &nd_namespace_attribute_group,
1687 &nd_numa_attribute_group,
1688 NULL,
1689 };
1690
1691 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1692 {
1693 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1694 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1695 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1696 struct nd_namespace_common *ndns = NULL;
1697 resource_size_t size;
1698
1699 if (nd_btt || nd_pfn || nd_dax) {
1700 if (nd_btt)
1701 ndns = nd_btt->ndns;
1702 else if (nd_pfn)
1703 ndns = nd_pfn->ndns;
1704 else if (nd_dax)
1705 ndns = nd_dax->nd_pfn.ndns;
1706
1707 if (!ndns)
1708 return ERR_PTR(-ENODEV);
1709
1710 /*
1711 * Flush any in-progess probes / removals in the driver
1712 * for the raw personality of this namespace.
1713 */
1714 device_lock(&ndns->dev);
1715 device_unlock(&ndns->dev);
1716 if (ndns->dev.driver) {
1717 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1718 dev_name(dev));
1719 return ERR_PTR(-EBUSY);
1720 }
1721 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1722 "host (%s) vs claim (%s) mismatch\n",
1723 dev_name(dev),
1724 dev_name(ndns->claim)))
1725 return ERR_PTR(-ENXIO);
1726 } else {
1727 ndns = to_ndns(dev);
1728 if (ndns->claim) {
1729 dev_dbg(dev, "claimed by %s, failing probe\n",
1730 dev_name(ndns->claim));
1731
1732 return ERR_PTR(-ENXIO);
1733 }
1734 }
1735
1736 if (nvdimm_namespace_locked(ndns))
1737 return ERR_PTR(-EACCES);
1738
1739 size = nvdimm_namespace_capacity(ndns);
1740 if (size < ND_MIN_NAMESPACE_SIZE) {
1741 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1742 &size, ND_MIN_NAMESPACE_SIZE);
1743 return ERR_PTR(-ENODEV);
1744 }
1745
1746 if (is_namespace_pmem(&ndns->dev)) {
1747 struct nd_namespace_pmem *nspm;
1748
1749 nspm = to_nd_namespace_pmem(&ndns->dev);
1750 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1751 return ERR_PTR(-ENODEV);
1752 } else if (is_namespace_blk(&ndns->dev)) {
1753 struct nd_namespace_blk *nsblk;
1754
1755 nsblk = to_nd_namespace_blk(&ndns->dev);
1756 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1757 return ERR_PTR(-ENODEV);
1758 if (!nsblk->lbasize) {
1759 dev_dbg(&ndns->dev, "sector size not set\n");
1760 return ERR_PTR(-ENODEV);
1761 }
1762 if (!nd_namespace_blk_validate(nsblk))
1763 return ERR_PTR(-ENODEV);
1764 }
1765
1766 return ndns;
1767 }
1768 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1769
1770 static struct device **create_namespace_io(struct nd_region *nd_region)
1771 {
1772 struct nd_namespace_io *nsio;
1773 struct device *dev, **devs;
1774 struct resource *res;
1775
1776 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1777 if (!nsio)
1778 return NULL;
1779
1780 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1781 if (!devs) {
1782 kfree(nsio);
1783 return NULL;
1784 }
1785
1786 dev = &nsio->common.dev;
1787 dev->type = &namespace_io_device_type;
1788 dev->parent = &nd_region->dev;
1789 res = &nsio->res;
1790 res->name = dev_name(&nd_region->dev);
1791 res->flags = IORESOURCE_MEM;
1792 res->start = nd_region->ndr_start;
1793 res->end = res->start + nd_region->ndr_size - 1;
1794
1795 devs[0] = dev;
1796 return devs;
1797 }
1798
1799 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1800 u64 cookie, u16 pos)
1801 {
1802 struct nd_namespace_label *found = NULL;
1803 int i;
1804
1805 for (i = 0; i < nd_region->ndr_mappings; i++) {
1806 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1807 struct nd_interleave_set *nd_set = nd_region->nd_set;
1808 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1809 struct nd_label_ent *label_ent;
1810 bool found_uuid = false;
1811
1812 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1813 struct nd_namespace_label *nd_label = label_ent->label;
1814 u16 position, nlabel;
1815 u64 isetcookie;
1816
1817 if (!nd_label)
1818 continue;
1819 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1820 position = __le16_to_cpu(nd_label->position);
1821 nlabel = __le16_to_cpu(nd_label->nlabel);
1822
1823 if (isetcookie != cookie)
1824 continue;
1825
1826 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1827 continue;
1828
1829 if (namespace_label_has(ndd, type_guid)
1830 && !guid_equal(&nd_set->type_guid,
1831 &nd_label->type_guid)) {
1832 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1833 nd_set->type_guid.b,
1834 nd_label->type_guid.b);
1835 continue;
1836 }
1837
1838 if (found_uuid) {
1839 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1840 return false;
1841 }
1842 found_uuid = true;
1843 if (nlabel != nd_region->ndr_mappings)
1844 continue;
1845 if (position != pos)
1846 continue;
1847 found = nd_label;
1848 break;
1849 }
1850 if (found)
1851 break;
1852 }
1853 return found != NULL;
1854 }
1855
1856 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1857 {
1858 int i;
1859
1860 if (!pmem_id)
1861 return -ENODEV;
1862
1863 for (i = 0; i < nd_region->ndr_mappings; i++) {
1864 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1865 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1866 struct nd_namespace_label *nd_label = NULL;
1867 u64 hw_start, hw_end, pmem_start, pmem_end;
1868 struct nd_label_ent *label_ent;
1869
1870 lockdep_assert_held(&nd_mapping->lock);
1871 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1872 nd_label = label_ent->label;
1873 if (!nd_label)
1874 continue;
1875 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1876 break;
1877 nd_label = NULL;
1878 }
1879
1880 if (!nd_label) {
1881 WARN_ON(1);
1882 return -EINVAL;
1883 }
1884
1885 /*
1886 * Check that this label is compliant with the dpa
1887 * range published in NFIT
1888 */
1889 hw_start = nd_mapping->start;
1890 hw_end = hw_start + nd_mapping->size;
1891 pmem_start = __le64_to_cpu(nd_label->dpa);
1892 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1893 if (pmem_start >= hw_start && pmem_start < hw_end
1894 && pmem_end <= hw_end && pmem_end > hw_start)
1895 /* pass */;
1896 else {
1897 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1898 dev_name(ndd->dev), nd_label->uuid);
1899 return -EINVAL;
1900 }
1901
1902 /* move recently validated label to the front of the list */
1903 list_move(&label_ent->list, &nd_mapping->labels);
1904 }
1905 return 0;
1906 }
1907
1908 /**
1909 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1910 * @nd_region: region with mappings to validate
1911 * @nspm: target namespace to create
1912 * @nd_label: target pmem namespace label to evaluate
1913 */
1914 static struct device *create_namespace_pmem(struct nd_region *nd_region,
1915 struct nd_namespace_index *nsindex,
1916 struct nd_namespace_label *nd_label)
1917 {
1918 u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1919 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1920 struct nd_label_ent *label_ent;
1921 struct nd_namespace_pmem *nspm;
1922 struct nd_mapping *nd_mapping;
1923 resource_size_t size = 0;
1924 struct resource *res;
1925 struct device *dev;
1926 int rc = 0;
1927 u16 i;
1928
1929 if (cookie == 0) {
1930 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1931 return ERR_PTR(-ENXIO);
1932 }
1933
1934 if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1935 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1936 nd_label->uuid);
1937 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1938 return ERR_PTR(-EAGAIN);
1939
1940 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1941 nd_label->uuid);
1942 }
1943
1944 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1945 if (!nspm)
1946 return ERR_PTR(-ENOMEM);
1947
1948 nspm->id = -1;
1949 dev = &nspm->nsio.common.dev;
1950 dev->type = &namespace_pmem_device_type;
1951 dev->parent = &nd_region->dev;
1952 res = &nspm->nsio.res;
1953 res->name = dev_name(&nd_region->dev);
1954 res->flags = IORESOURCE_MEM;
1955
1956 for (i = 0; i < nd_region->ndr_mappings; i++) {
1957 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1958 continue;
1959 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1960 continue;
1961 break;
1962 }
1963
1964 if (i < nd_region->ndr_mappings) {
1965 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1966
1967 /*
1968 * Give up if we don't find an instance of a uuid at each
1969 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1970 * find a dimm with two instances of the same uuid.
1971 */
1972 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1973 nvdimm_name(nvdimm), nd_label->uuid);
1974 rc = -EINVAL;
1975 goto err;
1976 }
1977
1978 /*
1979 * Fix up each mapping's 'labels' to have the validated pmem label for
1980 * that position at labels[0], and NULL at labels[1]. In the process,
1981 * check that the namespace aligns with interleave-set. We know
1982 * that it does not overlap with any blk namespaces by virtue of
1983 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1984 * succeeded).
1985 */
1986 rc = select_pmem_id(nd_region, nd_label->uuid);
1987 if (rc)
1988 goto err;
1989
1990 /* Calculate total size and populate namespace properties from label0 */
1991 for (i = 0; i < nd_region->ndr_mappings; i++) {
1992 struct nd_namespace_label *label0;
1993 struct nvdimm_drvdata *ndd;
1994
1995 nd_mapping = &nd_region->mapping[i];
1996 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1997 typeof(*label_ent), list);
1998 label0 = label_ent ? label_ent->label : 0;
1999
2000 if (!label0) {
2001 WARN_ON(1);
2002 continue;
2003 }
2004
2005 size += __le64_to_cpu(label0->rawsize);
2006 if (__le16_to_cpu(label0->position) != 0)
2007 continue;
2008 WARN_ON(nspm->alt_name || nspm->uuid);
2009 nspm->alt_name = kmemdup((void __force *) label0->name,
2010 NSLABEL_NAME_LEN, GFP_KERNEL);
2011 nspm->uuid = kmemdup((void __force *) label0->uuid,
2012 NSLABEL_UUID_LEN, GFP_KERNEL);
2013 nspm->lbasize = __le64_to_cpu(label0->lbasize);
2014 ndd = to_ndd(nd_mapping);
2015 if (namespace_label_has(ndd, abstraction_guid))
2016 nspm->nsio.common.claim_class
2017 = to_nvdimm_cclass(&label0->abstraction_guid);
2018
2019 }
2020
2021 if (!nspm->alt_name || !nspm->uuid) {
2022 rc = -ENOMEM;
2023 goto err;
2024 }
2025
2026 nd_namespace_pmem_set_resource(nd_region, nspm, size);
2027
2028 return dev;
2029 err:
2030 namespace_pmem_release(dev);
2031 switch (rc) {
2032 case -EINVAL:
2033 dev_dbg(&nd_region->dev, "invalid label(s)\n");
2034 break;
2035 case -ENODEV:
2036 dev_dbg(&nd_region->dev, "label not found\n");
2037 break;
2038 default:
2039 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
2040 break;
2041 }
2042 return ERR_PTR(rc);
2043 }
2044
2045 struct resource *nsblk_add_resource(struct nd_region *nd_region,
2046 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2047 resource_size_t start)
2048 {
2049 struct nd_label_id label_id;
2050 struct resource *res;
2051
2052 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2053 res = krealloc(nsblk->res,
2054 sizeof(void *) * (nsblk->num_resources + 1),
2055 GFP_KERNEL);
2056 if (!res)
2057 return NULL;
2058 nsblk->res = (struct resource **) res;
2059 for_each_dpa_resource(ndd, res)
2060 if (strcmp(res->name, label_id.id) == 0
2061 && res->start == start) {
2062 nsblk->res[nsblk->num_resources++] = res;
2063 return res;
2064 }
2065 return NULL;
2066 }
2067
2068 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2069 {
2070 struct nd_namespace_blk *nsblk;
2071 struct device *dev;
2072
2073 if (!is_nd_blk(&nd_region->dev))
2074 return NULL;
2075
2076 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2077 if (!nsblk)
2078 return NULL;
2079
2080 dev = &nsblk->common.dev;
2081 dev->type = &namespace_blk_device_type;
2082 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2083 if (nsblk->id < 0) {
2084 kfree(nsblk);
2085 return NULL;
2086 }
2087 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2088 dev->parent = &nd_region->dev;
2089 dev->groups = nd_namespace_attribute_groups;
2090
2091 return &nsblk->common.dev;
2092 }
2093
2094 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2095 {
2096 struct nd_namespace_pmem *nspm;
2097 struct resource *res;
2098 struct device *dev;
2099
2100 if (!is_memory(&nd_region->dev))
2101 return NULL;
2102
2103 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2104 if (!nspm)
2105 return NULL;
2106
2107 dev = &nspm->nsio.common.dev;
2108 dev->type = &namespace_pmem_device_type;
2109 dev->parent = &nd_region->dev;
2110 res = &nspm->nsio.res;
2111 res->name = dev_name(&nd_region->dev);
2112 res->flags = IORESOURCE_MEM;
2113
2114 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2115 if (nspm->id < 0) {
2116 kfree(nspm);
2117 return NULL;
2118 }
2119 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2120 dev->groups = nd_namespace_attribute_groups;
2121 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2122
2123 return dev;
2124 }
2125
2126 void nd_region_create_ns_seed(struct nd_region *nd_region)
2127 {
2128 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2129
2130 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2131 return;
2132
2133 if (is_nd_blk(&nd_region->dev))
2134 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2135 else
2136 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2137
2138 /*
2139 * Seed creation failures are not fatal, provisioning is simply
2140 * disabled until memory becomes available
2141 */
2142 if (!nd_region->ns_seed)
2143 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2144 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2145 else
2146 nd_device_register(nd_region->ns_seed);
2147 }
2148
2149 void nd_region_create_dax_seed(struct nd_region *nd_region)
2150 {
2151 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2152 nd_region->dax_seed = nd_dax_create(nd_region);
2153 /*
2154 * Seed creation failures are not fatal, provisioning is simply
2155 * disabled until memory becomes available
2156 */
2157 if (!nd_region->dax_seed)
2158 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2159 }
2160
2161 void nd_region_create_pfn_seed(struct nd_region *nd_region)
2162 {
2163 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2164 nd_region->pfn_seed = nd_pfn_create(nd_region);
2165 /*
2166 * Seed creation failures are not fatal, provisioning is simply
2167 * disabled until memory becomes available
2168 */
2169 if (!nd_region->pfn_seed)
2170 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2171 }
2172
2173 void nd_region_create_btt_seed(struct nd_region *nd_region)
2174 {
2175 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2176 nd_region->btt_seed = nd_btt_create(nd_region);
2177 /*
2178 * Seed creation failures are not fatal, provisioning is simply
2179 * disabled until memory becomes available
2180 */
2181 if (!nd_region->btt_seed)
2182 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2183 }
2184
2185 static int add_namespace_resource(struct nd_region *nd_region,
2186 struct nd_namespace_label *nd_label, struct device **devs,
2187 int count)
2188 {
2189 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2190 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2191 int i;
2192
2193 for (i = 0; i < count; i++) {
2194 u8 *uuid = namespace_to_uuid(devs[i]);
2195 struct resource *res;
2196
2197 if (IS_ERR_OR_NULL(uuid)) {
2198 WARN_ON(1);
2199 continue;
2200 }
2201
2202 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2203 continue;
2204 if (is_namespace_blk(devs[i])) {
2205 res = nsblk_add_resource(nd_region, ndd,
2206 to_nd_namespace_blk(devs[i]),
2207 __le64_to_cpu(nd_label->dpa));
2208 if (!res)
2209 return -ENXIO;
2210 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2211 } else {
2212 dev_err(&nd_region->dev,
2213 "error: conflicting extents for uuid: %pUb\n",
2214 nd_label->uuid);
2215 return -ENXIO;
2216 }
2217 break;
2218 }
2219
2220 return i;
2221 }
2222
2223 static struct device *create_namespace_blk(struct nd_region *nd_region,
2224 struct nd_namespace_label *nd_label, int count)
2225 {
2226
2227 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2228 struct nd_interleave_set *nd_set = nd_region->nd_set;
2229 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2230 struct nd_namespace_blk *nsblk;
2231 char name[NSLABEL_NAME_LEN];
2232 struct device *dev = NULL;
2233 struct resource *res;
2234
2235 if (namespace_label_has(ndd, type_guid)) {
2236 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2237 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2238 nd_set->type_guid.b,
2239 nd_label->type_guid.b);
2240 return ERR_PTR(-EAGAIN);
2241 }
2242
2243 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2244 dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2245 nd_set->cookie2,
2246 __le64_to_cpu(nd_label->isetcookie));
2247 return ERR_PTR(-EAGAIN);
2248 }
2249 }
2250
2251 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2252 if (!nsblk)
2253 return ERR_PTR(-ENOMEM);
2254 dev = &nsblk->common.dev;
2255 dev->type = &namespace_blk_device_type;
2256 dev->parent = &nd_region->dev;
2257 nsblk->id = -1;
2258 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2259 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2260 GFP_KERNEL);
2261 if (namespace_label_has(ndd, abstraction_guid))
2262 nsblk->common.claim_class
2263 = to_nvdimm_cclass(&nd_label->abstraction_guid);
2264 if (!nsblk->uuid)
2265 goto blk_err;
2266 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2267 if (name[0]) {
2268 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2269 GFP_KERNEL);
2270 if (!nsblk->alt_name)
2271 goto blk_err;
2272 }
2273 res = nsblk_add_resource(nd_region, ndd, nsblk,
2274 __le64_to_cpu(nd_label->dpa));
2275 if (!res)
2276 goto blk_err;
2277 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2278 return dev;
2279 blk_err:
2280 namespace_blk_release(dev);
2281 return ERR_PTR(-ENXIO);
2282 }
2283
2284 static int cmp_dpa(const void *a, const void *b)
2285 {
2286 const struct device *dev_a = *(const struct device **) a;
2287 const struct device *dev_b = *(const struct device **) b;
2288 struct nd_namespace_blk *nsblk_a, *nsblk_b;
2289 struct nd_namespace_pmem *nspm_a, *nspm_b;
2290
2291 if (is_namespace_io(dev_a))
2292 return 0;
2293
2294 if (is_namespace_blk(dev_a)) {
2295 nsblk_a = to_nd_namespace_blk(dev_a);
2296 nsblk_b = to_nd_namespace_blk(dev_b);
2297
2298 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2299 sizeof(resource_size_t));
2300 }
2301
2302 nspm_a = to_nd_namespace_pmem(dev_a);
2303 nspm_b = to_nd_namespace_pmem(dev_b);
2304
2305 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2306 sizeof(resource_size_t));
2307 }
2308
2309 static struct device **scan_labels(struct nd_region *nd_region)
2310 {
2311 int i, count = 0;
2312 struct device *dev, **devs = NULL;
2313 struct nd_label_ent *label_ent, *e;
2314 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2315 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2316
2317 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2318 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2319 struct nd_namespace_label *nd_label = label_ent->label;
2320 struct device **__devs;
2321 u32 flags;
2322
2323 if (!nd_label)
2324 continue;
2325 flags = __le32_to_cpu(nd_label->flags);
2326 if (is_nd_blk(&nd_region->dev)
2327 == !!(flags & NSLABEL_FLAG_LOCAL))
2328 /* pass, region matches label type */;
2329 else
2330 continue;
2331
2332 /* skip labels that describe extents outside of the region */
2333 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2334 continue;
2335
2336 i = add_namespace_resource(nd_region, nd_label, devs, count);
2337 if (i < 0)
2338 goto err;
2339 if (i < count)
2340 continue;
2341 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2342 if (!__devs)
2343 goto err;
2344 memcpy(__devs, devs, sizeof(dev) * count);
2345 kfree(devs);
2346 devs = __devs;
2347
2348 if (is_nd_blk(&nd_region->dev))
2349 dev = create_namespace_blk(nd_region, nd_label, count);
2350 else {
2351 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2352 struct nd_namespace_index *nsindex;
2353
2354 nsindex = to_namespace_index(ndd, ndd->ns_current);
2355 dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2356 }
2357
2358 if (IS_ERR(dev)) {
2359 switch (PTR_ERR(dev)) {
2360 case -EAGAIN:
2361 /* skip invalid labels */
2362 continue;
2363 case -ENODEV:
2364 /* fallthrough to seed creation */
2365 break;
2366 default:
2367 goto err;
2368 }
2369 } else
2370 devs[count++] = dev;
2371
2372 }
2373
2374 dev_dbg(&nd_region->dev, "discovered %d %s namespace%s\n",
2375 count, is_nd_blk(&nd_region->dev)
2376 ? "blk" : "pmem", count == 1 ? "" : "s");
2377
2378 if (count == 0) {
2379 /* Publish a zero-sized namespace for userspace to configure. */
2380 nd_mapping_free_labels(nd_mapping);
2381
2382 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2383 if (!devs)
2384 goto err;
2385 if (is_nd_blk(&nd_region->dev)) {
2386 struct nd_namespace_blk *nsblk;
2387
2388 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2389 if (!nsblk)
2390 goto err;
2391 dev = &nsblk->common.dev;
2392 dev->type = &namespace_blk_device_type;
2393 } else {
2394 struct nd_namespace_pmem *nspm;
2395
2396 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2397 if (!nspm)
2398 goto err;
2399 dev = &nspm->nsio.common.dev;
2400 dev->type = &namespace_pmem_device_type;
2401 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2402 }
2403 dev->parent = &nd_region->dev;
2404 devs[count++] = dev;
2405 } else if (is_memory(&nd_region->dev)) {
2406 /* clean unselected labels */
2407 for (i = 0; i < nd_region->ndr_mappings; i++) {
2408 struct list_head *l, *e;
2409 LIST_HEAD(list);
2410 int j;
2411
2412 nd_mapping = &nd_region->mapping[i];
2413 if (list_empty(&nd_mapping->labels)) {
2414 WARN_ON(1);
2415 continue;
2416 }
2417
2418 j = count;
2419 list_for_each_safe(l, e, &nd_mapping->labels) {
2420 if (!j--)
2421 break;
2422 list_move_tail(l, &list);
2423 }
2424 nd_mapping_free_labels(nd_mapping);
2425 list_splice_init(&list, &nd_mapping->labels);
2426 }
2427 }
2428
2429 if (count > 1)
2430 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2431
2432 return devs;
2433
2434 err:
2435 if (devs) {
2436 for (i = 0; devs[i]; i++)
2437 if (is_nd_blk(&nd_region->dev))
2438 namespace_blk_release(devs[i]);
2439 else
2440 namespace_pmem_release(devs[i]);
2441 kfree(devs);
2442 }
2443 return NULL;
2444 }
2445
2446 static struct device **create_namespaces(struct nd_region *nd_region)
2447 {
2448 struct nd_mapping *nd_mapping;
2449 struct device **devs;
2450 int i;
2451
2452 if (nd_region->ndr_mappings == 0)
2453 return NULL;
2454
2455 /* lock down all mappings while we scan labels */
2456 for (i = 0; i < nd_region->ndr_mappings; i++) {
2457 nd_mapping = &nd_region->mapping[i];
2458 mutex_lock_nested(&nd_mapping->lock, i);
2459 }
2460
2461 devs = scan_labels(nd_region);
2462
2463 for (i = 0; i < nd_region->ndr_mappings; i++) {
2464 int reverse = nd_region->ndr_mappings - 1 - i;
2465
2466 nd_mapping = &nd_region->mapping[reverse];
2467 mutex_unlock(&nd_mapping->lock);
2468 }
2469
2470 return devs;
2471 }
2472
2473 static int init_active_labels(struct nd_region *nd_region)
2474 {
2475 int i;
2476
2477 for (i = 0; i < nd_region->ndr_mappings; i++) {
2478 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2479 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2480 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2481 struct nd_label_ent *label_ent;
2482 int count, j;
2483
2484 /*
2485 * If the dimm is disabled then we may need to prevent
2486 * the region from being activated.
2487 */
2488 if (!ndd) {
2489 if (test_bit(NDD_LOCKED, &nvdimm->flags))
2490 /* fail, label data may be unreadable */;
2491 else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2492 /* fail, labels needed to disambiguate dpa */;
2493 else
2494 return 0;
2495
2496 dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2497 dev_name(&nd_mapping->nvdimm->dev),
2498 test_bit(NDD_LOCKED, &nvdimm->flags)
2499 ? "locked" : "disabled");
2500 return -ENXIO;
2501 }
2502 nd_mapping->ndd = ndd;
2503 atomic_inc(&nvdimm->busy);
2504 get_ndd(ndd);
2505
2506 count = nd_label_active_count(ndd);
2507 dev_dbg(ndd->dev, "count: %d\n", count);
2508 if (!count)
2509 continue;
2510 for (j = 0; j < count; j++) {
2511 struct nd_namespace_label *label;
2512
2513 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2514 if (!label_ent)
2515 break;
2516 label = nd_label_active(ndd, j);
2517 if (test_bit(NDD_NOBLK, &nvdimm->flags)) {
2518 u32 flags = __le32_to_cpu(label->flags);
2519
2520 flags &= ~NSLABEL_FLAG_LOCAL;
2521 label->flags = __cpu_to_le32(flags);
2522 }
2523 label_ent->label = label;
2524
2525 mutex_lock(&nd_mapping->lock);
2526 list_add_tail(&label_ent->list, &nd_mapping->labels);
2527 mutex_unlock(&nd_mapping->lock);
2528 }
2529
2530 if (j >= count)
2531 continue;
2532
2533 mutex_lock(&nd_mapping->lock);
2534 nd_mapping_free_labels(nd_mapping);
2535 mutex_unlock(&nd_mapping->lock);
2536 return -ENOMEM;
2537 }
2538
2539 return 0;
2540 }
2541
2542 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2543 {
2544 struct device **devs = NULL;
2545 int i, rc = 0, type;
2546
2547 *err = 0;
2548 nvdimm_bus_lock(&nd_region->dev);
2549 rc = init_active_labels(nd_region);
2550 if (rc) {
2551 nvdimm_bus_unlock(&nd_region->dev);
2552 return rc;
2553 }
2554
2555 type = nd_region_to_nstype(nd_region);
2556 switch (type) {
2557 case ND_DEVICE_NAMESPACE_IO:
2558 devs = create_namespace_io(nd_region);
2559 break;
2560 case ND_DEVICE_NAMESPACE_PMEM:
2561 case ND_DEVICE_NAMESPACE_BLK:
2562 devs = create_namespaces(nd_region);
2563 break;
2564 default:
2565 break;
2566 }
2567 nvdimm_bus_unlock(&nd_region->dev);
2568
2569 if (!devs)
2570 return -ENODEV;
2571
2572 for (i = 0; devs[i]; i++) {
2573 struct device *dev = devs[i];
2574 int id;
2575
2576 if (type == ND_DEVICE_NAMESPACE_BLK) {
2577 struct nd_namespace_blk *nsblk;
2578
2579 nsblk = to_nd_namespace_blk(dev);
2580 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2581 GFP_KERNEL);
2582 nsblk->id = id;
2583 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2584 struct nd_namespace_pmem *nspm;
2585
2586 nspm = to_nd_namespace_pmem(dev);
2587 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2588 GFP_KERNEL);
2589 nspm->id = id;
2590 } else
2591 id = i;
2592
2593 if (id < 0)
2594 break;
2595 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2596 dev->groups = nd_namespace_attribute_groups;
2597 nd_device_register(dev);
2598 }
2599 if (i)
2600 nd_region->ns_seed = devs[0];
2601
2602 if (devs[i]) {
2603 int j;
2604
2605 for (j = i; devs[j]; j++) {
2606 struct device *dev = devs[j];
2607
2608 device_initialize(dev);
2609 put_device(dev);
2610 }
2611 *err = j - i;
2612 /*
2613 * All of the namespaces we tried to register failed, so
2614 * fail region activation.
2615 */
2616 if (*err == 0)
2617 rc = -ENODEV;
2618 }
2619 kfree(devs);
2620
2621 if (rc == -ENODEV)
2622 return rc;
2623
2624 return i;
2625 }
Linux with added FPC-III support