]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/nvdimm/namespace_devs.c
coresight: etm3x: Set synchronisation frequencty to TRM default
[mirror_ubuntu-bionic-kernel.git] / drivers / nvdimm / namespace_devs.c
1 /*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13 #include <linux/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_io *nsio;
142
143 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
144 return false;
145
146 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
147 return false;
148
149 if (is_nd_pfn(dev) || is_nd_btt(dev))
150 return false;
151
152 nsio = to_nd_namespace_io(dev);
153 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
154 IORESOURCE_SYSTEM_RAM,
155 IORES_DESC_NONE) == REGION_MIXED)
156 return false;
157
158 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
159 }
160 EXPORT_SYMBOL(pmem_should_map_pages);
161
162 unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
163 {
164 if (is_namespace_pmem(&ndns->dev)) {
165 struct nd_namespace_pmem *nspm;
166
167 nspm = to_nd_namespace_pmem(&ndns->dev);
168 if (nspm->lbasize == 0 || nspm->lbasize == 512)
169 /* default */;
170 else if (nspm->lbasize == 4096)
171 return 4096;
172 else
173 dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
174 nspm->lbasize);
175 }
176
177 /*
178 * There is no namespace label (is_namespace_io()), or the label
179 * indicates the default sector size.
180 */
181 return 512;
182 }
183 EXPORT_SYMBOL(pmem_sector_size);
184
185 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
186 char *name)
187 {
188 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
189 const char *suffix = NULL;
190
191 if (ndns->claim && is_nd_btt(ndns->claim))
192 suffix = "s";
193
194 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
195 int nsidx = 0;
196
197 if (is_namespace_pmem(&ndns->dev)) {
198 struct nd_namespace_pmem *nspm;
199
200 nspm = to_nd_namespace_pmem(&ndns->dev);
201 nsidx = nspm->id;
202 }
203
204 if (nsidx)
205 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
206 suffix ? suffix : "");
207 else
208 sprintf(name, "pmem%d%s", nd_region->id,
209 suffix ? suffix : "");
210 } else if (is_namespace_blk(&ndns->dev)) {
211 struct nd_namespace_blk *nsblk;
212
213 nsblk = to_nd_namespace_blk(&ndns->dev);
214 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
215 suffix ? suffix : "");
216 } else {
217 return NULL;
218 }
219
220 return name;
221 }
222 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
223
224 const u8 *nd_dev_to_uuid(struct device *dev)
225 {
226 static const u8 null_uuid[16];
227
228 if (!dev)
229 return null_uuid;
230
231 if (is_namespace_pmem(dev)) {
232 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
233
234 return nspm->uuid;
235 } else if (is_namespace_blk(dev)) {
236 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
237
238 return nsblk->uuid;
239 } else
240 return null_uuid;
241 }
242 EXPORT_SYMBOL(nd_dev_to_uuid);
243
244 static ssize_t nstype_show(struct device *dev,
245 struct device_attribute *attr, char *buf)
246 {
247 struct nd_region *nd_region = to_nd_region(dev->parent);
248
249 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
250 }
251 static DEVICE_ATTR_RO(nstype);
252
253 static ssize_t __alt_name_store(struct device *dev, const char *buf,
254 const size_t len)
255 {
256 char *input, *pos, *alt_name, **ns_altname;
257 ssize_t rc;
258
259 if (is_namespace_pmem(dev)) {
260 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
261
262 ns_altname = &nspm->alt_name;
263 } else if (is_namespace_blk(dev)) {
264 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
265
266 ns_altname = &nsblk->alt_name;
267 } else
268 return -ENXIO;
269
270 if (dev->driver || to_ndns(dev)->claim)
271 return -EBUSY;
272
273 input = kmemdup(buf, len + 1, GFP_KERNEL);
274 if (!input)
275 return -ENOMEM;
276
277 input[len] = '\0';
278 pos = strim(input);
279 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
280 rc = -EINVAL;
281 goto out;
282 }
283
284 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
285 if (!alt_name) {
286 rc = -ENOMEM;
287 goto out;
288 }
289 kfree(*ns_altname);
290 *ns_altname = alt_name;
291 sprintf(*ns_altname, "%s", pos);
292 rc = len;
293
294 out:
295 kfree(input);
296 return rc;
297 }
298
299 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
300 {
301 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
302 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
303 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
304 struct nd_label_id label_id;
305 resource_size_t size = 0;
306 struct resource *res;
307
308 if (!nsblk->uuid)
309 return 0;
310 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
311 for_each_dpa_resource(ndd, res)
312 if (strcmp(res->name, label_id.id) == 0)
313 size += resource_size(res);
314 return size;
315 }
316
317 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
318 {
319 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
320 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
321 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
322 struct nd_label_id label_id;
323 struct resource *res;
324 int count, i;
325
326 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
327 return false;
328
329 count = 0;
330 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
331 for_each_dpa_resource(ndd, res) {
332 if (strcmp(res->name, label_id.id) != 0)
333 continue;
334 /*
335 * Resources with unacknowledged adjustments indicate a
336 * failure to update labels
337 */
338 if (res->flags & DPA_RESOURCE_ADJUSTED)
339 return false;
340 count++;
341 }
342
343 /* These values match after a successful label update */
344 if (count != nsblk->num_resources)
345 return false;
346
347 for (i = 0; i < nsblk->num_resources; i++) {
348 struct resource *found = NULL;
349
350 for_each_dpa_resource(ndd, res)
351 if (res == nsblk->res[i]) {
352 found = res;
353 break;
354 }
355 /* stale resource */
356 if (!found)
357 return false;
358 }
359
360 return true;
361 }
362
363 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
364 {
365 resource_size_t size;
366
367 nvdimm_bus_lock(&nsblk->common.dev);
368 size = __nd_namespace_blk_validate(nsblk);
369 nvdimm_bus_unlock(&nsblk->common.dev);
370
371 return size;
372 }
373 EXPORT_SYMBOL(nd_namespace_blk_validate);
374
375
376 static int nd_namespace_label_update(struct nd_region *nd_region,
377 struct device *dev)
378 {
379 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
380 "namespace must be idle during label update\n");
381 if (dev->driver || to_ndns(dev)->claim)
382 return 0;
383
384 /*
385 * Only allow label writes that will result in a valid namespace
386 * or deletion of an existing namespace.
387 */
388 if (is_namespace_pmem(dev)) {
389 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
390 resource_size_t size = resource_size(&nspm->nsio.res);
391
392 if (size == 0 && nspm->uuid)
393 /* delete allocation */;
394 else if (!nspm->uuid)
395 return 0;
396
397 return nd_pmem_namespace_label_update(nd_region, nspm, size);
398 } else if (is_namespace_blk(dev)) {
399 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
400 resource_size_t size = nd_namespace_blk_size(nsblk);
401
402 if (size == 0 && nsblk->uuid)
403 /* delete allocation */;
404 else if (!nsblk->uuid || !nsblk->lbasize)
405 return 0;
406
407 return nd_blk_namespace_label_update(nd_region, nsblk, size);
408 } else
409 return -ENXIO;
410 }
411
412 static ssize_t alt_name_store(struct device *dev,
413 struct device_attribute *attr, const char *buf, size_t len)
414 {
415 struct nd_region *nd_region = to_nd_region(dev->parent);
416 ssize_t rc;
417
418 device_lock(dev);
419 nvdimm_bus_lock(dev);
420 wait_nvdimm_bus_probe_idle(dev);
421 rc = __alt_name_store(dev, buf, len);
422 if (rc >= 0)
423 rc = nd_namespace_label_update(nd_region, dev);
424 dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
425 nvdimm_bus_unlock(dev);
426 device_unlock(dev);
427
428 return rc < 0 ? rc : len;
429 }
430
431 static ssize_t alt_name_show(struct device *dev,
432 struct device_attribute *attr, char *buf)
433 {
434 char *ns_altname;
435
436 if (is_namespace_pmem(dev)) {
437 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
438
439 ns_altname = nspm->alt_name;
440 } else if (is_namespace_blk(dev)) {
441 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
442
443 ns_altname = nsblk->alt_name;
444 } else
445 return -ENXIO;
446
447 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
448 }
449 static DEVICE_ATTR_RW(alt_name);
450
451 static int scan_free(struct nd_region *nd_region,
452 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
453 resource_size_t n)
454 {
455 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
456 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
457 int rc = 0;
458
459 while (n) {
460 struct resource *res, *last;
461 resource_size_t new_start;
462
463 last = NULL;
464 for_each_dpa_resource(ndd, res)
465 if (strcmp(res->name, label_id->id) == 0)
466 last = res;
467 res = last;
468 if (!res)
469 return 0;
470
471 if (n >= resource_size(res)) {
472 n -= resource_size(res);
473 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
474 nvdimm_free_dpa(ndd, res);
475 /* retry with last resource deleted */
476 continue;
477 }
478
479 /*
480 * Keep BLK allocations relegated to high DPA as much as
481 * possible
482 */
483 if (is_blk)
484 new_start = res->start + n;
485 else
486 new_start = res->start;
487
488 rc = adjust_resource(res, new_start, resource_size(res) - n);
489 if (rc == 0)
490 res->flags |= DPA_RESOURCE_ADJUSTED;
491 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
492 break;
493 }
494
495 return rc;
496 }
497
498 /**
499 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
500 * @nd_region: the set of dimms to reclaim @n bytes from
501 * @label_id: unique identifier for the namespace consuming this dpa range
502 * @n: number of bytes per-dimm to release
503 *
504 * Assumes resources are ordered. Starting from the end try to
505 * adjust_resource() the allocation to @n, but if @n is larger than the
506 * allocation delete it and find the 'new' last allocation in the label
507 * set.
508 */
509 static int shrink_dpa_allocation(struct nd_region *nd_region,
510 struct nd_label_id *label_id, resource_size_t n)
511 {
512 int i;
513
514 for (i = 0; i < nd_region->ndr_mappings; i++) {
515 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
516 int rc;
517
518 rc = scan_free(nd_region, nd_mapping, label_id, n);
519 if (rc)
520 return rc;
521 }
522
523 return 0;
524 }
525
526 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
527 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
528 resource_size_t n)
529 {
530 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
531 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
532 resource_size_t first_dpa;
533 struct resource *res;
534 int rc = 0;
535
536 /* allocate blk from highest dpa first */
537 if (is_blk)
538 first_dpa = nd_mapping->start + nd_mapping->size - n;
539 else
540 first_dpa = nd_mapping->start;
541
542 /* first resource allocation for this label-id or dimm */
543 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
544 if (!res)
545 rc = -EBUSY;
546
547 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
548 return rc ? n : 0;
549 }
550
551
552 /**
553 * space_valid() - validate free dpa space against constraints
554 * @nd_region: hosting region of the free space
555 * @ndd: dimm device data for debug
556 * @label_id: namespace id to allocate space
557 * @prev: potential allocation that precedes free space
558 * @next: allocation that follows the given free space range
559 * @exist: first allocation with same id in the mapping
560 * @n: range that must satisfied for pmem allocations
561 * @valid: free space range to validate
562 *
563 * BLK-space is valid as long as it does not precede a PMEM
564 * allocation in a given region. PMEM-space must be contiguous
565 * and adjacent to an existing existing allocation (if one
566 * exists). If reserving PMEM any space is valid.
567 */
568 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
569 struct nd_label_id *label_id, struct resource *prev,
570 struct resource *next, struct resource *exist,
571 resource_size_t n, struct resource *valid)
572 {
573 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
574 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
575
576 if (valid->start >= valid->end)
577 goto invalid;
578
579 if (is_reserve)
580 return;
581
582 if (!is_pmem) {
583 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
584 struct nvdimm_bus *nvdimm_bus;
585 struct blk_alloc_info info = {
586 .nd_mapping = nd_mapping,
587 .available = nd_mapping->size,
588 .res = valid,
589 };
590
591 WARN_ON(!is_nd_blk(&nd_region->dev));
592 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
593 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
594 return;
595 }
596
597 /* allocation needs to be contiguous, so this is all or nothing */
598 if (resource_size(valid) < n)
599 goto invalid;
600
601 /* we've got all the space we need and no existing allocation */
602 if (!exist)
603 return;
604
605 /* allocation needs to be contiguous with the existing namespace */
606 if (valid->start == exist->end + 1
607 || valid->end == exist->start - 1)
608 return;
609
610 invalid:
611 /* truncate @valid size to 0 */
612 valid->end = valid->start - 1;
613 }
614
615 enum alloc_loc {
616 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
617 };
618
619 static resource_size_t scan_allocate(struct nd_region *nd_region,
620 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
621 resource_size_t n)
622 {
623 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
624 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
625 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
626 struct resource *res, *exist = NULL, valid;
627 const resource_size_t to_allocate = n;
628 int first;
629
630 for_each_dpa_resource(ndd, res)
631 if (strcmp(label_id->id, res->name) == 0)
632 exist = res;
633
634 valid.start = nd_mapping->start;
635 valid.end = mapping_end;
636 valid.name = "free space";
637 retry:
638 first = 0;
639 for_each_dpa_resource(ndd, res) {
640 struct resource *next = res->sibling, *new_res = NULL;
641 resource_size_t allocate, available = 0;
642 enum alloc_loc loc = ALLOC_ERR;
643 const char *action;
644 int rc = 0;
645
646 /* ignore resources outside this nd_mapping */
647 if (res->start > mapping_end)
648 continue;
649 if (res->end < nd_mapping->start)
650 continue;
651
652 /* space at the beginning of the mapping */
653 if (!first++ && res->start > nd_mapping->start) {
654 valid.start = nd_mapping->start;
655 valid.end = res->start - 1;
656 space_valid(nd_region, ndd, label_id, NULL, next, exist,
657 to_allocate, &valid);
658 available = resource_size(&valid);
659 if (available)
660 loc = ALLOC_BEFORE;
661 }
662
663 /* space between allocations */
664 if (!loc && next) {
665 valid.start = res->start + resource_size(res);
666 valid.end = min(mapping_end, next->start - 1);
667 space_valid(nd_region, ndd, label_id, res, next, exist,
668 to_allocate, &valid);
669 available = resource_size(&valid);
670 if (available)
671 loc = ALLOC_MID;
672 }
673
674 /* space at the end of the mapping */
675 if (!loc && !next) {
676 valid.start = res->start + resource_size(res);
677 valid.end = mapping_end;
678 space_valid(nd_region, ndd, label_id, res, next, exist,
679 to_allocate, &valid);
680 available = resource_size(&valid);
681 if (available)
682 loc = ALLOC_AFTER;
683 }
684
685 if (!loc || !available)
686 continue;
687 allocate = min(available, n);
688 switch (loc) {
689 case ALLOC_BEFORE:
690 if (strcmp(res->name, label_id->id) == 0) {
691 /* adjust current resource up */
692 rc = adjust_resource(res, res->start - allocate,
693 resource_size(res) + allocate);
694 action = "cur grow up";
695 } else
696 action = "allocate";
697 break;
698 case ALLOC_MID:
699 if (strcmp(next->name, label_id->id) == 0) {
700 /* adjust next resource up */
701 rc = adjust_resource(next, next->start
702 - allocate, resource_size(next)
703 + allocate);
704 new_res = next;
705 action = "next grow up";
706 } else if (strcmp(res->name, label_id->id) == 0) {
707 action = "grow down";
708 } else
709 action = "allocate";
710 break;
711 case ALLOC_AFTER:
712 if (strcmp(res->name, label_id->id) == 0)
713 action = "grow down";
714 else
715 action = "allocate";
716 break;
717 default:
718 return n;
719 }
720
721 if (strcmp(action, "allocate") == 0) {
722 /* BLK allocate bottom up */
723 if (!is_pmem)
724 valid.start += available - allocate;
725
726 new_res = nvdimm_allocate_dpa(ndd, label_id,
727 valid.start, allocate);
728 if (!new_res)
729 rc = -EBUSY;
730 } else if (strcmp(action, "grow down") == 0) {
731 /* adjust current resource down */
732 rc = adjust_resource(res, res->start, resource_size(res)
733 + allocate);
734 if (rc == 0)
735 res->flags |= DPA_RESOURCE_ADJUSTED;
736 }
737
738 if (!new_res)
739 new_res = res;
740
741 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
742 action, loc, rc);
743
744 if (rc)
745 return n;
746
747 n -= allocate;
748 if (n) {
749 /*
750 * Retry scan with newly inserted resources.
751 * For example, if we did an ALLOC_BEFORE
752 * insertion there may also have been space
753 * available for an ALLOC_AFTER insertion, so we
754 * need to check this same resource again
755 */
756 goto retry;
757 } else
758 return 0;
759 }
760
761 /*
762 * If we allocated nothing in the BLK case it may be because we are in
763 * an initial "pmem-reserve pass". Only do an initial BLK allocation
764 * when none of the DPA space is reserved.
765 */
766 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
767 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
768 return n;
769 }
770
771 static int merge_dpa(struct nd_region *nd_region,
772 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
773 {
774 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
775 struct resource *res;
776
777 if (strncmp("pmem", label_id->id, 4) == 0)
778 return 0;
779 retry:
780 for_each_dpa_resource(ndd, res) {
781 int rc;
782 struct resource *next = res->sibling;
783 resource_size_t end = res->start + resource_size(res);
784
785 if (!next || strcmp(res->name, label_id->id) != 0
786 || strcmp(next->name, label_id->id) != 0
787 || end != next->start)
788 continue;
789 end += resource_size(next);
790 nvdimm_free_dpa(ndd, next);
791 rc = adjust_resource(res, res->start, end - res->start);
792 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
793 if (rc)
794 return rc;
795 res->flags |= DPA_RESOURCE_ADJUSTED;
796 goto retry;
797 }
798
799 return 0;
800 }
801
802 static int __reserve_free_pmem(struct device *dev, void *data)
803 {
804 struct nvdimm *nvdimm = data;
805 struct nd_region *nd_region;
806 struct nd_label_id label_id;
807 int i;
808
809 if (!is_memory(dev))
810 return 0;
811
812 nd_region = to_nd_region(dev);
813 if (nd_region->ndr_mappings == 0)
814 return 0;
815
816 memset(&label_id, 0, sizeof(label_id));
817 strcat(label_id.id, "pmem-reserve");
818 for (i = 0; i < nd_region->ndr_mappings; i++) {
819 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
820 resource_size_t n, rem = 0;
821
822 if (nd_mapping->nvdimm != nvdimm)
823 continue;
824
825 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
826 if (n == 0)
827 return 0;
828 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
829 dev_WARN_ONCE(&nd_region->dev, rem,
830 "pmem reserve underrun: %#llx of %#llx bytes\n",
831 (unsigned long long) n - rem,
832 (unsigned long long) n);
833 return rem ? -ENXIO : 0;
834 }
835
836 return 0;
837 }
838
839 static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
840 struct nd_mapping *nd_mapping)
841 {
842 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
843 struct resource *res, *_res;
844
845 for_each_dpa_resource_safe(ndd, res, _res)
846 if (strcmp(res->name, "pmem-reserve") == 0)
847 nvdimm_free_dpa(ndd, res);
848 }
849
850 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
851 struct nd_mapping *nd_mapping)
852 {
853 struct nvdimm *nvdimm = nd_mapping->nvdimm;
854 int rc;
855
856 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
857 __reserve_free_pmem);
858 if (rc)
859 release_free_pmem(nvdimm_bus, nd_mapping);
860 return rc;
861 }
862
863 /**
864 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
865 * @nd_region: the set of dimms to allocate @n more bytes from
866 * @label_id: unique identifier for the namespace consuming this dpa range
867 * @n: number of bytes per-dimm to add to the existing allocation
868 *
869 * Assumes resources are ordered. For BLK regions, first consume
870 * BLK-only available DPA free space, then consume PMEM-aliased DPA
871 * space starting at the highest DPA. For PMEM regions start
872 * allocations from the start of an interleave set and end at the first
873 * BLK allocation or the end of the interleave set, whichever comes
874 * first.
875 */
876 static int grow_dpa_allocation(struct nd_region *nd_region,
877 struct nd_label_id *label_id, resource_size_t n)
878 {
879 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
880 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
881 int i;
882
883 for (i = 0; i < nd_region->ndr_mappings; i++) {
884 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
885 resource_size_t rem = n;
886 int rc, j;
887
888 /*
889 * In the BLK case try once with all unallocated PMEM
890 * reserved, and once without
891 */
892 for (j = is_pmem; j < 2; j++) {
893 bool blk_only = j == 0;
894
895 if (blk_only) {
896 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
897 if (rc)
898 return rc;
899 }
900 rem = scan_allocate(nd_region, nd_mapping,
901 label_id, rem);
902 if (blk_only)
903 release_free_pmem(nvdimm_bus, nd_mapping);
904
905 /* try again and allow encroachments into PMEM */
906 if (rem == 0)
907 break;
908 }
909
910 dev_WARN_ONCE(&nd_region->dev, rem,
911 "allocation underrun: %#llx of %#llx bytes\n",
912 (unsigned long long) n - rem,
913 (unsigned long long) n);
914 if (rem)
915 return -ENXIO;
916
917 rc = merge_dpa(nd_region, nd_mapping, label_id);
918 if (rc)
919 return rc;
920 }
921
922 return 0;
923 }
924
925 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
926 struct nd_namespace_pmem *nspm, resource_size_t size)
927 {
928 struct resource *res = &nspm->nsio.res;
929 resource_size_t offset = 0;
930
931 if (size && !nspm->uuid) {
932 WARN_ON_ONCE(1);
933 size = 0;
934 }
935
936 if (size && nspm->uuid) {
937 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
938 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
939 struct nd_label_id label_id;
940 struct resource *res;
941
942 if (!ndd) {
943 size = 0;
944 goto out;
945 }
946
947 nd_label_gen_id(&label_id, nspm->uuid, 0);
948
949 /* calculate a spa offset from the dpa allocation offset */
950 for_each_dpa_resource(ndd, res)
951 if (strcmp(res->name, label_id.id) == 0) {
952 offset = (res->start - nd_mapping->start)
953 * nd_region->ndr_mappings;
954 goto out;
955 }
956
957 WARN_ON_ONCE(1);
958 size = 0;
959 }
960
961 out:
962 res->start = nd_region->ndr_start + offset;
963 res->end = res->start + size - 1;
964 }
965
966 static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
967 {
968 if (!uuid) {
969 dev_dbg(dev, "%s: uuid not set\n", where);
970 return true;
971 }
972 return false;
973 }
974
975 static ssize_t __size_store(struct device *dev, unsigned long long val)
976 {
977 resource_size_t allocated = 0, available = 0;
978 struct nd_region *nd_region = to_nd_region(dev->parent);
979 struct nd_namespace_common *ndns = to_ndns(dev);
980 struct nd_mapping *nd_mapping;
981 struct nvdimm_drvdata *ndd;
982 struct nd_label_id label_id;
983 u32 flags = 0, remainder;
984 int rc, i, id = -1;
985 u8 *uuid = NULL;
986
987 if (dev->driver || ndns->claim)
988 return -EBUSY;
989
990 if (is_namespace_pmem(dev)) {
991 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
992
993 uuid = nspm->uuid;
994 id = nspm->id;
995 } else if (is_namespace_blk(dev)) {
996 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
997
998 uuid = nsblk->uuid;
999 flags = NSLABEL_FLAG_LOCAL;
1000 id = nsblk->id;
1001 }
1002
1003 /*
1004 * We need a uuid for the allocation-label and dimm(s) on which
1005 * to store the label.
1006 */
1007 if (uuid_not_set(uuid, dev, __func__))
1008 return -ENXIO;
1009 if (nd_region->ndr_mappings == 0) {
1010 dev_dbg(dev, "%s: not associated with dimm(s)\n", __func__);
1011 return -ENXIO;
1012 }
1013
1014 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
1015 if (remainder) {
1016 dev_dbg(dev, "%llu is not %dK aligned\n", val,
1017 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
1018 return -EINVAL;
1019 }
1020
1021 nd_label_gen_id(&label_id, uuid, flags);
1022 for (i = 0; i < nd_region->ndr_mappings; i++) {
1023 nd_mapping = &nd_region->mapping[i];
1024 ndd = to_ndd(nd_mapping);
1025
1026 /*
1027 * All dimms in an interleave set, or the base dimm for a blk
1028 * region, need to be enabled for the size to be changed.
1029 */
1030 if (!ndd)
1031 return -ENXIO;
1032
1033 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1034 }
1035 available = nd_region_available_dpa(nd_region);
1036
1037 if (val > available + allocated)
1038 return -ENOSPC;
1039
1040 if (val == allocated)
1041 return 0;
1042
1043 val = div_u64(val, nd_region->ndr_mappings);
1044 allocated = div_u64(allocated, nd_region->ndr_mappings);
1045 if (val < allocated)
1046 rc = shrink_dpa_allocation(nd_region, &label_id,
1047 allocated - val);
1048 else
1049 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1050
1051 if (rc)
1052 return rc;
1053
1054 if (is_namespace_pmem(dev)) {
1055 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1056
1057 nd_namespace_pmem_set_resource(nd_region, nspm,
1058 val * nd_region->ndr_mappings);
1059 }
1060
1061 /*
1062 * Try to delete the namespace if we deleted all of its
1063 * allocation, this is not the seed or 0th device for the
1064 * region, and it is not actively claimed by a btt, pfn, or dax
1065 * instance.
1066 */
1067 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1068 nd_device_unregister(dev, ND_ASYNC);
1069
1070 return rc;
1071 }
1072
1073 static ssize_t size_store(struct device *dev,
1074 struct device_attribute *attr, const char *buf, size_t len)
1075 {
1076 struct nd_region *nd_region = to_nd_region(dev->parent);
1077 unsigned long long val;
1078 u8 **uuid = NULL;
1079 int rc;
1080
1081 rc = kstrtoull(buf, 0, &val);
1082 if (rc)
1083 return rc;
1084
1085 device_lock(dev);
1086 nvdimm_bus_lock(dev);
1087 wait_nvdimm_bus_probe_idle(dev);
1088 rc = __size_store(dev, val);
1089 if (rc >= 0)
1090 rc = nd_namespace_label_update(nd_region, dev);
1091
1092 if (is_namespace_pmem(dev)) {
1093 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1094
1095 uuid = &nspm->uuid;
1096 } else if (is_namespace_blk(dev)) {
1097 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1098
1099 uuid = &nsblk->uuid;
1100 }
1101
1102 if (rc == 0 && val == 0 && uuid) {
1103 /* setting size zero == 'delete namespace' */
1104 kfree(*uuid);
1105 *uuid = NULL;
1106 }
1107
1108 dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
1109 ? "fail" : "success", rc);
1110
1111 nvdimm_bus_unlock(dev);
1112 device_unlock(dev);
1113
1114 return rc < 0 ? rc : len;
1115 }
1116
1117 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1118 {
1119 struct device *dev = &ndns->dev;
1120
1121 if (is_namespace_pmem(dev)) {
1122 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1123
1124 return resource_size(&nspm->nsio.res);
1125 } else if (is_namespace_blk(dev)) {
1126 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1127 } else if (is_namespace_io(dev)) {
1128 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1129
1130 return resource_size(&nsio->res);
1131 } else
1132 WARN_ONCE(1, "unknown namespace type\n");
1133 return 0;
1134 }
1135
1136 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1137 {
1138 resource_size_t size;
1139
1140 nvdimm_bus_lock(&ndns->dev);
1141 size = __nvdimm_namespace_capacity(ndns);
1142 nvdimm_bus_unlock(&ndns->dev);
1143
1144 return size;
1145 }
1146 EXPORT_SYMBOL(nvdimm_namespace_capacity);
1147
1148 static ssize_t size_show(struct device *dev,
1149 struct device_attribute *attr, char *buf)
1150 {
1151 return sprintf(buf, "%llu\n", (unsigned long long)
1152 nvdimm_namespace_capacity(to_ndns(dev)));
1153 }
1154 static DEVICE_ATTR(size, 0444, size_show, size_store);
1155
1156 static u8 *namespace_to_uuid(struct device *dev)
1157 {
1158 if (is_namespace_pmem(dev)) {
1159 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1160
1161 return nspm->uuid;
1162 } else if (is_namespace_blk(dev)) {
1163 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1164
1165 return nsblk->uuid;
1166 } else
1167 return ERR_PTR(-ENXIO);
1168 }
1169
1170 static ssize_t uuid_show(struct device *dev,
1171 struct device_attribute *attr, char *buf)
1172 {
1173 u8 *uuid = namespace_to_uuid(dev);
1174
1175 if (IS_ERR(uuid))
1176 return PTR_ERR(uuid);
1177 if (uuid)
1178 return sprintf(buf, "%pUb\n", uuid);
1179 return sprintf(buf, "\n");
1180 }
1181
1182 /**
1183 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1184 * @nd_region: parent region so we can updates all dimms in the set
1185 * @dev: namespace type for generating label_id
1186 * @new_uuid: incoming uuid
1187 * @old_uuid: reference to the uuid storage location in the namespace object
1188 */
1189 static int namespace_update_uuid(struct nd_region *nd_region,
1190 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1191 {
1192 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1193 struct nd_label_id old_label_id;
1194 struct nd_label_id new_label_id;
1195 int i;
1196
1197 if (!nd_is_uuid_unique(dev, new_uuid))
1198 return -EINVAL;
1199
1200 if (*old_uuid == NULL)
1201 goto out;
1202
1203 /*
1204 * If we've already written a label with this uuid, then it's
1205 * too late to rename because we can't reliably update the uuid
1206 * without losing the old namespace. Userspace must delete this
1207 * namespace to abandon the old uuid.
1208 */
1209 for (i = 0; i < nd_region->ndr_mappings; i++) {
1210 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1211
1212 /*
1213 * This check by itself is sufficient because old_uuid
1214 * would be NULL above if this uuid did not exist in the
1215 * currently written set.
1216 *
1217 * FIXME: can we delete uuid with zero dpa allocated?
1218 */
1219 if (list_empty(&nd_mapping->labels))
1220 return -EBUSY;
1221 }
1222
1223 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1224 nd_label_gen_id(&new_label_id, new_uuid, flags);
1225 for (i = 0; i < nd_region->ndr_mappings; i++) {
1226 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1227 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1228 struct resource *res;
1229
1230 for_each_dpa_resource(ndd, res)
1231 if (strcmp(res->name, old_label_id.id) == 0)
1232 sprintf((void *) res->name, "%s",
1233 new_label_id.id);
1234 }
1235 kfree(*old_uuid);
1236 out:
1237 *old_uuid = new_uuid;
1238 return 0;
1239 }
1240
1241 static ssize_t uuid_store(struct device *dev,
1242 struct device_attribute *attr, const char *buf, size_t len)
1243 {
1244 struct nd_region *nd_region = to_nd_region(dev->parent);
1245 u8 *uuid = NULL;
1246 ssize_t rc = 0;
1247 u8 **ns_uuid;
1248
1249 if (is_namespace_pmem(dev)) {
1250 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1251
1252 ns_uuid = &nspm->uuid;
1253 } else if (is_namespace_blk(dev)) {
1254 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1255
1256 ns_uuid = &nsblk->uuid;
1257 } else
1258 return -ENXIO;
1259
1260 device_lock(dev);
1261 nvdimm_bus_lock(dev);
1262 wait_nvdimm_bus_probe_idle(dev);
1263 if (to_ndns(dev)->claim)
1264 rc = -EBUSY;
1265 if (rc >= 0)
1266 rc = nd_uuid_store(dev, &uuid, buf, len);
1267 if (rc >= 0)
1268 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1269 if (rc >= 0)
1270 rc = nd_namespace_label_update(nd_region, dev);
1271 else
1272 kfree(uuid);
1273 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1274 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1275 nvdimm_bus_unlock(dev);
1276 device_unlock(dev);
1277
1278 return rc < 0 ? rc : len;
1279 }
1280 static DEVICE_ATTR_RW(uuid);
1281
1282 static ssize_t resource_show(struct device *dev,
1283 struct device_attribute *attr, char *buf)
1284 {
1285 struct resource *res;
1286
1287 if (is_namespace_pmem(dev)) {
1288 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1289
1290 res = &nspm->nsio.res;
1291 } else if (is_namespace_io(dev)) {
1292 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1293
1294 res = &nsio->res;
1295 } else
1296 return -ENXIO;
1297
1298 /* no address to convey if the namespace has no allocation */
1299 if (resource_size(res) == 0)
1300 return -ENXIO;
1301 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1302 }
1303 static DEVICE_ATTR_RO(resource);
1304
1305 static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1306 4096, 4104, 4160, 4224, 0 };
1307
1308 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1309
1310 static ssize_t sector_size_show(struct device *dev,
1311 struct device_attribute *attr, char *buf)
1312 {
1313 if (is_namespace_blk(dev)) {
1314 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1315
1316 return nd_sector_size_show(nsblk->lbasize,
1317 blk_lbasize_supported, buf);
1318 }
1319
1320 if (is_namespace_pmem(dev)) {
1321 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1322
1323 return nd_sector_size_show(nspm->lbasize,
1324 pmem_lbasize_supported, buf);
1325 }
1326 return -ENXIO;
1327 }
1328
1329 static ssize_t sector_size_store(struct device *dev,
1330 struct device_attribute *attr, const char *buf, size_t len)
1331 {
1332 struct nd_region *nd_region = to_nd_region(dev->parent);
1333 const unsigned long *supported;
1334 unsigned long *lbasize;
1335 ssize_t rc = 0;
1336
1337 if (is_namespace_blk(dev)) {
1338 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1339
1340 lbasize = &nsblk->lbasize;
1341 supported = blk_lbasize_supported;
1342 } else if (is_namespace_pmem(dev)) {
1343 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1344
1345 lbasize = &nspm->lbasize;
1346 supported = pmem_lbasize_supported;
1347 } else
1348 return -ENXIO;
1349
1350 device_lock(dev);
1351 nvdimm_bus_lock(dev);
1352 if (to_ndns(dev)->claim)
1353 rc = -EBUSY;
1354 if (rc >= 0)
1355 rc = nd_sector_size_store(dev, buf, lbasize, supported);
1356 if (rc >= 0)
1357 rc = nd_namespace_label_update(nd_region, dev);
1358 dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1359 rc, rc < 0 ? "tried" : "wrote", buf,
1360 buf[len - 1] == '\n' ? "" : "\n");
1361 nvdimm_bus_unlock(dev);
1362 device_unlock(dev);
1363
1364 return rc ? rc : len;
1365 }
1366 static DEVICE_ATTR_RW(sector_size);
1367
1368 static ssize_t dpa_extents_show(struct device *dev,
1369 struct device_attribute *attr, char *buf)
1370 {
1371 struct nd_region *nd_region = to_nd_region(dev->parent);
1372 struct nd_label_id label_id;
1373 int count = 0, i;
1374 u8 *uuid = NULL;
1375 u32 flags = 0;
1376
1377 nvdimm_bus_lock(dev);
1378 if (is_namespace_pmem(dev)) {
1379 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1380
1381 uuid = nspm->uuid;
1382 flags = 0;
1383 } else if (is_namespace_blk(dev)) {
1384 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1385
1386 uuid = nsblk->uuid;
1387 flags = NSLABEL_FLAG_LOCAL;
1388 }
1389
1390 if (!uuid)
1391 goto out;
1392
1393 nd_label_gen_id(&label_id, uuid, flags);
1394 for (i = 0; i < nd_region->ndr_mappings; i++) {
1395 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1396 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1397 struct resource *res;
1398
1399 for_each_dpa_resource(ndd, res)
1400 if (strcmp(res->name, label_id.id) == 0)
1401 count++;
1402 }
1403 out:
1404 nvdimm_bus_unlock(dev);
1405
1406 return sprintf(buf, "%d\n", count);
1407 }
1408 static DEVICE_ATTR_RO(dpa_extents);
1409
1410 static int btt_claim_class(struct device *dev)
1411 {
1412 struct nd_region *nd_region = to_nd_region(dev->parent);
1413 int i, loop_bitmask = 0;
1414
1415 for (i = 0; i < nd_region->ndr_mappings; i++) {
1416 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1417 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1418 struct nd_namespace_index *nsindex;
1419
1420 nsindex = to_namespace_index(ndd, ndd->ns_current);
1421 if (nsindex == NULL)
1422 loop_bitmask |= 1;
1423 else {
1424 /* check whether existing labels are v1.1 or v1.2 */
1425 if (__le16_to_cpu(nsindex->major) == 1
1426 && __le16_to_cpu(nsindex->minor) == 1)
1427 loop_bitmask |= 2;
1428 else
1429 loop_bitmask |= 4;
1430 }
1431 }
1432 /*
1433 * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1434 * block is found, a v1.1 label for any mapping will set bit 1, and a
1435 * v1.2 label will set bit 2.
1436 *
1437 * At the end of the loop, at most one of the three bits must be set.
1438 * If multiple bits were set, it means the different mappings disagree
1439 * about their labels, and this must be cleaned up first.
1440 *
1441 * If all the label index blocks are found to agree, nsindex of NULL
1442 * implies labels haven't been initialized yet, and when they will,
1443 * they will be of the 1.2 format, so we can assume BTT2.0
1444 *
1445 * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1446 * found, we enforce BTT2.0
1447 *
1448 * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1449 */
1450 switch (loop_bitmask) {
1451 case 0:
1452 case 2:
1453 return NVDIMM_CCLASS_BTT;
1454 case 1:
1455 case 4:
1456 return NVDIMM_CCLASS_BTT2;
1457 default:
1458 return -ENXIO;
1459 }
1460 }
1461
1462 static ssize_t holder_show(struct device *dev,
1463 struct device_attribute *attr, char *buf)
1464 {
1465 struct nd_namespace_common *ndns = to_ndns(dev);
1466 ssize_t rc;
1467
1468 device_lock(dev);
1469 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1470 device_unlock(dev);
1471
1472 return rc;
1473 }
1474 static DEVICE_ATTR_RO(holder);
1475
1476 static ssize_t __holder_class_store(struct device *dev, const char *buf)
1477 {
1478 struct nd_namespace_common *ndns = to_ndns(dev);
1479
1480 if (dev->driver || ndns->claim)
1481 return -EBUSY;
1482
1483 if (strcmp(buf, "btt") == 0 || strcmp(buf, "btt\n") == 0)
1484 ndns->claim_class = btt_claim_class(dev);
1485 else if (strcmp(buf, "pfn") == 0 || strcmp(buf, "pfn\n") == 0)
1486 ndns->claim_class = NVDIMM_CCLASS_PFN;
1487 else if (strcmp(buf, "dax") == 0 || strcmp(buf, "dax\n") == 0)
1488 ndns->claim_class = NVDIMM_CCLASS_DAX;
1489 else if (strcmp(buf, "") == 0 || strcmp(buf, "\n") == 0)
1490 ndns->claim_class = NVDIMM_CCLASS_NONE;
1491 else
1492 return -EINVAL;
1493
1494 /* btt_claim_class() could've returned an error */
1495 if (ndns->claim_class < 0)
1496 return ndns->claim_class;
1497
1498 return 0;
1499 }
1500
1501 static ssize_t holder_class_store(struct device *dev,
1502 struct device_attribute *attr, const char *buf, size_t len)
1503 {
1504 struct nd_region *nd_region = to_nd_region(dev->parent);
1505 ssize_t rc;
1506
1507 device_lock(dev);
1508 nvdimm_bus_lock(dev);
1509 wait_nvdimm_bus_probe_idle(dev);
1510 rc = __holder_class_store(dev, buf);
1511 if (rc >= 0)
1512 rc = nd_namespace_label_update(nd_region, dev);
1513 dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
1514 nvdimm_bus_unlock(dev);
1515 device_unlock(dev);
1516
1517 return rc < 0 ? rc : len;
1518 }
1519
1520 static ssize_t holder_class_show(struct device *dev,
1521 struct device_attribute *attr, char *buf)
1522 {
1523 struct nd_namespace_common *ndns = to_ndns(dev);
1524 ssize_t rc;
1525
1526 device_lock(dev);
1527 if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1528 rc = sprintf(buf, "\n");
1529 else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1530 (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1531 rc = sprintf(buf, "btt\n");
1532 else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1533 rc = sprintf(buf, "pfn\n");
1534 else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1535 rc = sprintf(buf, "dax\n");
1536 else
1537 rc = sprintf(buf, "<unknown>\n");
1538 device_unlock(dev);
1539
1540 return rc;
1541 }
1542 static DEVICE_ATTR_RW(holder_class);
1543
1544 static ssize_t mode_show(struct device *dev,
1545 struct device_attribute *attr, char *buf)
1546 {
1547 struct nd_namespace_common *ndns = to_ndns(dev);
1548 struct device *claim;
1549 char *mode;
1550 ssize_t rc;
1551
1552 device_lock(dev);
1553 claim = ndns->claim;
1554 if (claim && is_nd_btt(claim))
1555 mode = "safe";
1556 else if (claim && is_nd_pfn(claim))
1557 mode = "memory";
1558 else if (claim && is_nd_dax(claim))
1559 mode = "dax";
1560 else if (!claim && pmem_should_map_pages(dev))
1561 mode = "memory";
1562 else
1563 mode = "raw";
1564 rc = sprintf(buf, "%s\n", mode);
1565 device_unlock(dev);
1566
1567 return rc;
1568 }
1569 static DEVICE_ATTR_RO(mode);
1570
1571 static ssize_t force_raw_store(struct device *dev,
1572 struct device_attribute *attr, const char *buf, size_t len)
1573 {
1574 bool force_raw;
1575 int rc = strtobool(buf, &force_raw);
1576
1577 if (rc)
1578 return rc;
1579
1580 to_ndns(dev)->force_raw = force_raw;
1581 return len;
1582 }
1583
1584 static ssize_t force_raw_show(struct device *dev,
1585 struct device_attribute *attr, char *buf)
1586 {
1587 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1588 }
1589 static DEVICE_ATTR_RW(force_raw);
1590
1591 static struct attribute *nd_namespace_attributes[] = {
1592 &dev_attr_nstype.attr,
1593 &dev_attr_size.attr,
1594 &dev_attr_mode.attr,
1595 &dev_attr_uuid.attr,
1596 &dev_attr_holder.attr,
1597 &dev_attr_resource.attr,
1598 &dev_attr_alt_name.attr,
1599 &dev_attr_force_raw.attr,
1600 &dev_attr_sector_size.attr,
1601 &dev_attr_dpa_extents.attr,
1602 &dev_attr_holder_class.attr,
1603 NULL,
1604 };
1605
1606 static umode_t namespace_visible(struct kobject *kobj,
1607 struct attribute *a, int n)
1608 {
1609 struct device *dev = container_of(kobj, struct device, kobj);
1610
1611 if (a == &dev_attr_resource.attr) {
1612 if (is_namespace_blk(dev))
1613 return 0;
1614 return a->mode;
1615 }
1616
1617 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1618 if (a == &dev_attr_size.attr)
1619 return 0644;
1620
1621 return a->mode;
1622 }
1623
1624 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1625 || a == &dev_attr_holder.attr
1626 || a == &dev_attr_holder_class.attr
1627 || a == &dev_attr_force_raw.attr
1628 || a == &dev_attr_mode.attr)
1629 return a->mode;
1630
1631 return 0;
1632 }
1633
1634 static struct attribute_group nd_namespace_attribute_group = {
1635 .attrs = nd_namespace_attributes,
1636 .is_visible = namespace_visible,
1637 };
1638
1639 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1640 &nd_device_attribute_group,
1641 &nd_namespace_attribute_group,
1642 &nd_numa_attribute_group,
1643 NULL,
1644 };
1645
1646 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1647 {
1648 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1649 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1650 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1651 struct nd_namespace_common *ndns = NULL;
1652 resource_size_t size;
1653
1654 if (nd_btt || nd_pfn || nd_dax) {
1655 if (nd_btt)
1656 ndns = nd_btt->ndns;
1657 else if (nd_pfn)
1658 ndns = nd_pfn->ndns;
1659 else if (nd_dax)
1660 ndns = nd_dax->nd_pfn.ndns;
1661
1662 if (!ndns)
1663 return ERR_PTR(-ENODEV);
1664
1665 /*
1666 * Flush any in-progess probes / removals in the driver
1667 * for the raw personality of this namespace.
1668 */
1669 device_lock(&ndns->dev);
1670 device_unlock(&ndns->dev);
1671 if (ndns->dev.driver) {
1672 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1673 dev_name(dev));
1674 return ERR_PTR(-EBUSY);
1675 }
1676 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1677 "host (%s) vs claim (%s) mismatch\n",
1678 dev_name(dev),
1679 dev_name(ndns->claim)))
1680 return ERR_PTR(-ENXIO);
1681 } else {
1682 ndns = to_ndns(dev);
1683 if (ndns->claim) {
1684 dev_dbg(dev, "claimed by %s, failing probe\n",
1685 dev_name(ndns->claim));
1686
1687 return ERR_PTR(-ENXIO);
1688 }
1689 }
1690
1691 size = nvdimm_namespace_capacity(ndns);
1692 if (size < ND_MIN_NAMESPACE_SIZE) {
1693 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1694 &size, ND_MIN_NAMESPACE_SIZE);
1695 return ERR_PTR(-ENODEV);
1696 }
1697
1698 if (is_namespace_pmem(&ndns->dev)) {
1699 struct nd_namespace_pmem *nspm;
1700
1701 nspm = to_nd_namespace_pmem(&ndns->dev);
1702 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1703 return ERR_PTR(-ENODEV);
1704 } else if (is_namespace_blk(&ndns->dev)) {
1705 struct nd_namespace_blk *nsblk;
1706
1707 nsblk = to_nd_namespace_blk(&ndns->dev);
1708 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1709 return ERR_PTR(-ENODEV);
1710 if (!nsblk->lbasize) {
1711 dev_dbg(&ndns->dev, "%s: sector size not set\n",
1712 __func__);
1713 return ERR_PTR(-ENODEV);
1714 }
1715 if (!nd_namespace_blk_validate(nsblk))
1716 return ERR_PTR(-ENODEV);
1717 }
1718
1719 return ndns;
1720 }
1721 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1722
1723 static struct device **create_namespace_io(struct nd_region *nd_region)
1724 {
1725 struct nd_namespace_io *nsio;
1726 struct device *dev, **devs;
1727 struct resource *res;
1728
1729 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1730 if (!nsio)
1731 return NULL;
1732
1733 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1734 if (!devs) {
1735 kfree(nsio);
1736 return NULL;
1737 }
1738
1739 dev = &nsio->common.dev;
1740 dev->type = &namespace_io_device_type;
1741 dev->parent = &nd_region->dev;
1742 res = &nsio->res;
1743 res->name = dev_name(&nd_region->dev);
1744 res->flags = IORESOURCE_MEM;
1745 res->start = nd_region->ndr_start;
1746 res->end = res->start + nd_region->ndr_size - 1;
1747
1748 devs[0] = dev;
1749 return devs;
1750 }
1751
1752 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1753 u64 cookie, u16 pos)
1754 {
1755 struct nd_namespace_label *found = NULL;
1756 int i;
1757
1758 for (i = 0; i < nd_region->ndr_mappings; i++) {
1759 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1760 struct nd_interleave_set *nd_set = nd_region->nd_set;
1761 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1762 struct nd_label_ent *label_ent;
1763 bool found_uuid = false;
1764
1765 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1766 struct nd_namespace_label *nd_label = label_ent->label;
1767 u16 position, nlabel;
1768 u64 isetcookie;
1769
1770 if (!nd_label)
1771 continue;
1772 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1773 position = __le16_to_cpu(nd_label->position);
1774 nlabel = __le16_to_cpu(nd_label->nlabel);
1775
1776 if (isetcookie != cookie)
1777 continue;
1778
1779 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1780 continue;
1781
1782 if (namespace_label_has(ndd, type_guid)
1783 && !guid_equal(&nd_set->type_guid,
1784 &nd_label->type_guid)) {
1785 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1786 nd_set->type_guid.b,
1787 nd_label->type_guid.b);
1788 continue;
1789 }
1790
1791 if (found_uuid) {
1792 dev_dbg(ndd->dev,
1793 "%s duplicate entry for uuid\n",
1794 __func__);
1795 return false;
1796 }
1797 found_uuid = true;
1798 if (nlabel != nd_region->ndr_mappings)
1799 continue;
1800 if (position != pos)
1801 continue;
1802 found = nd_label;
1803 break;
1804 }
1805 if (found)
1806 break;
1807 }
1808 return found != NULL;
1809 }
1810
1811 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1812 {
1813 int i;
1814
1815 if (!pmem_id)
1816 return -ENODEV;
1817
1818 for (i = 0; i < nd_region->ndr_mappings; i++) {
1819 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1820 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1821 struct nd_namespace_label *nd_label = NULL;
1822 u64 hw_start, hw_end, pmem_start, pmem_end;
1823 struct nd_label_ent *label_ent;
1824
1825 lockdep_assert_held(&nd_mapping->lock);
1826 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1827 nd_label = label_ent->label;
1828 if (!nd_label)
1829 continue;
1830 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1831 break;
1832 nd_label = NULL;
1833 }
1834
1835 if (!nd_label) {
1836 WARN_ON(1);
1837 return -EINVAL;
1838 }
1839
1840 /*
1841 * Check that this label is compliant with the dpa
1842 * range published in NFIT
1843 */
1844 hw_start = nd_mapping->start;
1845 hw_end = hw_start + nd_mapping->size;
1846 pmem_start = __le64_to_cpu(nd_label->dpa);
1847 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1848 if (pmem_start >= hw_start && pmem_start < hw_end
1849 && pmem_end <= hw_end && pmem_end > hw_start)
1850 /* pass */;
1851 else {
1852 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1853 dev_name(ndd->dev), nd_label->uuid);
1854 return -EINVAL;
1855 }
1856
1857 /* move recently validated label to the front of the list */
1858 list_move(&label_ent->list, &nd_mapping->labels);
1859 }
1860 return 0;
1861 }
1862
1863 /**
1864 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1865 * @nd_region: region with mappings to validate
1866 * @nspm: target namespace to create
1867 * @nd_label: target pmem namespace label to evaluate
1868 */
1869 struct device *create_namespace_pmem(struct nd_region *nd_region,
1870 struct nd_namespace_index *nsindex,
1871 struct nd_namespace_label *nd_label)
1872 {
1873 u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1874 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1875 struct nd_label_ent *label_ent;
1876 struct nd_namespace_pmem *nspm;
1877 struct nd_mapping *nd_mapping;
1878 resource_size_t size = 0;
1879 struct resource *res;
1880 struct device *dev;
1881 int rc = 0;
1882 u16 i;
1883
1884 if (cookie == 0) {
1885 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1886 return ERR_PTR(-ENXIO);
1887 }
1888
1889 if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1890 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1891 nd_label->uuid);
1892 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1893 return ERR_PTR(-EAGAIN);
1894
1895 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1896 nd_label->uuid);
1897 }
1898
1899 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1900 if (!nspm)
1901 return ERR_PTR(-ENOMEM);
1902
1903 nspm->id = -1;
1904 dev = &nspm->nsio.common.dev;
1905 dev->type = &namespace_pmem_device_type;
1906 dev->parent = &nd_region->dev;
1907 res = &nspm->nsio.res;
1908 res->name = dev_name(&nd_region->dev);
1909 res->flags = IORESOURCE_MEM;
1910
1911 for (i = 0; i < nd_region->ndr_mappings; i++) {
1912 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1913 continue;
1914 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1915 continue;
1916 break;
1917 }
1918
1919 if (i < nd_region->ndr_mappings) {
1920 struct nvdimm_drvdata *ndd = to_ndd(&nd_region->mapping[i]);
1921
1922 /*
1923 * Give up if we don't find an instance of a uuid at each
1924 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1925 * find a dimm with two instances of the same uuid.
1926 */
1927 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1928 dev_name(ndd->dev), nd_label->uuid);
1929 rc = -EINVAL;
1930 goto err;
1931 }
1932
1933 /*
1934 * Fix up each mapping's 'labels' to have the validated pmem label for
1935 * that position at labels[0], and NULL at labels[1]. In the process,
1936 * check that the namespace aligns with interleave-set. We know
1937 * that it does not overlap with any blk namespaces by virtue of
1938 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1939 * succeeded).
1940 */
1941 rc = select_pmem_id(nd_region, nd_label->uuid);
1942 if (rc)
1943 goto err;
1944
1945 /* Calculate total size and populate namespace properties from label0 */
1946 for (i = 0; i < nd_region->ndr_mappings; i++) {
1947 struct nd_namespace_label *label0;
1948 struct nvdimm_drvdata *ndd;
1949
1950 nd_mapping = &nd_region->mapping[i];
1951 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1952 typeof(*label_ent), list);
1953 label0 = label_ent ? label_ent->label : 0;
1954
1955 if (!label0) {
1956 WARN_ON(1);
1957 continue;
1958 }
1959
1960 size += __le64_to_cpu(label0->rawsize);
1961 if (__le16_to_cpu(label0->position) != 0)
1962 continue;
1963 WARN_ON(nspm->alt_name || nspm->uuid);
1964 nspm->alt_name = kmemdup((void __force *) label0->name,
1965 NSLABEL_NAME_LEN, GFP_KERNEL);
1966 nspm->uuid = kmemdup((void __force *) label0->uuid,
1967 NSLABEL_UUID_LEN, GFP_KERNEL);
1968 nspm->lbasize = __le64_to_cpu(label0->lbasize);
1969 ndd = to_ndd(nd_mapping);
1970 if (namespace_label_has(ndd, abstraction_guid))
1971 nspm->nsio.common.claim_class
1972 = to_nvdimm_cclass(&label0->abstraction_guid);
1973
1974 }
1975
1976 if (!nspm->alt_name || !nspm->uuid) {
1977 rc = -ENOMEM;
1978 goto err;
1979 }
1980
1981 nd_namespace_pmem_set_resource(nd_region, nspm, size);
1982
1983 return dev;
1984 err:
1985 namespace_pmem_release(dev);
1986 switch (rc) {
1987 case -EINVAL:
1988 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1989 break;
1990 case -ENODEV:
1991 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
1992 break;
1993 default:
1994 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
1995 __func__, rc);
1996 break;
1997 }
1998 return ERR_PTR(rc);
1999 }
2000
2001 struct resource *nsblk_add_resource(struct nd_region *nd_region,
2002 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2003 resource_size_t start)
2004 {
2005 struct nd_label_id label_id;
2006 struct resource *res;
2007
2008 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2009 res = krealloc(nsblk->res,
2010 sizeof(void *) * (nsblk->num_resources + 1),
2011 GFP_KERNEL);
2012 if (!res)
2013 return NULL;
2014 nsblk->res = (struct resource **) res;
2015 for_each_dpa_resource(ndd, res)
2016 if (strcmp(res->name, label_id.id) == 0
2017 && res->start == start) {
2018 nsblk->res[nsblk->num_resources++] = res;
2019 return res;
2020 }
2021 return NULL;
2022 }
2023
2024 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2025 {
2026 struct nd_namespace_blk *nsblk;
2027 struct device *dev;
2028
2029 if (!is_nd_blk(&nd_region->dev))
2030 return NULL;
2031
2032 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2033 if (!nsblk)
2034 return NULL;
2035
2036 dev = &nsblk->common.dev;
2037 dev->type = &namespace_blk_device_type;
2038 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2039 if (nsblk->id < 0) {
2040 kfree(nsblk);
2041 return NULL;
2042 }
2043 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2044 dev->parent = &nd_region->dev;
2045 dev->groups = nd_namespace_attribute_groups;
2046
2047 return &nsblk->common.dev;
2048 }
2049
2050 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2051 {
2052 struct nd_namespace_pmem *nspm;
2053 struct resource *res;
2054 struct device *dev;
2055
2056 if (!is_memory(&nd_region->dev))
2057 return NULL;
2058
2059 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2060 if (!nspm)
2061 return NULL;
2062
2063 dev = &nspm->nsio.common.dev;
2064 dev->type = &namespace_pmem_device_type;
2065 dev->parent = &nd_region->dev;
2066 res = &nspm->nsio.res;
2067 res->name = dev_name(&nd_region->dev);
2068 res->flags = IORESOURCE_MEM;
2069
2070 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2071 if (nspm->id < 0) {
2072 kfree(nspm);
2073 return NULL;
2074 }
2075 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2076 dev->parent = &nd_region->dev;
2077 dev->groups = nd_namespace_attribute_groups;
2078 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2079
2080 return dev;
2081 }
2082
2083 void nd_region_create_ns_seed(struct nd_region *nd_region)
2084 {
2085 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2086
2087 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2088 return;
2089
2090 if (is_nd_blk(&nd_region->dev))
2091 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2092 else
2093 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2094
2095 /*
2096 * Seed creation failures are not fatal, provisioning is simply
2097 * disabled until memory becomes available
2098 */
2099 if (!nd_region->ns_seed)
2100 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2101 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2102 else
2103 nd_device_register(nd_region->ns_seed);
2104 }
2105
2106 void nd_region_create_dax_seed(struct nd_region *nd_region)
2107 {
2108 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2109 nd_region->dax_seed = nd_dax_create(nd_region);
2110 /*
2111 * Seed creation failures are not fatal, provisioning is simply
2112 * disabled until memory becomes available
2113 */
2114 if (!nd_region->dax_seed)
2115 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2116 }
2117
2118 void nd_region_create_pfn_seed(struct nd_region *nd_region)
2119 {
2120 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2121 nd_region->pfn_seed = nd_pfn_create(nd_region);
2122 /*
2123 * Seed creation failures are not fatal, provisioning is simply
2124 * disabled until memory becomes available
2125 */
2126 if (!nd_region->pfn_seed)
2127 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2128 }
2129
2130 void nd_region_create_btt_seed(struct nd_region *nd_region)
2131 {
2132 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2133 nd_region->btt_seed = nd_btt_create(nd_region);
2134 /*
2135 * Seed creation failures are not fatal, provisioning is simply
2136 * disabled until memory becomes available
2137 */
2138 if (!nd_region->btt_seed)
2139 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2140 }
2141
2142 static int add_namespace_resource(struct nd_region *nd_region,
2143 struct nd_namespace_label *nd_label, struct device **devs,
2144 int count)
2145 {
2146 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2147 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2148 int i;
2149
2150 for (i = 0; i < count; i++) {
2151 u8 *uuid = namespace_to_uuid(devs[i]);
2152 struct resource *res;
2153
2154 if (IS_ERR_OR_NULL(uuid)) {
2155 WARN_ON(1);
2156 continue;
2157 }
2158
2159 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2160 continue;
2161 if (is_namespace_blk(devs[i])) {
2162 res = nsblk_add_resource(nd_region, ndd,
2163 to_nd_namespace_blk(devs[i]),
2164 __le64_to_cpu(nd_label->dpa));
2165 if (!res)
2166 return -ENXIO;
2167 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2168 } else {
2169 dev_err(&nd_region->dev,
2170 "error: conflicting extents for uuid: %pUb\n",
2171 nd_label->uuid);
2172 return -ENXIO;
2173 }
2174 break;
2175 }
2176
2177 return i;
2178 }
2179
2180 struct device *create_namespace_blk(struct nd_region *nd_region,
2181 struct nd_namespace_label *nd_label, int count)
2182 {
2183
2184 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2185 struct nd_interleave_set *nd_set = nd_region->nd_set;
2186 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2187 struct nd_namespace_blk *nsblk;
2188 char name[NSLABEL_NAME_LEN];
2189 struct device *dev = NULL;
2190 struct resource *res;
2191
2192 if (namespace_label_has(ndd, type_guid)) {
2193 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2194 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2195 nd_set->type_guid.b,
2196 nd_label->type_guid.b);
2197 return ERR_PTR(-EAGAIN);
2198 }
2199
2200 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2201 dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2202 nd_set->cookie2,
2203 __le64_to_cpu(nd_label->isetcookie));
2204 return ERR_PTR(-EAGAIN);
2205 }
2206 }
2207
2208 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2209 if (!nsblk)
2210 return ERR_PTR(-ENOMEM);
2211 dev = &nsblk->common.dev;
2212 dev->type = &namespace_blk_device_type;
2213 dev->parent = &nd_region->dev;
2214 nsblk->id = -1;
2215 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2216 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2217 GFP_KERNEL);
2218 if (namespace_label_has(ndd, abstraction_guid))
2219 nsblk->common.claim_class
2220 = to_nvdimm_cclass(&nd_label->abstraction_guid);
2221 if (!nsblk->uuid)
2222 goto blk_err;
2223 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2224 if (name[0])
2225 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2226 GFP_KERNEL);
2227 res = nsblk_add_resource(nd_region, ndd, nsblk,
2228 __le64_to_cpu(nd_label->dpa));
2229 if (!res)
2230 goto blk_err;
2231 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2232 return dev;
2233 blk_err:
2234 namespace_blk_release(dev);
2235 return ERR_PTR(-ENXIO);
2236 }
2237
2238 static int cmp_dpa(const void *a, const void *b)
2239 {
2240 const struct device *dev_a = *(const struct device **) a;
2241 const struct device *dev_b = *(const struct device **) b;
2242 struct nd_namespace_blk *nsblk_a, *nsblk_b;
2243 struct nd_namespace_pmem *nspm_a, *nspm_b;
2244
2245 if (is_namespace_io(dev_a))
2246 return 0;
2247
2248 if (is_namespace_blk(dev_a)) {
2249 nsblk_a = to_nd_namespace_blk(dev_a);
2250 nsblk_b = to_nd_namespace_blk(dev_b);
2251
2252 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2253 sizeof(resource_size_t));
2254 }
2255
2256 nspm_a = to_nd_namespace_pmem(dev_a);
2257 nspm_b = to_nd_namespace_pmem(dev_b);
2258
2259 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2260 sizeof(resource_size_t));
2261 }
2262
2263 static struct device **scan_labels(struct nd_region *nd_region)
2264 {
2265 int i, count = 0;
2266 struct device *dev, **devs = NULL;
2267 struct nd_label_ent *label_ent, *e;
2268 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2269 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2270
2271 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2272 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2273 struct nd_namespace_label *nd_label = label_ent->label;
2274 struct device **__devs;
2275 u32 flags;
2276
2277 if (!nd_label)
2278 continue;
2279 flags = __le32_to_cpu(nd_label->flags);
2280 if (is_nd_blk(&nd_region->dev)
2281 == !!(flags & NSLABEL_FLAG_LOCAL))
2282 /* pass, region matches label type */;
2283 else
2284 continue;
2285
2286 /* skip labels that describe extents outside of the region */
2287 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2288 continue;
2289
2290 i = add_namespace_resource(nd_region, nd_label, devs, count);
2291 if (i < 0)
2292 goto err;
2293 if (i < count)
2294 continue;
2295 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2296 if (!__devs)
2297 goto err;
2298 memcpy(__devs, devs, sizeof(dev) * count);
2299 kfree(devs);
2300 devs = __devs;
2301
2302 if (is_nd_blk(&nd_region->dev))
2303 dev = create_namespace_blk(nd_region, nd_label, count);
2304 else {
2305 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2306 struct nd_namespace_index *nsindex;
2307
2308 nsindex = to_namespace_index(ndd, ndd->ns_current);
2309 dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2310 }
2311
2312 if (IS_ERR(dev)) {
2313 switch (PTR_ERR(dev)) {
2314 case -EAGAIN:
2315 /* skip invalid labels */
2316 continue;
2317 case -ENODEV:
2318 /* fallthrough to seed creation */
2319 break;
2320 default:
2321 goto err;
2322 }
2323 } else
2324 devs[count++] = dev;
2325
2326 }
2327
2328 dev_dbg(&nd_region->dev, "%s: discovered %d %s namespace%s\n",
2329 __func__, count, is_nd_blk(&nd_region->dev)
2330 ? "blk" : "pmem", count == 1 ? "" : "s");
2331
2332 if (count == 0) {
2333 /* Publish a zero-sized namespace for userspace to configure. */
2334 nd_mapping_free_labels(nd_mapping);
2335
2336 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2337 if (!devs)
2338 goto err;
2339 if (is_nd_blk(&nd_region->dev)) {
2340 struct nd_namespace_blk *nsblk;
2341
2342 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2343 if (!nsblk)
2344 goto err;
2345 dev = &nsblk->common.dev;
2346 dev->type = &namespace_blk_device_type;
2347 } else {
2348 struct nd_namespace_pmem *nspm;
2349
2350 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2351 if (!nspm)
2352 goto err;
2353 dev = &nspm->nsio.common.dev;
2354 dev->type = &namespace_pmem_device_type;
2355 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2356 }
2357 dev->parent = &nd_region->dev;
2358 devs[count++] = dev;
2359 } else if (is_memory(&nd_region->dev)) {
2360 /* clean unselected labels */
2361 for (i = 0; i < nd_region->ndr_mappings; i++) {
2362 struct list_head *l, *e;
2363 LIST_HEAD(list);
2364 int j;
2365
2366 nd_mapping = &nd_region->mapping[i];
2367 if (list_empty(&nd_mapping->labels)) {
2368 WARN_ON(1);
2369 continue;
2370 }
2371
2372 j = count;
2373 list_for_each_safe(l, e, &nd_mapping->labels) {
2374 if (!j--)
2375 break;
2376 list_move_tail(l, &list);
2377 }
2378 nd_mapping_free_labels(nd_mapping);
2379 list_splice_init(&list, &nd_mapping->labels);
2380 }
2381 }
2382
2383 if (count > 1)
2384 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2385
2386 return devs;
2387
2388 err:
2389 if (devs) {
2390 for (i = 0; devs[i]; i++)
2391 if (is_nd_blk(&nd_region->dev))
2392 namespace_blk_release(devs[i]);
2393 else
2394 namespace_pmem_release(devs[i]);
2395 kfree(devs);
2396 }
2397 return NULL;
2398 }
2399
2400 static struct device **create_namespaces(struct nd_region *nd_region)
2401 {
2402 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2403 struct device **devs;
2404 int i;
2405
2406 if (nd_region->ndr_mappings == 0)
2407 return NULL;
2408
2409 /* lock down all mappings while we scan labels */
2410 for (i = 0; i < nd_region->ndr_mappings; i++) {
2411 nd_mapping = &nd_region->mapping[i];
2412 mutex_lock_nested(&nd_mapping->lock, i);
2413 }
2414
2415 devs = scan_labels(nd_region);
2416
2417 for (i = 0; i < nd_region->ndr_mappings; i++) {
2418 int reverse = nd_region->ndr_mappings - 1 - i;
2419
2420 nd_mapping = &nd_region->mapping[reverse];
2421 mutex_unlock(&nd_mapping->lock);
2422 }
2423
2424 return devs;
2425 }
2426
2427 static int init_active_labels(struct nd_region *nd_region)
2428 {
2429 int i;
2430
2431 for (i = 0; i < nd_region->ndr_mappings; i++) {
2432 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2433 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2434 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2435 struct nd_label_ent *label_ent;
2436 int count, j;
2437
2438 /*
2439 * If the dimm is disabled then we may need to prevent
2440 * the region from being activated.
2441 */
2442 if (!ndd) {
2443 if (test_bit(NDD_LOCKED, &nvdimm->flags))
2444 /* fail, label data may be unreadable */;
2445 else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2446 /* fail, labels needed to disambiguate dpa */;
2447 else
2448 return 0;
2449
2450 dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2451 dev_name(&nd_mapping->nvdimm->dev),
2452 test_bit(NDD_LOCKED, &nvdimm->flags)
2453 ? "locked" : "disabled");
2454 return -ENXIO;
2455 }
2456 nd_mapping->ndd = ndd;
2457 atomic_inc(&nvdimm->busy);
2458 get_ndd(ndd);
2459
2460 count = nd_label_active_count(ndd);
2461 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
2462 if (!count)
2463 continue;
2464 for (j = 0; j < count; j++) {
2465 struct nd_namespace_label *label;
2466
2467 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2468 if (!label_ent)
2469 break;
2470 label = nd_label_active(ndd, j);
2471 label_ent->label = label;
2472
2473 mutex_lock(&nd_mapping->lock);
2474 list_add_tail(&label_ent->list, &nd_mapping->labels);
2475 mutex_unlock(&nd_mapping->lock);
2476 }
2477
2478 if (j >= count)
2479 continue;
2480
2481 mutex_lock(&nd_mapping->lock);
2482 nd_mapping_free_labels(nd_mapping);
2483 mutex_unlock(&nd_mapping->lock);
2484 return -ENOMEM;
2485 }
2486
2487 return 0;
2488 }
2489
2490 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2491 {
2492 struct device **devs = NULL;
2493 int i, rc = 0, type;
2494
2495 *err = 0;
2496 nvdimm_bus_lock(&nd_region->dev);
2497 rc = init_active_labels(nd_region);
2498 if (rc) {
2499 nvdimm_bus_unlock(&nd_region->dev);
2500 return rc;
2501 }
2502
2503 type = nd_region_to_nstype(nd_region);
2504 switch (type) {
2505 case ND_DEVICE_NAMESPACE_IO:
2506 devs = create_namespace_io(nd_region);
2507 break;
2508 case ND_DEVICE_NAMESPACE_PMEM:
2509 case ND_DEVICE_NAMESPACE_BLK:
2510 devs = create_namespaces(nd_region);
2511 break;
2512 default:
2513 break;
2514 }
2515 nvdimm_bus_unlock(&nd_region->dev);
2516
2517 if (!devs)
2518 return -ENODEV;
2519
2520 for (i = 0; devs[i]; i++) {
2521 struct device *dev = devs[i];
2522 int id;
2523
2524 if (type == ND_DEVICE_NAMESPACE_BLK) {
2525 struct nd_namespace_blk *nsblk;
2526
2527 nsblk = to_nd_namespace_blk(dev);
2528 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2529 GFP_KERNEL);
2530 nsblk->id = id;
2531 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2532 struct nd_namespace_pmem *nspm;
2533
2534 nspm = to_nd_namespace_pmem(dev);
2535 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2536 GFP_KERNEL);
2537 nspm->id = id;
2538 } else
2539 id = i;
2540
2541 if (id < 0)
2542 break;
2543 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2544 dev->groups = nd_namespace_attribute_groups;
2545 nd_device_register(dev);
2546 }
2547 if (i)
2548 nd_region->ns_seed = devs[0];
2549
2550 if (devs[i]) {
2551 int j;
2552
2553 for (j = i; devs[j]; j++) {
2554 struct device *dev = devs[j];
2555
2556 device_initialize(dev);
2557 put_device(dev);
2558 }
2559 *err = j - i;
2560 /*
2561 * All of the namespaces we tried to register failed, so
2562 * fail region activation.
2563 */
2564 if (*err == 0)
2565 rc = -ENODEV;
2566 }
2567 kfree(devs);
2568
2569 if (rc == -ENODEV)
2570 return rc;
2571
2572 return i;
2573 }