]>
Commit | Line | Data |
---|---|---|
5212e11f VV |
1 | /* |
2 | * Block Translation Table | |
3 | * Copyright (c) 2014-2015, Intel Corporation. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | */ | |
14 | #include <linux/highmem.h> | |
15 | #include <linux/debugfs.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/device.h> | |
19 | #include <linux/mutex.h> | |
20 | #include <linux/hdreg.h> | |
21 | #include <linux/genhd.h> | |
22 | #include <linux/sizes.h> | |
23 | #include <linux/ndctl.h> | |
24 | #include <linux/fs.h> | |
25 | #include <linux/nd.h> | |
26 | #include "btt.h" | |
27 | #include "nd.h" | |
28 | ||
29 | enum log_ent_request { | |
30 | LOG_NEW_ENT = 0, | |
31 | LOG_OLD_ENT | |
32 | }; | |
33 | ||
5212e11f | 34 | static int arena_read_bytes(struct arena_info *arena, resource_size_t offset, |
3ae3d67b | 35 | void *buf, size_t n, unsigned long flags) |
5212e11f VV |
36 | { |
37 | struct nd_btt *nd_btt = arena->nd_btt; | |
38 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
39 | ||
14e49454 VV |
40 | /* arena offsets may be shifted from the base of the device */ |
41 | offset += arena->nd_btt->initial_offset; | |
3ae3d67b | 42 | return nvdimm_read_bytes(ndns, offset, buf, n, flags); |
5212e11f VV |
43 | } |
44 | ||
45 | static int arena_write_bytes(struct arena_info *arena, resource_size_t offset, | |
3ae3d67b | 46 | void *buf, size_t n, unsigned long flags) |
5212e11f VV |
47 | { |
48 | struct nd_btt *nd_btt = arena->nd_btt; | |
49 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
50 | ||
14e49454 VV |
51 | /* arena offsets may be shifted from the base of the device */ |
52 | offset += arena->nd_btt->initial_offset; | |
3ae3d67b | 53 | return nvdimm_write_bytes(ndns, offset, buf, n, flags); |
5212e11f VV |
54 | } |
55 | ||
56 | static int btt_info_write(struct arena_info *arena, struct btt_sb *super) | |
57 | { | |
58 | int ret; | |
59 | ||
b177fe85 VV |
60 | /* |
61 | * infooff and info2off should always be at least 512B aligned. | |
62 | * We rely on that to make sure rw_bytes does error clearing | |
63 | * correctly, so make sure that is the case. | |
64 | */ | |
65 | WARN_ON_ONCE(!IS_ALIGNED(arena->infooff, 512)); | |
66 | WARN_ON_ONCE(!IS_ALIGNED(arena->info2off, 512)); | |
67 | ||
5212e11f | 68 | ret = arena_write_bytes(arena, arena->info2off, super, |
3ae3d67b | 69 | sizeof(struct btt_sb), 0); |
5212e11f VV |
70 | if (ret) |
71 | return ret; | |
72 | ||
73 | return arena_write_bytes(arena, arena->infooff, super, | |
3ae3d67b | 74 | sizeof(struct btt_sb), 0); |
5212e11f VV |
75 | } |
76 | ||
77 | static int btt_info_read(struct arena_info *arena, struct btt_sb *super) | |
78 | { | |
79 | WARN_ON(!super); | |
80 | return arena_read_bytes(arena, arena->infooff, super, | |
3ae3d67b | 81 | sizeof(struct btt_sb), 0); |
5212e11f VV |
82 | } |
83 | ||
84 | /* | |
85 | * 'raw' version of btt_map write | |
86 | * Assumptions: | |
87 | * mapping is in little-endian | |
88 | * mapping contains 'E' and 'Z' flags as desired | |
89 | */ | |
3ae3d67b VV |
90 | static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping, |
91 | unsigned long flags) | |
5212e11f VV |
92 | { |
93 | u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE); | |
94 | ||
95 | WARN_ON(lba >= arena->external_nlba); | |
3ae3d67b | 96 | return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags); |
5212e11f VV |
97 | } |
98 | ||
99 | static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping, | |
3ae3d67b | 100 | u32 z_flag, u32 e_flag, unsigned long rwb_flags) |
5212e11f VV |
101 | { |
102 | u32 ze; | |
103 | __le32 mapping_le; | |
104 | ||
105 | /* | |
106 | * This 'mapping' is supposed to be just the LBA mapping, without | |
107 | * any flags set, so strip the flag bits. | |
108 | */ | |
0595d539 | 109 | mapping = ent_lba(mapping); |
5212e11f VV |
110 | |
111 | ze = (z_flag << 1) + e_flag; | |
112 | switch (ze) { | |
113 | case 0: | |
114 | /* | |
115 | * We want to set neither of the Z or E flags, and | |
116 | * in the actual layout, this means setting the bit | |
117 | * positions of both to '1' to indicate a 'normal' | |
118 | * map entry | |
119 | */ | |
120 | mapping |= MAP_ENT_NORMAL; | |
121 | break; | |
122 | case 1: | |
123 | mapping |= (1 << MAP_ERR_SHIFT); | |
124 | break; | |
125 | case 2: | |
126 | mapping |= (1 << MAP_TRIM_SHIFT); | |
127 | break; | |
128 | default: | |
129 | /* | |
130 | * The case where Z and E are both sent in as '1' could be | |
131 | * construed as a valid 'normal' case, but we decide not to, | |
132 | * to avoid confusion | |
133 | */ | |
134 | WARN_ONCE(1, "Invalid use of Z and E flags\n"); | |
135 | return -EIO; | |
136 | } | |
137 | ||
138 | mapping_le = cpu_to_le32(mapping); | |
3ae3d67b | 139 | return __btt_map_write(arena, lba, mapping_le, rwb_flags); |
5212e11f VV |
140 | } |
141 | ||
142 | static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping, | |
3ae3d67b | 143 | int *trim, int *error, unsigned long rwb_flags) |
5212e11f VV |
144 | { |
145 | int ret; | |
146 | __le32 in; | |
147 | u32 raw_mapping, postmap, ze, z_flag, e_flag; | |
148 | u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE); | |
149 | ||
150 | WARN_ON(lba >= arena->external_nlba); | |
151 | ||
3ae3d67b | 152 | ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags); |
5212e11f VV |
153 | if (ret) |
154 | return ret; | |
155 | ||
156 | raw_mapping = le32_to_cpu(in); | |
157 | ||
0595d539 VV |
158 | z_flag = ent_z_flag(raw_mapping); |
159 | e_flag = ent_e_flag(raw_mapping); | |
5212e11f | 160 | ze = (z_flag << 1) + e_flag; |
0595d539 | 161 | postmap = ent_lba(raw_mapping); |
5212e11f VV |
162 | |
163 | /* Reuse the {z,e}_flag variables for *trim and *error */ | |
164 | z_flag = 0; | |
165 | e_flag = 0; | |
166 | ||
167 | switch (ze) { | |
168 | case 0: | |
169 | /* Initial state. Return postmap = premap */ | |
170 | *mapping = lba; | |
171 | break; | |
172 | case 1: | |
173 | *mapping = postmap; | |
174 | e_flag = 1; | |
175 | break; | |
176 | case 2: | |
177 | *mapping = postmap; | |
178 | z_flag = 1; | |
179 | break; | |
180 | case 3: | |
181 | *mapping = postmap; | |
182 | break; | |
183 | default: | |
184 | return -EIO; | |
185 | } | |
186 | ||
187 | if (trim) | |
188 | *trim = z_flag; | |
189 | if (error) | |
190 | *error = e_flag; | |
191 | ||
192 | return ret; | |
193 | } | |
194 | ||
195 | static int btt_log_read_pair(struct arena_info *arena, u32 lane, | |
196 | struct log_entry *ent) | |
197 | { | |
198 | WARN_ON(!ent); | |
199 | return arena_read_bytes(arena, | |
200 | arena->logoff + (2 * lane * LOG_ENT_SIZE), ent, | |
3ae3d67b | 201 | 2 * LOG_ENT_SIZE, 0); |
5212e11f VV |
202 | } |
203 | ||
204 | static struct dentry *debugfs_root; | |
205 | ||
206 | static void arena_debugfs_init(struct arena_info *a, struct dentry *parent, | |
207 | int idx) | |
208 | { | |
209 | char dirname[32]; | |
210 | struct dentry *d; | |
211 | ||
212 | /* If for some reason, parent bttN was not created, exit */ | |
213 | if (!parent) | |
214 | return; | |
215 | ||
216 | snprintf(dirname, 32, "arena%d", idx); | |
217 | d = debugfs_create_dir(dirname, parent); | |
218 | if (IS_ERR_OR_NULL(d)) | |
219 | return; | |
220 | a->debugfs_dir = d; | |
221 | ||
222 | debugfs_create_x64("size", S_IRUGO, d, &a->size); | |
223 | debugfs_create_x64("external_lba_start", S_IRUGO, d, | |
224 | &a->external_lba_start); | |
225 | debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba); | |
226 | debugfs_create_u32("internal_lbasize", S_IRUGO, d, | |
227 | &a->internal_lbasize); | |
228 | debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba); | |
229 | debugfs_create_u32("external_lbasize", S_IRUGO, d, | |
230 | &a->external_lbasize); | |
231 | debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree); | |
232 | debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major); | |
233 | debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor); | |
234 | debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff); | |
235 | debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff); | |
236 | debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff); | |
237 | debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff); | |
238 | debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff); | |
239 | debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off); | |
240 | debugfs_create_x32("flags", S_IRUGO, d, &a->flags); | |
241 | } | |
242 | ||
243 | static void btt_debugfs_init(struct btt *btt) | |
244 | { | |
245 | int i = 0; | |
246 | struct arena_info *arena; | |
247 | ||
248 | btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev), | |
249 | debugfs_root); | |
250 | if (IS_ERR_OR_NULL(btt->debugfs_dir)) | |
251 | return; | |
252 | ||
253 | list_for_each_entry(arena, &btt->arena_list, list) { | |
254 | arena_debugfs_init(arena, btt->debugfs_dir, i); | |
255 | i++; | |
256 | } | |
257 | } | |
258 | ||
259 | /* | |
260 | * This function accepts two log entries, and uses the | |
261 | * sequence number to find the 'older' entry. | |
262 | * It also updates the sequence number in this old entry to | |
263 | * make it the 'new' one if the mark_flag is set. | |
264 | * Finally, it returns which of the entries was the older one. | |
265 | * | |
266 | * TODO The logic feels a bit kludge-y. make it better.. | |
267 | */ | |
268 | static int btt_log_get_old(struct log_entry *ent) | |
269 | { | |
270 | int old; | |
271 | ||
272 | /* | |
273 | * the first ever time this is seen, the entry goes into [0] | |
274 | * the next time, the following logic works out to put this | |
275 | * (next) entry into [1] | |
276 | */ | |
277 | if (ent[0].seq == 0) { | |
278 | ent[0].seq = cpu_to_le32(1); | |
279 | return 0; | |
280 | } | |
281 | ||
282 | if (ent[0].seq == ent[1].seq) | |
283 | return -EINVAL; | |
284 | if (le32_to_cpu(ent[0].seq) + le32_to_cpu(ent[1].seq) > 5) | |
285 | return -EINVAL; | |
286 | ||
287 | if (le32_to_cpu(ent[0].seq) < le32_to_cpu(ent[1].seq)) { | |
288 | if (le32_to_cpu(ent[1].seq) - le32_to_cpu(ent[0].seq) == 1) | |
289 | old = 0; | |
290 | else | |
291 | old = 1; | |
292 | } else { | |
293 | if (le32_to_cpu(ent[0].seq) - le32_to_cpu(ent[1].seq) == 1) | |
294 | old = 1; | |
295 | else | |
296 | old = 0; | |
297 | } | |
298 | ||
299 | return old; | |
300 | } | |
301 | ||
302 | static struct device *to_dev(struct arena_info *arena) | |
303 | { | |
304 | return &arena->nd_btt->dev; | |
305 | } | |
306 | ||
307 | /* | |
308 | * This function copies the desired (old/new) log entry into ent if | |
309 | * it is not NULL. It returns the sub-slot number (0 or 1) | |
310 | * where the desired log entry was found. Negative return values | |
311 | * indicate errors. | |
312 | */ | |
313 | static int btt_log_read(struct arena_info *arena, u32 lane, | |
314 | struct log_entry *ent, int old_flag) | |
315 | { | |
316 | int ret; | |
317 | int old_ent, ret_ent; | |
318 | struct log_entry log[2]; | |
319 | ||
320 | ret = btt_log_read_pair(arena, lane, log); | |
321 | if (ret) | |
322 | return -EIO; | |
323 | ||
324 | old_ent = btt_log_get_old(log); | |
325 | if (old_ent < 0 || old_ent > 1) { | |
e6be2dcb | 326 | dev_err(to_dev(arena), |
5212e11f VV |
327 | "log corruption (%d): lane %d seq [%d, %d]\n", |
328 | old_ent, lane, log[0].seq, log[1].seq); | |
329 | /* TODO set error state? */ | |
330 | return -EIO; | |
331 | } | |
332 | ||
333 | ret_ent = (old_flag ? old_ent : (1 - old_ent)); | |
334 | ||
335 | if (ent != NULL) | |
336 | memcpy(ent, &log[ret_ent], LOG_ENT_SIZE); | |
337 | ||
338 | return ret_ent; | |
339 | } | |
340 | ||
341 | /* | |
342 | * This function commits a log entry to media | |
343 | * It does _not_ prepare the freelist entry for the next write | |
344 | * btt_flog_write is the wrapper for updating the freelist elements | |
345 | */ | |
346 | static int __btt_log_write(struct arena_info *arena, u32 lane, | |
3ae3d67b | 347 | u32 sub, struct log_entry *ent, unsigned long flags) |
5212e11f VV |
348 | { |
349 | int ret; | |
350 | /* | |
351 | * Ignore the padding in log_entry for calculating log_half. | |
352 | * The entry is 'committed' when we write the sequence number, | |
353 | * and we want to ensure that that is the last thing written. | |
354 | * We don't bother writing the padding as that would be extra | |
355 | * media wear and write amplification | |
356 | */ | |
357 | unsigned int log_half = (LOG_ENT_SIZE - 2 * sizeof(u64)) / 2; | |
358 | u64 ns_off = arena->logoff + (((2 * lane) + sub) * LOG_ENT_SIZE); | |
359 | void *src = ent; | |
360 | ||
361 | /* split the 16B write into atomic, durable halves */ | |
3ae3d67b | 362 | ret = arena_write_bytes(arena, ns_off, src, log_half, flags); |
5212e11f VV |
363 | if (ret) |
364 | return ret; | |
365 | ||
366 | ns_off += log_half; | |
367 | src += log_half; | |
3ae3d67b | 368 | return arena_write_bytes(arena, ns_off, src, log_half, flags); |
5212e11f VV |
369 | } |
370 | ||
371 | static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub, | |
372 | struct log_entry *ent) | |
373 | { | |
374 | int ret; | |
375 | ||
3ae3d67b | 376 | ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC); |
5212e11f VV |
377 | if (ret) |
378 | return ret; | |
379 | ||
380 | /* prepare the next free entry */ | |
381 | arena->freelist[lane].sub = 1 - arena->freelist[lane].sub; | |
382 | if (++(arena->freelist[lane].seq) == 4) | |
383 | arena->freelist[lane].seq = 1; | |
384 | arena->freelist[lane].block = le32_to_cpu(ent->old_map); | |
385 | ||
386 | return ret; | |
387 | } | |
388 | ||
389 | /* | |
390 | * This function initializes the BTT map to the initial state, which is | |
391 | * all-zeroes, and indicates an identity mapping | |
392 | */ | |
393 | static int btt_map_init(struct arena_info *arena) | |
394 | { | |
395 | int ret = -EINVAL; | |
396 | void *zerobuf; | |
397 | size_t offset = 0; | |
398 | size_t chunk_size = SZ_2M; | |
399 | size_t mapsize = arena->logoff - arena->mapoff; | |
400 | ||
401 | zerobuf = kzalloc(chunk_size, GFP_KERNEL); | |
402 | if (!zerobuf) | |
403 | return -ENOMEM; | |
404 | ||
b177fe85 VV |
405 | /* |
406 | * mapoff should always be at least 512B aligned. We rely on that to | |
407 | * make sure rw_bytes does error clearing correctly, so make sure that | |
408 | * is the case. | |
409 | */ | |
410 | WARN_ON_ONCE(!IS_ALIGNED(arena->mapoff, 512)); | |
411 | ||
5212e11f VV |
412 | while (mapsize) { |
413 | size_t size = min(mapsize, chunk_size); | |
414 | ||
b177fe85 | 415 | WARN_ON_ONCE(size < 512); |
5212e11f | 416 | ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf, |
3ae3d67b | 417 | size, 0); |
5212e11f VV |
418 | if (ret) |
419 | goto free; | |
420 | ||
421 | offset += size; | |
422 | mapsize -= size; | |
423 | cond_resched(); | |
424 | } | |
425 | ||
426 | free: | |
427 | kfree(zerobuf); | |
428 | return ret; | |
429 | } | |
430 | ||
431 | /* | |
432 | * This function initializes the BTT log with 'fake' entries pointing | |
433 | * to the initial reserved set of blocks as being free | |
434 | */ | |
435 | static int btt_log_init(struct arena_info *arena) | |
436 | { | |
b177fe85 VV |
437 | size_t logsize = arena->info2off - arena->logoff; |
438 | size_t chunk_size = SZ_4K, offset = 0; | |
439 | struct log_entry log; | |
440 | void *zerobuf; | |
5212e11f VV |
441 | int ret; |
442 | u32 i; | |
5212e11f | 443 | |
b177fe85 VV |
444 | zerobuf = kzalloc(chunk_size, GFP_KERNEL); |
445 | if (!zerobuf) | |
446 | return -ENOMEM; | |
447 | /* | |
448 | * logoff should always be at least 512B aligned. We rely on that to | |
449 | * make sure rw_bytes does error clearing correctly, so make sure that | |
450 | * is the case. | |
451 | */ | |
452 | WARN_ON_ONCE(!IS_ALIGNED(arena->logoff, 512)); | |
453 | ||
454 | while (logsize) { | |
455 | size_t size = min(logsize, chunk_size); | |
456 | ||
457 | WARN_ON_ONCE(size < 512); | |
458 | ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf, | |
459 | size, 0); | |
460 | if (ret) | |
461 | goto free; | |
462 | ||
463 | offset += size; | |
464 | logsize -= size; | |
465 | cond_resched(); | |
466 | } | |
5212e11f VV |
467 | |
468 | for (i = 0; i < arena->nfree; i++) { | |
469 | log.lba = cpu_to_le32(i); | |
470 | log.old_map = cpu_to_le32(arena->external_nlba + i); | |
471 | log.new_map = cpu_to_le32(arena->external_nlba + i); | |
472 | log.seq = cpu_to_le32(LOG_SEQ_INIT); | |
3ae3d67b | 473 | ret = __btt_log_write(arena, i, 0, &log, 0); |
5212e11f | 474 | if (ret) |
b177fe85 | 475 | goto free; |
5212e11f VV |
476 | } |
477 | ||
b177fe85 VV |
478 | free: |
479 | kfree(zerobuf); | |
480 | return ret; | |
5212e11f VV |
481 | } |
482 | ||
483 | static int btt_freelist_init(struct arena_info *arena) | |
484 | { | |
485 | int old, new, ret; | |
486 | u32 i, map_entry; | |
487 | struct log_entry log_new, log_old; | |
488 | ||
489 | arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry), | |
490 | GFP_KERNEL); | |
491 | if (!arena->freelist) | |
492 | return -ENOMEM; | |
493 | ||
494 | for (i = 0; i < arena->nfree; i++) { | |
495 | old = btt_log_read(arena, i, &log_old, LOG_OLD_ENT); | |
496 | if (old < 0) | |
497 | return old; | |
498 | ||
499 | new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT); | |
500 | if (new < 0) | |
501 | return new; | |
502 | ||
503 | /* sub points to the next one to be overwritten */ | |
504 | arena->freelist[i].sub = 1 - new; | |
505 | arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq)); | |
506 | arena->freelist[i].block = le32_to_cpu(log_new.old_map); | |
507 | ||
508 | /* This implies a newly created or untouched flog entry */ | |
509 | if (log_new.old_map == log_new.new_map) | |
510 | continue; | |
511 | ||
512 | /* Check if map recovery is needed */ | |
513 | ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry, | |
3ae3d67b | 514 | NULL, NULL, 0); |
5212e11f VV |
515 | if (ret) |
516 | return ret; | |
517 | if ((le32_to_cpu(log_new.new_map) != map_entry) && | |
518 | (le32_to_cpu(log_new.old_map) == map_entry)) { | |
519 | /* | |
520 | * Last transaction wrote the flog, but wasn't able | |
521 | * to complete the map write. So fix up the map. | |
522 | */ | |
523 | ret = btt_map_write(arena, le32_to_cpu(log_new.lba), | |
3ae3d67b | 524 | le32_to_cpu(log_new.new_map), 0, 0, 0); |
5212e11f VV |
525 | if (ret) |
526 | return ret; | |
527 | } | |
528 | ||
529 | } | |
530 | ||
531 | return 0; | |
532 | } | |
533 | ||
534 | static int btt_rtt_init(struct arena_info *arena) | |
535 | { | |
536 | arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL); | |
537 | if (arena->rtt == NULL) | |
538 | return -ENOMEM; | |
539 | ||
540 | return 0; | |
541 | } | |
542 | ||
543 | static int btt_maplocks_init(struct arena_info *arena) | |
544 | { | |
545 | u32 i; | |
546 | ||
547 | arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock), | |
548 | GFP_KERNEL); | |
549 | if (!arena->map_locks) | |
550 | return -ENOMEM; | |
551 | ||
552 | for (i = 0; i < arena->nfree; i++) | |
553 | spin_lock_init(&arena->map_locks[i].lock); | |
554 | ||
555 | return 0; | |
556 | } | |
557 | ||
558 | static struct arena_info *alloc_arena(struct btt *btt, size_t size, | |
559 | size_t start, size_t arena_off) | |
560 | { | |
561 | struct arena_info *arena; | |
562 | u64 logsize, mapsize, datasize; | |
563 | u64 available = size; | |
564 | ||
565 | arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL); | |
566 | if (!arena) | |
567 | return NULL; | |
568 | arena->nd_btt = btt->nd_btt; | |
75892004 | 569 | arena->sector_size = btt->sector_size; |
5212e11f VV |
570 | |
571 | if (!size) | |
572 | return arena; | |
573 | ||
574 | arena->size = size; | |
575 | arena->external_lba_start = start; | |
576 | arena->external_lbasize = btt->lbasize; | |
577 | arena->internal_lbasize = roundup(arena->external_lbasize, | |
578 | INT_LBASIZE_ALIGNMENT); | |
579 | arena->nfree = BTT_DEFAULT_NFREE; | |
14e49454 VV |
580 | arena->version_major = btt->nd_btt->version_major; |
581 | arena->version_minor = btt->nd_btt->version_minor; | |
5212e11f VV |
582 | |
583 | if (available % BTT_PG_SIZE) | |
584 | available -= (available % BTT_PG_SIZE); | |
585 | ||
586 | /* Two pages are reserved for the super block and its copy */ | |
587 | available -= 2 * BTT_PG_SIZE; | |
588 | ||
589 | /* The log takes a fixed amount of space based on nfree */ | |
590 | logsize = roundup(2 * arena->nfree * sizeof(struct log_entry), | |
591 | BTT_PG_SIZE); | |
592 | available -= logsize; | |
593 | ||
594 | /* Calculate optimal split between map and data area */ | |
595 | arena->internal_nlba = div_u64(available - BTT_PG_SIZE, | |
596 | arena->internal_lbasize + MAP_ENT_SIZE); | |
597 | arena->external_nlba = arena->internal_nlba - arena->nfree; | |
598 | ||
599 | mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE); | |
600 | datasize = available - mapsize; | |
601 | ||
602 | /* 'Absolute' values, relative to start of storage space */ | |
603 | arena->infooff = arena_off; | |
604 | arena->dataoff = arena->infooff + BTT_PG_SIZE; | |
605 | arena->mapoff = arena->dataoff + datasize; | |
606 | arena->logoff = arena->mapoff + mapsize; | |
607 | arena->info2off = arena->logoff + logsize; | |
608 | return arena; | |
609 | } | |
610 | ||
611 | static void free_arenas(struct btt *btt) | |
612 | { | |
613 | struct arena_info *arena, *next; | |
614 | ||
615 | list_for_each_entry_safe(arena, next, &btt->arena_list, list) { | |
616 | list_del(&arena->list); | |
617 | kfree(arena->rtt); | |
618 | kfree(arena->map_locks); | |
619 | kfree(arena->freelist); | |
620 | debugfs_remove_recursive(arena->debugfs_dir); | |
621 | kfree(arena); | |
622 | } | |
623 | } | |
624 | ||
5212e11f VV |
625 | /* |
626 | * This function reads an existing valid btt superblock and | |
627 | * populates the corresponding arena_info struct | |
628 | */ | |
629 | static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super, | |
630 | u64 arena_off) | |
631 | { | |
632 | arena->internal_nlba = le32_to_cpu(super->internal_nlba); | |
633 | arena->internal_lbasize = le32_to_cpu(super->internal_lbasize); | |
634 | arena->external_nlba = le32_to_cpu(super->external_nlba); | |
635 | arena->external_lbasize = le32_to_cpu(super->external_lbasize); | |
636 | arena->nfree = le32_to_cpu(super->nfree); | |
637 | arena->version_major = le16_to_cpu(super->version_major); | |
638 | arena->version_minor = le16_to_cpu(super->version_minor); | |
639 | ||
640 | arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off + | |
641 | le64_to_cpu(super->nextoff)); | |
642 | arena->infooff = arena_off; | |
643 | arena->dataoff = arena_off + le64_to_cpu(super->dataoff); | |
644 | arena->mapoff = arena_off + le64_to_cpu(super->mapoff); | |
645 | arena->logoff = arena_off + le64_to_cpu(super->logoff); | |
646 | arena->info2off = arena_off + le64_to_cpu(super->info2off); | |
647 | ||
5e329406 DW |
648 | arena->size = (le64_to_cpu(super->nextoff) > 0) |
649 | ? (le64_to_cpu(super->nextoff)) | |
650 | : (arena->info2off - arena->infooff + BTT_PG_SIZE); | |
5212e11f VV |
651 | |
652 | arena->flags = le32_to_cpu(super->flags); | |
653 | } | |
654 | ||
655 | static int discover_arenas(struct btt *btt) | |
656 | { | |
657 | int ret = 0; | |
658 | struct arena_info *arena; | |
659 | struct btt_sb *super; | |
660 | size_t remaining = btt->rawsize; | |
661 | u64 cur_nlba = 0; | |
662 | size_t cur_off = 0; | |
663 | int num_arenas = 0; | |
664 | ||
665 | super = kzalloc(sizeof(*super), GFP_KERNEL); | |
666 | if (!super) | |
667 | return -ENOMEM; | |
668 | ||
669 | while (remaining) { | |
670 | /* Alloc memory for arena */ | |
671 | arena = alloc_arena(btt, 0, 0, 0); | |
672 | if (!arena) { | |
673 | ret = -ENOMEM; | |
674 | goto out_super; | |
675 | } | |
676 | ||
677 | arena->infooff = cur_off; | |
678 | ret = btt_info_read(arena, super); | |
679 | if (ret) | |
680 | goto out; | |
681 | ||
ab45e763 | 682 | if (!nd_btt_arena_is_valid(btt->nd_btt, super)) { |
5212e11f VV |
683 | if (remaining == btt->rawsize) { |
684 | btt->init_state = INIT_NOTFOUND; | |
685 | dev_info(to_dev(arena), "No existing arenas\n"); | |
686 | goto out; | |
687 | } else { | |
e6be2dcb | 688 | dev_err(to_dev(arena), |
5212e11f VV |
689 | "Found corrupted metadata!\n"); |
690 | ret = -ENODEV; | |
691 | goto out; | |
692 | } | |
693 | } | |
694 | ||
695 | arena->external_lba_start = cur_nlba; | |
696 | parse_arena_meta(arena, super, cur_off); | |
697 | ||
698 | ret = btt_freelist_init(arena); | |
699 | if (ret) | |
700 | goto out; | |
701 | ||
702 | ret = btt_rtt_init(arena); | |
703 | if (ret) | |
704 | goto out; | |
705 | ||
706 | ret = btt_maplocks_init(arena); | |
707 | if (ret) | |
708 | goto out; | |
709 | ||
710 | list_add_tail(&arena->list, &btt->arena_list); | |
711 | ||
712 | remaining -= arena->size; | |
713 | cur_off += arena->size; | |
714 | cur_nlba += arena->external_nlba; | |
715 | num_arenas++; | |
716 | ||
717 | if (arena->nextoff == 0) | |
718 | break; | |
719 | } | |
720 | btt->num_arenas = num_arenas; | |
721 | btt->nlba = cur_nlba; | |
722 | btt->init_state = INIT_READY; | |
723 | ||
724 | kfree(super); | |
725 | return ret; | |
726 | ||
727 | out: | |
728 | kfree(arena); | |
729 | free_arenas(btt); | |
730 | out_super: | |
731 | kfree(super); | |
732 | return ret; | |
733 | } | |
734 | ||
735 | static int create_arenas(struct btt *btt) | |
736 | { | |
737 | size_t remaining = btt->rawsize; | |
738 | size_t cur_off = 0; | |
739 | ||
740 | while (remaining) { | |
741 | struct arena_info *arena; | |
742 | size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining); | |
743 | ||
744 | remaining -= arena_size; | |
745 | if (arena_size < ARENA_MIN_SIZE) | |
746 | break; | |
747 | ||
748 | arena = alloc_arena(btt, arena_size, btt->nlba, cur_off); | |
749 | if (!arena) { | |
750 | free_arenas(btt); | |
751 | return -ENOMEM; | |
752 | } | |
753 | btt->nlba += arena->external_nlba; | |
754 | if (remaining >= ARENA_MIN_SIZE) | |
755 | arena->nextoff = arena->size; | |
756 | else | |
757 | arena->nextoff = 0; | |
758 | cur_off += arena_size; | |
759 | list_add_tail(&arena->list, &btt->arena_list); | |
760 | } | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | /* | |
766 | * This function completes arena initialization by writing | |
767 | * all the metadata. | |
768 | * It is only called for an uninitialized arena when a write | |
769 | * to that arena occurs for the first time. | |
770 | */ | |
fbde1414 | 771 | static int btt_arena_write_layout(struct arena_info *arena) |
5212e11f VV |
772 | { |
773 | int ret; | |
e1455744 | 774 | u64 sum; |
5212e11f | 775 | struct btt_sb *super; |
fbde1414 | 776 | struct nd_btt *nd_btt = arena->nd_btt; |
6ec68954 | 777 | const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev); |
5212e11f VV |
778 | |
779 | ret = btt_map_init(arena); | |
780 | if (ret) | |
781 | return ret; | |
782 | ||
783 | ret = btt_log_init(arena); | |
784 | if (ret) | |
785 | return ret; | |
786 | ||
787 | super = kzalloc(sizeof(struct btt_sb), GFP_NOIO); | |
788 | if (!super) | |
789 | return -ENOMEM; | |
790 | ||
791 | strncpy(super->signature, BTT_SIG, BTT_SIG_LEN); | |
fbde1414 | 792 | memcpy(super->uuid, nd_btt->uuid, 16); |
6ec68954 | 793 | memcpy(super->parent_uuid, parent_uuid, 16); |
5212e11f VV |
794 | super->flags = cpu_to_le32(arena->flags); |
795 | super->version_major = cpu_to_le16(arena->version_major); | |
796 | super->version_minor = cpu_to_le16(arena->version_minor); | |
797 | super->external_lbasize = cpu_to_le32(arena->external_lbasize); | |
798 | super->external_nlba = cpu_to_le32(arena->external_nlba); | |
799 | super->internal_lbasize = cpu_to_le32(arena->internal_lbasize); | |
800 | super->internal_nlba = cpu_to_le32(arena->internal_nlba); | |
801 | super->nfree = cpu_to_le32(arena->nfree); | |
802 | super->infosize = cpu_to_le32(sizeof(struct btt_sb)); | |
803 | super->nextoff = cpu_to_le64(arena->nextoff); | |
804 | /* | |
805 | * Subtract arena->infooff (arena start) so numbers are relative | |
806 | * to 'this' arena | |
807 | */ | |
808 | super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff); | |
809 | super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff); | |
810 | super->logoff = cpu_to_le64(arena->logoff - arena->infooff); | |
811 | super->info2off = cpu_to_le64(arena->info2off - arena->infooff); | |
812 | ||
813 | super->flags = 0; | |
e1455744 DW |
814 | sum = nd_sb_checksum((struct nd_gen_sb *) super); |
815 | super->checksum = cpu_to_le64(sum); | |
5212e11f VV |
816 | |
817 | ret = btt_info_write(arena, super); | |
818 | ||
819 | kfree(super); | |
820 | return ret; | |
821 | } | |
822 | ||
823 | /* | |
824 | * This function completes the initialization for the BTT namespace | |
825 | * such that it is ready to accept IOs | |
826 | */ | |
827 | static int btt_meta_init(struct btt *btt) | |
828 | { | |
829 | int ret = 0; | |
830 | struct arena_info *arena; | |
831 | ||
832 | mutex_lock(&btt->init_lock); | |
833 | list_for_each_entry(arena, &btt->arena_list, list) { | |
fbde1414 | 834 | ret = btt_arena_write_layout(arena); |
5212e11f VV |
835 | if (ret) |
836 | goto unlock; | |
837 | ||
838 | ret = btt_freelist_init(arena); | |
839 | if (ret) | |
840 | goto unlock; | |
841 | ||
842 | ret = btt_rtt_init(arena); | |
843 | if (ret) | |
844 | goto unlock; | |
845 | ||
846 | ret = btt_maplocks_init(arena); | |
847 | if (ret) | |
848 | goto unlock; | |
849 | } | |
850 | ||
851 | btt->init_state = INIT_READY; | |
852 | ||
853 | unlock: | |
854 | mutex_unlock(&btt->init_lock); | |
855 | return ret; | |
856 | } | |
857 | ||
41cd8b70 VV |
858 | static u32 btt_meta_size(struct btt *btt) |
859 | { | |
860 | return btt->lbasize - btt->sector_size; | |
861 | } | |
862 | ||
5212e11f VV |
863 | /* |
864 | * This function calculates the arena in which the given LBA lies | |
865 | * by doing a linear walk. This is acceptable since we expect only | |
866 | * a few arenas. If we have backing devices that get much larger, | |
867 | * we can construct a balanced binary tree of arenas at init time | |
868 | * so that this range search becomes faster. | |
869 | */ | |
870 | static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap, | |
871 | struct arena_info **arena) | |
872 | { | |
873 | struct arena_info *arena_list; | |
874 | __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size); | |
875 | ||
876 | list_for_each_entry(arena_list, &btt->arena_list, list) { | |
877 | if (lba < arena_list->external_nlba) { | |
878 | *arena = arena_list; | |
879 | *premap = lba; | |
880 | return 0; | |
881 | } | |
882 | lba -= arena_list->external_nlba; | |
883 | } | |
884 | ||
885 | return -EIO; | |
886 | } | |
887 | ||
888 | /* | |
889 | * The following (lock_map, unlock_map) are mostly just to improve | |
890 | * readability, since they index into an array of locks | |
891 | */ | |
892 | static void lock_map(struct arena_info *arena, u32 premap) | |
893 | __acquires(&arena->map_locks[idx].lock) | |
894 | { | |
895 | u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree; | |
896 | ||
897 | spin_lock(&arena->map_locks[idx].lock); | |
898 | } | |
899 | ||
900 | static void unlock_map(struct arena_info *arena, u32 premap) | |
901 | __releases(&arena->map_locks[idx].lock) | |
902 | { | |
903 | u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree; | |
904 | ||
905 | spin_unlock(&arena->map_locks[idx].lock); | |
906 | } | |
907 | ||
908 | static u64 to_namespace_offset(struct arena_info *arena, u64 lba) | |
909 | { | |
910 | return arena->dataoff + ((u64)lba * arena->internal_lbasize); | |
911 | } | |
912 | ||
913 | static int btt_data_read(struct arena_info *arena, struct page *page, | |
914 | unsigned int off, u32 lba, u32 len) | |
915 | { | |
916 | int ret; | |
917 | u64 nsoff = to_namespace_offset(arena, lba); | |
918 | void *mem = kmap_atomic(page); | |
919 | ||
3ae3d67b | 920 | ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC); |
5212e11f VV |
921 | kunmap_atomic(mem); |
922 | ||
923 | return ret; | |
924 | } | |
925 | ||
926 | static int btt_data_write(struct arena_info *arena, u32 lba, | |
927 | struct page *page, unsigned int off, u32 len) | |
928 | { | |
929 | int ret; | |
930 | u64 nsoff = to_namespace_offset(arena, lba); | |
931 | void *mem = kmap_atomic(page); | |
932 | ||
3ae3d67b | 933 | ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC); |
5212e11f VV |
934 | kunmap_atomic(mem); |
935 | ||
936 | return ret; | |
937 | } | |
938 | ||
939 | static void zero_fill_data(struct page *page, unsigned int off, u32 len) | |
940 | { | |
941 | void *mem = kmap_atomic(page); | |
942 | ||
943 | memset(mem + off, 0, len); | |
944 | kunmap_atomic(mem); | |
945 | } | |
946 | ||
41cd8b70 VV |
947 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
948 | static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip, | |
949 | struct arena_info *arena, u32 postmap, int rw) | |
950 | { | |
951 | unsigned int len = btt_meta_size(btt); | |
952 | u64 meta_nsoff; | |
953 | int ret = 0; | |
954 | ||
955 | if (bip == NULL) | |
956 | return 0; | |
957 | ||
958 | meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size; | |
959 | ||
960 | while (len) { | |
961 | unsigned int cur_len; | |
962 | struct bio_vec bv; | |
963 | void *mem; | |
964 | ||
965 | bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter); | |
966 | /* | |
967 | * The 'bv' obtained from bvec_iter_bvec has its .bv_len and | |
968 | * .bv_offset already adjusted for iter->bi_bvec_done, and we | |
969 | * can use those directly | |
970 | */ | |
971 | ||
972 | cur_len = min(len, bv.bv_len); | |
973 | mem = kmap_atomic(bv.bv_page); | |
974 | if (rw) | |
975 | ret = arena_write_bytes(arena, meta_nsoff, | |
3ae3d67b VV |
976 | mem + bv.bv_offset, cur_len, |
977 | NVDIMM_IO_ATOMIC); | |
41cd8b70 VV |
978 | else |
979 | ret = arena_read_bytes(arena, meta_nsoff, | |
3ae3d67b VV |
980 | mem + bv.bv_offset, cur_len, |
981 | NVDIMM_IO_ATOMIC); | |
41cd8b70 VV |
982 | |
983 | kunmap_atomic(mem); | |
984 | if (ret) | |
985 | return ret; | |
986 | ||
987 | len -= cur_len; | |
988 | meta_nsoff += cur_len; | |
b1fb2c52 DM |
989 | if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len)) |
990 | return -EIO; | |
41cd8b70 VV |
991 | } |
992 | ||
993 | return ret; | |
994 | } | |
995 | ||
996 | #else /* CONFIG_BLK_DEV_INTEGRITY */ | |
997 | static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip, | |
998 | struct arena_info *arena, u32 postmap, int rw) | |
999 | { | |
1000 | return 0; | |
1001 | } | |
1002 | #endif | |
1003 | ||
1004 | static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip, | |
1005 | struct page *page, unsigned int off, sector_t sector, | |
1006 | unsigned int len) | |
5212e11f VV |
1007 | { |
1008 | int ret = 0; | |
1009 | int t_flag, e_flag; | |
1010 | struct arena_info *arena = NULL; | |
1011 | u32 lane = 0, premap, postmap; | |
1012 | ||
1013 | while (len) { | |
1014 | u32 cur_len; | |
1015 | ||
1016 | lane = nd_region_acquire_lane(btt->nd_region); | |
1017 | ||
1018 | ret = lba_to_arena(btt, sector, &premap, &arena); | |
1019 | if (ret) | |
1020 | goto out_lane; | |
1021 | ||
1022 | cur_len = min(btt->sector_size, len); | |
1023 | ||
3ae3d67b VV |
1024 | ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag, |
1025 | NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1026 | if (ret) |
1027 | goto out_lane; | |
1028 | ||
1029 | /* | |
1030 | * We loop to make sure that the post map LBA didn't change | |
1031 | * from under us between writing the RTT and doing the actual | |
1032 | * read. | |
1033 | */ | |
1034 | while (1) { | |
1035 | u32 new_map; | |
1398199d | 1036 | int new_t, new_e; |
5212e11f VV |
1037 | |
1038 | if (t_flag) { | |
1039 | zero_fill_data(page, off, cur_len); | |
1040 | goto out_lane; | |
1041 | } | |
1042 | ||
1043 | if (e_flag) { | |
1044 | ret = -EIO; | |
1045 | goto out_lane; | |
1046 | } | |
1047 | ||
1048 | arena->rtt[lane] = RTT_VALID | postmap; | |
1049 | /* | |
1050 | * Barrier to make sure this write is not reordered | |
1051 | * to do the verification map_read before the RTT store | |
1052 | */ | |
1053 | barrier(); | |
1054 | ||
1398199d VV |
1055 | ret = btt_map_read(arena, premap, &new_map, &new_t, |
1056 | &new_e, NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1057 | if (ret) |
1058 | goto out_rtt; | |
1059 | ||
1398199d VV |
1060 | if ((postmap == new_map) && (t_flag == new_t) && |
1061 | (e_flag == new_e)) | |
5212e11f VV |
1062 | break; |
1063 | ||
1064 | postmap = new_map; | |
1398199d VV |
1065 | t_flag = new_t; |
1066 | e_flag = new_e; | |
5212e11f VV |
1067 | } |
1068 | ||
1069 | ret = btt_data_read(arena, page, off, postmap, cur_len); | |
1070 | if (ret) | |
1071 | goto out_rtt; | |
1072 | ||
41cd8b70 VV |
1073 | if (bip) { |
1074 | ret = btt_rw_integrity(btt, bip, arena, postmap, READ); | |
1075 | if (ret) | |
1076 | goto out_rtt; | |
1077 | } | |
1078 | ||
5212e11f VV |
1079 | arena->rtt[lane] = RTT_INVALID; |
1080 | nd_region_release_lane(btt->nd_region, lane); | |
1081 | ||
1082 | len -= cur_len; | |
1083 | off += cur_len; | |
1084 | sector += btt->sector_size >> SECTOR_SHIFT; | |
1085 | } | |
1086 | ||
1087 | return 0; | |
1088 | ||
1089 | out_rtt: | |
1090 | arena->rtt[lane] = RTT_INVALID; | |
1091 | out_lane: | |
1092 | nd_region_release_lane(btt->nd_region, lane); | |
1093 | return ret; | |
1094 | } | |
1095 | ||
41cd8b70 VV |
1096 | static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip, |
1097 | sector_t sector, struct page *page, unsigned int off, | |
1098 | unsigned int len) | |
5212e11f VV |
1099 | { |
1100 | int ret = 0; | |
1101 | struct arena_info *arena = NULL; | |
1102 | u32 premap = 0, old_postmap, new_postmap, lane = 0, i; | |
1103 | struct log_entry log; | |
1104 | int sub; | |
1105 | ||
1106 | while (len) { | |
1107 | u32 cur_len; | |
1108 | ||
1109 | lane = nd_region_acquire_lane(btt->nd_region); | |
1110 | ||
1111 | ret = lba_to_arena(btt, sector, &premap, &arena); | |
1112 | if (ret) | |
1113 | goto out_lane; | |
1114 | cur_len = min(btt->sector_size, len); | |
1115 | ||
1116 | if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) { | |
1117 | ret = -EIO; | |
1118 | goto out_lane; | |
1119 | } | |
1120 | ||
1121 | new_postmap = arena->freelist[lane].block; | |
1122 | ||
1123 | /* Wait if the new block is being read from */ | |
1124 | for (i = 0; i < arena->nfree; i++) | |
1125 | while (arena->rtt[i] == (RTT_VALID | new_postmap)) | |
1126 | cpu_relax(); | |
1127 | ||
1128 | ||
1129 | if (new_postmap >= arena->internal_nlba) { | |
1130 | ret = -EIO; | |
1131 | goto out_lane; | |
41cd8b70 VV |
1132 | } |
1133 | ||
1134 | ret = btt_data_write(arena, new_postmap, page, off, cur_len); | |
5212e11f VV |
1135 | if (ret) |
1136 | goto out_lane; | |
1137 | ||
41cd8b70 VV |
1138 | if (bip) { |
1139 | ret = btt_rw_integrity(btt, bip, arena, new_postmap, | |
1140 | WRITE); | |
1141 | if (ret) | |
1142 | goto out_lane; | |
1143 | } | |
1144 | ||
5212e11f | 1145 | lock_map(arena, premap); |
3ae3d67b VV |
1146 | ret = btt_map_read(arena, premap, &old_postmap, NULL, NULL, |
1147 | NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1148 | if (ret) |
1149 | goto out_map; | |
1150 | if (old_postmap >= arena->internal_nlba) { | |
1151 | ret = -EIO; | |
1152 | goto out_map; | |
1153 | } | |
1154 | ||
1155 | log.lba = cpu_to_le32(premap); | |
1156 | log.old_map = cpu_to_le32(old_postmap); | |
1157 | log.new_map = cpu_to_le32(new_postmap); | |
1158 | log.seq = cpu_to_le32(arena->freelist[lane].seq); | |
1159 | sub = arena->freelist[lane].sub; | |
1160 | ret = btt_flog_write(arena, lane, sub, &log); | |
1161 | if (ret) | |
1162 | goto out_map; | |
1163 | ||
1db1f3ce VV |
1164 | ret = btt_map_write(arena, premap, new_postmap, 0, 0, |
1165 | NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1166 | if (ret) |
1167 | goto out_map; | |
1168 | ||
1169 | unlock_map(arena, premap); | |
1170 | nd_region_release_lane(btt->nd_region, lane); | |
1171 | ||
1172 | len -= cur_len; | |
1173 | off += cur_len; | |
1174 | sector += btt->sector_size >> SECTOR_SHIFT; | |
1175 | } | |
1176 | ||
1177 | return 0; | |
1178 | ||
1179 | out_map: | |
1180 | unlock_map(arena, premap); | |
1181 | out_lane: | |
1182 | nd_region_release_lane(btt->nd_region, lane); | |
1183 | return ret; | |
1184 | } | |
1185 | ||
41cd8b70 VV |
1186 | static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip, |
1187 | struct page *page, unsigned int len, unsigned int off, | |
c11f0c0b | 1188 | bool is_write, sector_t sector) |
5212e11f VV |
1189 | { |
1190 | int ret; | |
1191 | ||
c11f0c0b | 1192 | if (!is_write) { |
41cd8b70 | 1193 | ret = btt_read_pg(btt, bip, page, off, sector, len); |
5212e11f VV |
1194 | flush_dcache_page(page); |
1195 | } else { | |
1196 | flush_dcache_page(page); | |
41cd8b70 | 1197 | ret = btt_write_pg(btt, bip, sector, page, off, len); |
5212e11f VV |
1198 | } |
1199 | ||
1200 | return ret; | |
1201 | } | |
1202 | ||
dece1635 | 1203 | static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio) |
5212e11f | 1204 | { |
41cd8b70 | 1205 | struct bio_integrity_payload *bip = bio_integrity(bio); |
5212e11f VV |
1206 | struct btt *btt = q->queuedata; |
1207 | struct bvec_iter iter; | |
f0dc089c | 1208 | unsigned long start; |
5212e11f | 1209 | struct bio_vec bvec; |
abf54548 | 1210 | int err = 0; |
f0dc089c | 1211 | bool do_acct; |
5212e11f | 1212 | |
e23947bd DM |
1213 | if (!bio_integrity_prep(bio)) |
1214 | return BLK_QC_T_NONE; | |
41cd8b70 | 1215 | |
f0dc089c | 1216 | do_acct = nd_iostat_start(bio, &start); |
5212e11f VV |
1217 | bio_for_each_segment(bvec, bio, iter) { |
1218 | unsigned int len = bvec.bv_len; | |
1219 | ||
1220 | BUG_ON(len > PAGE_SIZE); | |
1221 | /* Make sure len is in multiples of sector size. */ | |
1222 | /* XXX is this right? */ | |
1223 | BUG_ON(len < btt->sector_size); | |
1224 | BUG_ON(len % btt->sector_size); | |
1225 | ||
41cd8b70 | 1226 | err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset, |
c11f0c0b | 1227 | op_is_write(bio_op(bio)), iter.bi_sector); |
5212e11f | 1228 | if (err) { |
e6be2dcb | 1229 | dev_err(&btt->nd_btt->dev, |
5212e11f | 1230 | "io error in %s sector %lld, len %d,\n", |
abf54548 MC |
1231 | (op_is_write(bio_op(bio))) ? "WRITE" : |
1232 | "READ", | |
5212e11f | 1233 | (unsigned long long) iter.bi_sector, len); |
4e4cbee9 | 1234 | bio->bi_status = errno_to_blk_status(err); |
f0dc089c | 1235 | break; |
5212e11f VV |
1236 | } |
1237 | } | |
f0dc089c DW |
1238 | if (do_acct) |
1239 | nd_iostat_end(bio, start); | |
5212e11f | 1240 | |
4246a0b6 | 1241 | bio_endio(bio); |
dece1635 | 1242 | return BLK_QC_T_NONE; |
5212e11f VV |
1243 | } |
1244 | ||
1245 | static int btt_rw_page(struct block_device *bdev, sector_t sector, | |
c11f0c0b | 1246 | struct page *page, bool is_write) |
5212e11f VV |
1247 | { |
1248 | struct btt *btt = bdev->bd_disk->private_data; | |
c13c43d5 | 1249 | int rc; |
5212e11f | 1250 | |
c13c43d5 VV |
1251 | rc = btt_do_bvec(btt, NULL, page, PAGE_SIZE, 0, is_write, sector); |
1252 | if (rc == 0) | |
1253 | page_endio(page, is_write, 0); | |
1254 | ||
1255 | return rc; | |
5212e11f VV |
1256 | } |
1257 | ||
1258 | ||
1259 | static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
1260 | { | |
1261 | /* some standard values */ | |
1262 | geo->heads = 1 << 6; | |
1263 | geo->sectors = 1 << 5; | |
1264 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | static const struct block_device_operations btt_fops = { | |
1269 | .owner = THIS_MODULE, | |
1270 | .rw_page = btt_rw_page, | |
1271 | .getgeo = btt_getgeo, | |
58138820 | 1272 | .revalidate_disk = nvdimm_revalidate_disk, |
5212e11f VV |
1273 | }; |
1274 | ||
1275 | static int btt_blk_init(struct btt *btt) | |
1276 | { | |
1277 | struct nd_btt *nd_btt = btt->nd_btt; | |
1278 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
1279 | ||
1280 | /* create a new disk and request queue for btt */ | |
1281 | btt->btt_queue = blk_alloc_queue(GFP_KERNEL); | |
1282 | if (!btt->btt_queue) | |
1283 | return -ENOMEM; | |
1284 | ||
1285 | btt->btt_disk = alloc_disk(0); | |
1286 | if (!btt->btt_disk) { | |
1287 | blk_cleanup_queue(btt->btt_queue); | |
1288 | return -ENOMEM; | |
1289 | } | |
1290 | ||
1291 | nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name); | |
5212e11f VV |
1292 | btt->btt_disk->first_minor = 0; |
1293 | btt->btt_disk->fops = &btt_fops; | |
1294 | btt->btt_disk->private_data = btt; | |
1295 | btt->btt_disk->queue = btt->btt_queue; | |
1296 | btt->btt_disk->flags = GENHD_FL_EXT_DEVT; | |
1297 | ||
1298 | blk_queue_make_request(btt->btt_queue, btt_make_request); | |
1299 | blk_queue_logical_block_size(btt->btt_queue, btt->sector_size); | |
1300 | blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX); | |
5212e11f VV |
1301 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, btt->btt_queue); |
1302 | btt->btt_queue->queuedata = btt; | |
1303 | ||
41cd8b70 | 1304 | set_capacity(btt->btt_disk, 0); |
0d52c756 | 1305 | device_add_disk(&btt->nd_btt->dev, btt->btt_disk); |
41cd8b70 VV |
1306 | if (btt_meta_size(btt)) { |
1307 | int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt)); | |
1308 | ||
1309 | if (rc) { | |
1310 | del_gendisk(btt->btt_disk); | |
1311 | put_disk(btt->btt_disk); | |
1312 | blk_cleanup_queue(btt->btt_queue); | |
1313 | return rc; | |
1314 | } | |
1315 | } | |
1316 | set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9); | |
abe8b4e3 | 1317 | btt->nd_btt->size = btt->nlba * (u64)btt->sector_size; |
58138820 | 1318 | revalidate_disk(btt->btt_disk); |
5212e11f VV |
1319 | |
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | static void btt_blk_cleanup(struct btt *btt) | |
1324 | { | |
1325 | del_gendisk(btt->btt_disk); | |
1326 | put_disk(btt->btt_disk); | |
1327 | blk_cleanup_queue(btt->btt_queue); | |
1328 | } | |
1329 | ||
1330 | /** | |
1331 | * btt_init - initialize a block translation table for the given device | |
1332 | * @nd_btt: device with BTT geometry and backing device info | |
1333 | * @rawsize: raw size in bytes of the backing device | |
1334 | * @lbasize: lba size of the backing device | |
1335 | * @uuid: A uuid for the backing device - this is stored on media | |
1336 | * @maxlane: maximum number of parallel requests the device can handle | |
1337 | * | |
1338 | * Initialize a Block Translation Table on a backing device to provide | |
1339 | * single sector power fail atomicity. | |
1340 | * | |
1341 | * Context: | |
1342 | * Might sleep. | |
1343 | * | |
1344 | * Returns: | |
1345 | * Pointer to a new struct btt on success, NULL on failure. | |
1346 | */ | |
1347 | static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize, | |
1348 | u32 lbasize, u8 *uuid, struct nd_region *nd_region) | |
1349 | { | |
1350 | int ret; | |
1351 | struct btt *btt; | |
1352 | struct device *dev = &nd_btt->dev; | |
1353 | ||
e32bc729 | 1354 | btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL); |
5212e11f VV |
1355 | if (!btt) |
1356 | return NULL; | |
1357 | ||
1358 | btt->nd_btt = nd_btt; | |
1359 | btt->rawsize = rawsize; | |
1360 | btt->lbasize = lbasize; | |
1361 | btt->sector_size = ((lbasize >= 4096) ? 4096 : 512); | |
1362 | INIT_LIST_HEAD(&btt->arena_list); | |
1363 | mutex_init(&btt->init_lock); | |
1364 | btt->nd_region = nd_region; | |
1365 | ||
1366 | ret = discover_arenas(btt); | |
1367 | if (ret) { | |
1368 | dev_err(dev, "init: error in arena_discover: %d\n", ret); | |
e32bc729 | 1369 | return NULL; |
5212e11f VV |
1370 | } |
1371 | ||
58138820 | 1372 | if (btt->init_state != INIT_READY && nd_region->ro) { |
e6be2dcb | 1373 | dev_warn(dev, "%s is read-only, unable to init btt metadata\n", |
58138820 | 1374 | dev_name(&nd_region->dev)); |
e32bc729 | 1375 | return NULL; |
58138820 | 1376 | } else if (btt->init_state != INIT_READY) { |
5212e11f VV |
1377 | btt->num_arenas = (rawsize / ARENA_MAX_SIZE) + |
1378 | ((rawsize % ARENA_MAX_SIZE) ? 1 : 0); | |
1379 | dev_dbg(dev, "init: %d arenas for %llu rawsize\n", | |
1380 | btt->num_arenas, rawsize); | |
1381 | ||
1382 | ret = create_arenas(btt); | |
1383 | if (ret) { | |
1384 | dev_info(dev, "init: create_arenas: %d\n", ret); | |
e32bc729 | 1385 | return NULL; |
5212e11f VV |
1386 | } |
1387 | ||
1388 | ret = btt_meta_init(btt); | |
1389 | if (ret) { | |
1390 | dev_err(dev, "init: error in meta_init: %d\n", ret); | |
e32bc729 | 1391 | return NULL; |
5212e11f VV |
1392 | } |
1393 | } | |
1394 | ||
1395 | ret = btt_blk_init(btt); | |
1396 | if (ret) { | |
1397 | dev_err(dev, "init: error in blk_init: %d\n", ret); | |
e32bc729 | 1398 | return NULL; |
5212e11f VV |
1399 | } |
1400 | ||
1401 | btt_debugfs_init(btt); | |
1402 | ||
1403 | return btt; | |
5212e11f VV |
1404 | } |
1405 | ||
1406 | /** | |
1407 | * btt_fini - de-initialize a BTT | |
1408 | * @btt: the BTT handle that was generated by btt_init | |
1409 | * | |
1410 | * De-initialize a Block Translation Table on device removal | |
1411 | * | |
1412 | * Context: | |
1413 | * Might sleep. | |
1414 | */ | |
1415 | static void btt_fini(struct btt *btt) | |
1416 | { | |
1417 | if (btt) { | |
1418 | btt_blk_cleanup(btt); | |
1419 | free_arenas(btt); | |
1420 | debugfs_remove_recursive(btt->debugfs_dir); | |
5212e11f VV |
1421 | } |
1422 | } | |
1423 | ||
1424 | int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns) | |
1425 | { | |
1426 | struct nd_btt *nd_btt = to_nd_btt(ndns->claim); | |
1427 | struct nd_region *nd_region; | |
14e49454 | 1428 | struct btt_sb *btt_sb; |
5212e11f VV |
1429 | struct btt *btt; |
1430 | size_t rawsize; | |
1431 | ||
9dec4892 DW |
1432 | if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) { |
1433 | dev_dbg(&nd_btt->dev, "incomplete btt configuration\n"); | |
5212e11f | 1434 | return -ENODEV; |
9dec4892 | 1435 | } |
5212e11f | 1436 | |
14e49454 | 1437 | btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL); |
ed36b4db CJ |
1438 | if (!btt_sb) |
1439 | return -ENOMEM; | |
14e49454 VV |
1440 | |
1441 | /* | |
1442 | * If this returns < 0, that is ok as it just means there wasn't | |
1443 | * an existing BTT, and we're creating a new one. We still need to | |
1444 | * call this as we need the version dependent fields in nd_btt to be | |
1445 | * set correctly based on the holder class | |
1446 | */ | |
1447 | nd_btt_version(nd_btt, ndns, btt_sb); | |
1448 | ||
1449 | rawsize = nvdimm_namespace_capacity(ndns) - nd_btt->initial_offset; | |
5212e11f | 1450 | if (rawsize < ARENA_MIN_SIZE) { |
9dec4892 | 1451 | dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n", |
14e49454 VV |
1452 | dev_name(&ndns->dev), |
1453 | ARENA_MIN_SIZE + nd_btt->initial_offset); | |
5212e11f VV |
1454 | return -ENXIO; |
1455 | } | |
1456 | nd_region = to_nd_region(nd_btt->dev.parent); | |
1457 | btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid, | |
1458 | nd_region); | |
1459 | if (!btt) | |
1460 | return -ENOMEM; | |
1461 | nd_btt->btt = btt; | |
1462 | ||
1463 | return 0; | |
1464 | } | |
1465 | EXPORT_SYMBOL(nvdimm_namespace_attach_btt); | |
1466 | ||
298f2bc5 | 1467 | int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt) |
5212e11f | 1468 | { |
5212e11f VV |
1469 | struct btt *btt = nd_btt->btt; |
1470 | ||
1471 | btt_fini(btt); | |
1472 | nd_btt->btt = NULL; | |
1473 | ||
1474 | return 0; | |
1475 | } | |
1476 | EXPORT_SYMBOL(nvdimm_namespace_detach_btt); | |
1477 | ||
1478 | static int __init nd_btt_init(void) | |
1479 | { | |
ff8e92d5 | 1480 | int rc = 0; |
5212e11f VV |
1481 | |
1482 | debugfs_root = debugfs_create_dir("btt", NULL); | |
ff8e92d5 | 1483 | if (IS_ERR_OR_NULL(debugfs_root)) |
5212e11f | 1484 | rc = -ENXIO; |
5212e11f VV |
1485 | |
1486 | return rc; | |
1487 | } | |
1488 | ||
1489 | static void __exit nd_btt_exit(void) | |
1490 | { | |
1491 | debugfs_remove_recursive(debugfs_root); | |
5212e11f VV |
1492 | } |
1493 | ||
1494 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT); | |
1495 | MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>"); | |
1496 | MODULE_LICENSE("GPL v2"); | |
1497 | module_init(nd_btt_init); | |
1498 | module_exit(nd_btt_exit); |