]>
Commit | Line | Data |
---|---|---|
3d88002e DW |
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/slab.h> | |
16 | #include <linux/nd.h> | |
bf9bccc1 | 17 | #include "nd-core.h" |
3d88002e DW |
18 | #include "nd.h" |
19 | ||
20 | static void namespace_io_release(struct device *dev) | |
21 | { | |
22 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); | |
23 | ||
24 | kfree(nsio); | |
25 | } | |
26 | ||
bf9bccc1 DW |
27 | static void namespace_pmem_release(struct device *dev) |
28 | { | |
29 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
30 | ||
31 | kfree(nspm->alt_name); | |
32 | kfree(nspm->uuid); | |
33 | kfree(nspm); | |
34 | } | |
35 | ||
36 | static void namespace_blk_release(struct device *dev) | |
37 | { | |
1b40e09a DW |
38 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
39 | struct nd_region *nd_region = to_nd_region(dev->parent); | |
40 | ||
41 | if (nsblk->id >= 0) | |
42 | ida_simple_remove(&nd_region->ns_ida, nsblk->id); | |
43 | kfree(nsblk->alt_name); | |
44 | kfree(nsblk->uuid); | |
45 | kfree(nsblk->res); | |
46 | kfree(nsblk); | |
bf9bccc1 DW |
47 | } |
48 | ||
3d88002e DW |
49 | static struct device_type namespace_io_device_type = { |
50 | .name = "nd_namespace_io", | |
51 | .release = namespace_io_release, | |
52 | }; | |
53 | ||
bf9bccc1 DW |
54 | static struct device_type namespace_pmem_device_type = { |
55 | .name = "nd_namespace_pmem", | |
56 | .release = namespace_pmem_release, | |
57 | }; | |
58 | ||
59 | static struct device_type namespace_blk_device_type = { | |
60 | .name = "nd_namespace_blk", | |
61 | .release = namespace_blk_release, | |
62 | }; | |
63 | ||
64 | static bool is_namespace_pmem(struct device *dev) | |
65 | { | |
66 | return dev ? dev->type == &namespace_pmem_device_type : false; | |
67 | } | |
68 | ||
69 | static bool is_namespace_blk(struct device *dev) | |
70 | { | |
71 | return dev ? dev->type == &namespace_blk_device_type : false; | |
72 | } | |
73 | ||
74 | static bool is_namespace_io(struct device *dev) | |
75 | { | |
76 | return dev ? dev->type == &namespace_io_device_type : false; | |
77 | } | |
78 | ||
3d88002e DW |
79 | static ssize_t nstype_show(struct device *dev, |
80 | struct device_attribute *attr, char *buf) | |
81 | { | |
82 | struct nd_region *nd_region = to_nd_region(dev->parent); | |
83 | ||
84 | return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region)); | |
85 | } | |
86 | static DEVICE_ATTR_RO(nstype); | |
87 | ||
bf9bccc1 DW |
88 | static ssize_t __alt_name_store(struct device *dev, const char *buf, |
89 | const size_t len) | |
90 | { | |
91 | char *input, *pos, *alt_name, **ns_altname; | |
92 | ssize_t rc; | |
93 | ||
94 | if (is_namespace_pmem(dev)) { | |
95 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
96 | ||
97 | ns_altname = &nspm->alt_name; | |
98 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
99 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
100 | ||
101 | ns_altname = &nsblk->alt_name; | |
bf9bccc1 DW |
102 | } else |
103 | return -ENXIO; | |
104 | ||
105 | if (dev->driver) | |
106 | return -EBUSY; | |
107 | ||
108 | input = kmemdup(buf, len + 1, GFP_KERNEL); | |
109 | if (!input) | |
110 | return -ENOMEM; | |
111 | ||
112 | input[len] = '\0'; | |
113 | pos = strim(input); | |
114 | if (strlen(pos) + 1 > NSLABEL_NAME_LEN) { | |
115 | rc = -EINVAL; | |
116 | goto out; | |
117 | } | |
118 | ||
119 | alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL); | |
120 | if (!alt_name) { | |
121 | rc = -ENOMEM; | |
122 | goto out; | |
123 | } | |
124 | kfree(*ns_altname); | |
125 | *ns_altname = alt_name; | |
126 | sprintf(*ns_altname, "%s", pos); | |
127 | rc = len; | |
128 | ||
129 | out: | |
130 | kfree(input); | |
131 | return rc; | |
132 | } | |
133 | ||
1b40e09a DW |
134 | static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk) |
135 | { | |
136 | struct nd_region *nd_region = to_nd_region(nsblk->dev.parent); | |
137 | struct nd_mapping *nd_mapping = &nd_region->mapping[0]; | |
138 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
139 | struct nd_label_id label_id; | |
140 | resource_size_t size = 0; | |
141 | struct resource *res; | |
142 | ||
143 | if (!nsblk->uuid) | |
144 | return 0; | |
145 | nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL); | |
146 | for_each_dpa_resource(ndd, res) | |
147 | if (strcmp(res->name, label_id.id) == 0) | |
148 | size += resource_size(res); | |
149 | return size; | |
150 | } | |
151 | ||
f524bf27 DW |
152 | static int nd_namespace_label_update(struct nd_region *nd_region, |
153 | struct device *dev) | |
154 | { | |
155 | dev_WARN_ONCE(dev, dev->driver, | |
156 | "namespace must be idle during label update\n"); | |
157 | if (dev->driver) | |
158 | return 0; | |
159 | ||
160 | /* | |
161 | * Only allow label writes that will result in a valid namespace | |
162 | * or deletion of an existing namespace. | |
163 | */ | |
164 | if (is_namespace_pmem(dev)) { | |
165 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
166 | struct resource *res = &nspm->nsio.res; | |
167 | resource_size_t size = resource_size(res); | |
168 | ||
169 | if (size == 0 && nspm->uuid) | |
170 | /* delete allocation */; | |
171 | else if (!nspm->uuid) | |
172 | return 0; | |
173 | ||
174 | return nd_pmem_namespace_label_update(nd_region, nspm, size); | |
175 | } else if (is_namespace_blk(dev)) { | |
176 | /* TODO: implement blk labels */ | |
177 | return 0; | |
178 | } else | |
179 | return -ENXIO; | |
180 | } | |
181 | ||
bf9bccc1 DW |
182 | static ssize_t alt_name_store(struct device *dev, |
183 | struct device_attribute *attr, const char *buf, size_t len) | |
184 | { | |
f524bf27 | 185 | struct nd_region *nd_region = to_nd_region(dev->parent); |
bf9bccc1 DW |
186 | ssize_t rc; |
187 | ||
188 | device_lock(dev); | |
189 | nvdimm_bus_lock(dev); | |
190 | wait_nvdimm_bus_probe_idle(dev); | |
191 | rc = __alt_name_store(dev, buf, len); | |
f524bf27 DW |
192 | if (rc >= 0) |
193 | rc = nd_namespace_label_update(nd_region, dev); | |
bf9bccc1 DW |
194 | dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc); |
195 | nvdimm_bus_unlock(dev); | |
196 | device_unlock(dev); | |
197 | ||
f524bf27 | 198 | return rc < 0 ? rc : len; |
bf9bccc1 DW |
199 | } |
200 | ||
201 | static ssize_t alt_name_show(struct device *dev, | |
202 | struct device_attribute *attr, char *buf) | |
203 | { | |
204 | char *ns_altname; | |
205 | ||
206 | if (is_namespace_pmem(dev)) { | |
207 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
208 | ||
209 | ns_altname = nspm->alt_name; | |
210 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
211 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
212 | ||
213 | ns_altname = nsblk->alt_name; | |
bf9bccc1 DW |
214 | } else |
215 | return -ENXIO; | |
216 | ||
217 | return sprintf(buf, "%s\n", ns_altname ? ns_altname : ""); | |
218 | } | |
219 | static DEVICE_ATTR_RW(alt_name); | |
220 | ||
221 | static int scan_free(struct nd_region *nd_region, | |
222 | struct nd_mapping *nd_mapping, struct nd_label_id *label_id, | |
223 | resource_size_t n) | |
224 | { | |
225 | bool is_blk = strncmp(label_id->id, "blk", 3) == 0; | |
226 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
227 | int rc = 0; | |
228 | ||
229 | while (n) { | |
230 | struct resource *res, *last; | |
231 | resource_size_t new_start; | |
232 | ||
233 | last = NULL; | |
234 | for_each_dpa_resource(ndd, res) | |
235 | if (strcmp(res->name, label_id->id) == 0) | |
236 | last = res; | |
237 | res = last; | |
238 | if (!res) | |
239 | return 0; | |
240 | ||
241 | if (n >= resource_size(res)) { | |
242 | n -= resource_size(res); | |
243 | nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc); | |
244 | nvdimm_free_dpa(ndd, res); | |
245 | /* retry with last resource deleted */ | |
246 | continue; | |
247 | } | |
248 | ||
249 | /* | |
250 | * Keep BLK allocations relegated to high DPA as much as | |
251 | * possible | |
252 | */ | |
253 | if (is_blk) | |
254 | new_start = res->start + n; | |
255 | else | |
256 | new_start = res->start; | |
257 | ||
258 | rc = adjust_resource(res, new_start, resource_size(res) - n); | |
1b40e09a DW |
259 | if (rc == 0) |
260 | res->flags |= DPA_RESOURCE_ADJUSTED; | |
bf9bccc1 DW |
261 | nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc); |
262 | break; | |
263 | } | |
264 | ||
265 | return rc; | |
266 | } | |
267 | ||
268 | /** | |
269 | * shrink_dpa_allocation - for each dimm in region free n bytes for label_id | |
270 | * @nd_region: the set of dimms to reclaim @n bytes from | |
271 | * @label_id: unique identifier for the namespace consuming this dpa range | |
272 | * @n: number of bytes per-dimm to release | |
273 | * | |
274 | * Assumes resources are ordered. Starting from the end try to | |
275 | * adjust_resource() the allocation to @n, but if @n is larger than the | |
276 | * allocation delete it and find the 'new' last allocation in the label | |
277 | * set. | |
278 | */ | |
279 | static int shrink_dpa_allocation(struct nd_region *nd_region, | |
280 | struct nd_label_id *label_id, resource_size_t n) | |
281 | { | |
282 | int i; | |
283 | ||
284 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
285 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
286 | int rc; | |
287 | ||
288 | rc = scan_free(nd_region, nd_mapping, label_id, n); | |
289 | if (rc) | |
290 | return rc; | |
291 | } | |
292 | ||
293 | return 0; | |
294 | } | |
295 | ||
296 | static resource_size_t init_dpa_allocation(struct nd_label_id *label_id, | |
297 | struct nd_region *nd_region, struct nd_mapping *nd_mapping, | |
298 | resource_size_t n) | |
299 | { | |
300 | bool is_blk = strncmp(label_id->id, "blk", 3) == 0; | |
301 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
302 | resource_size_t first_dpa; | |
303 | struct resource *res; | |
304 | int rc = 0; | |
305 | ||
306 | /* allocate blk from highest dpa first */ | |
307 | if (is_blk) | |
308 | first_dpa = nd_mapping->start + nd_mapping->size - n; | |
309 | else | |
310 | first_dpa = nd_mapping->start; | |
311 | ||
312 | /* first resource allocation for this label-id or dimm */ | |
313 | res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n); | |
314 | if (!res) | |
315 | rc = -EBUSY; | |
316 | ||
317 | nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc); | |
318 | return rc ? n : 0; | |
319 | } | |
320 | ||
1b40e09a DW |
321 | static bool space_valid(bool is_pmem, bool is_reserve, |
322 | struct nd_label_id *label_id, struct resource *res) | |
bf9bccc1 DW |
323 | { |
324 | /* | |
325 | * For BLK-space any space is valid, for PMEM-space, it must be | |
1b40e09a DW |
326 | * contiguous with an existing allocation unless we are |
327 | * reserving pmem. | |
bf9bccc1 | 328 | */ |
1b40e09a | 329 | if (is_reserve || !is_pmem) |
bf9bccc1 DW |
330 | return true; |
331 | if (!res || strcmp(res->name, label_id->id) == 0) | |
332 | return true; | |
333 | return false; | |
334 | } | |
335 | ||
336 | enum alloc_loc { | |
337 | ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER, | |
338 | }; | |
339 | ||
340 | static resource_size_t scan_allocate(struct nd_region *nd_region, | |
341 | struct nd_mapping *nd_mapping, struct nd_label_id *label_id, | |
342 | resource_size_t n) | |
343 | { | |
344 | resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1; | |
1b40e09a | 345 | bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0; |
bf9bccc1 DW |
346 | bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0; |
347 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
348 | const resource_size_t to_allocate = n; | |
349 | struct resource *res; | |
350 | int first; | |
351 | ||
352 | retry: | |
353 | first = 0; | |
354 | for_each_dpa_resource(ndd, res) { | |
355 | resource_size_t allocate, available = 0, free_start, free_end; | |
356 | struct resource *next = res->sibling, *new_res = NULL; | |
357 | enum alloc_loc loc = ALLOC_ERR; | |
358 | const char *action; | |
359 | int rc = 0; | |
360 | ||
361 | /* ignore resources outside this nd_mapping */ | |
362 | if (res->start > mapping_end) | |
363 | continue; | |
364 | if (res->end < nd_mapping->start) | |
365 | continue; | |
366 | ||
367 | /* space at the beginning of the mapping */ | |
368 | if (!first++ && res->start > nd_mapping->start) { | |
369 | free_start = nd_mapping->start; | |
370 | available = res->start - free_start; | |
1b40e09a | 371 | if (space_valid(is_pmem, is_reserve, label_id, NULL)) |
bf9bccc1 DW |
372 | loc = ALLOC_BEFORE; |
373 | } | |
374 | ||
375 | /* space between allocations */ | |
376 | if (!loc && next) { | |
377 | free_start = res->start + resource_size(res); | |
378 | free_end = min(mapping_end, next->start - 1); | |
1b40e09a | 379 | if (space_valid(is_pmem, is_reserve, label_id, res) |
bf9bccc1 DW |
380 | && free_start < free_end) { |
381 | available = free_end + 1 - free_start; | |
382 | loc = ALLOC_MID; | |
383 | } | |
384 | } | |
385 | ||
386 | /* space at the end of the mapping */ | |
387 | if (!loc && !next) { | |
388 | free_start = res->start + resource_size(res); | |
389 | free_end = mapping_end; | |
1b40e09a | 390 | if (space_valid(is_pmem, is_reserve, label_id, res) |
bf9bccc1 DW |
391 | && free_start < free_end) { |
392 | available = free_end + 1 - free_start; | |
393 | loc = ALLOC_AFTER; | |
394 | } | |
395 | } | |
396 | ||
397 | if (!loc || !available) | |
398 | continue; | |
399 | allocate = min(available, n); | |
400 | switch (loc) { | |
401 | case ALLOC_BEFORE: | |
402 | if (strcmp(res->name, label_id->id) == 0) { | |
403 | /* adjust current resource up */ | |
1b40e09a | 404 | if (is_pmem && !is_reserve) |
bf9bccc1 DW |
405 | return n; |
406 | rc = adjust_resource(res, res->start - allocate, | |
407 | resource_size(res) + allocate); | |
408 | action = "cur grow up"; | |
409 | } else | |
410 | action = "allocate"; | |
411 | break; | |
412 | case ALLOC_MID: | |
413 | if (strcmp(next->name, label_id->id) == 0) { | |
414 | /* adjust next resource up */ | |
1b40e09a | 415 | if (is_pmem && !is_reserve) |
bf9bccc1 DW |
416 | return n; |
417 | rc = adjust_resource(next, next->start | |
418 | - allocate, resource_size(next) | |
419 | + allocate); | |
420 | new_res = next; | |
421 | action = "next grow up"; | |
422 | } else if (strcmp(res->name, label_id->id) == 0) { | |
423 | action = "grow down"; | |
424 | } else | |
425 | action = "allocate"; | |
426 | break; | |
427 | case ALLOC_AFTER: | |
428 | if (strcmp(res->name, label_id->id) == 0) | |
429 | action = "grow down"; | |
430 | else | |
431 | action = "allocate"; | |
432 | break; | |
433 | default: | |
434 | return n; | |
435 | } | |
436 | ||
437 | if (strcmp(action, "allocate") == 0) { | |
438 | /* BLK allocate bottom up */ | |
439 | if (!is_pmem) | |
440 | free_start += available - allocate; | |
1b40e09a | 441 | else if (!is_reserve && free_start != nd_mapping->start) |
bf9bccc1 DW |
442 | return n; |
443 | ||
444 | new_res = nvdimm_allocate_dpa(ndd, label_id, | |
445 | free_start, allocate); | |
446 | if (!new_res) | |
447 | rc = -EBUSY; | |
448 | } else if (strcmp(action, "grow down") == 0) { | |
449 | /* adjust current resource down */ | |
450 | rc = adjust_resource(res, res->start, resource_size(res) | |
451 | + allocate); | |
1b40e09a DW |
452 | if (rc == 0) |
453 | res->flags |= DPA_RESOURCE_ADJUSTED; | |
bf9bccc1 DW |
454 | } |
455 | ||
456 | if (!new_res) | |
457 | new_res = res; | |
458 | ||
459 | nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n", | |
460 | action, loc, rc); | |
461 | ||
462 | if (rc) | |
463 | return n; | |
464 | ||
465 | n -= allocate; | |
466 | if (n) { | |
467 | /* | |
468 | * Retry scan with newly inserted resources. | |
469 | * For example, if we did an ALLOC_BEFORE | |
470 | * insertion there may also have been space | |
471 | * available for an ALLOC_AFTER insertion, so we | |
472 | * need to check this same resource again | |
473 | */ | |
474 | goto retry; | |
475 | } else | |
476 | return 0; | |
477 | } | |
478 | ||
1b40e09a DW |
479 | /* |
480 | * If we allocated nothing in the BLK case it may be because we are in | |
481 | * an initial "pmem-reserve pass". Only do an initial BLK allocation | |
482 | * when none of the DPA space is reserved. | |
483 | */ | |
484 | if ((is_pmem || !ndd->dpa.child) && n == to_allocate) | |
bf9bccc1 DW |
485 | return init_dpa_allocation(label_id, nd_region, nd_mapping, n); |
486 | return n; | |
487 | } | |
488 | ||
1b40e09a DW |
489 | static int merge_dpa(struct nd_region *nd_region, |
490 | struct nd_mapping *nd_mapping, struct nd_label_id *label_id) | |
491 | { | |
492 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
493 | struct resource *res; | |
494 | ||
495 | if (strncmp("pmem", label_id->id, 4) == 0) | |
496 | return 0; | |
497 | retry: | |
498 | for_each_dpa_resource(ndd, res) { | |
499 | int rc; | |
500 | struct resource *next = res->sibling; | |
501 | resource_size_t end = res->start + resource_size(res); | |
502 | ||
503 | if (!next || strcmp(res->name, label_id->id) != 0 | |
504 | || strcmp(next->name, label_id->id) != 0 | |
505 | || end != next->start) | |
506 | continue; | |
507 | end += resource_size(next); | |
508 | nvdimm_free_dpa(ndd, next); | |
509 | rc = adjust_resource(res, res->start, end - res->start); | |
510 | nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc); | |
511 | if (rc) | |
512 | return rc; | |
513 | res->flags |= DPA_RESOURCE_ADJUSTED; | |
514 | goto retry; | |
515 | } | |
516 | ||
517 | return 0; | |
518 | } | |
519 | ||
520 | static int __reserve_free_pmem(struct device *dev, void *data) | |
521 | { | |
522 | struct nvdimm *nvdimm = data; | |
523 | struct nd_region *nd_region; | |
524 | struct nd_label_id label_id; | |
525 | int i; | |
526 | ||
527 | if (!is_nd_pmem(dev)) | |
528 | return 0; | |
529 | ||
530 | nd_region = to_nd_region(dev); | |
531 | if (nd_region->ndr_mappings == 0) | |
532 | return 0; | |
533 | ||
534 | memset(&label_id, 0, sizeof(label_id)); | |
535 | strcat(label_id.id, "pmem-reserve"); | |
536 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
537 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
538 | resource_size_t n, rem = 0; | |
539 | ||
540 | if (nd_mapping->nvdimm != nvdimm) | |
541 | continue; | |
542 | ||
543 | n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem); | |
544 | if (n == 0) | |
545 | return 0; | |
546 | rem = scan_allocate(nd_region, nd_mapping, &label_id, n); | |
547 | dev_WARN_ONCE(&nd_region->dev, rem, | |
548 | "pmem reserve underrun: %#llx of %#llx bytes\n", | |
549 | (unsigned long long) n - rem, | |
550 | (unsigned long long) n); | |
551 | return rem ? -ENXIO : 0; | |
552 | } | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
557 | static void release_free_pmem(struct nvdimm_bus *nvdimm_bus, | |
558 | struct nd_mapping *nd_mapping) | |
559 | { | |
560 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
561 | struct resource *res, *_res; | |
562 | ||
563 | for_each_dpa_resource_safe(ndd, res, _res) | |
564 | if (strcmp(res->name, "pmem-reserve") == 0) | |
565 | nvdimm_free_dpa(ndd, res); | |
566 | } | |
567 | ||
568 | static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus, | |
569 | struct nd_mapping *nd_mapping) | |
570 | { | |
571 | struct nvdimm *nvdimm = nd_mapping->nvdimm; | |
572 | int rc; | |
573 | ||
574 | rc = device_for_each_child(&nvdimm_bus->dev, nvdimm, | |
575 | __reserve_free_pmem); | |
576 | if (rc) | |
577 | release_free_pmem(nvdimm_bus, nd_mapping); | |
578 | return rc; | |
579 | } | |
580 | ||
bf9bccc1 DW |
581 | /** |
582 | * grow_dpa_allocation - for each dimm allocate n bytes for @label_id | |
583 | * @nd_region: the set of dimms to allocate @n more bytes from | |
584 | * @label_id: unique identifier for the namespace consuming this dpa range | |
585 | * @n: number of bytes per-dimm to add to the existing allocation | |
586 | * | |
587 | * Assumes resources are ordered. For BLK regions, first consume | |
588 | * BLK-only available DPA free space, then consume PMEM-aliased DPA | |
589 | * space starting at the highest DPA. For PMEM regions start | |
590 | * allocations from the start of an interleave set and end at the first | |
591 | * BLK allocation or the end of the interleave set, whichever comes | |
592 | * first. | |
593 | */ | |
594 | static int grow_dpa_allocation(struct nd_region *nd_region, | |
595 | struct nd_label_id *label_id, resource_size_t n) | |
596 | { | |
1b40e09a DW |
597 | struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev); |
598 | bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0; | |
bf9bccc1 DW |
599 | int i; |
600 | ||
601 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
602 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1b40e09a DW |
603 | resource_size_t rem = n; |
604 | int rc, j; | |
605 | ||
606 | /* | |
607 | * In the BLK case try once with all unallocated PMEM | |
608 | * reserved, and once without | |
609 | */ | |
610 | for (j = is_pmem; j < 2; j++) { | |
611 | bool blk_only = j == 0; | |
612 | ||
613 | if (blk_only) { | |
614 | rc = reserve_free_pmem(nvdimm_bus, nd_mapping); | |
615 | if (rc) | |
616 | return rc; | |
617 | } | |
618 | rem = scan_allocate(nd_region, nd_mapping, | |
619 | label_id, rem); | |
620 | if (blk_only) | |
621 | release_free_pmem(nvdimm_bus, nd_mapping); | |
bf9bccc1 | 622 | |
1b40e09a DW |
623 | /* try again and allow encroachments into PMEM */ |
624 | if (rem == 0) | |
625 | break; | |
626 | } | |
627 | ||
628 | dev_WARN_ONCE(&nd_region->dev, rem, | |
629 | "allocation underrun: %#llx of %#llx bytes\n", | |
630 | (unsigned long long) n - rem, | |
631 | (unsigned long long) n); | |
632 | if (rem) | |
633 | return -ENXIO; | |
634 | ||
635 | rc = merge_dpa(nd_region, nd_mapping, label_id); | |
bf9bccc1 DW |
636 | if (rc) |
637 | return rc; | |
638 | } | |
639 | ||
640 | return 0; | |
641 | } | |
642 | ||
643 | static void nd_namespace_pmem_set_size(struct nd_region *nd_region, | |
644 | struct nd_namespace_pmem *nspm, resource_size_t size) | |
645 | { | |
646 | struct resource *res = &nspm->nsio.res; | |
647 | ||
648 | res->start = nd_region->ndr_start; | |
649 | res->end = nd_region->ndr_start + size - 1; | |
650 | } | |
651 | ||
652 | static ssize_t __size_store(struct device *dev, unsigned long long val) | |
653 | { | |
654 | resource_size_t allocated = 0, available = 0; | |
655 | struct nd_region *nd_region = to_nd_region(dev->parent); | |
656 | struct nd_mapping *nd_mapping; | |
657 | struct nvdimm_drvdata *ndd; | |
658 | struct nd_label_id label_id; | |
659 | u32 flags = 0, remainder; | |
660 | u8 *uuid = NULL; | |
661 | int rc, i; | |
662 | ||
663 | if (dev->driver) | |
664 | return -EBUSY; | |
665 | ||
666 | if (is_namespace_pmem(dev)) { | |
667 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
668 | ||
669 | uuid = nspm->uuid; | |
670 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
671 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
672 | ||
673 | uuid = nsblk->uuid; | |
674 | flags = NSLABEL_FLAG_LOCAL; | |
bf9bccc1 DW |
675 | } |
676 | ||
677 | /* | |
678 | * We need a uuid for the allocation-label and dimm(s) on which | |
679 | * to store the label. | |
680 | */ | |
681 | if (!uuid || nd_region->ndr_mappings == 0) | |
682 | return -ENXIO; | |
683 | ||
684 | div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder); | |
685 | if (remainder) { | |
686 | dev_dbg(dev, "%llu is not %dK aligned\n", val, | |
687 | (SZ_4K * nd_region->ndr_mappings) / SZ_1K); | |
688 | return -EINVAL; | |
689 | } | |
690 | ||
691 | nd_label_gen_id(&label_id, uuid, flags); | |
692 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
693 | nd_mapping = &nd_region->mapping[i]; | |
694 | ndd = to_ndd(nd_mapping); | |
695 | ||
696 | /* | |
697 | * All dimms in an interleave set, or the base dimm for a blk | |
698 | * region, need to be enabled for the size to be changed. | |
699 | */ | |
700 | if (!ndd) | |
701 | return -ENXIO; | |
702 | ||
703 | allocated += nvdimm_allocated_dpa(ndd, &label_id); | |
704 | } | |
705 | available = nd_region_available_dpa(nd_region); | |
706 | ||
707 | if (val > available + allocated) | |
708 | return -ENOSPC; | |
709 | ||
710 | if (val == allocated) | |
711 | return 0; | |
712 | ||
713 | val = div_u64(val, nd_region->ndr_mappings); | |
714 | allocated = div_u64(allocated, nd_region->ndr_mappings); | |
715 | if (val < allocated) | |
716 | rc = shrink_dpa_allocation(nd_region, &label_id, | |
717 | allocated - val); | |
718 | else | |
719 | rc = grow_dpa_allocation(nd_region, &label_id, val - allocated); | |
720 | ||
721 | if (rc) | |
722 | return rc; | |
723 | ||
724 | if (is_namespace_pmem(dev)) { | |
725 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
726 | ||
727 | nd_namespace_pmem_set_size(nd_region, nspm, | |
728 | val * nd_region->ndr_mappings); | |
1b40e09a DW |
729 | } else if (is_namespace_blk(dev)) { |
730 | /* | |
731 | * Try to delete the namespace if we deleted all of its | |
732 | * allocation and this is not the seed device for the | |
733 | * region. | |
734 | */ | |
735 | if (val == 0 && nd_region->ns_seed != dev) | |
736 | nd_device_unregister(dev, ND_ASYNC); | |
bf9bccc1 DW |
737 | } |
738 | ||
739 | return rc; | |
740 | } | |
741 | ||
742 | static ssize_t size_store(struct device *dev, | |
743 | struct device_attribute *attr, const char *buf, size_t len) | |
744 | { | |
f524bf27 | 745 | struct nd_region *nd_region = to_nd_region(dev->parent); |
bf9bccc1 DW |
746 | unsigned long long val; |
747 | u8 **uuid = NULL; | |
748 | int rc; | |
749 | ||
750 | rc = kstrtoull(buf, 0, &val); | |
751 | if (rc) | |
752 | return rc; | |
753 | ||
754 | device_lock(dev); | |
755 | nvdimm_bus_lock(dev); | |
756 | wait_nvdimm_bus_probe_idle(dev); | |
757 | rc = __size_store(dev, val); | |
f524bf27 DW |
758 | if (rc >= 0) |
759 | rc = nd_namespace_label_update(nd_region, dev); | |
bf9bccc1 DW |
760 | |
761 | if (is_namespace_pmem(dev)) { | |
762 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
763 | ||
764 | uuid = &nspm->uuid; | |
765 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
766 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
767 | ||
768 | uuid = &nsblk->uuid; | |
bf9bccc1 DW |
769 | } |
770 | ||
771 | if (rc == 0 && val == 0 && uuid) { | |
772 | /* setting size zero == 'delete namespace' */ | |
773 | kfree(*uuid); | |
774 | *uuid = NULL; | |
775 | } | |
776 | ||
777 | dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0 | |
778 | ? "fail" : "success", rc); | |
779 | ||
780 | nvdimm_bus_unlock(dev); | |
781 | device_unlock(dev); | |
782 | ||
f524bf27 | 783 | return rc < 0 ? rc : len; |
bf9bccc1 DW |
784 | } |
785 | ||
786 | static ssize_t size_show(struct device *dev, | |
787 | struct device_attribute *attr, char *buf) | |
788 | { | |
1b40e09a DW |
789 | unsigned long long size = 0; |
790 | ||
791 | nvdimm_bus_lock(dev); | |
bf9bccc1 DW |
792 | if (is_namespace_pmem(dev)) { |
793 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
794 | ||
1b40e09a | 795 | size = resource_size(&nspm->nsio.res); |
bf9bccc1 | 796 | } else if (is_namespace_blk(dev)) { |
1b40e09a | 797 | size = nd_namespace_blk_size(to_nd_namespace_blk(dev)); |
bf9bccc1 DW |
798 | } else if (is_namespace_io(dev)) { |
799 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); | |
800 | ||
1b40e09a DW |
801 | size = resource_size(&nsio->res); |
802 | } | |
803 | nvdimm_bus_unlock(dev); | |
804 | ||
805 | return sprintf(buf, "%llu\n", size); | |
bf9bccc1 DW |
806 | } |
807 | static DEVICE_ATTR(size, S_IRUGO, size_show, size_store); | |
808 | ||
809 | static ssize_t uuid_show(struct device *dev, | |
810 | struct device_attribute *attr, char *buf) | |
811 | { | |
812 | u8 *uuid; | |
813 | ||
814 | if (is_namespace_pmem(dev)) { | |
815 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
816 | ||
817 | uuid = nspm->uuid; | |
818 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
819 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
820 | ||
821 | uuid = nsblk->uuid; | |
bf9bccc1 DW |
822 | } else |
823 | return -ENXIO; | |
824 | ||
825 | if (uuid) | |
826 | return sprintf(buf, "%pUb\n", uuid); | |
827 | return sprintf(buf, "\n"); | |
828 | } | |
829 | ||
830 | /** | |
831 | * namespace_update_uuid - check for a unique uuid and whether we're "renaming" | |
832 | * @nd_region: parent region so we can updates all dimms in the set | |
833 | * @dev: namespace type for generating label_id | |
834 | * @new_uuid: incoming uuid | |
835 | * @old_uuid: reference to the uuid storage location in the namespace object | |
836 | */ | |
837 | static int namespace_update_uuid(struct nd_region *nd_region, | |
838 | struct device *dev, u8 *new_uuid, u8 **old_uuid) | |
839 | { | |
840 | u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0; | |
841 | struct nd_label_id old_label_id; | |
842 | struct nd_label_id new_label_id; | |
f524bf27 | 843 | int i; |
bf9bccc1 | 844 | |
f524bf27 DW |
845 | if (!nd_is_uuid_unique(dev, new_uuid)) |
846 | return -EINVAL; | |
bf9bccc1 DW |
847 | |
848 | if (*old_uuid == NULL) | |
849 | goto out; | |
850 | ||
f524bf27 DW |
851 | /* |
852 | * If we've already written a label with this uuid, then it's | |
853 | * too late to rename because we can't reliably update the uuid | |
854 | * without losing the old namespace. Userspace must delete this | |
855 | * namespace to abandon the old uuid. | |
856 | */ | |
857 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
858 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
859 | ||
860 | /* | |
861 | * This check by itself is sufficient because old_uuid | |
862 | * would be NULL above if this uuid did not exist in the | |
863 | * currently written set. | |
864 | * | |
865 | * FIXME: can we delete uuid with zero dpa allocated? | |
866 | */ | |
867 | if (nd_mapping->labels) | |
868 | return -EBUSY; | |
869 | } | |
870 | ||
bf9bccc1 DW |
871 | nd_label_gen_id(&old_label_id, *old_uuid, flags); |
872 | nd_label_gen_id(&new_label_id, new_uuid, flags); | |
873 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
874 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
875 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
876 | struct resource *res; | |
877 | ||
878 | for_each_dpa_resource(ndd, res) | |
879 | if (strcmp(res->name, old_label_id.id) == 0) | |
880 | sprintf((void *) res->name, "%s", | |
881 | new_label_id.id); | |
882 | } | |
883 | kfree(*old_uuid); | |
884 | out: | |
885 | *old_uuid = new_uuid; | |
886 | return 0; | |
887 | } | |
888 | ||
889 | static ssize_t uuid_store(struct device *dev, | |
890 | struct device_attribute *attr, const char *buf, size_t len) | |
891 | { | |
892 | struct nd_region *nd_region = to_nd_region(dev->parent); | |
893 | u8 *uuid = NULL; | |
894 | u8 **ns_uuid; | |
895 | ssize_t rc; | |
896 | ||
897 | if (is_namespace_pmem(dev)) { | |
898 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
899 | ||
900 | ns_uuid = &nspm->uuid; | |
901 | } else if (is_namespace_blk(dev)) { | |
1b40e09a DW |
902 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); |
903 | ||
904 | ns_uuid = &nsblk->uuid; | |
bf9bccc1 DW |
905 | } else |
906 | return -ENXIO; | |
907 | ||
908 | device_lock(dev); | |
909 | nvdimm_bus_lock(dev); | |
910 | wait_nvdimm_bus_probe_idle(dev); | |
911 | rc = nd_uuid_store(dev, &uuid, buf, len); | |
912 | if (rc >= 0) | |
913 | rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid); | |
f524bf27 DW |
914 | if (rc >= 0) |
915 | rc = nd_namespace_label_update(nd_region, dev); | |
916 | else | |
917 | kfree(uuid); | |
bf9bccc1 DW |
918 | dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__, |
919 | rc, buf, buf[len - 1] == '\n' ? "" : "\n"); | |
920 | nvdimm_bus_unlock(dev); | |
921 | device_unlock(dev); | |
922 | ||
f524bf27 | 923 | return rc < 0 ? rc : len; |
bf9bccc1 DW |
924 | } |
925 | static DEVICE_ATTR_RW(uuid); | |
926 | ||
927 | static ssize_t resource_show(struct device *dev, | |
928 | struct device_attribute *attr, char *buf) | |
929 | { | |
930 | struct resource *res; | |
931 | ||
932 | if (is_namespace_pmem(dev)) { | |
933 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); | |
934 | ||
935 | res = &nspm->nsio.res; | |
936 | } else if (is_namespace_io(dev)) { | |
937 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); | |
938 | ||
939 | res = &nsio->res; | |
940 | } else | |
941 | return -ENXIO; | |
942 | ||
943 | /* no address to convey if the namespace has no allocation */ | |
944 | if (resource_size(res) == 0) | |
945 | return -ENXIO; | |
946 | return sprintf(buf, "%#llx\n", (unsigned long long) res->start); | |
947 | } | |
948 | static DEVICE_ATTR_RO(resource); | |
949 | ||
1b40e09a DW |
950 | static const unsigned long ns_lbasize_supported[] = { 512, 0 }; |
951 | ||
952 | static ssize_t sector_size_show(struct device *dev, | |
953 | struct device_attribute *attr, char *buf) | |
954 | { | |
955 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); | |
956 | ||
957 | if (!is_namespace_blk(dev)) | |
958 | return -ENXIO; | |
959 | ||
960 | return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf); | |
961 | } | |
962 | ||
963 | static ssize_t sector_size_store(struct device *dev, | |
964 | struct device_attribute *attr, const char *buf, size_t len) | |
965 | { | |
966 | struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); | |
f524bf27 | 967 | struct nd_region *nd_region = to_nd_region(dev->parent); |
1b40e09a DW |
968 | ssize_t rc; |
969 | ||
970 | if (!is_namespace_blk(dev)) | |
971 | return -ENXIO; | |
972 | ||
973 | device_lock(dev); | |
974 | nvdimm_bus_lock(dev); | |
975 | rc = nd_sector_size_store(dev, buf, &nsblk->lbasize, | |
976 | ns_lbasize_supported); | |
f524bf27 DW |
977 | if (rc >= 0) |
978 | rc = nd_namespace_label_update(nd_region, dev); | |
979 | dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__, | |
980 | rc, rc < 0 ? "tried" : "wrote", buf, | |
981 | buf[len - 1] == '\n' ? "" : "\n"); | |
1b40e09a DW |
982 | nvdimm_bus_unlock(dev); |
983 | device_unlock(dev); | |
984 | ||
985 | return rc ? rc : len; | |
986 | } | |
987 | static DEVICE_ATTR_RW(sector_size); | |
988 | ||
3d88002e DW |
989 | static struct attribute *nd_namespace_attributes[] = { |
990 | &dev_attr_nstype.attr, | |
bf9bccc1 DW |
991 | &dev_attr_size.attr, |
992 | &dev_attr_uuid.attr, | |
993 | &dev_attr_resource.attr, | |
994 | &dev_attr_alt_name.attr, | |
1b40e09a | 995 | &dev_attr_sector_size.attr, |
3d88002e DW |
996 | NULL, |
997 | }; | |
998 | ||
bf9bccc1 DW |
999 | static umode_t namespace_visible(struct kobject *kobj, |
1000 | struct attribute *a, int n) | |
1001 | { | |
1002 | struct device *dev = container_of(kobj, struct device, kobj); | |
1003 | ||
1004 | if (a == &dev_attr_resource.attr) { | |
1005 | if (is_namespace_blk(dev)) | |
1006 | return 0; | |
1007 | return a->mode; | |
1008 | } | |
1009 | ||
1010 | if (is_namespace_pmem(dev) || is_namespace_blk(dev)) { | |
1011 | if (a == &dev_attr_size.attr) | |
1012 | return S_IWUSR | S_IRUGO; | |
1b40e09a DW |
1013 | |
1014 | if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr) | |
1015 | return 0; | |
1016 | ||
bf9bccc1 DW |
1017 | return a->mode; |
1018 | } | |
1019 | ||
1020 | if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr) | |
1021 | return a->mode; | |
1022 | ||
1023 | return 0; | |
1024 | } | |
1025 | ||
3d88002e DW |
1026 | static struct attribute_group nd_namespace_attribute_group = { |
1027 | .attrs = nd_namespace_attributes, | |
bf9bccc1 | 1028 | .is_visible = namespace_visible, |
3d88002e DW |
1029 | }; |
1030 | ||
1031 | static const struct attribute_group *nd_namespace_attribute_groups[] = { | |
1032 | &nd_device_attribute_group, | |
1033 | &nd_namespace_attribute_group, | |
1034 | NULL, | |
1035 | }; | |
1036 | ||
1037 | static struct device **create_namespace_io(struct nd_region *nd_region) | |
1038 | { | |
1039 | struct nd_namespace_io *nsio; | |
1040 | struct device *dev, **devs; | |
1041 | struct resource *res; | |
1042 | ||
1043 | nsio = kzalloc(sizeof(*nsio), GFP_KERNEL); | |
1044 | if (!nsio) | |
1045 | return NULL; | |
1046 | ||
1047 | devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL); | |
1048 | if (!devs) { | |
1049 | kfree(nsio); | |
1050 | return NULL; | |
1051 | } | |
1052 | ||
1053 | dev = &nsio->dev; | |
1054 | dev->type = &namespace_io_device_type; | |
1055 | dev->parent = &nd_region->dev; | |
1056 | res = &nsio->res; | |
1057 | res->name = dev_name(&nd_region->dev); | |
1058 | res->flags = IORESOURCE_MEM; | |
1059 | res->start = nd_region->ndr_start; | |
1060 | res->end = res->start + nd_region->ndr_size - 1; | |
1061 | ||
1062 | devs[0] = dev; | |
1063 | return devs; | |
1064 | } | |
1065 | ||
bf9bccc1 DW |
1066 | static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid, |
1067 | u64 cookie, u16 pos) | |
1068 | { | |
1069 | struct nd_namespace_label *found = NULL; | |
1070 | int i; | |
1071 | ||
1072 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1073 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1074 | struct nd_namespace_label *nd_label; | |
1075 | bool found_uuid = false; | |
1076 | int l; | |
1077 | ||
1078 | for_each_label(l, nd_label, nd_mapping->labels) { | |
1079 | u64 isetcookie = __le64_to_cpu(nd_label->isetcookie); | |
1080 | u16 position = __le16_to_cpu(nd_label->position); | |
1081 | u16 nlabel = __le16_to_cpu(nd_label->nlabel); | |
1082 | ||
1083 | if (isetcookie != cookie) | |
1084 | continue; | |
1085 | ||
1086 | if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0) | |
1087 | continue; | |
1088 | ||
1089 | if (found_uuid) { | |
1090 | dev_dbg(to_ndd(nd_mapping)->dev, | |
1091 | "%s duplicate entry for uuid\n", | |
1092 | __func__); | |
1093 | return false; | |
1094 | } | |
1095 | found_uuid = true; | |
1096 | if (nlabel != nd_region->ndr_mappings) | |
1097 | continue; | |
1098 | if (position != pos) | |
1099 | continue; | |
1100 | found = nd_label; | |
1101 | break; | |
1102 | } | |
1103 | if (found) | |
1104 | break; | |
1105 | } | |
1106 | return found != NULL; | |
1107 | } | |
1108 | ||
1109 | static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id) | |
1110 | { | |
1111 | struct nd_namespace_label *select = NULL; | |
1112 | int i; | |
1113 | ||
1114 | if (!pmem_id) | |
1115 | return -ENODEV; | |
1116 | ||
1117 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1118 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1119 | struct nd_namespace_label *nd_label; | |
1120 | u64 hw_start, hw_end, pmem_start, pmem_end; | |
1121 | int l; | |
1122 | ||
1123 | for_each_label(l, nd_label, nd_mapping->labels) | |
1124 | if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0) | |
1125 | break; | |
1126 | ||
1127 | if (!nd_label) { | |
1128 | WARN_ON(1); | |
1129 | return -EINVAL; | |
1130 | } | |
1131 | ||
1132 | select = nd_label; | |
1133 | /* | |
1134 | * Check that this label is compliant with the dpa | |
1135 | * range published in NFIT | |
1136 | */ | |
1137 | hw_start = nd_mapping->start; | |
1138 | hw_end = hw_start + nd_mapping->size; | |
1139 | pmem_start = __le64_to_cpu(select->dpa); | |
1140 | pmem_end = pmem_start + __le64_to_cpu(select->rawsize); | |
1141 | if (pmem_start == hw_start && pmem_end <= hw_end) | |
1142 | /* pass */; | |
1143 | else | |
1144 | return -EINVAL; | |
1145 | ||
1146 | nd_mapping->labels[0] = select; | |
1147 | nd_mapping->labels[1] = NULL; | |
1148 | } | |
1149 | return 0; | |
1150 | } | |
1151 | ||
1152 | /** | |
1153 | * find_pmem_label_set - validate interleave set labelling, retrieve label0 | |
1154 | * @nd_region: region with mappings to validate | |
1155 | */ | |
1156 | static int find_pmem_label_set(struct nd_region *nd_region, | |
1157 | struct nd_namespace_pmem *nspm) | |
1158 | { | |
1159 | u64 cookie = nd_region_interleave_set_cookie(nd_region); | |
1160 | struct nd_namespace_label *nd_label; | |
1161 | u8 select_id[NSLABEL_UUID_LEN]; | |
1162 | resource_size_t size = 0; | |
1163 | u8 *pmem_id = NULL; | |
1164 | int rc = -ENODEV, l; | |
1165 | u16 i; | |
1166 | ||
1167 | if (cookie == 0) | |
1168 | return -ENXIO; | |
1169 | ||
1170 | /* | |
1171 | * Find a complete set of labels by uuid. By definition we can start | |
1172 | * with any mapping as the reference label | |
1173 | */ | |
1174 | for_each_label(l, nd_label, nd_region->mapping[0].labels) { | |
1175 | u64 isetcookie = __le64_to_cpu(nd_label->isetcookie); | |
1176 | ||
1177 | if (isetcookie != cookie) | |
1178 | continue; | |
1179 | ||
1180 | for (i = 0; nd_region->ndr_mappings; i++) | |
1181 | if (!has_uuid_at_pos(nd_region, nd_label->uuid, | |
1182 | cookie, i)) | |
1183 | break; | |
1184 | if (i < nd_region->ndr_mappings) { | |
1185 | /* | |
1186 | * Give up if we don't find an instance of a | |
1187 | * uuid at each position (from 0 to | |
1188 | * nd_region->ndr_mappings - 1), or if we find a | |
1189 | * dimm with two instances of the same uuid. | |
1190 | */ | |
1191 | rc = -EINVAL; | |
1192 | goto err; | |
1193 | } else if (pmem_id) { | |
1194 | /* | |
1195 | * If there is more than one valid uuid set, we | |
1196 | * need userspace to clean this up. | |
1197 | */ | |
1198 | rc = -EBUSY; | |
1199 | goto err; | |
1200 | } | |
1201 | memcpy(select_id, nd_label->uuid, NSLABEL_UUID_LEN); | |
1202 | pmem_id = select_id; | |
1203 | } | |
1204 | ||
1205 | /* | |
1206 | * Fix up each mapping's 'labels' to have the validated pmem label for | |
1207 | * that position at labels[0], and NULL at labels[1]. In the process, | |
1208 | * check that the namespace aligns with interleave-set. We know | |
1209 | * that it does not overlap with any blk namespaces by virtue of | |
1210 | * the dimm being enabled (i.e. nd_label_reserve_dpa() | |
1211 | * succeeded). | |
1212 | */ | |
1213 | rc = select_pmem_id(nd_region, pmem_id); | |
1214 | if (rc) | |
1215 | goto err; | |
1216 | ||
1217 | /* Calculate total size and populate namespace properties from label0 */ | |
1218 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1219 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1220 | struct nd_namespace_label *label0 = nd_mapping->labels[0]; | |
1221 | ||
1222 | size += __le64_to_cpu(label0->rawsize); | |
1223 | if (__le16_to_cpu(label0->position) != 0) | |
1224 | continue; | |
1225 | WARN_ON(nspm->alt_name || nspm->uuid); | |
1226 | nspm->alt_name = kmemdup((void __force *) label0->name, | |
1227 | NSLABEL_NAME_LEN, GFP_KERNEL); | |
1228 | nspm->uuid = kmemdup((void __force *) label0->uuid, | |
1229 | NSLABEL_UUID_LEN, GFP_KERNEL); | |
1230 | } | |
1231 | ||
1232 | if (!nspm->alt_name || !nspm->uuid) { | |
1233 | rc = -ENOMEM; | |
1234 | goto err; | |
1235 | } | |
1236 | ||
1237 | nd_namespace_pmem_set_size(nd_region, nspm, size); | |
1238 | ||
1239 | return 0; | |
1240 | err: | |
1241 | switch (rc) { | |
1242 | case -EINVAL: | |
1243 | dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__); | |
1244 | break; | |
1245 | case -ENODEV: | |
1246 | dev_dbg(&nd_region->dev, "%s: label not found\n", __func__); | |
1247 | break; | |
1248 | default: | |
1249 | dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n", | |
1250 | __func__, rc); | |
1251 | break; | |
1252 | } | |
1253 | return rc; | |
1254 | } | |
1255 | ||
1256 | static struct device **create_namespace_pmem(struct nd_region *nd_region) | |
1257 | { | |
1258 | struct nd_namespace_pmem *nspm; | |
1259 | struct device *dev, **devs; | |
1260 | struct resource *res; | |
1261 | int rc; | |
1262 | ||
1263 | nspm = kzalloc(sizeof(*nspm), GFP_KERNEL); | |
1264 | if (!nspm) | |
1265 | return NULL; | |
1266 | ||
1267 | dev = &nspm->nsio.dev; | |
1268 | dev->type = &namespace_pmem_device_type; | |
1269 | dev->parent = &nd_region->dev; | |
1270 | res = &nspm->nsio.res; | |
1271 | res->name = dev_name(&nd_region->dev); | |
1272 | res->flags = IORESOURCE_MEM; | |
1273 | rc = find_pmem_label_set(nd_region, nspm); | |
1274 | if (rc == -ENODEV) { | |
1275 | int i; | |
1276 | ||
1277 | /* Pass, try to permit namespace creation... */ | |
1278 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1279 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1280 | ||
1281 | kfree(nd_mapping->labels); | |
1282 | nd_mapping->labels = NULL; | |
1283 | } | |
1284 | ||
1285 | /* Publish a zero-sized namespace for userspace to configure. */ | |
1286 | nd_namespace_pmem_set_size(nd_region, nspm, 0); | |
1287 | ||
1288 | rc = 0; | |
1289 | } else if (rc) | |
1290 | goto err; | |
1291 | ||
1292 | devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL); | |
1293 | if (!devs) | |
1294 | goto err; | |
1295 | ||
1296 | devs[0] = dev; | |
1297 | return devs; | |
1298 | ||
1299 | err: | |
1300 | namespace_pmem_release(&nspm->nsio.dev); | |
1301 | return NULL; | |
1302 | } | |
1303 | ||
1b40e09a DW |
1304 | struct resource *nsblk_add_resource(struct nd_region *nd_region, |
1305 | struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk, | |
1306 | resource_size_t start) | |
1307 | { | |
1308 | struct nd_label_id label_id; | |
1309 | struct resource *res; | |
1310 | ||
1311 | nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL); | |
1312 | res = krealloc(nsblk->res, | |
1313 | sizeof(void *) * (nsblk->num_resources + 1), | |
1314 | GFP_KERNEL); | |
1315 | if (!res) | |
1316 | return NULL; | |
1317 | nsblk->res = (struct resource **) res; | |
1318 | for_each_dpa_resource(ndd, res) | |
1319 | if (strcmp(res->name, label_id.id) == 0 | |
1320 | && res->start == start) { | |
1321 | nsblk->res[nsblk->num_resources++] = res; | |
1322 | return res; | |
1323 | } | |
1324 | return NULL; | |
1325 | } | |
1326 | ||
1327 | static struct device *nd_namespace_blk_create(struct nd_region *nd_region) | |
1328 | { | |
1329 | struct nd_namespace_blk *nsblk; | |
1330 | struct device *dev; | |
1331 | ||
1332 | if (!is_nd_blk(&nd_region->dev)) | |
1333 | return NULL; | |
1334 | ||
1335 | nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); | |
1336 | if (!nsblk) | |
1337 | return NULL; | |
1338 | ||
1339 | dev = &nsblk->dev; | |
1340 | dev->type = &namespace_blk_device_type; | |
1341 | nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL); | |
1342 | if (nsblk->id < 0) { | |
1343 | kfree(nsblk); | |
1344 | return NULL; | |
1345 | } | |
1346 | dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id); | |
1347 | dev->parent = &nd_region->dev; | |
1348 | dev->groups = nd_namespace_attribute_groups; | |
1349 | ||
1350 | return &nsblk->dev; | |
1351 | } | |
1352 | ||
1353 | void nd_region_create_blk_seed(struct nd_region *nd_region) | |
1354 | { | |
1355 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); | |
1356 | nd_region->ns_seed = nd_namespace_blk_create(nd_region); | |
1357 | /* | |
1358 | * Seed creation failures are not fatal, provisioning is simply | |
1359 | * disabled until memory becomes available | |
1360 | */ | |
1361 | if (!nd_region->ns_seed) | |
1362 | dev_err(&nd_region->dev, "failed to create blk namespace\n"); | |
1363 | else | |
1364 | nd_device_register(nd_region->ns_seed); | |
1365 | } | |
1366 | ||
1367 | static struct device **create_namespace_blk(struct nd_region *nd_region) | |
1368 | { | |
1369 | struct nd_mapping *nd_mapping = &nd_region->mapping[0]; | |
1370 | struct nd_namespace_label *nd_label; | |
1371 | struct device *dev, **devs = NULL; | |
1372 | struct nd_namespace_blk *nsblk; | |
1373 | struct nvdimm_drvdata *ndd; | |
1374 | int i, l, count = 0; | |
1375 | struct resource *res; | |
1376 | ||
1377 | if (nd_region->ndr_mappings == 0) | |
1378 | return NULL; | |
1379 | ||
1380 | ndd = to_ndd(nd_mapping); | |
1381 | for_each_label(l, nd_label, nd_mapping->labels) { | |
1382 | u32 flags = __le32_to_cpu(nd_label->flags); | |
1383 | char *name[NSLABEL_NAME_LEN]; | |
1384 | struct device **__devs; | |
1385 | ||
1386 | if (flags & NSLABEL_FLAG_LOCAL) | |
1387 | /* pass */; | |
1388 | else | |
1389 | continue; | |
1390 | ||
1391 | for (i = 0; i < count; i++) { | |
1392 | nsblk = to_nd_namespace_blk(devs[i]); | |
1393 | if (memcmp(nsblk->uuid, nd_label->uuid, | |
1394 | NSLABEL_UUID_LEN) == 0) { | |
1395 | res = nsblk_add_resource(nd_region, ndd, nsblk, | |
1396 | __le64_to_cpu(nd_label->dpa)); | |
1397 | if (!res) | |
1398 | goto err; | |
1399 | nd_dbg_dpa(nd_region, ndd, res, "%s assign\n", | |
1400 | dev_name(&nsblk->dev)); | |
1401 | break; | |
1402 | } | |
1403 | } | |
1404 | if (i < count) | |
1405 | continue; | |
1406 | __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL); | |
1407 | if (!__devs) | |
1408 | goto err; | |
1409 | memcpy(__devs, devs, sizeof(dev) * count); | |
1410 | kfree(devs); | |
1411 | devs = __devs; | |
1412 | ||
1413 | nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); | |
1414 | if (!nsblk) | |
1415 | goto err; | |
1416 | dev = &nsblk->dev; | |
1417 | dev->type = &namespace_blk_device_type; | |
1418 | dev->parent = &nd_region->dev; | |
1419 | dev_set_name(dev, "namespace%d.%d", nd_region->id, count); | |
1420 | devs[count++] = dev; | |
1421 | nsblk->id = -1; | |
1422 | nsblk->lbasize = __le64_to_cpu(nd_label->lbasize); | |
1423 | nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN, | |
1424 | GFP_KERNEL); | |
1425 | if (!nsblk->uuid) | |
1426 | goto err; | |
1427 | memcpy(name, nd_label->name, NSLABEL_NAME_LEN); | |
1428 | if (name[0]) | |
1429 | nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN, | |
1430 | GFP_KERNEL); | |
1431 | res = nsblk_add_resource(nd_region, ndd, nsblk, | |
1432 | __le64_to_cpu(nd_label->dpa)); | |
1433 | if (!res) | |
1434 | goto err; | |
1435 | nd_dbg_dpa(nd_region, ndd, res, "%s assign\n", | |
1436 | dev_name(&nsblk->dev)); | |
1437 | } | |
1438 | ||
1439 | dev_dbg(&nd_region->dev, "%s: discovered %d blk namespace%s\n", | |
1440 | __func__, count, count == 1 ? "" : "s"); | |
1441 | ||
1442 | if (count == 0) { | |
1443 | /* Publish a zero-sized namespace for userspace to configure. */ | |
1444 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1445 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1446 | ||
1447 | kfree(nd_mapping->labels); | |
1448 | nd_mapping->labels = NULL; | |
1449 | } | |
1450 | ||
1451 | devs = kcalloc(2, sizeof(dev), GFP_KERNEL); | |
1452 | if (!devs) | |
1453 | goto err; | |
1454 | nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); | |
1455 | if (!nsblk) | |
1456 | goto err; | |
1457 | dev = &nsblk->dev; | |
1458 | dev->type = &namespace_blk_device_type; | |
1459 | dev->parent = &nd_region->dev; | |
1460 | devs[count++] = dev; | |
1461 | } | |
1462 | ||
1463 | return devs; | |
1464 | ||
1465 | err: | |
1466 | for (i = 0; i < count; i++) { | |
1467 | nsblk = to_nd_namespace_blk(devs[i]); | |
1468 | namespace_blk_release(&nsblk->dev); | |
1469 | } | |
1470 | kfree(devs); | |
1471 | return NULL; | |
1472 | } | |
1473 | ||
bf9bccc1 DW |
1474 | static int init_active_labels(struct nd_region *nd_region) |
1475 | { | |
1476 | int i; | |
1477 | ||
1478 | for (i = 0; i < nd_region->ndr_mappings; i++) { | |
1479 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; | |
1480 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); | |
1481 | struct nvdimm *nvdimm = nd_mapping->nvdimm; | |
1482 | int count, j; | |
1483 | ||
1484 | /* | |
1485 | * If the dimm is disabled then prevent the region from | |
1486 | * being activated if it aliases DPA. | |
1487 | */ | |
1488 | if (!ndd) { | |
1489 | if ((nvdimm->flags & NDD_ALIASING) == 0) | |
1490 | return 0; | |
1491 | dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n", | |
1492 | dev_name(&nd_mapping->nvdimm->dev)); | |
1493 | return -ENXIO; | |
1494 | } | |
1495 | nd_mapping->ndd = ndd; | |
1496 | atomic_inc(&nvdimm->busy); | |
1497 | get_ndd(ndd); | |
1498 | ||
1499 | count = nd_label_active_count(ndd); | |
1500 | dev_dbg(ndd->dev, "%s: %d\n", __func__, count); | |
1501 | if (!count) | |
1502 | continue; | |
1503 | nd_mapping->labels = kcalloc(count + 1, sizeof(void *), | |
1504 | GFP_KERNEL); | |
1505 | if (!nd_mapping->labels) | |
1506 | return -ENOMEM; | |
1507 | for (j = 0; j < count; j++) { | |
1508 | struct nd_namespace_label *label; | |
1509 | ||
1510 | label = nd_label_active(ndd, j); | |
1511 | nd_mapping->labels[j] = label; | |
1512 | } | |
1513 | } | |
1514 | ||
1515 | return 0; | |
1516 | } | |
1517 | ||
3d88002e DW |
1518 | int nd_region_register_namespaces(struct nd_region *nd_region, int *err) |
1519 | { | |
1520 | struct device **devs = NULL; | |
bf9bccc1 | 1521 | int i, rc = 0, type; |
3d88002e DW |
1522 | |
1523 | *err = 0; | |
bf9bccc1 DW |
1524 | nvdimm_bus_lock(&nd_region->dev); |
1525 | rc = init_active_labels(nd_region); | |
1526 | if (rc) { | |
1527 | nvdimm_bus_unlock(&nd_region->dev); | |
1528 | return rc; | |
1529 | } | |
1530 | ||
1531 | type = nd_region_to_nstype(nd_region); | |
1532 | switch (type) { | |
3d88002e DW |
1533 | case ND_DEVICE_NAMESPACE_IO: |
1534 | devs = create_namespace_io(nd_region); | |
1535 | break; | |
bf9bccc1 DW |
1536 | case ND_DEVICE_NAMESPACE_PMEM: |
1537 | devs = create_namespace_pmem(nd_region); | |
1538 | break; | |
1b40e09a DW |
1539 | case ND_DEVICE_NAMESPACE_BLK: |
1540 | devs = create_namespace_blk(nd_region); | |
1541 | break; | |
3d88002e DW |
1542 | default: |
1543 | break; | |
1544 | } | |
bf9bccc1 | 1545 | nvdimm_bus_unlock(&nd_region->dev); |
3d88002e DW |
1546 | |
1547 | if (!devs) | |
1548 | return -ENODEV; | |
1549 | ||
1550 | for (i = 0; devs[i]; i++) { | |
1551 | struct device *dev = devs[i]; | |
1b40e09a | 1552 | int id; |
3d88002e | 1553 | |
1b40e09a DW |
1554 | if (type == ND_DEVICE_NAMESPACE_BLK) { |
1555 | struct nd_namespace_blk *nsblk; | |
1556 | ||
1557 | nsblk = to_nd_namespace_blk(dev); | |
1558 | id = ida_simple_get(&nd_region->ns_ida, 0, 0, | |
1559 | GFP_KERNEL); | |
1560 | nsblk->id = id; | |
1561 | } else | |
1562 | id = i; | |
1563 | ||
1564 | if (id < 0) | |
1565 | break; | |
1566 | dev_set_name(dev, "namespace%d.%d", nd_region->id, id); | |
3d88002e DW |
1567 | dev->groups = nd_namespace_attribute_groups; |
1568 | nd_device_register(dev); | |
1569 | } | |
1b40e09a DW |
1570 | if (i) |
1571 | nd_region->ns_seed = devs[0]; | |
1572 | ||
1573 | if (devs[i]) { | |
1574 | int j; | |
1575 | ||
1576 | for (j = i; devs[j]; j++) { | |
1577 | struct device *dev = devs[j]; | |
1578 | ||
1579 | device_initialize(dev); | |
1580 | put_device(dev); | |
1581 | } | |
1582 | *err = j - i; | |
1583 | /* | |
1584 | * All of the namespaces we tried to register failed, so | |
1585 | * fail region activation. | |
1586 | */ | |
1587 | if (*err == 0) | |
1588 | rc = -ENODEV; | |
1589 | } | |
3d88002e DW |
1590 | kfree(devs); |
1591 | ||
1b40e09a DW |
1592 | if (rc == -ENODEV) |
1593 | return rc; | |
1594 | ||
3d88002e DW |
1595 | return i; |
1596 | } |