Commit | Line | Data |
---|---|---|
2025cf9e | 1 | // SPDX-License-Identifier: GPL-2.0-only |
5212e11f VV |
2 | /* |
3 | * Block Translation Table | |
4 | * Copyright (c) 2014-2015, Intel Corporation. | |
5212e11f VV |
5 | */ |
6 | #include <linux/highmem.h> | |
7 | #include <linux/debugfs.h> | |
8 | #include <linux/blkdev.h> | |
4ee60ec1 | 9 | #include <linux/pagemap.h> |
5212e11f VV |
10 | #include <linux/module.h> |
11 | #include <linux/device.h> | |
12 | #include <linux/mutex.h> | |
13 | #include <linux/hdreg.h> | |
14 | #include <linux/genhd.h> | |
15 | #include <linux/sizes.h> | |
16 | #include <linux/ndctl.h> | |
17 | #include <linux/fs.h> | |
18 | #include <linux/nd.h> | |
23c47d2a | 19 | #include <linux/backing-dev.h> |
5212e11f VV |
20 | #include "btt.h" |
21 | #include "nd.h" | |
22 | ||
23 | enum log_ent_request { | |
24 | LOG_NEW_ENT = 0, | |
25 | LOG_OLD_ENT | |
26 | }; | |
27 | ||
86652d2e VV |
28 | static struct device *to_dev(struct arena_info *arena) |
29 | { | |
30 | return &arena->nd_btt->dev; | |
31 | } | |
32 | ||
d9b83c75 VV |
33 | static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset) |
34 | { | |
35 | return offset + nd_btt->initial_offset; | |
36 | } | |
37 | ||
5212e11f | 38 | static int arena_read_bytes(struct arena_info *arena, resource_size_t offset, |
3ae3d67b | 39 | void *buf, size_t n, unsigned long flags) |
5212e11f VV |
40 | { |
41 | struct nd_btt *nd_btt = arena->nd_btt; | |
42 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
43 | ||
14e49454 | 44 | /* arena offsets may be shifted from the base of the device */ |
d9b83c75 | 45 | offset = adjust_initial_offset(nd_btt, offset); |
3ae3d67b | 46 | return nvdimm_read_bytes(ndns, offset, buf, n, flags); |
5212e11f VV |
47 | } |
48 | ||
49 | static int arena_write_bytes(struct arena_info *arena, resource_size_t offset, | |
3ae3d67b | 50 | void *buf, size_t n, unsigned long flags) |
5212e11f VV |
51 | { |
52 | struct nd_btt *nd_btt = arena->nd_btt; | |
53 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
54 | ||
14e49454 | 55 | /* arena offsets may be shifted from the base of the device */ |
d9b83c75 | 56 | offset = adjust_initial_offset(nd_btt, offset); |
3ae3d67b | 57 | return nvdimm_write_bytes(ndns, offset, buf, n, flags); |
5212e11f VV |
58 | } |
59 | ||
60 | static int btt_info_write(struct arena_info *arena, struct btt_sb *super) | |
61 | { | |
62 | int ret; | |
63 | ||
b177fe85 VV |
64 | /* |
65 | * infooff and info2off should always be at least 512B aligned. | |
66 | * We rely on that to make sure rw_bytes does error clearing | |
67 | * correctly, so make sure that is the case. | |
68 | */ | |
86652d2e VV |
69 | dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512), |
70 | "arena->infooff: %#llx is unaligned\n", arena->infooff); | |
71 | dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512), | |
72 | "arena->info2off: %#llx is unaligned\n", arena->info2off); | |
b177fe85 | 73 | |
5212e11f | 74 | ret = arena_write_bytes(arena, arena->info2off, super, |
3ae3d67b | 75 | sizeof(struct btt_sb), 0); |
5212e11f VV |
76 | if (ret) |
77 | return ret; | |
78 | ||
79 | return arena_write_bytes(arena, arena->infooff, super, | |
3ae3d67b | 80 | sizeof(struct btt_sb), 0); |
5212e11f VV |
81 | } |
82 | ||
83 | static int btt_info_read(struct arena_info *arena, struct btt_sb *super) | |
84 | { | |
5212e11f | 85 | return arena_read_bytes(arena, arena->infooff, super, |
3ae3d67b | 86 | sizeof(struct btt_sb), 0); |
5212e11f VV |
87 | } |
88 | ||
89 | /* | |
90 | * 'raw' version of btt_map write | |
91 | * Assumptions: | |
92 | * mapping is in little-endian | |
93 | * mapping contains 'E' and 'Z' flags as desired | |
94 | */ | |
3ae3d67b VV |
95 | static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping, |
96 | unsigned long flags) | |
5212e11f VV |
97 | { |
98 | u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE); | |
99 | ||
86652d2e VV |
100 | if (unlikely(lba >= arena->external_nlba)) |
101 | dev_err_ratelimited(to_dev(arena), | |
102 | "%s: lba %#x out of range (max: %#x)\n", | |
103 | __func__, lba, arena->external_nlba); | |
3ae3d67b | 104 | return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags); |
5212e11f VV |
105 | } |
106 | ||
107 | static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping, | |
3ae3d67b | 108 | u32 z_flag, u32 e_flag, unsigned long rwb_flags) |
5212e11f VV |
109 | { |
110 | u32 ze; | |
111 | __le32 mapping_le; | |
112 | ||
113 | /* | |
114 | * This 'mapping' is supposed to be just the LBA mapping, without | |
115 | * any flags set, so strip the flag bits. | |
116 | */ | |
0595d539 | 117 | mapping = ent_lba(mapping); |
5212e11f VV |
118 | |
119 | ze = (z_flag << 1) + e_flag; | |
120 | switch (ze) { | |
121 | case 0: | |
122 | /* | |
123 | * We want to set neither of the Z or E flags, and | |
124 | * in the actual layout, this means setting the bit | |
125 | * positions of both to '1' to indicate a 'normal' | |
126 | * map entry | |
127 | */ | |
128 | mapping |= MAP_ENT_NORMAL; | |
129 | break; | |
130 | case 1: | |
131 | mapping |= (1 << MAP_ERR_SHIFT); | |
132 | break; | |
133 | case 2: | |
134 | mapping |= (1 << MAP_TRIM_SHIFT); | |
135 | break; | |
136 | default: | |
137 | /* | |
138 | * The case where Z and E are both sent in as '1' could be | |
139 | * construed as a valid 'normal' case, but we decide not to, | |
140 | * to avoid confusion | |
141 | */ | |
86652d2e VV |
142 | dev_err_ratelimited(to_dev(arena), |
143 | "Invalid use of Z and E flags\n"); | |
5212e11f VV |
144 | return -EIO; |
145 | } | |
146 | ||
147 | mapping_le = cpu_to_le32(mapping); | |
3ae3d67b | 148 | return __btt_map_write(arena, lba, mapping_le, rwb_flags); |
5212e11f VV |
149 | } |
150 | ||
151 | static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping, | |
3ae3d67b | 152 | int *trim, int *error, unsigned long rwb_flags) |
5212e11f VV |
153 | { |
154 | int ret; | |
155 | __le32 in; | |
156 | u32 raw_mapping, postmap, ze, z_flag, e_flag; | |
157 | u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE); | |
158 | ||
86652d2e VV |
159 | if (unlikely(lba >= arena->external_nlba)) |
160 | dev_err_ratelimited(to_dev(arena), | |
161 | "%s: lba %#x out of range (max: %#x)\n", | |
162 | __func__, lba, arena->external_nlba); | |
5212e11f | 163 | |
3ae3d67b | 164 | ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags); |
5212e11f VV |
165 | if (ret) |
166 | return ret; | |
167 | ||
168 | raw_mapping = le32_to_cpu(in); | |
169 | ||
0595d539 VV |
170 | z_flag = ent_z_flag(raw_mapping); |
171 | e_flag = ent_e_flag(raw_mapping); | |
5212e11f | 172 | ze = (z_flag << 1) + e_flag; |
0595d539 | 173 | postmap = ent_lba(raw_mapping); |
5212e11f VV |
174 | |
175 | /* Reuse the {z,e}_flag variables for *trim and *error */ | |
176 | z_flag = 0; | |
177 | e_flag = 0; | |
178 | ||
179 | switch (ze) { | |
180 | case 0: | |
181 | /* Initial state. Return postmap = premap */ | |
182 | *mapping = lba; | |
183 | break; | |
184 | case 1: | |
185 | *mapping = postmap; | |
186 | e_flag = 1; | |
187 | break; | |
188 | case 2: | |
189 | *mapping = postmap; | |
190 | z_flag = 1; | |
191 | break; | |
192 | case 3: | |
193 | *mapping = postmap; | |
194 | break; | |
195 | default: | |
196 | return -EIO; | |
197 | } | |
198 | ||
199 | if (trim) | |
200 | *trim = z_flag; | |
201 | if (error) | |
202 | *error = e_flag; | |
203 | ||
204 | return ret; | |
205 | } | |
206 | ||
24e3a7fb VV |
207 | static int btt_log_group_read(struct arena_info *arena, u32 lane, |
208 | struct log_group *log) | |
5212e11f | 209 | { |
5212e11f | 210 | return arena_read_bytes(arena, |
24e3a7fb VV |
211 | arena->logoff + (lane * LOG_GRP_SIZE), log, |
212 | LOG_GRP_SIZE, 0); | |
5212e11f VV |
213 | } |
214 | ||
215 | static struct dentry *debugfs_root; | |
216 | ||
217 | static void arena_debugfs_init(struct arena_info *a, struct dentry *parent, | |
218 | int idx) | |
219 | { | |
220 | char dirname[32]; | |
221 | struct dentry *d; | |
222 | ||
223 | /* If for some reason, parent bttN was not created, exit */ | |
224 | if (!parent) | |
225 | return; | |
226 | ||
227 | snprintf(dirname, 32, "arena%d", idx); | |
228 | d = debugfs_create_dir(dirname, parent); | |
229 | if (IS_ERR_OR_NULL(d)) | |
230 | return; | |
231 | a->debugfs_dir = d; | |
232 | ||
233 | debugfs_create_x64("size", S_IRUGO, d, &a->size); | |
234 | debugfs_create_x64("external_lba_start", S_IRUGO, d, | |
235 | &a->external_lba_start); | |
236 | debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba); | |
237 | debugfs_create_u32("internal_lbasize", S_IRUGO, d, | |
238 | &a->internal_lbasize); | |
239 | debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba); | |
240 | debugfs_create_u32("external_lbasize", S_IRUGO, d, | |
241 | &a->external_lbasize); | |
242 | debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree); | |
243 | debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major); | |
244 | debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor); | |
245 | debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff); | |
246 | debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff); | |
247 | debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff); | |
248 | debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff); | |
249 | debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff); | |
250 | debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off); | |
251 | debugfs_create_x32("flags", S_IRUGO, d, &a->flags); | |
24e3a7fb VV |
252 | debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]); |
253 | debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]); | |
5212e11f VV |
254 | } |
255 | ||
256 | static void btt_debugfs_init(struct btt *btt) | |
257 | { | |
258 | int i = 0; | |
259 | struct arena_info *arena; | |
260 | ||
261 | btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev), | |
262 | debugfs_root); | |
263 | if (IS_ERR_OR_NULL(btt->debugfs_dir)) | |
264 | return; | |
265 | ||
266 | list_for_each_entry(arena, &btt->arena_list, list) { | |
267 | arena_debugfs_init(arena, btt->debugfs_dir, i); | |
268 | i++; | |
269 | } | |
270 | } | |
271 | ||
24e3a7fb VV |
272 | static u32 log_seq(struct log_group *log, int log_idx) |
273 | { | |
274 | return le32_to_cpu(log->ent[log_idx].seq); | |
275 | } | |
276 | ||
5212e11f VV |
277 | /* |
278 | * This function accepts two log entries, and uses the | |
279 | * sequence number to find the 'older' entry. | |
280 | * It also updates the sequence number in this old entry to | |
281 | * make it the 'new' one if the mark_flag is set. | |
282 | * Finally, it returns which of the entries was the older one. | |
283 | * | |
284 | * TODO The logic feels a bit kludge-y. make it better.. | |
285 | */ | |
24e3a7fb | 286 | static int btt_log_get_old(struct arena_info *a, struct log_group *log) |
5212e11f | 287 | { |
24e3a7fb VV |
288 | int idx0 = a->log_index[0]; |
289 | int idx1 = a->log_index[1]; | |
5212e11f VV |
290 | int old; |
291 | ||
292 | /* | |
293 | * the first ever time this is seen, the entry goes into [0] | |
294 | * the next time, the following logic works out to put this | |
295 | * (next) entry into [1] | |
296 | */ | |
24e3a7fb VV |
297 | if (log_seq(log, idx0) == 0) { |
298 | log->ent[idx0].seq = cpu_to_le32(1); | |
5212e11f VV |
299 | return 0; |
300 | } | |
301 | ||
24e3a7fb | 302 | if (log_seq(log, idx0) == log_seq(log, idx1)) |
5212e11f | 303 | return -EINVAL; |
24e3a7fb | 304 | if (log_seq(log, idx0) + log_seq(log, idx1) > 5) |
5212e11f VV |
305 | return -EINVAL; |
306 | ||
24e3a7fb VV |
307 | if (log_seq(log, idx0) < log_seq(log, idx1)) { |
308 | if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1) | |
5212e11f VV |
309 | old = 0; |
310 | else | |
311 | old = 1; | |
312 | } else { | |
24e3a7fb | 313 | if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1) |
5212e11f VV |
314 | old = 1; |
315 | else | |
316 | old = 0; | |
317 | } | |
318 | ||
319 | return old; | |
320 | } | |
321 | ||
5212e11f VV |
322 | /* |
323 | * This function copies the desired (old/new) log entry into ent if | |
324 | * it is not NULL. It returns the sub-slot number (0 or 1) | |
325 | * where the desired log entry was found. Negative return values | |
326 | * indicate errors. | |
327 | */ | |
328 | static int btt_log_read(struct arena_info *arena, u32 lane, | |
329 | struct log_entry *ent, int old_flag) | |
330 | { | |
331 | int ret; | |
332 | int old_ent, ret_ent; | |
24e3a7fb | 333 | struct log_group log; |
5212e11f | 334 | |
24e3a7fb | 335 | ret = btt_log_group_read(arena, lane, &log); |
5212e11f VV |
336 | if (ret) |
337 | return -EIO; | |
338 | ||
24e3a7fb | 339 | old_ent = btt_log_get_old(arena, &log); |
5212e11f | 340 | if (old_ent < 0 || old_ent > 1) { |
e6be2dcb | 341 | dev_err(to_dev(arena), |
5212e11f | 342 | "log corruption (%d): lane %d seq [%d, %d]\n", |
24e3a7fb VV |
343 | old_ent, lane, log.ent[arena->log_index[0]].seq, |
344 | log.ent[arena->log_index[1]].seq); | |
5212e11f VV |
345 | /* TODO set error state? */ |
346 | return -EIO; | |
347 | } | |
348 | ||
349 | ret_ent = (old_flag ? old_ent : (1 - old_ent)); | |
350 | ||
351 | if (ent != NULL) | |
24e3a7fb | 352 | memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE); |
5212e11f VV |
353 | |
354 | return ret_ent; | |
355 | } | |
356 | ||
357 | /* | |
358 | * This function commits a log entry to media | |
359 | * It does _not_ prepare the freelist entry for the next write | |
360 | * btt_flog_write is the wrapper for updating the freelist elements | |
361 | */ | |
362 | static int __btt_log_write(struct arena_info *arena, u32 lane, | |
3ae3d67b | 363 | u32 sub, struct log_entry *ent, unsigned long flags) |
5212e11f VV |
364 | { |
365 | int ret; | |
24e3a7fb VV |
366 | u32 group_slot = arena->log_index[sub]; |
367 | unsigned int log_half = LOG_ENT_SIZE / 2; | |
5212e11f | 368 | void *src = ent; |
24e3a7fb | 369 | u64 ns_off; |
5212e11f | 370 | |
24e3a7fb VV |
371 | ns_off = arena->logoff + (lane * LOG_GRP_SIZE) + |
372 | (group_slot * LOG_ENT_SIZE); | |
5212e11f | 373 | /* split the 16B write into atomic, durable halves */ |
3ae3d67b | 374 | ret = arena_write_bytes(arena, ns_off, src, log_half, flags); |
5212e11f VV |
375 | if (ret) |
376 | return ret; | |
377 | ||
378 | ns_off += log_half; | |
379 | src += log_half; | |
3ae3d67b | 380 | return arena_write_bytes(arena, ns_off, src, log_half, flags); |
5212e11f VV |
381 | } |
382 | ||
383 | static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub, | |
384 | struct log_entry *ent) | |
385 | { | |
386 | int ret; | |
387 | ||
3ae3d67b | 388 | ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC); |
5212e11f VV |
389 | if (ret) |
390 | return ret; | |
391 | ||
392 | /* prepare the next free entry */ | |
393 | arena->freelist[lane].sub = 1 - arena->freelist[lane].sub; | |
394 | if (++(arena->freelist[lane].seq) == 4) | |
395 | arena->freelist[lane].seq = 1; | |
86aa6668 | 396 | if (ent_e_flag(le32_to_cpu(ent->old_map))) |
d9b83c75 | 397 | arena->freelist[lane].has_err = 1; |
86aa6668 | 398 | arena->freelist[lane].block = ent_lba(le32_to_cpu(ent->old_map)); |
5212e11f VV |
399 | |
400 | return ret; | |
401 | } | |
402 | ||
403 | /* | |
404 | * This function initializes the BTT map to the initial state, which is | |
405 | * all-zeroes, and indicates an identity mapping | |
406 | */ | |
407 | static int btt_map_init(struct arena_info *arena) | |
408 | { | |
409 | int ret = -EINVAL; | |
410 | void *zerobuf; | |
411 | size_t offset = 0; | |
412 | size_t chunk_size = SZ_2M; | |
413 | size_t mapsize = arena->logoff - arena->mapoff; | |
414 | ||
415 | zerobuf = kzalloc(chunk_size, GFP_KERNEL); | |
416 | if (!zerobuf) | |
417 | return -ENOMEM; | |
418 | ||
b177fe85 VV |
419 | /* |
420 | * mapoff should always be at least 512B aligned. We rely on that to | |
421 | * make sure rw_bytes does error clearing correctly, so make sure that | |
422 | * is the case. | |
423 | */ | |
86652d2e VV |
424 | dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512), |
425 | "arena->mapoff: %#llx is unaligned\n", arena->mapoff); | |
b177fe85 | 426 | |
5212e11f VV |
427 | while (mapsize) { |
428 | size_t size = min(mapsize, chunk_size); | |
429 | ||
86652d2e | 430 | dev_WARN_ONCE(to_dev(arena), size < 512, |
04c3c982 | 431 | "chunk size: %#zx is unaligned\n", size); |
5212e11f | 432 | ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf, |
3ae3d67b | 433 | size, 0); |
5212e11f VV |
434 | if (ret) |
435 | goto free; | |
436 | ||
437 | offset += size; | |
438 | mapsize -= size; | |
439 | cond_resched(); | |
440 | } | |
441 | ||
442 | free: | |
443 | kfree(zerobuf); | |
444 | return ret; | |
445 | } | |
446 | ||
447 | /* | |
448 | * This function initializes the BTT log with 'fake' entries pointing | |
449 | * to the initial reserved set of blocks as being free | |
450 | */ | |
451 | static int btt_log_init(struct arena_info *arena) | |
452 | { | |
b177fe85 VV |
453 | size_t logsize = arena->info2off - arena->logoff; |
454 | size_t chunk_size = SZ_4K, offset = 0; | |
24e3a7fb | 455 | struct log_entry ent; |
b177fe85 | 456 | void *zerobuf; |
5212e11f VV |
457 | int ret; |
458 | u32 i; | |
5212e11f | 459 | |
b177fe85 VV |
460 | zerobuf = kzalloc(chunk_size, GFP_KERNEL); |
461 | if (!zerobuf) | |
462 | return -ENOMEM; | |
463 | /* | |
464 | * logoff should always be at least 512B aligned. We rely on that to | |
465 | * make sure rw_bytes does error clearing correctly, so make sure that | |
466 | * is the case. | |
467 | */ | |
86652d2e VV |
468 | dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512), |
469 | "arena->logoff: %#llx is unaligned\n", arena->logoff); | |
b177fe85 VV |
470 | |
471 | while (logsize) { | |
472 | size_t size = min(logsize, chunk_size); | |
473 | ||
86652d2e | 474 | dev_WARN_ONCE(to_dev(arena), size < 512, |
04c3c982 | 475 | "chunk size: %#zx is unaligned\n", size); |
b177fe85 VV |
476 | ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf, |
477 | size, 0); | |
478 | if (ret) | |
479 | goto free; | |
480 | ||
481 | offset += size; | |
482 | logsize -= size; | |
483 | cond_resched(); | |
484 | } | |
5212e11f VV |
485 | |
486 | for (i = 0; i < arena->nfree; i++) { | |
24e3a7fb VV |
487 | ent.lba = cpu_to_le32(i); |
488 | ent.old_map = cpu_to_le32(arena->external_nlba + i); | |
489 | ent.new_map = cpu_to_le32(arena->external_nlba + i); | |
490 | ent.seq = cpu_to_le32(LOG_SEQ_INIT); | |
491 | ret = __btt_log_write(arena, i, 0, &ent, 0); | |
5212e11f | 492 | if (ret) |
b177fe85 | 493 | goto free; |
5212e11f VV |
494 | } |
495 | ||
b177fe85 VV |
496 | free: |
497 | kfree(zerobuf); | |
498 | return ret; | |
5212e11f VV |
499 | } |
500 | ||
d9b83c75 VV |
501 | static u64 to_namespace_offset(struct arena_info *arena, u64 lba) |
502 | { | |
503 | return arena->dataoff + ((u64)lba * arena->internal_lbasize); | |
504 | } | |
505 | ||
506 | static int arena_clear_freelist_error(struct arena_info *arena, u32 lane) | |
507 | { | |
508 | int ret = 0; | |
509 | ||
510 | if (arena->freelist[lane].has_err) { | |
511 | void *zero_page = page_address(ZERO_PAGE(0)); | |
512 | u32 lba = arena->freelist[lane].block; | |
513 | u64 nsoff = to_namespace_offset(arena, lba); | |
514 | unsigned long len = arena->sector_size; | |
515 | ||
516 | mutex_lock(&arena->err_lock); | |
517 | ||
518 | while (len) { | |
519 | unsigned long chunk = min(len, PAGE_SIZE); | |
520 | ||
521 | ret = arena_write_bytes(arena, nsoff, zero_page, | |
522 | chunk, 0); | |
523 | if (ret) | |
524 | break; | |
525 | len -= chunk; | |
526 | nsoff += chunk; | |
527 | if (len == 0) | |
528 | arena->freelist[lane].has_err = 0; | |
529 | } | |
530 | mutex_unlock(&arena->err_lock); | |
531 | } | |
532 | return ret; | |
533 | } | |
534 | ||
5212e11f VV |
535 | static int btt_freelist_init(struct arena_info *arena) |
536 | { | |
2f8c9011 | 537 | int new, ret; |
2f8c9011 | 538 | struct log_entry log_new; |
9dedc73a | 539 | u32 i, map_entry, log_oldmap, log_newmap; |
5212e11f VV |
540 | |
541 | arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry), | |
542 | GFP_KERNEL); | |
543 | if (!arena->freelist) | |
544 | return -ENOMEM; | |
545 | ||
546 | for (i = 0; i < arena->nfree; i++) { | |
5212e11f VV |
547 | new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT); |
548 | if (new < 0) | |
549 | return new; | |
550 | ||
9dedc73a VV |
551 | /* old and new map entries with any flags stripped out */ |
552 | log_oldmap = ent_lba(le32_to_cpu(log_new.old_map)); | |
553 | log_newmap = ent_lba(le32_to_cpu(log_new.new_map)); | |
554 | ||
5212e11f VV |
555 | /* sub points to the next one to be overwritten */ |
556 | arena->freelist[i].sub = 1 - new; | |
557 | arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq)); | |
9dedc73a | 558 | arena->freelist[i].block = log_oldmap; |
5212e11f | 559 | |
d9b83c75 VV |
560 | /* |
561 | * FIXME: if error clearing fails during init, we want to make | |
562 | * the BTT read-only | |
563 | */ | |
86aa6668 AK |
564 | if (ent_e_flag(le32_to_cpu(log_new.old_map)) && |
565 | !ent_normal(le32_to_cpu(log_new.old_map))) { | |
9dedc73a | 566 | arena->freelist[i].has_err = 1; |
d9b83c75 VV |
567 | ret = arena_clear_freelist_error(arena, i); |
568 | if (ret) | |
86652d2e VV |
569 | dev_err_ratelimited(to_dev(arena), |
570 | "Unable to clear known errors\n"); | |
d9b83c75 VV |
571 | } |
572 | ||
5212e11f | 573 | /* This implies a newly created or untouched flog entry */ |
9dedc73a | 574 | if (log_oldmap == log_newmap) |
5212e11f VV |
575 | continue; |
576 | ||
577 | /* Check if map recovery is needed */ | |
578 | ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry, | |
3ae3d67b | 579 | NULL, NULL, 0); |
5212e11f VV |
580 | if (ret) |
581 | return ret; | |
9dedc73a VV |
582 | |
583 | /* | |
584 | * The map_entry from btt_read_map is stripped of any flag bits, | |
585 | * so use the stripped out versions from the log as well for | |
586 | * testing whether recovery is needed. For restoration, use the | |
587 | * 'raw' version of the log entries as that captured what we | |
588 | * were going to write originally. | |
589 | */ | |
590 | if ((log_newmap != map_entry) && (log_oldmap == map_entry)) { | |
5212e11f VV |
591 | /* |
592 | * Last transaction wrote the flog, but wasn't able | |
593 | * to complete the map write. So fix up the map. | |
594 | */ | |
595 | ret = btt_map_write(arena, le32_to_cpu(log_new.lba), | |
3ae3d67b | 596 | le32_to_cpu(log_new.new_map), 0, 0, 0); |
5212e11f VV |
597 | if (ret) |
598 | return ret; | |
599 | } | |
5212e11f VV |
600 | } |
601 | ||
602 | return 0; | |
603 | } | |
604 | ||
24e3a7fb VV |
605 | static bool ent_is_padding(struct log_entry *ent) |
606 | { | |
607 | return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0) | |
608 | && (ent->seq == 0); | |
609 | } | |
610 | ||
611 | /* | |
612 | * Detecting valid log indices: We read a log group (see the comments in btt.h | |
613 | * for a description of a 'log_group' and its 'slots'), and iterate over its | |
614 | * four slots. We expect that a padding slot will be all-zeroes, and use this | |
615 | * to detect a padding slot vs. an actual entry. | |
616 | * | |
617 | * If a log_group is in the initial state, i.e. hasn't been used since the | |
618 | * creation of this BTT layout, it will have three of the four slots with | |
619 | * zeroes. We skip over these log_groups for the detection of log_index. If | |
620 | * all log_groups are in the initial state (i.e. the BTT has never been | |
621 | * written to), it is safe to assume the 'new format' of log entries in slots | |
622 | * (0, 1). | |
623 | */ | |
624 | static int log_set_indices(struct arena_info *arena) | |
625 | { | |
626 | bool idx_set = false, initial_state = true; | |
627 | int ret, log_index[2] = {-1, -1}; | |
628 | u32 i, j, next_idx = 0; | |
629 | struct log_group log; | |
630 | u32 pad_count = 0; | |
631 | ||
632 | for (i = 0; i < arena->nfree; i++) { | |
633 | ret = btt_log_group_read(arena, i, &log); | |
634 | if (ret < 0) | |
635 | return ret; | |
636 | ||
637 | for (j = 0; j < 4; j++) { | |
638 | if (!idx_set) { | |
639 | if (ent_is_padding(&log.ent[j])) { | |
640 | pad_count++; | |
641 | continue; | |
642 | } else { | |
643 | /* Skip if index has been recorded */ | |
644 | if ((next_idx == 1) && | |
645 | (j == log_index[0])) | |
646 | continue; | |
647 | /* valid entry, record index */ | |
648 | log_index[next_idx] = j; | |
649 | next_idx++; | |
650 | } | |
651 | if (next_idx == 2) { | |
652 | /* two valid entries found */ | |
653 | idx_set = true; | |
654 | } else if (next_idx > 2) { | |
655 | /* too many valid indices */ | |
656 | return -ENXIO; | |
657 | } | |
658 | } else { | |
659 | /* | |
660 | * once the indices have been set, just verify | |
661 | * that all subsequent log groups are either in | |
662 | * their initial state or follow the same | |
663 | * indices. | |
664 | */ | |
665 | if (j == log_index[0]) { | |
666 | /* entry must be 'valid' */ | |
667 | if (ent_is_padding(&log.ent[j])) | |
668 | return -ENXIO; | |
669 | } else if (j == log_index[1]) { | |
670 | ; | |
671 | /* | |
672 | * log_index[1] can be padding if the | |
673 | * lane never got used and it is still | |
674 | * in the initial state (three 'padding' | |
675 | * entries) | |
676 | */ | |
677 | } else { | |
678 | /* entry must be invalid (padding) */ | |
679 | if (!ent_is_padding(&log.ent[j])) | |
680 | return -ENXIO; | |
681 | } | |
682 | } | |
683 | } | |
684 | /* | |
685 | * If any of the log_groups have more than one valid, | |
686 | * non-padding entry, then the we are no longer in the | |
687 | * initial_state | |
688 | */ | |
689 | if (pad_count < 3) | |
690 | initial_state = false; | |
691 | pad_count = 0; | |
692 | } | |
693 | ||
694 | if (!initial_state && !idx_set) | |
695 | return -ENXIO; | |
696 | ||
697 | /* | |
698 | * If all the entries in the log were in the initial state, | |
699 | * assume new padding scheme | |
700 | */ | |
701 | if (initial_state) | |
702 | log_index[1] = 1; | |
703 | ||
704 | /* | |
705 | * Only allow the known permutations of log/padding indices, | |
706 | * i.e. (0, 1), and (0, 2) | |
707 | */ | |
708 | if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2))) | |
709 | ; /* known index possibilities */ | |
710 | else { | |
711 | dev_err(to_dev(arena), "Found an unknown padding scheme\n"); | |
712 | return -ENXIO; | |
713 | } | |
714 | ||
715 | arena->log_index[0] = log_index[0]; | |
716 | arena->log_index[1] = log_index[1]; | |
717 | dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]); | |
718 | dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]); | |
719 | return 0; | |
720 | } | |
721 | ||
5212e11f VV |
722 | static int btt_rtt_init(struct arena_info *arena) |
723 | { | |
724 | arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL); | |
725 | if (arena->rtt == NULL) | |
726 | return -ENOMEM; | |
727 | ||
728 | return 0; | |
729 | } | |
730 | ||
731 | static int btt_maplocks_init(struct arena_info *arena) | |
732 | { | |
733 | u32 i; | |
734 | ||
735 | arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock), | |
736 | GFP_KERNEL); | |
737 | if (!arena->map_locks) | |
738 | return -ENOMEM; | |
739 | ||
740 | for (i = 0; i < arena->nfree; i++) | |
741 | spin_lock_init(&arena->map_locks[i].lock); | |
742 | ||
743 | return 0; | |
744 | } | |
745 | ||
746 | static struct arena_info *alloc_arena(struct btt *btt, size_t size, | |
747 | size_t start, size_t arena_off) | |
748 | { | |
749 | struct arena_info *arena; | |
750 | u64 logsize, mapsize, datasize; | |
751 | u64 available = size; | |
752 | ||
753 | arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL); | |
754 | if (!arena) | |
755 | return NULL; | |
756 | arena->nd_btt = btt->nd_btt; | |
75892004 | 757 | arena->sector_size = btt->sector_size; |
d08cd5e0 | 758 | mutex_init(&arena->err_lock); |
5212e11f VV |
759 | |
760 | if (!size) | |
761 | return arena; | |
762 | ||
763 | arena->size = size; | |
764 | arena->external_lba_start = start; | |
765 | arena->external_lbasize = btt->lbasize; | |
766 | arena->internal_lbasize = roundup(arena->external_lbasize, | |
767 | INT_LBASIZE_ALIGNMENT); | |
768 | arena->nfree = BTT_DEFAULT_NFREE; | |
14e49454 VV |
769 | arena->version_major = btt->nd_btt->version_major; |
770 | arena->version_minor = btt->nd_btt->version_minor; | |
5212e11f VV |
771 | |
772 | if (available % BTT_PG_SIZE) | |
773 | available -= (available % BTT_PG_SIZE); | |
774 | ||
775 | /* Two pages are reserved for the super block and its copy */ | |
776 | available -= 2 * BTT_PG_SIZE; | |
777 | ||
778 | /* The log takes a fixed amount of space based on nfree */ | |
24e3a7fb | 779 | logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE); |
5212e11f VV |
780 | available -= logsize; |
781 | ||
782 | /* Calculate optimal split between map and data area */ | |
783 | arena->internal_nlba = div_u64(available - BTT_PG_SIZE, | |
784 | arena->internal_lbasize + MAP_ENT_SIZE); | |
785 | arena->external_nlba = arena->internal_nlba - arena->nfree; | |
786 | ||
787 | mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE); | |
788 | datasize = available - mapsize; | |
789 | ||
790 | /* 'Absolute' values, relative to start of storage space */ | |
791 | arena->infooff = arena_off; | |
792 | arena->dataoff = arena->infooff + BTT_PG_SIZE; | |
793 | arena->mapoff = arena->dataoff + datasize; | |
794 | arena->logoff = arena->mapoff + mapsize; | |
795 | arena->info2off = arena->logoff + logsize; | |
24e3a7fb VV |
796 | |
797 | /* Default log indices are (0,1) */ | |
798 | arena->log_index[0] = 0; | |
799 | arena->log_index[1] = 1; | |
5212e11f VV |
800 | return arena; |
801 | } | |
802 | ||
803 | static void free_arenas(struct btt *btt) | |
804 | { | |
805 | struct arena_info *arena, *next; | |
806 | ||
807 | list_for_each_entry_safe(arena, next, &btt->arena_list, list) { | |
808 | list_del(&arena->list); | |
809 | kfree(arena->rtt); | |
810 | kfree(arena->map_locks); | |
811 | kfree(arena->freelist); | |
812 | debugfs_remove_recursive(arena->debugfs_dir); | |
813 | kfree(arena); | |
814 | } | |
815 | } | |
816 | ||
5212e11f VV |
817 | /* |
818 | * This function reads an existing valid btt superblock and | |
819 | * populates the corresponding arena_info struct | |
820 | */ | |
821 | static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super, | |
822 | u64 arena_off) | |
823 | { | |
824 | arena->internal_nlba = le32_to_cpu(super->internal_nlba); | |
825 | arena->internal_lbasize = le32_to_cpu(super->internal_lbasize); | |
826 | arena->external_nlba = le32_to_cpu(super->external_nlba); | |
827 | arena->external_lbasize = le32_to_cpu(super->external_lbasize); | |
828 | arena->nfree = le32_to_cpu(super->nfree); | |
829 | arena->version_major = le16_to_cpu(super->version_major); | |
830 | arena->version_minor = le16_to_cpu(super->version_minor); | |
831 | ||
832 | arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off + | |
833 | le64_to_cpu(super->nextoff)); | |
834 | arena->infooff = arena_off; | |
835 | arena->dataoff = arena_off + le64_to_cpu(super->dataoff); | |
836 | arena->mapoff = arena_off + le64_to_cpu(super->mapoff); | |
837 | arena->logoff = arena_off + le64_to_cpu(super->logoff); | |
838 | arena->info2off = arena_off + le64_to_cpu(super->info2off); | |
839 | ||
5e329406 DW |
840 | arena->size = (le64_to_cpu(super->nextoff) > 0) |
841 | ? (le64_to_cpu(super->nextoff)) | |
842 | : (arena->info2off - arena->infooff + BTT_PG_SIZE); | |
5212e11f VV |
843 | |
844 | arena->flags = le32_to_cpu(super->flags); | |
845 | } | |
846 | ||
847 | static int discover_arenas(struct btt *btt) | |
848 | { | |
849 | int ret = 0; | |
850 | struct arena_info *arena; | |
851 | struct btt_sb *super; | |
852 | size_t remaining = btt->rawsize; | |
853 | u64 cur_nlba = 0; | |
854 | size_t cur_off = 0; | |
855 | int num_arenas = 0; | |
856 | ||
857 | super = kzalloc(sizeof(*super), GFP_KERNEL); | |
858 | if (!super) | |
859 | return -ENOMEM; | |
860 | ||
861 | while (remaining) { | |
862 | /* Alloc memory for arena */ | |
863 | arena = alloc_arena(btt, 0, 0, 0); | |
864 | if (!arena) { | |
865 | ret = -ENOMEM; | |
866 | goto out_super; | |
867 | } | |
868 | ||
869 | arena->infooff = cur_off; | |
870 | ret = btt_info_read(arena, super); | |
871 | if (ret) | |
872 | goto out; | |
873 | ||
ab45e763 | 874 | if (!nd_btt_arena_is_valid(btt->nd_btt, super)) { |
5212e11f VV |
875 | if (remaining == btt->rawsize) { |
876 | btt->init_state = INIT_NOTFOUND; | |
877 | dev_info(to_dev(arena), "No existing arenas\n"); | |
878 | goto out; | |
879 | } else { | |
e6be2dcb | 880 | dev_err(to_dev(arena), |
5212e11f VV |
881 | "Found corrupted metadata!\n"); |
882 | ret = -ENODEV; | |
883 | goto out; | |
884 | } | |
885 | } | |
886 | ||
887 | arena->external_lba_start = cur_nlba; | |
888 | parse_arena_meta(arena, super, cur_off); | |
889 | ||
24e3a7fb VV |
890 | ret = log_set_indices(arena); |
891 | if (ret) { | |
892 | dev_err(to_dev(arena), | |
893 | "Unable to deduce log/padding indices\n"); | |
894 | goto out; | |
895 | } | |
896 | ||
5212e11f VV |
897 | ret = btt_freelist_init(arena); |
898 | if (ret) | |
899 | goto out; | |
900 | ||
901 | ret = btt_rtt_init(arena); | |
902 | if (ret) | |
903 | goto out; | |
904 | ||
905 | ret = btt_maplocks_init(arena); | |
906 | if (ret) | |
907 | goto out; | |
908 | ||
909 | list_add_tail(&arena->list, &btt->arena_list); | |
910 | ||
911 | remaining -= arena->size; | |
912 | cur_off += arena->size; | |
913 | cur_nlba += arena->external_nlba; | |
914 | num_arenas++; | |
915 | ||
916 | if (arena->nextoff == 0) | |
917 | break; | |
918 | } | |
919 | btt->num_arenas = num_arenas; | |
920 | btt->nlba = cur_nlba; | |
921 | btt->init_state = INIT_READY; | |
922 | ||
923 | kfree(super); | |
924 | return ret; | |
925 | ||
926 | out: | |
927 | kfree(arena); | |
928 | free_arenas(btt); | |
929 | out_super: | |
930 | kfree(super); | |
931 | return ret; | |
932 | } | |
933 | ||
934 | static int create_arenas(struct btt *btt) | |
935 | { | |
936 | size_t remaining = btt->rawsize; | |
937 | size_t cur_off = 0; | |
938 | ||
939 | while (remaining) { | |
940 | struct arena_info *arena; | |
941 | size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining); | |
942 | ||
943 | remaining -= arena_size; | |
944 | if (arena_size < ARENA_MIN_SIZE) | |
945 | break; | |
946 | ||
947 | arena = alloc_arena(btt, arena_size, btt->nlba, cur_off); | |
948 | if (!arena) { | |
949 | free_arenas(btt); | |
950 | return -ENOMEM; | |
951 | } | |
952 | btt->nlba += arena->external_nlba; | |
953 | if (remaining >= ARENA_MIN_SIZE) | |
954 | arena->nextoff = arena->size; | |
955 | else | |
956 | arena->nextoff = 0; | |
957 | cur_off += arena_size; | |
958 | list_add_tail(&arena->list, &btt->arena_list); | |
959 | } | |
960 | ||
961 | return 0; | |
962 | } | |
963 | ||
964 | /* | |
965 | * This function completes arena initialization by writing | |
966 | * all the metadata. | |
967 | * It is only called for an uninitialized arena when a write | |
968 | * to that arena occurs for the first time. | |
969 | */ | |
fbde1414 | 970 | static int btt_arena_write_layout(struct arena_info *arena) |
5212e11f VV |
971 | { |
972 | int ret; | |
e1455744 | 973 | u64 sum; |
5212e11f | 974 | struct btt_sb *super; |
fbde1414 | 975 | struct nd_btt *nd_btt = arena->nd_btt; |
d1c6e08e | 976 | const uuid_t *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev); |
5212e11f VV |
977 | |
978 | ret = btt_map_init(arena); | |
979 | if (ret) | |
980 | return ret; | |
981 | ||
982 | ret = btt_log_init(arena); | |
983 | if (ret) | |
984 | return ret; | |
985 | ||
986 | super = kzalloc(sizeof(struct btt_sb), GFP_NOIO); | |
987 | if (!super) | |
988 | return -ENOMEM; | |
989 | ||
990 | strncpy(super->signature, BTT_SIG, BTT_SIG_LEN); | |
d1c6e08e DW |
991 | export_uuid(super->uuid, nd_btt->uuid); |
992 | export_uuid(super->parent_uuid, parent_uuid); | |
5212e11f VV |
993 | super->flags = cpu_to_le32(arena->flags); |
994 | super->version_major = cpu_to_le16(arena->version_major); | |
995 | super->version_minor = cpu_to_le16(arena->version_minor); | |
996 | super->external_lbasize = cpu_to_le32(arena->external_lbasize); | |
997 | super->external_nlba = cpu_to_le32(arena->external_nlba); | |
998 | super->internal_lbasize = cpu_to_le32(arena->internal_lbasize); | |
999 | super->internal_nlba = cpu_to_le32(arena->internal_nlba); | |
1000 | super->nfree = cpu_to_le32(arena->nfree); | |
1001 | super->infosize = cpu_to_le32(sizeof(struct btt_sb)); | |
1002 | super->nextoff = cpu_to_le64(arena->nextoff); | |
1003 | /* | |
1004 | * Subtract arena->infooff (arena start) so numbers are relative | |
1005 | * to 'this' arena | |
1006 | */ | |
1007 | super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff); | |
1008 | super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff); | |
1009 | super->logoff = cpu_to_le64(arena->logoff - arena->infooff); | |
1010 | super->info2off = cpu_to_le64(arena->info2off - arena->infooff); | |
1011 | ||
1012 | super->flags = 0; | |
e1455744 DW |
1013 | sum = nd_sb_checksum((struct nd_gen_sb *) super); |
1014 | super->checksum = cpu_to_le64(sum); | |
5212e11f VV |
1015 | |
1016 | ret = btt_info_write(arena, super); | |
1017 | ||
1018 | kfree(super); | |
1019 | return ret; | |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * This function completes the initialization for the BTT namespace | |
1024 | * such that it is ready to accept IOs | |
1025 | */ | |
1026 | static int btt_meta_init(struct btt *btt) | |
1027 | { | |
1028 | int ret = 0; | |
1029 | struct arena_info *arena; | |
1030 | ||
1031 | mutex_lock(&btt->init_lock); | |
1032 | list_for_each_entry(arena, &btt->arena_list, list) { | |
fbde1414 | 1033 | ret = btt_arena_write_layout(arena); |
5212e11f VV |
1034 | if (ret) |
1035 | goto unlock; | |
1036 | ||
1037 | ret = btt_freelist_init(arena); | |
1038 | if (ret) | |
1039 | goto unlock; | |
1040 | ||
1041 | ret = btt_rtt_init(arena); | |
1042 | if (ret) | |
1043 | goto unlock; | |
1044 | ||
1045 | ret = btt_maplocks_init(arena); | |
1046 | if (ret) | |
1047 | goto unlock; | |
1048 | } | |
1049 | ||
1050 | btt->init_state = INIT_READY; | |
1051 | ||
1052 | unlock: | |
1053 | mutex_unlock(&btt->init_lock); | |
1054 | return ret; | |
1055 | } | |
1056 | ||
41cd8b70 VV |
1057 | static u32 btt_meta_size(struct btt *btt) |
1058 | { | |
1059 | return btt->lbasize - btt->sector_size; | |
1060 | } | |
1061 | ||
5212e11f VV |
1062 | /* |
1063 | * This function calculates the arena in which the given LBA lies | |
1064 | * by doing a linear walk. This is acceptable since we expect only | |
1065 | * a few arenas. If we have backing devices that get much larger, | |
1066 | * we can construct a balanced binary tree of arenas at init time | |
1067 | * so that this range search becomes faster. | |
1068 | */ | |
1069 | static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap, | |
1070 | struct arena_info **arena) | |
1071 | { | |
1072 | struct arena_info *arena_list; | |
1073 | __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size); | |
1074 | ||
1075 | list_for_each_entry(arena_list, &btt->arena_list, list) { | |
1076 | if (lba < arena_list->external_nlba) { | |
1077 | *arena = arena_list; | |
1078 | *premap = lba; | |
1079 | return 0; | |
1080 | } | |
1081 | lba -= arena_list->external_nlba; | |
1082 | } | |
1083 | ||
1084 | return -EIO; | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * The following (lock_map, unlock_map) are mostly just to improve | |
1089 | * readability, since they index into an array of locks | |
1090 | */ | |
1091 | static void lock_map(struct arena_info *arena, u32 premap) | |
1092 | __acquires(&arena->map_locks[idx].lock) | |
1093 | { | |
1094 | u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree; | |
1095 | ||
1096 | spin_lock(&arena->map_locks[idx].lock); | |
1097 | } | |
1098 | ||
1099 | static void unlock_map(struct arena_info *arena, u32 premap) | |
1100 | __releases(&arena->map_locks[idx].lock) | |
1101 | { | |
1102 | u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree; | |
1103 | ||
1104 | spin_unlock(&arena->map_locks[idx].lock); | |
1105 | } | |
1106 | ||
5212e11f VV |
1107 | static int btt_data_read(struct arena_info *arena, struct page *page, |
1108 | unsigned int off, u32 lba, u32 len) | |
1109 | { | |
1110 | int ret; | |
1111 | u64 nsoff = to_namespace_offset(arena, lba); | |
1112 | void *mem = kmap_atomic(page); | |
1113 | ||
3ae3d67b | 1114 | ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC); |
5212e11f VV |
1115 | kunmap_atomic(mem); |
1116 | ||
1117 | return ret; | |
1118 | } | |
1119 | ||
1120 | static int btt_data_write(struct arena_info *arena, u32 lba, | |
1121 | struct page *page, unsigned int off, u32 len) | |
1122 | { | |
1123 | int ret; | |
1124 | u64 nsoff = to_namespace_offset(arena, lba); | |
1125 | void *mem = kmap_atomic(page); | |
1126 | ||
3ae3d67b | 1127 | ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC); |
5212e11f VV |
1128 | kunmap_atomic(mem); |
1129 | ||
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | static void zero_fill_data(struct page *page, unsigned int off, u32 len) | |
1134 | { | |
1135 | void *mem = kmap_atomic(page); | |
1136 | ||
1137 | memset(mem + off, 0, len); | |
1138 | kunmap_atomic(mem); | |
1139 | } | |
1140 | ||
41cd8b70 VV |
1141 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
1142 | static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip, | |
1143 | struct arena_info *arena, u32 postmap, int rw) | |
1144 | { | |
1145 | unsigned int len = btt_meta_size(btt); | |
1146 | u64 meta_nsoff; | |
1147 | int ret = 0; | |
1148 | ||
1149 | if (bip == NULL) | |
1150 | return 0; | |
1151 | ||
1152 | meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size; | |
1153 | ||
1154 | while (len) { | |
1155 | unsigned int cur_len; | |
1156 | struct bio_vec bv; | |
1157 | void *mem; | |
1158 | ||
1159 | bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter); | |
1160 | /* | |
1161 | * The 'bv' obtained from bvec_iter_bvec has its .bv_len and | |
1162 | * .bv_offset already adjusted for iter->bi_bvec_done, and we | |
1163 | * can use those directly | |
1164 | */ | |
1165 | ||
1166 | cur_len = min(len, bv.bv_len); | |
1167 | mem = kmap_atomic(bv.bv_page); | |
1168 | if (rw) | |
1169 | ret = arena_write_bytes(arena, meta_nsoff, | |
3ae3d67b VV |
1170 | mem + bv.bv_offset, cur_len, |
1171 | NVDIMM_IO_ATOMIC); | |
41cd8b70 VV |
1172 | else |
1173 | ret = arena_read_bytes(arena, meta_nsoff, | |
3ae3d67b VV |
1174 | mem + bv.bv_offset, cur_len, |
1175 | NVDIMM_IO_ATOMIC); | |
41cd8b70 VV |
1176 | |
1177 | kunmap_atomic(mem); | |
1178 | if (ret) | |
1179 | return ret; | |
1180 | ||
1181 | len -= cur_len; | |
1182 | meta_nsoff += cur_len; | |
b1fb2c52 DM |
1183 | if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len)) |
1184 | return -EIO; | |
41cd8b70 VV |
1185 | } |
1186 | ||
1187 | return ret; | |
1188 | } | |
1189 | ||
1190 | #else /* CONFIG_BLK_DEV_INTEGRITY */ | |
1191 | static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip, | |
1192 | struct arena_info *arena, u32 postmap, int rw) | |
1193 | { | |
1194 | return 0; | |
1195 | } | |
1196 | #endif | |
1197 | ||
1198 | static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip, | |
1199 | struct page *page, unsigned int off, sector_t sector, | |
1200 | unsigned int len) | |
5212e11f VV |
1201 | { |
1202 | int ret = 0; | |
1203 | int t_flag, e_flag; | |
1204 | struct arena_info *arena = NULL; | |
1205 | u32 lane = 0, premap, postmap; | |
1206 | ||
1207 | while (len) { | |
1208 | u32 cur_len; | |
1209 | ||
1210 | lane = nd_region_acquire_lane(btt->nd_region); | |
1211 | ||
1212 | ret = lba_to_arena(btt, sector, &premap, &arena); | |
1213 | if (ret) | |
1214 | goto out_lane; | |
1215 | ||
1216 | cur_len = min(btt->sector_size, len); | |
1217 | ||
3ae3d67b VV |
1218 | ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag, |
1219 | NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1220 | if (ret) |
1221 | goto out_lane; | |
1222 | ||
1223 | /* | |
1224 | * We loop to make sure that the post map LBA didn't change | |
1225 | * from under us between writing the RTT and doing the actual | |
1226 | * read. | |
1227 | */ | |
1228 | while (1) { | |
1229 | u32 new_map; | |
1398199d | 1230 | int new_t, new_e; |
5212e11f VV |
1231 | |
1232 | if (t_flag) { | |
1233 | zero_fill_data(page, off, cur_len); | |
1234 | goto out_lane; | |
1235 | } | |
1236 | ||
1237 | if (e_flag) { | |
1238 | ret = -EIO; | |
1239 | goto out_lane; | |
1240 | } | |
1241 | ||
1242 | arena->rtt[lane] = RTT_VALID | postmap; | |
1243 | /* | |
1244 | * Barrier to make sure this write is not reordered | |
1245 | * to do the verification map_read before the RTT store | |
1246 | */ | |
1247 | barrier(); | |
1248 | ||
1398199d VV |
1249 | ret = btt_map_read(arena, premap, &new_map, &new_t, |
1250 | &new_e, NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1251 | if (ret) |
1252 | goto out_rtt; | |
1253 | ||
1398199d VV |
1254 | if ((postmap == new_map) && (t_flag == new_t) && |
1255 | (e_flag == new_e)) | |
5212e11f VV |
1256 | break; |
1257 | ||
1258 | postmap = new_map; | |
1398199d VV |
1259 | t_flag = new_t; |
1260 | e_flag = new_e; | |
5212e11f VV |
1261 | } |
1262 | ||
1263 | ret = btt_data_read(arena, page, off, postmap, cur_len); | |
d9b83c75 | 1264 | if (ret) { |
d9b83c75 | 1265 | /* Media error - set the e_flag */ |
4e24e37d QC |
1266 | if (btt_map_write(arena, premap, postmap, 0, 1, NVDIMM_IO_ATOMIC)) |
1267 | dev_warn_ratelimited(to_dev(arena), | |
1268 | "Error persistently tracking bad blocks at %#x\n", | |
1269 | premap); | |
5212e11f | 1270 | goto out_rtt; |
d9b83c75 | 1271 | } |
5212e11f | 1272 | |
41cd8b70 VV |
1273 | if (bip) { |
1274 | ret = btt_rw_integrity(btt, bip, arena, postmap, READ); | |
1275 | if (ret) | |
1276 | goto out_rtt; | |
1277 | } | |
1278 | ||
5212e11f VV |
1279 | arena->rtt[lane] = RTT_INVALID; |
1280 | nd_region_release_lane(btt->nd_region, lane); | |
1281 | ||
1282 | len -= cur_len; | |
1283 | off += cur_len; | |
1284 | sector += btt->sector_size >> SECTOR_SHIFT; | |
1285 | } | |
1286 | ||
1287 | return 0; | |
1288 | ||
1289 | out_rtt: | |
1290 | arena->rtt[lane] = RTT_INVALID; | |
1291 | out_lane: | |
1292 | nd_region_release_lane(btt->nd_region, lane); | |
1293 | return ret; | |
1294 | } | |
1295 | ||
d9b83c75 VV |
1296 | /* |
1297 | * Normally, arena_{read,write}_bytes will take care of the initial offset | |
1298 | * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem, | |
1299 | * we need the final, raw namespace offset here | |
1300 | */ | |
1301 | static bool btt_is_badblock(struct btt *btt, struct arena_info *arena, | |
1302 | u32 postmap) | |
1303 | { | |
1304 | u64 nsoff = adjust_initial_offset(arena->nd_btt, | |
1305 | to_namespace_offset(arena, postmap)); | |
1306 | sector_t phys_sector = nsoff >> 9; | |
1307 | ||
1308 | return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize); | |
1309 | } | |
1310 | ||
41cd8b70 VV |
1311 | static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip, |
1312 | sector_t sector, struct page *page, unsigned int off, | |
1313 | unsigned int len) | |
5212e11f VV |
1314 | { |
1315 | int ret = 0; | |
1316 | struct arena_info *arena = NULL; | |
1317 | u32 premap = 0, old_postmap, new_postmap, lane = 0, i; | |
1318 | struct log_entry log; | |
1319 | int sub; | |
1320 | ||
1321 | while (len) { | |
1322 | u32 cur_len; | |
d9b83c75 | 1323 | int e_flag; |
5212e11f | 1324 | |
d9b83c75 | 1325 | retry: |
5212e11f VV |
1326 | lane = nd_region_acquire_lane(btt->nd_region); |
1327 | ||
1328 | ret = lba_to_arena(btt, sector, &premap, &arena); | |
1329 | if (ret) | |
1330 | goto out_lane; | |
1331 | cur_len = min(btt->sector_size, len); | |
1332 | ||
1333 | if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) { | |
1334 | ret = -EIO; | |
1335 | goto out_lane; | |
1336 | } | |
1337 | ||
d9b83c75 VV |
1338 | if (btt_is_badblock(btt, arena, arena->freelist[lane].block)) |
1339 | arena->freelist[lane].has_err = 1; | |
1340 | ||
1341 | if (mutex_is_locked(&arena->err_lock) | |
1342 | || arena->freelist[lane].has_err) { | |
1343 | nd_region_release_lane(btt->nd_region, lane); | |
1344 | ||
1345 | ret = arena_clear_freelist_error(arena, lane); | |
1346 | if (ret) | |
1347 | return ret; | |
1348 | ||
1349 | /* OK to acquire a different lane/free block */ | |
1350 | goto retry; | |
1351 | } | |
1352 | ||
5212e11f VV |
1353 | new_postmap = arena->freelist[lane].block; |
1354 | ||
1355 | /* Wait if the new block is being read from */ | |
1356 | for (i = 0; i < arena->nfree; i++) | |
1357 | while (arena->rtt[i] == (RTT_VALID | new_postmap)) | |
1358 | cpu_relax(); | |
1359 | ||
1360 | ||
1361 | if (new_postmap >= arena->internal_nlba) { | |
1362 | ret = -EIO; | |
1363 | goto out_lane; | |
41cd8b70 VV |
1364 | } |
1365 | ||
1366 | ret = btt_data_write(arena, new_postmap, page, off, cur_len); | |
5212e11f VV |
1367 | if (ret) |
1368 | goto out_lane; | |
1369 | ||
41cd8b70 VV |
1370 | if (bip) { |
1371 | ret = btt_rw_integrity(btt, bip, arena, new_postmap, | |
1372 | WRITE); | |
1373 | if (ret) | |
1374 | goto out_lane; | |
1375 | } | |
1376 | ||
5212e11f | 1377 | lock_map(arena, premap); |
d9b83c75 | 1378 | ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag, |
3ae3d67b | 1379 | NVDIMM_IO_ATOMIC); |
5212e11f VV |
1380 | if (ret) |
1381 | goto out_map; | |
1382 | if (old_postmap >= arena->internal_nlba) { | |
1383 | ret = -EIO; | |
1384 | goto out_map; | |
1385 | } | |
d9b83c75 VV |
1386 | if (e_flag) |
1387 | set_e_flag(old_postmap); | |
5212e11f VV |
1388 | |
1389 | log.lba = cpu_to_le32(premap); | |
1390 | log.old_map = cpu_to_le32(old_postmap); | |
1391 | log.new_map = cpu_to_le32(new_postmap); | |
1392 | log.seq = cpu_to_le32(arena->freelist[lane].seq); | |
1393 | sub = arena->freelist[lane].sub; | |
1394 | ret = btt_flog_write(arena, lane, sub, &log); | |
1395 | if (ret) | |
1396 | goto out_map; | |
1397 | ||
1db1f3ce VV |
1398 | ret = btt_map_write(arena, premap, new_postmap, 0, 0, |
1399 | NVDIMM_IO_ATOMIC); | |
5212e11f VV |
1400 | if (ret) |
1401 | goto out_map; | |
1402 | ||
1403 | unlock_map(arena, premap); | |
1404 | nd_region_release_lane(btt->nd_region, lane); | |
1405 | ||
d9b83c75 VV |
1406 | if (e_flag) { |
1407 | ret = arena_clear_freelist_error(arena, lane); | |
1408 | if (ret) | |
1409 | return ret; | |
1410 | } | |
1411 | ||
5212e11f VV |
1412 | len -= cur_len; |
1413 | off += cur_len; | |
1414 | sector += btt->sector_size >> SECTOR_SHIFT; | |
1415 | } | |
1416 | ||
1417 | return 0; | |
1418 | ||
1419 | out_map: | |
1420 | unlock_map(arena, premap); | |
1421 | out_lane: | |
1422 | nd_region_release_lane(btt->nd_region, lane); | |
1423 | return ret; | |
1424 | } | |
1425 | ||
41cd8b70 VV |
1426 | static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip, |
1427 | struct page *page, unsigned int len, unsigned int off, | |
3f289dcb | 1428 | unsigned int op, sector_t sector) |
5212e11f VV |
1429 | { |
1430 | int ret; | |
1431 | ||
3f289dcb | 1432 | if (!op_is_write(op)) { |
41cd8b70 | 1433 | ret = btt_read_pg(btt, bip, page, off, sector, len); |
5212e11f VV |
1434 | flush_dcache_page(page); |
1435 | } else { | |
1436 | flush_dcache_page(page); | |
41cd8b70 | 1437 | ret = btt_write_pg(btt, bip, sector, page, off, len); |
5212e11f VV |
1438 | } |
1439 | ||
1440 | return ret; | |
1441 | } | |
1442 | ||
3e08773c | 1443 | static void btt_submit_bio(struct bio *bio) |
5212e11f | 1444 | { |
41cd8b70 | 1445 | struct bio_integrity_payload *bip = bio_integrity(bio); |
309dca30 | 1446 | struct btt *btt = bio->bi_bdev->bd_disk->private_data; |
5212e11f | 1447 | struct bvec_iter iter; |
f0dc089c | 1448 | unsigned long start; |
5212e11f | 1449 | struct bio_vec bvec; |
abf54548 | 1450 | int err = 0; |
f0dc089c | 1451 | bool do_acct; |
5212e11f | 1452 | |
e23947bd | 1453 | if (!bio_integrity_prep(bio)) |
3e08773c | 1454 | return; |
41cd8b70 | 1455 | |
309dca30 | 1456 | do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue); |
0fd92f89 CH |
1457 | if (do_acct) |
1458 | start = bio_start_io_acct(bio); | |
5212e11f VV |
1459 | bio_for_each_segment(bvec, bio, iter) { |
1460 | unsigned int len = bvec.bv_len; | |
1461 | ||
86652d2e VV |
1462 | if (len > PAGE_SIZE || len < btt->sector_size || |
1463 | len % btt->sector_size) { | |
1464 | dev_err_ratelimited(&btt->nd_btt->dev, | |
1465 | "unaligned bio segment (len: %d)\n", len); | |
1466 | bio->bi_status = BLK_STS_IOERR; | |
1467 | break; | |
1468 | } | |
5212e11f | 1469 | |
41cd8b70 | 1470 | err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset, |
3f289dcb | 1471 | bio_op(bio), iter.bi_sector); |
5212e11f | 1472 | if (err) { |
e6be2dcb | 1473 | dev_err(&btt->nd_btt->dev, |
5212e11f | 1474 | "io error in %s sector %lld, len %d,\n", |
abf54548 MC |
1475 | (op_is_write(bio_op(bio))) ? "WRITE" : |
1476 | "READ", | |
5212e11f | 1477 | (unsigned long long) iter.bi_sector, len); |
4e4cbee9 | 1478 | bio->bi_status = errno_to_blk_status(err); |
f0dc089c | 1479 | break; |
5212e11f VV |
1480 | } |
1481 | } | |
f0dc089c | 1482 | if (do_acct) |
0fd92f89 | 1483 | bio_end_io_acct(bio, start); |
5212e11f | 1484 | |
4246a0b6 | 1485 | bio_endio(bio); |
5212e11f VV |
1486 | } |
1487 | ||
1488 | static int btt_rw_page(struct block_device *bdev, sector_t sector, | |
3f289dcb | 1489 | struct page *page, unsigned int op) |
5212e11f VV |
1490 | { |
1491 | struct btt *btt = bdev->bd_disk->private_data; | |
c13c43d5 | 1492 | int rc; |
5212e11f | 1493 | |
af3bbc12 | 1494 | rc = btt_do_bvec(btt, NULL, page, thp_size(page), 0, op, sector); |
c13c43d5 | 1495 | if (rc == 0) |
3f289dcb | 1496 | page_endio(page, op_is_write(op), 0); |
c13c43d5 VV |
1497 | |
1498 | return rc; | |
5212e11f VV |
1499 | } |
1500 | ||
1501 | ||
1502 | static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
1503 | { | |
1504 | /* some standard values */ | |
1505 | geo->heads = 1 << 6; | |
1506 | geo->sectors = 1 << 5; | |
1507 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
1508 | return 0; | |
1509 | } | |
1510 | ||
1511 | static const struct block_device_operations btt_fops = { | |
1512 | .owner = THIS_MODULE, | |
c62b37d9 | 1513 | .submit_bio = btt_submit_bio, |
5212e11f VV |
1514 | .rw_page = btt_rw_page, |
1515 | .getgeo = btt_getgeo, | |
1516 | }; | |
1517 | ||
1518 | static int btt_blk_init(struct btt *btt) | |
1519 | { | |
1520 | struct nd_btt *nd_btt = btt->nd_btt; | |
1521 | struct nd_namespace_common *ndns = nd_btt->ndns; | |
2762ff06 | 1522 | int rc = -ENOMEM; |
5212e11f | 1523 | |
d4e4e583 CH |
1524 | btt->btt_disk = blk_alloc_disk(NUMA_NO_NODE); |
1525 | if (!btt->btt_disk) | |
5212e11f VV |
1526 | return -ENOMEM; |
1527 | ||
5212e11f | 1528 | nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name); |
5212e11f VV |
1529 | btt->btt_disk->first_minor = 0; |
1530 | btt->btt_disk->fops = &btt_fops; | |
1531 | btt->btt_disk->private_data = btt; | |
5212e11f | 1532 | |
d4e4e583 CH |
1533 | blk_queue_logical_block_size(btt->btt_disk->queue, btt->sector_size); |
1534 | blk_queue_max_hw_sectors(btt->btt_disk->queue, UINT_MAX); | |
1535 | blk_queue_flag_set(QUEUE_FLAG_NONROT, btt->btt_disk->queue); | |
5212e11f | 1536 | |
41cd8b70 | 1537 | if (btt_meta_size(btt)) { |
2762ff06 LC |
1538 | rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt)); |
1539 | if (rc) | |
1540 | goto out_cleanup_disk; | |
41cd8b70 | 1541 | } |
2762ff06 | 1542 | |
41cd8b70 | 1543 | set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9); |
16be7974 LC |
1544 | rc = device_add_disk(&btt->nd_btt->dev, btt->btt_disk, NULL); |
1545 | if (rc) | |
1546 | goto out_cleanup_disk; | |
2762ff06 | 1547 | |
abe8b4e3 | 1548 | btt->nd_btt->size = btt->nlba * (u64)btt->sector_size; |
32f61d67 | 1549 | nvdimm_check_and_set_ro(btt->btt_disk); |
5212e11f VV |
1550 | |
1551 | return 0; | |
2762ff06 LC |
1552 | |
1553 | out_cleanup_disk: | |
1554 | blk_cleanup_disk(btt->btt_disk); | |
1555 | return rc; | |
5212e11f VV |
1556 | } |
1557 | ||
1558 | static void btt_blk_cleanup(struct btt *btt) | |
1559 | { | |
1560 | del_gendisk(btt->btt_disk); | |
d4e4e583 | 1561 | blk_cleanup_disk(btt->btt_disk); |
5212e11f VV |
1562 | } |
1563 | ||
1564 | /** | |
1565 | * btt_init - initialize a block translation table for the given device | |
1566 | * @nd_btt: device with BTT geometry and backing device info | |
1567 | * @rawsize: raw size in bytes of the backing device | |
1568 | * @lbasize: lba size of the backing device | |
1569 | * @uuid: A uuid for the backing device - this is stored on media | |
1570 | * @maxlane: maximum number of parallel requests the device can handle | |
1571 | * | |
1572 | * Initialize a Block Translation Table on a backing device to provide | |
1573 | * single sector power fail atomicity. | |
1574 | * | |
1575 | * Context: | |
1576 | * Might sleep. | |
1577 | * | |
1578 | * Returns: | |
1579 | * Pointer to a new struct btt on success, NULL on failure. | |
1580 | */ | |
1581 | static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize, | |
d1c6e08e DW |
1582 | u32 lbasize, uuid_t *uuid, |
1583 | struct nd_region *nd_region) | |
5212e11f VV |
1584 | { |
1585 | int ret; | |
1586 | struct btt *btt; | |
d9b83c75 | 1587 | struct nd_namespace_io *nsio; |
5212e11f VV |
1588 | struct device *dev = &nd_btt->dev; |
1589 | ||
e32bc729 | 1590 | btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL); |
5212e11f VV |
1591 | if (!btt) |
1592 | return NULL; | |
1593 | ||
1594 | btt->nd_btt = nd_btt; | |
1595 | btt->rawsize = rawsize; | |
1596 | btt->lbasize = lbasize; | |
1597 | btt->sector_size = ((lbasize >= 4096) ? 4096 : 512); | |
1598 | INIT_LIST_HEAD(&btt->arena_list); | |
1599 | mutex_init(&btt->init_lock); | |
1600 | btt->nd_region = nd_region; | |
d9b83c75 VV |
1601 | nsio = to_nd_namespace_io(&nd_btt->ndns->dev); |
1602 | btt->phys_bb = &nsio->bb; | |
5212e11f VV |
1603 | |
1604 | ret = discover_arenas(btt); | |
1605 | if (ret) { | |
1606 | dev_err(dev, "init: error in arena_discover: %d\n", ret); | |
e32bc729 | 1607 | return NULL; |
5212e11f VV |
1608 | } |
1609 | ||
58138820 | 1610 | if (btt->init_state != INIT_READY && nd_region->ro) { |
e6be2dcb | 1611 | dev_warn(dev, "%s is read-only, unable to init btt metadata\n", |
58138820 | 1612 | dev_name(&nd_region->dev)); |
e32bc729 | 1613 | return NULL; |
58138820 | 1614 | } else if (btt->init_state != INIT_READY) { |
5212e11f VV |
1615 | btt->num_arenas = (rawsize / ARENA_MAX_SIZE) + |
1616 | ((rawsize % ARENA_MAX_SIZE) ? 1 : 0); | |
1617 | dev_dbg(dev, "init: %d arenas for %llu rawsize\n", | |
1618 | btt->num_arenas, rawsize); | |
1619 | ||
1620 | ret = create_arenas(btt); | |
1621 | if (ret) { | |
1622 | dev_info(dev, "init: create_arenas: %d\n", ret); | |
e32bc729 | 1623 | return NULL; |
5212e11f VV |
1624 | } |
1625 | ||
1626 | ret = btt_meta_init(btt); | |
1627 | if (ret) { | |
1628 | dev_err(dev, "init: error in meta_init: %d\n", ret); | |
e32bc729 | 1629 | return NULL; |
5212e11f VV |
1630 | } |
1631 | } | |
1632 | ||
1633 | ret = btt_blk_init(btt); | |
1634 | if (ret) { | |
1635 | dev_err(dev, "init: error in blk_init: %d\n", ret); | |
e32bc729 | 1636 | return NULL; |
5212e11f VV |
1637 | } |
1638 | ||
1639 | btt_debugfs_init(btt); | |
1640 | ||
1641 | return btt; | |
5212e11f VV |
1642 | } |
1643 | ||
1644 | /** | |
1645 | * btt_fini - de-initialize a BTT | |
1646 | * @btt: the BTT handle that was generated by btt_init | |
1647 | * | |
1648 | * De-initialize a Block Translation Table on device removal | |
1649 | * | |
1650 | * Context: | |
1651 | * Might sleep. | |
1652 | */ | |
1653 | static void btt_fini(struct btt *btt) | |
1654 | { | |
1655 | if (btt) { | |
1656 | btt_blk_cleanup(btt); | |
1657 | free_arenas(btt); | |
1658 | debugfs_remove_recursive(btt->debugfs_dir); | |
5212e11f VV |
1659 | } |
1660 | } | |
1661 | ||
1662 | int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns) | |
1663 | { | |
1664 | struct nd_btt *nd_btt = to_nd_btt(ndns->claim); | |
1665 | struct nd_region *nd_region; | |
14e49454 | 1666 | struct btt_sb *btt_sb; |
5212e11f | 1667 | struct btt *btt; |
8f4b01fc AK |
1668 | size_t size, rawsize; |
1669 | int rc; | |
5212e11f | 1670 | |
9dec4892 DW |
1671 | if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) { |
1672 | dev_dbg(&nd_btt->dev, "incomplete btt configuration\n"); | |
5212e11f | 1673 | return -ENODEV; |
9dec4892 | 1674 | } |
5212e11f | 1675 | |
14e49454 | 1676 | btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL); |
ed36b4db CJ |
1677 | if (!btt_sb) |
1678 | return -ENOMEM; | |
14e49454 | 1679 | |
8f4b01fc AK |
1680 | size = nvdimm_namespace_capacity(ndns); |
1681 | rc = devm_namespace_enable(&nd_btt->dev, ndns, size); | |
1682 | if (rc) | |
1683 | return rc; | |
1684 | ||
14e49454 VV |
1685 | /* |
1686 | * If this returns < 0, that is ok as it just means there wasn't | |
1687 | * an existing BTT, and we're creating a new one. We still need to | |
1688 | * call this as we need the version dependent fields in nd_btt to be | |
1689 | * set correctly based on the holder class | |
1690 | */ | |
1691 | nd_btt_version(nd_btt, ndns, btt_sb); | |
1692 | ||
8f4b01fc | 1693 | rawsize = size - nd_btt->initial_offset; |
5212e11f | 1694 | if (rawsize < ARENA_MIN_SIZE) { |
9dec4892 | 1695 | dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n", |
14e49454 VV |
1696 | dev_name(&ndns->dev), |
1697 | ARENA_MIN_SIZE + nd_btt->initial_offset); | |
5212e11f VV |
1698 | return -ENXIO; |
1699 | } | |
1700 | nd_region = to_nd_region(nd_btt->dev.parent); | |
1701 | btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid, | |
d1c6e08e | 1702 | nd_region); |
5212e11f VV |
1703 | if (!btt) |
1704 | return -ENOMEM; | |
1705 | nd_btt->btt = btt; | |
1706 | ||
1707 | return 0; | |
1708 | } | |
1709 | EXPORT_SYMBOL(nvdimm_namespace_attach_btt); | |
1710 | ||
298f2bc5 | 1711 | int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt) |
5212e11f | 1712 | { |
5212e11f VV |
1713 | struct btt *btt = nd_btt->btt; |
1714 | ||
1715 | btt_fini(btt); | |
1716 | nd_btt->btt = NULL; | |
1717 | ||
1718 | return 0; | |
1719 | } | |
1720 | EXPORT_SYMBOL(nvdimm_namespace_detach_btt); | |
1721 | ||
1722 | static int __init nd_btt_init(void) | |
1723 | { | |
ff8e92d5 | 1724 | int rc = 0; |
5212e11f VV |
1725 | |
1726 | debugfs_root = debugfs_create_dir("btt", NULL); | |
ff8e92d5 | 1727 | if (IS_ERR_OR_NULL(debugfs_root)) |
5212e11f | 1728 | rc = -ENXIO; |
5212e11f VV |
1729 | |
1730 | return rc; | |
1731 | } | |
1732 | ||
1733 | static void __exit nd_btt_exit(void) | |
1734 | { | |
1735 | debugfs_remove_recursive(debugfs_root); | |
5212e11f VV |
1736 | } |
1737 | ||
1738 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT); | |
1739 | MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>"); | |
1740 | MODULE_LICENSE("GPL v2"); | |
1741 | module_init(nd_btt_init); | |
1742 | module_exit(nd_btt_exit); |