2 * Block Translation Table
3 * Copyright (c) 2014-2015, Intel Corporation.
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.
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
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>
26 #include <linux/backing-dev.h>
30 enum log_ent_request {
35 static struct device *to_dev(struct arena_info *arena)
37 return &arena->nd_btt->dev;
40 static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
42 return offset + nd_btt->initial_offset;
45 static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
46 void *buf, size_t n, unsigned long flags)
48 struct nd_btt *nd_btt = arena->nd_btt;
49 struct nd_namespace_common *ndns = nd_btt->ndns;
51 /* arena offsets may be shifted from the base of the device */
52 offset = adjust_initial_offset(nd_btt, offset);
53 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
56 static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
57 void *buf, size_t n, unsigned long flags)
59 struct nd_btt *nd_btt = arena->nd_btt;
60 struct nd_namespace_common *ndns = nd_btt->ndns;
62 /* arena offsets may be shifted from the base of the device */
63 offset = adjust_initial_offset(nd_btt, offset);
64 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
67 static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
72 * infooff and info2off should always be at least 512B aligned.
73 * We rely on that to make sure rw_bytes does error clearing
74 * correctly, so make sure that is the case.
76 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
77 "arena->infooff: %#llx is unaligned\n", arena->infooff);
78 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
79 "arena->info2off: %#llx is unaligned\n", arena->info2off);
81 ret = arena_write_bytes(arena, arena->info2off, super,
82 sizeof(struct btt_sb), 0);
86 return arena_write_bytes(arena, arena->infooff, super,
87 sizeof(struct btt_sb), 0);
90 static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
92 return arena_read_bytes(arena, arena->infooff, super,
93 sizeof(struct btt_sb), 0);
97 * 'raw' version of btt_map write
99 * mapping is in little-endian
100 * mapping contains 'E' and 'Z' flags as desired
102 static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
105 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
107 if (unlikely(lba >= arena->external_nlba))
108 dev_err_ratelimited(to_dev(arena),
109 "%s: lba %#x out of range (max: %#x)\n",
110 __func__, lba, arena->external_nlba);
111 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
114 static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
115 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
121 * This 'mapping' is supposed to be just the LBA mapping, without
122 * any flags set, so strip the flag bits.
124 mapping = ent_lba(mapping);
126 ze = (z_flag << 1) + e_flag;
130 * We want to set neither of the Z or E flags, and
131 * in the actual layout, this means setting the bit
132 * positions of both to '1' to indicate a 'normal'
135 mapping |= MAP_ENT_NORMAL;
138 mapping |= (1 << MAP_ERR_SHIFT);
141 mapping |= (1 << MAP_TRIM_SHIFT);
145 * The case where Z and E are both sent in as '1' could be
146 * construed as a valid 'normal' case, but we decide not to,
149 dev_err_ratelimited(to_dev(arena),
150 "Invalid use of Z and E flags\n");
154 mapping_le = cpu_to_le32(mapping);
155 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
158 static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
159 int *trim, int *error, unsigned long rwb_flags)
163 u32 raw_mapping, postmap, ze, z_flag, e_flag;
164 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
166 if (unlikely(lba >= arena->external_nlba))
167 dev_err_ratelimited(to_dev(arena),
168 "%s: lba %#x out of range (max: %#x)\n",
169 __func__, lba, arena->external_nlba);
171 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
175 raw_mapping = le32_to_cpu(in);
177 z_flag = ent_z_flag(raw_mapping);
178 e_flag = ent_e_flag(raw_mapping);
179 ze = (z_flag << 1) + e_flag;
180 postmap = ent_lba(raw_mapping);
182 /* Reuse the {z,e}_flag variables for *trim and *error */
188 /* Initial state. Return postmap = premap */
214 static int btt_log_read_pair(struct arena_info *arena, u32 lane,
215 struct log_entry *ent)
217 return arena_read_bytes(arena,
218 arena->logoff + (2 * lane * LOG_ENT_SIZE), ent,
219 2 * LOG_ENT_SIZE, 0);
222 static struct dentry *debugfs_root;
224 static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
230 /* If for some reason, parent bttN was not created, exit */
234 snprintf(dirname, 32, "arena%d", idx);
235 d = debugfs_create_dir(dirname, parent);
236 if (IS_ERR_OR_NULL(d))
240 debugfs_create_x64("size", S_IRUGO, d, &a->size);
241 debugfs_create_x64("external_lba_start", S_IRUGO, d,
242 &a->external_lba_start);
243 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
244 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
245 &a->internal_lbasize);
246 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
247 debugfs_create_u32("external_lbasize", S_IRUGO, d,
248 &a->external_lbasize);
249 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
250 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
251 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
252 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
253 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
254 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
255 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
256 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
257 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
258 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
261 static void btt_debugfs_init(struct btt *btt)
264 struct arena_info *arena;
266 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
268 if (IS_ERR_OR_NULL(btt->debugfs_dir))
271 list_for_each_entry(arena, &btt->arena_list, list) {
272 arena_debugfs_init(arena, btt->debugfs_dir, i);
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.
284 * TODO The logic feels a bit kludge-y. make it better..
286 static int btt_log_get_old(struct log_entry *ent)
291 * the first ever time this is seen, the entry goes into [0]
292 * the next time, the following logic works out to put this
293 * (next) entry into [1]
295 if (ent[0].seq == 0) {
296 ent[0].seq = cpu_to_le32(1);
300 if (ent[0].seq == ent[1].seq)
302 if (le32_to_cpu(ent[0].seq) + le32_to_cpu(ent[1].seq) > 5)
305 if (le32_to_cpu(ent[0].seq) < le32_to_cpu(ent[1].seq)) {
306 if (le32_to_cpu(ent[1].seq) - le32_to_cpu(ent[0].seq) == 1)
311 if (le32_to_cpu(ent[0].seq) - le32_to_cpu(ent[1].seq) == 1)
321 * This function copies the desired (old/new) log entry into ent if
322 * it is not NULL. It returns the sub-slot number (0 or 1)
323 * where the desired log entry was found. Negative return values
326 static int btt_log_read(struct arena_info *arena, u32 lane,
327 struct log_entry *ent, int old_flag)
330 int old_ent, ret_ent;
331 struct log_entry log[2];
333 ret = btt_log_read_pair(arena, lane, log);
337 old_ent = btt_log_get_old(log);
338 if (old_ent < 0 || old_ent > 1) {
339 dev_err(to_dev(arena),
340 "log corruption (%d): lane %d seq [%d, %d]\n",
341 old_ent, lane, log[0].seq, log[1].seq);
342 /* TODO set error state? */
346 ret_ent = (old_flag ? old_ent : (1 - old_ent));
349 memcpy(ent, &log[ret_ent], LOG_ENT_SIZE);
355 * This function commits a log entry to media
356 * It does _not_ prepare the freelist entry for the next write
357 * btt_flog_write is the wrapper for updating the freelist elements
359 static int __btt_log_write(struct arena_info *arena, u32 lane,
360 u32 sub, struct log_entry *ent, unsigned long flags)
364 * Ignore the padding in log_entry for calculating log_half.
365 * The entry is 'committed' when we write the sequence number,
366 * and we want to ensure that that is the last thing written.
367 * We don't bother writing the padding as that would be extra
368 * media wear and write amplification
370 unsigned int log_half = (LOG_ENT_SIZE - 2 * sizeof(u64)) / 2;
371 u64 ns_off = arena->logoff + (((2 * lane) + sub) * LOG_ENT_SIZE);
374 /* split the 16B write into atomic, durable halves */
375 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
381 return arena_write_bytes(arena, ns_off, src, log_half, flags);
384 static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
385 struct log_entry *ent)
389 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
393 /* prepare the next free entry */
394 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
395 if (++(arena->freelist[lane].seq) == 4)
396 arena->freelist[lane].seq = 1;
397 if (ent_e_flag(ent->old_map))
398 arena->freelist[lane].has_err = 1;
399 arena->freelist[lane].block = le32_to_cpu(ent_lba(ent->old_map));
405 * This function initializes the BTT map to the initial state, which is
406 * all-zeroes, and indicates an identity mapping
408 static int btt_map_init(struct arena_info *arena)
413 size_t chunk_size = SZ_2M;
414 size_t mapsize = arena->logoff - arena->mapoff;
416 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
421 * mapoff should always be at least 512B aligned. We rely on that to
422 * make sure rw_bytes does error clearing correctly, so make sure that
425 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
426 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
429 size_t size = min(mapsize, chunk_size);
431 dev_WARN_ONCE(to_dev(arena), size < 512,
432 "chunk size: %#zx is unaligned\n", size);
433 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
449 * This function initializes the BTT log with 'fake' entries pointing
450 * to the initial reserved set of blocks as being free
452 static int btt_log_init(struct arena_info *arena)
454 size_t logsize = arena->info2off - arena->logoff;
455 size_t chunk_size = SZ_4K, offset = 0;
456 struct log_entry log;
461 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
465 * logoff should always be at least 512B aligned. We rely on that to
466 * make sure rw_bytes does error clearing correctly, so make sure that
469 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
470 "arena->logoff: %#llx is unaligned\n", arena->logoff);
473 size_t size = min(logsize, chunk_size);
475 dev_WARN_ONCE(to_dev(arena), size < 512,
476 "chunk size: %#zx is unaligned\n", size);
477 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
487 for (i = 0; i < arena->nfree; i++) {
488 log.lba = cpu_to_le32(i);
489 log.old_map = cpu_to_le32(arena->external_nlba + i);
490 log.new_map = cpu_to_le32(arena->external_nlba + i);
491 log.seq = cpu_to_le32(LOG_SEQ_INIT);
492 ret = __btt_log_write(arena, i, 0, &log, 0);
502 static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
504 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
507 static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
511 if (arena->freelist[lane].has_err) {
512 void *zero_page = page_address(ZERO_PAGE(0));
513 u32 lba = arena->freelist[lane].block;
514 u64 nsoff = to_namespace_offset(arena, lba);
515 unsigned long len = arena->sector_size;
517 mutex_lock(&arena->err_lock);
520 unsigned long chunk = min(len, PAGE_SIZE);
522 ret = arena_write_bytes(arena, nsoff, zero_page,
529 arena->freelist[lane].has_err = 0;
531 mutex_unlock(&arena->err_lock);
536 static int btt_freelist_init(struct arena_info *arena)
540 struct log_entry log_new, log_old;
542 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
544 if (!arena->freelist)
547 for (i = 0; i < arena->nfree; i++) {
548 old = btt_log_read(arena, i, &log_old, LOG_OLD_ENT);
552 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
556 /* sub points to the next one to be overwritten */
557 arena->freelist[i].sub = 1 - new;
558 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
559 arena->freelist[i].block = le32_to_cpu(log_new.old_map);
562 * FIXME: if error clearing fails during init, we want to make
565 if (ent_e_flag(log_new.old_map)) {
566 ret = arena_clear_freelist_error(arena, i);
568 dev_err_ratelimited(to_dev(arena),
569 "Unable to clear known errors\n");
572 /* This implies a newly created or untouched flog entry */
573 if (log_new.old_map == log_new.new_map)
576 /* Check if map recovery is needed */
577 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
581 if ((le32_to_cpu(log_new.new_map) != map_entry) &&
582 (le32_to_cpu(log_new.old_map) == map_entry)) {
584 * Last transaction wrote the flog, but wasn't able
585 * to complete the map write. So fix up the map.
587 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
588 le32_to_cpu(log_new.new_map), 0, 0, 0);
597 static int btt_rtt_init(struct arena_info *arena)
599 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
600 if (arena->rtt == NULL)
606 static int btt_maplocks_init(struct arena_info *arena)
610 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
612 if (!arena->map_locks)
615 for (i = 0; i < arena->nfree; i++)
616 spin_lock_init(&arena->map_locks[i].lock);
621 static struct arena_info *alloc_arena(struct btt *btt, size_t size,
622 size_t start, size_t arena_off)
624 struct arena_info *arena;
625 u64 logsize, mapsize, datasize;
626 u64 available = size;
628 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
631 arena->nd_btt = btt->nd_btt;
632 arena->sector_size = btt->sector_size;
638 arena->external_lba_start = start;
639 arena->external_lbasize = btt->lbasize;
640 arena->internal_lbasize = roundup(arena->external_lbasize,
641 INT_LBASIZE_ALIGNMENT);
642 arena->nfree = BTT_DEFAULT_NFREE;
643 arena->version_major = btt->nd_btt->version_major;
644 arena->version_minor = btt->nd_btt->version_minor;
646 if (available % BTT_PG_SIZE)
647 available -= (available % BTT_PG_SIZE);
649 /* Two pages are reserved for the super block and its copy */
650 available -= 2 * BTT_PG_SIZE;
652 /* The log takes a fixed amount of space based on nfree */
653 logsize = roundup(2 * arena->nfree * sizeof(struct log_entry),
655 available -= logsize;
657 /* Calculate optimal split between map and data area */
658 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
659 arena->internal_lbasize + MAP_ENT_SIZE);
660 arena->external_nlba = arena->internal_nlba - arena->nfree;
662 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
663 datasize = available - mapsize;
665 /* 'Absolute' values, relative to start of storage space */
666 arena->infooff = arena_off;
667 arena->dataoff = arena->infooff + BTT_PG_SIZE;
668 arena->mapoff = arena->dataoff + datasize;
669 arena->logoff = arena->mapoff + mapsize;
670 arena->info2off = arena->logoff + logsize;
674 static void free_arenas(struct btt *btt)
676 struct arena_info *arena, *next;
678 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
679 list_del(&arena->list);
681 kfree(arena->map_locks);
682 kfree(arena->freelist);
683 debugfs_remove_recursive(arena->debugfs_dir);
689 * This function reads an existing valid btt superblock and
690 * populates the corresponding arena_info struct
692 static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
695 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
696 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
697 arena->external_nlba = le32_to_cpu(super->external_nlba);
698 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
699 arena->nfree = le32_to_cpu(super->nfree);
700 arena->version_major = le16_to_cpu(super->version_major);
701 arena->version_minor = le16_to_cpu(super->version_minor);
703 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
704 le64_to_cpu(super->nextoff));
705 arena->infooff = arena_off;
706 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
707 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
708 arena->logoff = arena_off + le64_to_cpu(super->logoff);
709 arena->info2off = arena_off + le64_to_cpu(super->info2off);
711 arena->size = (le64_to_cpu(super->nextoff) > 0)
712 ? (le64_to_cpu(super->nextoff))
713 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
715 arena->flags = le32_to_cpu(super->flags);
718 static int discover_arenas(struct btt *btt)
721 struct arena_info *arena;
722 struct btt_sb *super;
723 size_t remaining = btt->rawsize;
728 super = kzalloc(sizeof(*super), GFP_KERNEL);
733 /* Alloc memory for arena */
734 arena = alloc_arena(btt, 0, 0, 0);
740 arena->infooff = cur_off;
741 ret = btt_info_read(arena, super);
745 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
746 if (remaining == btt->rawsize) {
747 btt->init_state = INIT_NOTFOUND;
748 dev_info(to_dev(arena), "No existing arenas\n");
751 dev_err(to_dev(arena),
752 "Found corrupted metadata!\n");
758 arena->external_lba_start = cur_nlba;
759 parse_arena_meta(arena, super, cur_off);
761 mutex_init(&arena->err_lock);
762 ret = btt_freelist_init(arena);
766 ret = btt_rtt_init(arena);
770 ret = btt_maplocks_init(arena);
774 list_add_tail(&arena->list, &btt->arena_list);
776 remaining -= arena->size;
777 cur_off += arena->size;
778 cur_nlba += arena->external_nlba;
781 if (arena->nextoff == 0)
784 btt->num_arenas = num_arenas;
785 btt->nlba = cur_nlba;
786 btt->init_state = INIT_READY;
799 static int create_arenas(struct btt *btt)
801 size_t remaining = btt->rawsize;
805 struct arena_info *arena;
806 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
808 remaining -= arena_size;
809 if (arena_size < ARENA_MIN_SIZE)
812 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
817 btt->nlba += arena->external_nlba;
818 if (remaining >= ARENA_MIN_SIZE)
819 arena->nextoff = arena->size;
822 cur_off += arena_size;
823 list_add_tail(&arena->list, &btt->arena_list);
830 * This function completes arena initialization by writing
832 * It is only called for an uninitialized arena when a write
833 * to that arena occurs for the first time.
835 static int btt_arena_write_layout(struct arena_info *arena)
839 struct btt_sb *super;
840 struct nd_btt *nd_btt = arena->nd_btt;
841 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
843 ret = btt_map_init(arena);
847 ret = btt_log_init(arena);
851 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
855 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
856 memcpy(super->uuid, nd_btt->uuid, 16);
857 memcpy(super->parent_uuid, parent_uuid, 16);
858 super->flags = cpu_to_le32(arena->flags);
859 super->version_major = cpu_to_le16(arena->version_major);
860 super->version_minor = cpu_to_le16(arena->version_minor);
861 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
862 super->external_nlba = cpu_to_le32(arena->external_nlba);
863 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
864 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
865 super->nfree = cpu_to_le32(arena->nfree);
866 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
867 super->nextoff = cpu_to_le64(arena->nextoff);
869 * Subtract arena->infooff (arena start) so numbers are relative
872 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
873 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
874 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
875 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
878 sum = nd_sb_checksum((struct nd_gen_sb *) super);
879 super->checksum = cpu_to_le64(sum);
881 ret = btt_info_write(arena, super);
888 * This function completes the initialization for the BTT namespace
889 * such that it is ready to accept IOs
891 static int btt_meta_init(struct btt *btt)
894 struct arena_info *arena;
896 mutex_lock(&btt->init_lock);
897 list_for_each_entry(arena, &btt->arena_list, list) {
898 ret = btt_arena_write_layout(arena);
902 ret = btt_freelist_init(arena);
906 ret = btt_rtt_init(arena);
910 ret = btt_maplocks_init(arena);
915 btt->init_state = INIT_READY;
918 mutex_unlock(&btt->init_lock);
922 static u32 btt_meta_size(struct btt *btt)
924 return btt->lbasize - btt->sector_size;
928 * This function calculates the arena in which the given LBA lies
929 * by doing a linear walk. This is acceptable since we expect only
930 * a few arenas. If we have backing devices that get much larger,
931 * we can construct a balanced binary tree of arenas at init time
932 * so that this range search becomes faster.
934 static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
935 struct arena_info **arena)
937 struct arena_info *arena_list;
938 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
940 list_for_each_entry(arena_list, &btt->arena_list, list) {
941 if (lba < arena_list->external_nlba) {
946 lba -= arena_list->external_nlba;
953 * The following (lock_map, unlock_map) are mostly just to improve
954 * readability, since they index into an array of locks
956 static void lock_map(struct arena_info *arena, u32 premap)
957 __acquires(&arena->map_locks[idx].lock)
959 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
961 spin_lock(&arena->map_locks[idx].lock);
964 static void unlock_map(struct arena_info *arena, u32 premap)
965 __releases(&arena->map_locks[idx].lock)
967 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
969 spin_unlock(&arena->map_locks[idx].lock);
972 static int btt_data_read(struct arena_info *arena, struct page *page,
973 unsigned int off, u32 lba, u32 len)
976 u64 nsoff = to_namespace_offset(arena, lba);
977 void *mem = kmap_atomic(page);
979 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
985 static int btt_data_write(struct arena_info *arena, u32 lba,
986 struct page *page, unsigned int off, u32 len)
989 u64 nsoff = to_namespace_offset(arena, lba);
990 void *mem = kmap_atomic(page);
992 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
998 static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1000 void *mem = kmap_atomic(page);
1002 memset(mem + off, 0, len);
1006 #ifdef CONFIG_BLK_DEV_INTEGRITY
1007 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1008 struct arena_info *arena, u32 postmap, int rw)
1010 unsigned int len = btt_meta_size(btt);
1017 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1020 unsigned int cur_len;
1024 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1026 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1027 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1028 * can use those directly
1031 cur_len = min(len, bv.bv_len);
1032 mem = kmap_atomic(bv.bv_page);
1034 ret = arena_write_bytes(arena, meta_nsoff,
1035 mem + bv.bv_offset, cur_len,
1038 ret = arena_read_bytes(arena, meta_nsoff,
1039 mem + bv.bv_offset, cur_len,
1047 meta_nsoff += cur_len;
1048 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1055 #else /* CONFIG_BLK_DEV_INTEGRITY */
1056 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1057 struct arena_info *arena, u32 postmap, int rw)
1063 static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1064 struct page *page, unsigned int off, sector_t sector,
1069 struct arena_info *arena = NULL;
1070 u32 lane = 0, premap, postmap;
1075 lane = nd_region_acquire_lane(btt->nd_region);
1077 ret = lba_to_arena(btt, sector, &premap, &arena);
1081 cur_len = min(btt->sector_size, len);
1083 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1089 * We loop to make sure that the post map LBA didn't change
1090 * from under us between writing the RTT and doing the actual
1098 zero_fill_data(page, off, cur_len);
1107 arena->rtt[lane] = RTT_VALID | postmap;
1109 * Barrier to make sure this write is not reordered
1110 * to do the verification map_read before the RTT store
1114 ret = btt_map_read(arena, premap, &new_map, &new_t,
1115 &new_e, NVDIMM_IO_ATOMIC);
1119 if ((postmap == new_map) && (t_flag == new_t) &&
1128 ret = btt_data_read(arena, page, off, postmap, cur_len);
1132 /* Media error - set the e_flag */
1133 rc = btt_map_write(arena, premap, postmap, 0, 1,
1139 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1144 arena->rtt[lane] = RTT_INVALID;
1145 nd_region_release_lane(btt->nd_region, lane);
1149 sector += btt->sector_size >> SECTOR_SHIFT;
1155 arena->rtt[lane] = RTT_INVALID;
1157 nd_region_release_lane(btt->nd_region, lane);
1162 * Normally, arena_{read,write}_bytes will take care of the initial offset
1163 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1164 * we need the final, raw namespace offset here
1166 static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1169 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1170 to_namespace_offset(arena, postmap));
1171 sector_t phys_sector = nsoff >> 9;
1173 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1176 static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1177 sector_t sector, struct page *page, unsigned int off,
1181 struct arena_info *arena = NULL;
1182 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1183 struct log_entry log;
1191 lane = nd_region_acquire_lane(btt->nd_region);
1193 ret = lba_to_arena(btt, sector, &premap, &arena);
1196 cur_len = min(btt->sector_size, len);
1198 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1203 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1204 arena->freelist[lane].has_err = 1;
1206 if (mutex_is_locked(&arena->err_lock)
1207 || arena->freelist[lane].has_err) {
1208 nd_region_release_lane(btt->nd_region, lane);
1210 ret = arena_clear_freelist_error(arena, lane);
1214 /* OK to acquire a different lane/free block */
1218 new_postmap = arena->freelist[lane].block;
1220 /* Wait if the new block is being read from */
1221 for (i = 0; i < arena->nfree; i++)
1222 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1226 if (new_postmap >= arena->internal_nlba) {
1231 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1236 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1242 lock_map(arena, premap);
1243 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1247 if (old_postmap >= arena->internal_nlba) {
1252 set_e_flag(old_postmap);
1254 log.lba = cpu_to_le32(premap);
1255 log.old_map = cpu_to_le32(old_postmap);
1256 log.new_map = cpu_to_le32(new_postmap);
1257 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1258 sub = arena->freelist[lane].sub;
1259 ret = btt_flog_write(arena, lane, sub, &log);
1263 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1268 unlock_map(arena, premap);
1269 nd_region_release_lane(btt->nd_region, lane);
1272 ret = arena_clear_freelist_error(arena, lane);
1279 sector += btt->sector_size >> SECTOR_SHIFT;
1285 unlock_map(arena, premap);
1287 nd_region_release_lane(btt->nd_region, lane);
1291 static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1292 struct page *page, unsigned int len, unsigned int off,
1293 bool is_write, sector_t sector)
1298 ret = btt_read_pg(btt, bip, page, off, sector, len);
1299 flush_dcache_page(page);
1301 flush_dcache_page(page);
1302 ret = btt_write_pg(btt, bip, sector, page, off, len);
1308 static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio)
1310 struct bio_integrity_payload *bip = bio_integrity(bio);
1311 struct btt *btt = q->queuedata;
1312 struct bvec_iter iter;
1313 unsigned long start;
1314 struct bio_vec bvec;
1318 if (!bio_integrity_prep(bio))
1319 return BLK_QC_T_NONE;
1321 do_acct = nd_iostat_start(bio, &start);
1322 bio_for_each_segment(bvec, bio, iter) {
1323 unsigned int len = bvec.bv_len;
1325 if (len > PAGE_SIZE || len < btt->sector_size ||
1326 len % btt->sector_size) {
1327 dev_err_ratelimited(&btt->nd_btt->dev,
1328 "unaligned bio segment (len: %d)\n", len);
1329 bio->bi_status = BLK_STS_IOERR;
1333 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1334 op_is_write(bio_op(bio)), iter.bi_sector);
1336 dev_err(&btt->nd_btt->dev,
1337 "io error in %s sector %lld, len %d,\n",
1338 (op_is_write(bio_op(bio))) ? "WRITE" :
1340 (unsigned long long) iter.bi_sector, len);
1341 bio->bi_status = errno_to_blk_status(err);
1346 nd_iostat_end(bio, start);
1349 return BLK_QC_T_NONE;
1352 static int btt_rw_page(struct block_device *bdev, sector_t sector,
1353 struct page *page, bool is_write)
1355 struct btt *btt = bdev->bd_disk->private_data;
1359 len = hpage_nr_pages(page) * PAGE_SIZE;
1360 rc = btt_do_bvec(btt, NULL, page, len, 0, is_write, sector);
1362 page_endio(page, is_write, 0);
1368 static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1370 /* some standard values */
1371 geo->heads = 1 << 6;
1372 geo->sectors = 1 << 5;
1373 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1377 static const struct block_device_operations btt_fops = {
1378 .owner = THIS_MODULE,
1379 .rw_page = btt_rw_page,
1380 .getgeo = btt_getgeo,
1381 .revalidate_disk = nvdimm_revalidate_disk,
1384 static int btt_blk_init(struct btt *btt)
1386 struct nd_btt *nd_btt = btt->nd_btt;
1387 struct nd_namespace_common *ndns = nd_btt->ndns;
1389 /* create a new disk and request queue for btt */
1390 btt->btt_queue = blk_alloc_queue(GFP_KERNEL);
1391 if (!btt->btt_queue)
1394 btt->btt_disk = alloc_disk(0);
1395 if (!btt->btt_disk) {
1396 blk_cleanup_queue(btt->btt_queue);
1400 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1401 btt->btt_disk->first_minor = 0;
1402 btt->btt_disk->fops = &btt_fops;
1403 btt->btt_disk->private_data = btt;
1404 btt->btt_disk->queue = btt->btt_queue;
1405 btt->btt_disk->flags = GENHD_FL_EXT_DEVT;
1406 btt->btt_disk->queue->backing_dev_info->capabilities |=
1407 BDI_CAP_SYNCHRONOUS_IO;
1409 blk_queue_make_request(btt->btt_queue, btt_make_request);
1410 blk_queue_logical_block_size(btt->btt_queue, btt->sector_size);
1411 blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX);
1412 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, btt->btt_queue);
1413 btt->btt_queue->queuedata = btt;
1415 set_capacity(btt->btt_disk, 0);
1416 device_add_disk(&btt->nd_btt->dev, btt->btt_disk);
1417 if (btt_meta_size(btt)) {
1418 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1421 del_gendisk(btt->btt_disk);
1422 put_disk(btt->btt_disk);
1423 blk_cleanup_queue(btt->btt_queue);
1427 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1428 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1429 revalidate_disk(btt->btt_disk);
1434 static void btt_blk_cleanup(struct btt *btt)
1436 del_gendisk(btt->btt_disk);
1437 put_disk(btt->btt_disk);
1438 blk_cleanup_queue(btt->btt_queue);
1442 * btt_init - initialize a block translation table for the given device
1443 * @nd_btt: device with BTT geometry and backing device info
1444 * @rawsize: raw size in bytes of the backing device
1445 * @lbasize: lba size of the backing device
1446 * @uuid: A uuid for the backing device - this is stored on media
1447 * @maxlane: maximum number of parallel requests the device can handle
1449 * Initialize a Block Translation Table on a backing device to provide
1450 * single sector power fail atomicity.
1456 * Pointer to a new struct btt on success, NULL on failure.
1458 static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1459 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1463 struct nd_namespace_io *nsio;
1464 struct device *dev = &nd_btt->dev;
1466 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1470 btt->nd_btt = nd_btt;
1471 btt->rawsize = rawsize;
1472 btt->lbasize = lbasize;
1473 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1474 INIT_LIST_HEAD(&btt->arena_list);
1475 mutex_init(&btt->init_lock);
1476 btt->nd_region = nd_region;
1477 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1478 btt->phys_bb = &nsio->bb;
1480 ret = discover_arenas(btt);
1482 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1486 if (btt->init_state != INIT_READY && nd_region->ro) {
1487 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1488 dev_name(&nd_region->dev));
1490 } else if (btt->init_state != INIT_READY) {
1491 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1492 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1493 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1494 btt->num_arenas, rawsize);
1496 ret = create_arenas(btt);
1498 dev_info(dev, "init: create_arenas: %d\n", ret);
1502 ret = btt_meta_init(btt);
1504 dev_err(dev, "init: error in meta_init: %d\n", ret);
1509 ret = btt_blk_init(btt);
1511 dev_err(dev, "init: error in blk_init: %d\n", ret);
1515 btt_debugfs_init(btt);
1521 * btt_fini - de-initialize a BTT
1522 * @btt: the BTT handle that was generated by btt_init
1524 * De-initialize a Block Translation Table on device removal
1529 static void btt_fini(struct btt *btt)
1532 btt_blk_cleanup(btt);
1534 debugfs_remove_recursive(btt->debugfs_dir);
1538 int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1540 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1541 struct nd_region *nd_region;
1542 struct btt_sb *btt_sb;
1546 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1547 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1551 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1556 * If this returns < 0, that is ok as it just means there wasn't
1557 * an existing BTT, and we're creating a new one. We still need to
1558 * call this as we need the version dependent fields in nd_btt to be
1559 * set correctly based on the holder class
1561 nd_btt_version(nd_btt, ndns, btt_sb);
1563 rawsize = nvdimm_namespace_capacity(ndns) - nd_btt->initial_offset;
1564 if (rawsize < ARENA_MIN_SIZE) {
1565 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1566 dev_name(&ndns->dev),
1567 ARENA_MIN_SIZE + nd_btt->initial_offset);
1570 nd_region = to_nd_region(nd_btt->dev.parent);
1571 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1579 EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1581 int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1583 struct btt *btt = nd_btt->btt;
1590 EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1592 static int __init nd_btt_init(void)
1596 debugfs_root = debugfs_create_dir("btt", NULL);
1597 if (IS_ERR_OR_NULL(debugfs_root))
1603 static void __exit nd_btt_exit(void)
1605 debugfs_remove_recursive(debugfs_root);
1608 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1610 MODULE_LICENSE("GPL v2");
1611 module_init(nd_btt_init);
1612 module_exit(nd_btt_exit);