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>
29 enum log_ent_request {
34 static struct device *to_dev(struct arena_info *arena)
36 return &arena->nd_btt->dev;
39 static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
41 return offset + nd_btt->initial_offset;
44 static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
45 void *buf, size_t n, unsigned long flags)
47 struct nd_btt *nd_btt = arena->nd_btt;
48 struct nd_namespace_common *ndns = nd_btt->ndns;
50 /* arena offsets may be shifted from the base of the device */
51 offset = adjust_initial_offset(nd_btt, offset);
52 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
55 static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
56 void *buf, size_t n, unsigned long flags)
58 struct nd_btt *nd_btt = arena->nd_btt;
59 struct nd_namespace_common *ndns = nd_btt->ndns;
61 /* arena offsets may be shifted from the base of the device */
62 offset = adjust_initial_offset(nd_btt, offset);
63 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
66 static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
71 * infooff and info2off should always be at least 512B aligned.
72 * We rely on that to make sure rw_bytes does error clearing
73 * correctly, so make sure that is the case.
75 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
76 "arena->infooff: %#llx is unaligned\n", arena->infooff);
77 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
78 "arena->info2off: %#llx is unaligned\n", arena->info2off);
80 ret = arena_write_bytes(arena, arena->info2off, super,
81 sizeof(struct btt_sb), 0);
85 return arena_write_bytes(arena, arena->infooff, super,
86 sizeof(struct btt_sb), 0);
89 static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
91 return arena_read_bytes(arena, arena->infooff, super,
92 sizeof(struct btt_sb), 0);
96 * 'raw' version of btt_map write
98 * mapping is in little-endian
99 * mapping contains 'E' and 'Z' flags as desired
101 static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
104 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
106 if (unlikely(lba >= arena->external_nlba))
107 dev_err_ratelimited(to_dev(arena),
108 "%s: lba %#x out of range (max: %#x)\n",
109 __func__, lba, arena->external_nlba);
110 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
113 static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
114 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
120 * This 'mapping' is supposed to be just the LBA mapping, without
121 * any flags set, so strip the flag bits.
123 mapping = ent_lba(mapping);
125 ze = (z_flag << 1) + e_flag;
129 * We want to set neither of the Z or E flags, and
130 * in the actual layout, this means setting the bit
131 * positions of both to '1' to indicate a 'normal'
134 mapping |= MAP_ENT_NORMAL;
137 mapping |= (1 << MAP_ERR_SHIFT);
140 mapping |= (1 << MAP_TRIM_SHIFT);
144 * The case where Z and E are both sent in as '1' could be
145 * construed as a valid 'normal' case, but we decide not to,
148 dev_err_ratelimited(to_dev(arena),
149 "Invalid use of Z and E flags\n");
153 mapping_le = cpu_to_le32(mapping);
154 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
157 static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
158 int *trim, int *error, unsigned long rwb_flags)
162 u32 raw_mapping, postmap, ze, z_flag, e_flag;
163 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
165 if (unlikely(lba >= arena->external_nlba))
166 dev_err_ratelimited(to_dev(arena),
167 "%s: lba %#x out of range (max: %#x)\n",
168 __func__, lba, arena->external_nlba);
170 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
174 raw_mapping = le32_to_cpu(in);
176 z_flag = ent_z_flag(raw_mapping);
177 e_flag = ent_e_flag(raw_mapping);
178 ze = (z_flag << 1) + e_flag;
179 postmap = ent_lba(raw_mapping);
181 /* Reuse the {z,e}_flag variables for *trim and *error */
187 /* Initial state. Return postmap = premap */
213 static int btt_log_read_pair(struct arena_info *arena, u32 lane,
214 struct log_entry *ent)
216 return arena_read_bytes(arena,
217 arena->logoff + (2 * lane * LOG_ENT_SIZE), ent,
218 2 * LOG_ENT_SIZE, 0);
221 static struct dentry *debugfs_root;
223 static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
229 /* If for some reason, parent bttN was not created, exit */
233 snprintf(dirname, 32, "arena%d", idx);
234 d = debugfs_create_dir(dirname, parent);
235 if (IS_ERR_OR_NULL(d))
239 debugfs_create_x64("size", S_IRUGO, d, &a->size);
240 debugfs_create_x64("external_lba_start", S_IRUGO, d,
241 &a->external_lba_start);
242 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
243 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
244 &a->internal_lbasize);
245 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
246 debugfs_create_u32("external_lbasize", S_IRUGO, d,
247 &a->external_lbasize);
248 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
249 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
250 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
251 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
252 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
253 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
254 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
255 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
256 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
257 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
260 static void btt_debugfs_init(struct btt *btt)
263 struct arena_info *arena;
265 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
267 if (IS_ERR_OR_NULL(btt->debugfs_dir))
270 list_for_each_entry(arena, &btt->arena_list, list) {
271 arena_debugfs_init(arena, btt->debugfs_dir, i);
277 * This function accepts two log entries, and uses the
278 * sequence number to find the 'older' entry.
279 * It also updates the sequence number in this old entry to
280 * make it the 'new' one if the mark_flag is set.
281 * Finally, it returns which of the entries was the older one.
283 * TODO The logic feels a bit kludge-y. make it better..
285 static int btt_log_get_old(struct log_entry *ent)
290 * the first ever time this is seen, the entry goes into [0]
291 * the next time, the following logic works out to put this
292 * (next) entry into [1]
294 if (ent[0].seq == 0) {
295 ent[0].seq = cpu_to_le32(1);
299 if (ent[0].seq == ent[1].seq)
301 if (le32_to_cpu(ent[0].seq) + le32_to_cpu(ent[1].seq) > 5)
304 if (le32_to_cpu(ent[0].seq) < le32_to_cpu(ent[1].seq)) {
305 if (le32_to_cpu(ent[1].seq) - le32_to_cpu(ent[0].seq) == 1)
310 if (le32_to_cpu(ent[0].seq) - le32_to_cpu(ent[1].seq) == 1)
320 * This function copies the desired (old/new) log entry into ent if
321 * it is not NULL. It returns the sub-slot number (0 or 1)
322 * where the desired log entry was found. Negative return values
325 static int btt_log_read(struct arena_info *arena, u32 lane,
326 struct log_entry *ent, int old_flag)
329 int old_ent, ret_ent;
330 struct log_entry log[2];
332 ret = btt_log_read_pair(arena, lane, log);
336 old_ent = btt_log_get_old(log);
337 if (old_ent < 0 || old_ent > 1) {
338 dev_err(to_dev(arena),
339 "log corruption (%d): lane %d seq [%d, %d]\n",
340 old_ent, lane, log[0].seq, log[1].seq);
341 /* TODO set error state? */
345 ret_ent = (old_flag ? old_ent : (1 - old_ent));
348 memcpy(ent, &log[ret_ent], LOG_ENT_SIZE);
354 * This function commits a log entry to media
355 * It does _not_ prepare the freelist entry for the next write
356 * btt_flog_write is the wrapper for updating the freelist elements
358 static int __btt_log_write(struct arena_info *arena, u32 lane,
359 u32 sub, struct log_entry *ent, unsigned long flags)
363 * Ignore the padding in log_entry for calculating log_half.
364 * The entry is 'committed' when we write the sequence number,
365 * and we want to ensure that that is the last thing written.
366 * We don't bother writing the padding as that would be extra
367 * media wear and write amplification
369 unsigned int log_half = (LOG_ENT_SIZE - 2 * sizeof(u64)) / 2;
370 u64 ns_off = arena->logoff + (((2 * lane) + sub) * LOG_ENT_SIZE);
373 /* split the 16B write into atomic, durable halves */
374 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
380 return arena_write_bytes(arena, ns_off, src, log_half, flags);
383 static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
384 struct log_entry *ent)
388 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
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;
396 if (ent_e_flag(ent->old_map))
397 arena->freelist[lane].has_err = 1;
398 arena->freelist[lane].block = le32_to_cpu(ent_lba(ent->old_map));
404 * This function initializes the BTT map to the initial state, which is
405 * all-zeroes, and indicates an identity mapping
407 static int btt_map_init(struct arena_info *arena)
412 size_t chunk_size = SZ_2M;
413 size_t mapsize = arena->logoff - arena->mapoff;
415 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
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
424 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
425 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
428 size_t size = min(mapsize, chunk_size);
430 dev_WARN_ONCE(to_dev(arena), size < 512,
431 "chunk size: %#zx is unaligned\n", size);
432 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
448 * This function initializes the BTT log with 'fake' entries pointing
449 * to the initial reserved set of blocks as being free
451 static int btt_log_init(struct arena_info *arena)
453 size_t logsize = arena->info2off - arena->logoff;
454 size_t chunk_size = SZ_4K, offset = 0;
455 struct log_entry log;
460 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
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
468 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
469 "arena->logoff: %#llx is unaligned\n", arena->logoff);
472 size_t size = min(logsize, chunk_size);
474 dev_WARN_ONCE(to_dev(arena), size < 512,
475 "chunk size: %#zx is unaligned\n", size);
476 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
486 for (i = 0; i < arena->nfree; i++) {
487 log.lba = cpu_to_le32(i);
488 log.old_map = cpu_to_le32(arena->external_nlba + i);
489 log.new_map = cpu_to_le32(arena->external_nlba + i);
490 log.seq = cpu_to_le32(LOG_SEQ_INIT);
491 ret = __btt_log_write(arena, i, 0, &log, 0);
501 static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
503 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
506 static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
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;
516 mutex_lock(&arena->err_lock);
519 unsigned long chunk = min(len, PAGE_SIZE);
521 ret = arena_write_bytes(arena, nsoff, zero_page,
528 arena->freelist[lane].has_err = 0;
530 mutex_unlock(&arena->err_lock);
535 static int btt_freelist_init(struct arena_info *arena)
539 struct log_entry log_new, log_old;
541 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
543 if (!arena->freelist)
546 for (i = 0; i < arena->nfree; i++) {
547 old = btt_log_read(arena, i, &log_old, LOG_OLD_ENT);
551 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
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));
558 arena->freelist[i].block = le32_to_cpu(log_new.old_map);
561 * FIXME: if error clearing fails during init, we want to make
564 if (ent_e_flag(log_new.old_map)) {
565 ret = arena_clear_freelist_error(arena, i);
567 dev_err_ratelimited(to_dev(arena),
568 "Unable to clear known errors\n");
571 /* This implies a newly created or untouched flog entry */
572 if (log_new.old_map == log_new.new_map)
575 /* Check if map recovery is needed */
576 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
580 if ((le32_to_cpu(log_new.new_map) != map_entry) &&
581 (le32_to_cpu(log_new.old_map) == map_entry)) {
583 * Last transaction wrote the flog, but wasn't able
584 * to complete the map write. So fix up the map.
586 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
587 le32_to_cpu(log_new.new_map), 0, 0, 0);
596 static int btt_rtt_init(struct arena_info *arena)
598 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
599 if (arena->rtt == NULL)
605 static int btt_maplocks_init(struct arena_info *arena)
609 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
611 if (!arena->map_locks)
614 for (i = 0; i < arena->nfree; i++)
615 spin_lock_init(&arena->map_locks[i].lock);
620 static struct arena_info *alloc_arena(struct btt *btt, size_t size,
621 size_t start, size_t arena_off)
623 struct arena_info *arena;
624 u64 logsize, mapsize, datasize;
625 u64 available = size;
627 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
630 arena->nd_btt = btt->nd_btt;
631 arena->sector_size = btt->sector_size;
637 arena->external_lba_start = start;
638 arena->external_lbasize = btt->lbasize;
639 arena->internal_lbasize = roundup(arena->external_lbasize,
640 INT_LBASIZE_ALIGNMENT);
641 arena->nfree = BTT_DEFAULT_NFREE;
642 arena->version_major = btt->nd_btt->version_major;
643 arena->version_minor = btt->nd_btt->version_minor;
645 if (available % BTT_PG_SIZE)
646 available -= (available % BTT_PG_SIZE);
648 /* Two pages are reserved for the super block and its copy */
649 available -= 2 * BTT_PG_SIZE;
651 /* The log takes a fixed amount of space based on nfree */
652 logsize = roundup(2 * arena->nfree * sizeof(struct log_entry),
654 available -= logsize;
656 /* Calculate optimal split between map and data area */
657 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
658 arena->internal_lbasize + MAP_ENT_SIZE);
659 arena->external_nlba = arena->internal_nlba - arena->nfree;
661 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
662 datasize = available - mapsize;
664 /* 'Absolute' values, relative to start of storage space */
665 arena->infooff = arena_off;
666 arena->dataoff = arena->infooff + BTT_PG_SIZE;
667 arena->mapoff = arena->dataoff + datasize;
668 arena->logoff = arena->mapoff + mapsize;
669 arena->info2off = arena->logoff + logsize;
673 static void free_arenas(struct btt *btt)
675 struct arena_info *arena, *next;
677 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
678 list_del(&arena->list);
680 kfree(arena->map_locks);
681 kfree(arena->freelist);
682 debugfs_remove_recursive(arena->debugfs_dir);
688 * This function reads an existing valid btt superblock and
689 * populates the corresponding arena_info struct
691 static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
694 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
695 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
696 arena->external_nlba = le32_to_cpu(super->external_nlba);
697 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
698 arena->nfree = le32_to_cpu(super->nfree);
699 arena->version_major = le16_to_cpu(super->version_major);
700 arena->version_minor = le16_to_cpu(super->version_minor);
702 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
703 le64_to_cpu(super->nextoff));
704 arena->infooff = arena_off;
705 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
706 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
707 arena->logoff = arena_off + le64_to_cpu(super->logoff);
708 arena->info2off = arena_off + le64_to_cpu(super->info2off);
710 arena->size = (le64_to_cpu(super->nextoff) > 0)
711 ? (le64_to_cpu(super->nextoff))
712 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
714 arena->flags = le32_to_cpu(super->flags);
717 static int discover_arenas(struct btt *btt)
720 struct arena_info *arena;
721 struct btt_sb *super;
722 size_t remaining = btt->rawsize;
727 super = kzalloc(sizeof(*super), GFP_KERNEL);
732 /* Alloc memory for arena */
733 arena = alloc_arena(btt, 0, 0, 0);
739 arena->infooff = cur_off;
740 ret = btt_info_read(arena, super);
744 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
745 if (remaining == btt->rawsize) {
746 btt->init_state = INIT_NOTFOUND;
747 dev_info(to_dev(arena), "No existing arenas\n");
750 dev_err(to_dev(arena),
751 "Found corrupted metadata!\n");
757 arena->external_lba_start = cur_nlba;
758 parse_arena_meta(arena, super, cur_off);
760 mutex_init(&arena->err_lock);
761 ret = btt_freelist_init(arena);
765 ret = btt_rtt_init(arena);
769 ret = btt_maplocks_init(arena);
773 list_add_tail(&arena->list, &btt->arena_list);
775 remaining -= arena->size;
776 cur_off += arena->size;
777 cur_nlba += arena->external_nlba;
780 if (arena->nextoff == 0)
783 btt->num_arenas = num_arenas;
784 btt->nlba = cur_nlba;
785 btt->init_state = INIT_READY;
798 static int create_arenas(struct btt *btt)
800 size_t remaining = btt->rawsize;
804 struct arena_info *arena;
805 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
807 remaining -= arena_size;
808 if (arena_size < ARENA_MIN_SIZE)
811 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
816 btt->nlba += arena->external_nlba;
817 if (remaining >= ARENA_MIN_SIZE)
818 arena->nextoff = arena->size;
821 cur_off += arena_size;
822 list_add_tail(&arena->list, &btt->arena_list);
829 * This function completes arena initialization by writing
831 * It is only called for an uninitialized arena when a write
832 * to that arena occurs for the first time.
834 static int btt_arena_write_layout(struct arena_info *arena)
838 struct btt_sb *super;
839 struct nd_btt *nd_btt = arena->nd_btt;
840 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
842 ret = btt_map_init(arena);
846 ret = btt_log_init(arena);
850 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
854 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
855 memcpy(super->uuid, nd_btt->uuid, 16);
856 memcpy(super->parent_uuid, parent_uuid, 16);
857 super->flags = cpu_to_le32(arena->flags);
858 super->version_major = cpu_to_le16(arena->version_major);
859 super->version_minor = cpu_to_le16(arena->version_minor);
860 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
861 super->external_nlba = cpu_to_le32(arena->external_nlba);
862 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
863 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
864 super->nfree = cpu_to_le32(arena->nfree);
865 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
866 super->nextoff = cpu_to_le64(arena->nextoff);
868 * Subtract arena->infooff (arena start) so numbers are relative
871 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
872 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
873 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
874 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
877 sum = nd_sb_checksum((struct nd_gen_sb *) super);
878 super->checksum = cpu_to_le64(sum);
880 ret = btt_info_write(arena, super);
887 * This function completes the initialization for the BTT namespace
888 * such that it is ready to accept IOs
890 static int btt_meta_init(struct btt *btt)
893 struct arena_info *arena;
895 mutex_lock(&btt->init_lock);
896 list_for_each_entry(arena, &btt->arena_list, list) {
897 ret = btt_arena_write_layout(arena);
901 ret = btt_freelist_init(arena);
905 ret = btt_rtt_init(arena);
909 ret = btt_maplocks_init(arena);
914 btt->init_state = INIT_READY;
917 mutex_unlock(&btt->init_lock);
921 static u32 btt_meta_size(struct btt *btt)
923 return btt->lbasize - btt->sector_size;
927 * This function calculates the arena in which the given LBA lies
928 * by doing a linear walk. This is acceptable since we expect only
929 * a few arenas. If we have backing devices that get much larger,
930 * we can construct a balanced binary tree of arenas at init time
931 * so that this range search becomes faster.
933 static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
934 struct arena_info **arena)
936 struct arena_info *arena_list;
937 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
939 list_for_each_entry(arena_list, &btt->arena_list, list) {
940 if (lba < arena_list->external_nlba) {
945 lba -= arena_list->external_nlba;
952 * The following (lock_map, unlock_map) are mostly just to improve
953 * readability, since they index into an array of locks
955 static void lock_map(struct arena_info *arena, u32 premap)
956 __acquires(&arena->map_locks[idx].lock)
958 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
960 spin_lock(&arena->map_locks[idx].lock);
963 static void unlock_map(struct arena_info *arena, u32 premap)
964 __releases(&arena->map_locks[idx].lock)
966 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
968 spin_unlock(&arena->map_locks[idx].lock);
971 static int btt_data_read(struct arena_info *arena, struct page *page,
972 unsigned int off, u32 lba, u32 len)
975 u64 nsoff = to_namespace_offset(arena, lba);
976 void *mem = kmap_atomic(page);
978 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
984 static int btt_data_write(struct arena_info *arena, u32 lba,
985 struct page *page, unsigned int off, u32 len)
988 u64 nsoff = to_namespace_offset(arena, lba);
989 void *mem = kmap_atomic(page);
991 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
997 static void zero_fill_data(struct page *page, unsigned int off, u32 len)
999 void *mem = kmap_atomic(page);
1001 memset(mem + off, 0, len);
1005 #ifdef CONFIG_BLK_DEV_INTEGRITY
1006 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1007 struct arena_info *arena, u32 postmap, int rw)
1009 unsigned int len = btt_meta_size(btt);
1016 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1019 unsigned int cur_len;
1023 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1025 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1026 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1027 * can use those directly
1030 cur_len = min(len, bv.bv_len);
1031 mem = kmap_atomic(bv.bv_page);
1033 ret = arena_write_bytes(arena, meta_nsoff,
1034 mem + bv.bv_offset, cur_len,
1037 ret = arena_read_bytes(arena, meta_nsoff,
1038 mem + bv.bv_offset, cur_len,
1046 meta_nsoff += cur_len;
1047 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1054 #else /* CONFIG_BLK_DEV_INTEGRITY */
1055 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1056 struct arena_info *arena, u32 postmap, int rw)
1062 static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1063 struct page *page, unsigned int off, sector_t sector,
1068 struct arena_info *arena = NULL;
1069 u32 lane = 0, premap, postmap;
1074 lane = nd_region_acquire_lane(btt->nd_region);
1076 ret = lba_to_arena(btt, sector, &premap, &arena);
1080 cur_len = min(btt->sector_size, len);
1082 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1088 * We loop to make sure that the post map LBA didn't change
1089 * from under us between writing the RTT and doing the actual
1097 zero_fill_data(page, off, cur_len);
1106 arena->rtt[lane] = RTT_VALID | postmap;
1108 * Barrier to make sure this write is not reordered
1109 * to do the verification map_read before the RTT store
1113 ret = btt_map_read(arena, premap, &new_map, &new_t,
1114 &new_e, NVDIMM_IO_ATOMIC);
1118 if ((postmap == new_map) && (t_flag == new_t) &&
1127 ret = btt_data_read(arena, page, off, postmap, cur_len);
1131 /* Media error - set the e_flag */
1132 rc = btt_map_write(arena, premap, postmap, 0, 1,
1138 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1143 arena->rtt[lane] = RTT_INVALID;
1144 nd_region_release_lane(btt->nd_region, lane);
1148 sector += btt->sector_size >> SECTOR_SHIFT;
1154 arena->rtt[lane] = RTT_INVALID;
1156 nd_region_release_lane(btt->nd_region, lane);
1161 * Normally, arena_{read,write}_bytes will take care of the initial offset
1162 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1163 * we need the final, raw namespace offset here
1165 static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1168 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1169 to_namespace_offset(arena, postmap));
1170 sector_t phys_sector = nsoff >> 9;
1172 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1175 static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1176 sector_t sector, struct page *page, unsigned int off,
1180 struct arena_info *arena = NULL;
1181 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1182 struct log_entry log;
1190 lane = nd_region_acquire_lane(btt->nd_region);
1192 ret = lba_to_arena(btt, sector, &premap, &arena);
1195 cur_len = min(btt->sector_size, len);
1197 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1202 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1203 arena->freelist[lane].has_err = 1;
1205 if (mutex_is_locked(&arena->err_lock)
1206 || arena->freelist[lane].has_err) {
1207 nd_region_release_lane(btt->nd_region, lane);
1209 ret = arena_clear_freelist_error(arena, lane);
1213 /* OK to acquire a different lane/free block */
1217 new_postmap = arena->freelist[lane].block;
1219 /* Wait if the new block is being read from */
1220 for (i = 0; i < arena->nfree; i++)
1221 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1225 if (new_postmap >= arena->internal_nlba) {
1230 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1235 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1241 lock_map(arena, premap);
1242 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1246 if (old_postmap >= arena->internal_nlba) {
1251 set_e_flag(old_postmap);
1253 log.lba = cpu_to_le32(premap);
1254 log.old_map = cpu_to_le32(old_postmap);
1255 log.new_map = cpu_to_le32(new_postmap);
1256 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1257 sub = arena->freelist[lane].sub;
1258 ret = btt_flog_write(arena, lane, sub, &log);
1262 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1267 unlock_map(arena, premap);
1268 nd_region_release_lane(btt->nd_region, lane);
1271 ret = arena_clear_freelist_error(arena, lane);
1278 sector += btt->sector_size >> SECTOR_SHIFT;
1284 unlock_map(arena, premap);
1286 nd_region_release_lane(btt->nd_region, lane);
1290 static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1291 struct page *page, unsigned int len, unsigned int off,
1292 bool is_write, sector_t sector)
1297 ret = btt_read_pg(btt, bip, page, off, sector, len);
1298 flush_dcache_page(page);
1300 flush_dcache_page(page);
1301 ret = btt_write_pg(btt, bip, sector, page, off, len);
1307 static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio)
1309 struct bio_integrity_payload *bip = bio_integrity(bio);
1310 struct btt *btt = q->queuedata;
1311 struct bvec_iter iter;
1312 unsigned long start;
1313 struct bio_vec bvec;
1317 if (!bio_integrity_prep(bio))
1318 return BLK_QC_T_NONE;
1320 do_acct = nd_iostat_start(bio, &start);
1321 bio_for_each_segment(bvec, bio, iter) {
1322 unsigned int len = bvec.bv_len;
1324 if (len > PAGE_SIZE || len < btt->sector_size ||
1325 len % btt->sector_size) {
1326 dev_err_ratelimited(&btt->nd_btt->dev,
1327 "unaligned bio segment (len: %d)\n", len);
1328 bio->bi_status = BLK_STS_IOERR;
1332 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1333 op_is_write(bio_op(bio)), iter.bi_sector);
1335 dev_err(&btt->nd_btt->dev,
1336 "io error in %s sector %lld, len %d,\n",
1337 (op_is_write(bio_op(bio))) ? "WRITE" :
1339 (unsigned long long) iter.bi_sector, len);
1340 bio->bi_status = errno_to_blk_status(err);
1345 nd_iostat_end(bio, start);
1348 return BLK_QC_T_NONE;
1351 static int btt_rw_page(struct block_device *bdev, sector_t sector,
1352 struct page *page, bool is_write)
1354 struct btt *btt = bdev->bd_disk->private_data;
1358 len = hpage_nr_pages(page) * PAGE_SIZE;
1359 rc = btt_do_bvec(btt, NULL, page, len, 0, is_write, sector);
1361 page_endio(page, is_write, 0);
1367 static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1369 /* some standard values */
1370 geo->heads = 1 << 6;
1371 geo->sectors = 1 << 5;
1372 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1376 static const struct block_device_operations btt_fops = {
1377 .owner = THIS_MODULE,
1378 .rw_page = btt_rw_page,
1379 .getgeo = btt_getgeo,
1380 .revalidate_disk = nvdimm_revalidate_disk,
1383 static int btt_blk_init(struct btt *btt)
1385 struct nd_btt *nd_btt = btt->nd_btt;
1386 struct nd_namespace_common *ndns = nd_btt->ndns;
1388 /* create a new disk and request queue for btt */
1389 btt->btt_queue = blk_alloc_queue(GFP_KERNEL);
1390 if (!btt->btt_queue)
1393 btt->btt_disk = alloc_disk(0);
1394 if (!btt->btt_disk) {
1395 blk_cleanup_queue(btt->btt_queue);
1399 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1400 btt->btt_disk->first_minor = 0;
1401 btt->btt_disk->fops = &btt_fops;
1402 btt->btt_disk->private_data = btt;
1403 btt->btt_disk->queue = btt->btt_queue;
1404 btt->btt_disk->flags = GENHD_FL_EXT_DEVT;
1406 blk_queue_make_request(btt->btt_queue, btt_make_request);
1407 blk_queue_logical_block_size(btt->btt_queue, btt->sector_size);
1408 blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX);
1409 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, btt->btt_queue);
1410 btt->btt_queue->queuedata = btt;
1412 set_capacity(btt->btt_disk, 0);
1413 device_add_disk(&btt->nd_btt->dev, btt->btt_disk);
1414 if (btt_meta_size(btt)) {
1415 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1418 del_gendisk(btt->btt_disk);
1419 put_disk(btt->btt_disk);
1420 blk_cleanup_queue(btt->btt_queue);
1424 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1425 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1426 revalidate_disk(btt->btt_disk);
1431 static void btt_blk_cleanup(struct btt *btt)
1433 del_gendisk(btt->btt_disk);
1434 put_disk(btt->btt_disk);
1435 blk_cleanup_queue(btt->btt_queue);
1439 * btt_init - initialize a block translation table for the given device
1440 * @nd_btt: device with BTT geometry and backing device info
1441 * @rawsize: raw size in bytes of the backing device
1442 * @lbasize: lba size of the backing device
1443 * @uuid: A uuid for the backing device - this is stored on media
1444 * @maxlane: maximum number of parallel requests the device can handle
1446 * Initialize a Block Translation Table on a backing device to provide
1447 * single sector power fail atomicity.
1453 * Pointer to a new struct btt on success, NULL on failure.
1455 static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1456 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1460 struct nd_namespace_io *nsio;
1461 struct device *dev = &nd_btt->dev;
1463 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1467 btt->nd_btt = nd_btt;
1468 btt->rawsize = rawsize;
1469 btt->lbasize = lbasize;
1470 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1471 INIT_LIST_HEAD(&btt->arena_list);
1472 mutex_init(&btt->init_lock);
1473 btt->nd_region = nd_region;
1474 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1475 btt->phys_bb = &nsio->bb;
1477 ret = discover_arenas(btt);
1479 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1483 if (btt->init_state != INIT_READY && nd_region->ro) {
1484 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1485 dev_name(&nd_region->dev));
1487 } else if (btt->init_state != INIT_READY) {
1488 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1489 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1490 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1491 btt->num_arenas, rawsize);
1493 ret = create_arenas(btt);
1495 dev_info(dev, "init: create_arenas: %d\n", ret);
1499 ret = btt_meta_init(btt);
1501 dev_err(dev, "init: error in meta_init: %d\n", ret);
1506 ret = btt_blk_init(btt);
1508 dev_err(dev, "init: error in blk_init: %d\n", ret);
1512 btt_debugfs_init(btt);
1518 * btt_fini - de-initialize a BTT
1519 * @btt: the BTT handle that was generated by btt_init
1521 * De-initialize a Block Translation Table on device removal
1526 static void btt_fini(struct btt *btt)
1529 btt_blk_cleanup(btt);
1531 debugfs_remove_recursive(btt->debugfs_dir);
1535 int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1537 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1538 struct nd_region *nd_region;
1539 struct btt_sb *btt_sb;
1543 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1544 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1548 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1553 * If this returns < 0, that is ok as it just means there wasn't
1554 * an existing BTT, and we're creating a new one. We still need to
1555 * call this as we need the version dependent fields in nd_btt to be
1556 * set correctly based on the holder class
1558 nd_btt_version(nd_btt, ndns, btt_sb);
1560 rawsize = nvdimm_namespace_capacity(ndns) - nd_btt->initial_offset;
1561 if (rawsize < ARENA_MIN_SIZE) {
1562 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1563 dev_name(&ndns->dev),
1564 ARENA_MIN_SIZE + nd_btt->initial_offset);
1567 nd_region = to_nd_region(nd_btt->dev.parent);
1568 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1576 EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1578 int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1580 struct btt *btt = nd_btt->btt;
1587 EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1589 static int __init nd_btt_init(void)
1593 debugfs_root = debugfs_create_dir("btt", NULL);
1594 if (IS_ERR_OR_NULL(debugfs_root))
1600 static void __exit nd_btt_exit(void)
1602 debugfs_remove_recursive(debugfs_root);
1605 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1607 MODULE_LICENSE("GPL v2");
1608 module_init(nd_btt_init);
1609 module_exit(nd_btt_exit);