1 // SPDX-License-Identifier: GPL-2.0-only
3 * Block Translation Table
4 * Copyright (c) 2014-2015, Intel Corporation.
6 #include <linux/highmem.h>
7 #include <linux/debugfs.h>
8 #include <linux/blkdev.h>
9 #include <linux/pagemap.h>
10 #include <linux/module.h>
11 #include <linux/device.h>
12 #include <linux/mutex.h>
13 #include <linux/hdreg.h>
14 #include <linux/sizes.h>
15 #include <linux/ndctl.h>
18 #include <linux/backing-dev.h>
22 enum log_ent_request {
27 static struct device *to_dev(struct arena_info *arena)
29 return &arena->nd_btt->dev;
32 static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
34 return offset + nd_btt->initial_offset;
37 static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
38 void *buf, size_t n, unsigned long flags)
40 struct nd_btt *nd_btt = arena->nd_btt;
41 struct nd_namespace_common *ndns = nd_btt->ndns;
43 /* arena offsets may be shifted from the base of the device */
44 offset = adjust_initial_offset(nd_btt, offset);
45 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
48 static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
49 void *buf, size_t n, unsigned long flags)
51 struct nd_btt *nd_btt = arena->nd_btt;
52 struct nd_namespace_common *ndns = nd_btt->ndns;
54 /* arena offsets may be shifted from the base of the device */
55 offset = adjust_initial_offset(nd_btt, offset);
56 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
59 static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
64 * infooff and info2off should always be at least 512B aligned.
65 * We rely on that to make sure rw_bytes does error clearing
66 * correctly, so make sure that is the case.
68 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
69 "arena->infooff: %#llx is unaligned\n", arena->infooff);
70 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
71 "arena->info2off: %#llx is unaligned\n", arena->info2off);
73 ret = arena_write_bytes(arena, arena->info2off, super,
74 sizeof(struct btt_sb), 0);
78 return arena_write_bytes(arena, arena->infooff, super,
79 sizeof(struct btt_sb), 0);
82 static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
84 return arena_read_bytes(arena, arena->infooff, super,
85 sizeof(struct btt_sb), 0);
89 * 'raw' version of btt_map write
91 * mapping is in little-endian
92 * mapping contains 'E' and 'Z' flags as desired
94 static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
97 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
99 if (unlikely(lba >= arena->external_nlba))
100 dev_err_ratelimited(to_dev(arena),
101 "%s: lba %#x out of range (max: %#x)\n",
102 __func__, lba, arena->external_nlba);
103 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
106 static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
107 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
113 * This 'mapping' is supposed to be just the LBA mapping, without
114 * any flags set, so strip the flag bits.
116 mapping = ent_lba(mapping);
118 ze = (z_flag << 1) + e_flag;
122 * We want to set neither of the Z or E flags, and
123 * in the actual layout, this means setting the bit
124 * positions of both to '1' to indicate a 'normal'
127 mapping |= MAP_ENT_NORMAL;
130 mapping |= (1 << MAP_ERR_SHIFT);
133 mapping |= (1 << MAP_TRIM_SHIFT);
137 * The case where Z and E are both sent in as '1' could be
138 * construed as a valid 'normal' case, but we decide not to,
141 dev_err_ratelimited(to_dev(arena),
142 "Invalid use of Z and E flags\n");
146 mapping_le = cpu_to_le32(mapping);
147 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
150 static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
151 int *trim, int *error, unsigned long rwb_flags)
155 u32 raw_mapping, postmap, ze, z_flag, e_flag;
156 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
158 if (unlikely(lba >= arena->external_nlba))
159 dev_err_ratelimited(to_dev(arena),
160 "%s: lba %#x out of range (max: %#x)\n",
161 __func__, lba, arena->external_nlba);
163 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
167 raw_mapping = le32_to_cpu(in);
169 z_flag = ent_z_flag(raw_mapping);
170 e_flag = ent_e_flag(raw_mapping);
171 ze = (z_flag << 1) + e_flag;
172 postmap = ent_lba(raw_mapping);
174 /* Reuse the {z,e}_flag variables for *trim and *error */
180 /* Initial state. Return postmap = premap */
206 static int btt_log_group_read(struct arena_info *arena, u32 lane,
207 struct log_group *log)
209 return arena_read_bytes(arena,
210 arena->logoff + (lane * LOG_GRP_SIZE), log,
214 static struct dentry *debugfs_root;
216 static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
222 /* If for some reason, parent bttN was not created, exit */
226 snprintf(dirname, 32, "arena%d", idx);
227 d = debugfs_create_dir(dirname, parent);
228 if (IS_ERR_OR_NULL(d))
232 debugfs_create_x64("size", S_IRUGO, d, &a->size);
233 debugfs_create_x64("external_lba_start", S_IRUGO, d,
234 &a->external_lba_start);
235 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
236 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
237 &a->internal_lbasize);
238 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
239 debugfs_create_u32("external_lbasize", S_IRUGO, d,
240 &a->external_lbasize);
241 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
242 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
243 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
244 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
245 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
246 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
247 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
248 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
249 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
250 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
251 debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]);
252 debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]);
255 static void btt_debugfs_init(struct btt *btt)
258 struct arena_info *arena;
260 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
262 if (IS_ERR_OR_NULL(btt->debugfs_dir))
265 list_for_each_entry(arena, &btt->arena_list, list) {
266 arena_debugfs_init(arena, btt->debugfs_dir, i);
271 static u32 log_seq(struct log_group *log, int log_idx)
273 return le32_to_cpu(log->ent[log_idx].seq);
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 arena_info *a, struct log_group *log)
287 int idx0 = a->log_index[0];
288 int idx1 = a->log_index[1];
292 * the first ever time this is seen, the entry goes into [0]
293 * the next time, the following logic works out to put this
294 * (next) entry into [1]
296 if (log_seq(log, idx0) == 0) {
297 log->ent[idx0].seq = cpu_to_le32(1);
301 if (log_seq(log, idx0) == log_seq(log, idx1))
303 if (log_seq(log, idx0) + log_seq(log, idx1) > 5)
306 if (log_seq(log, idx0) < log_seq(log, idx1)) {
307 if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1)
312 if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1)
322 * This function copies the desired (old/new) log entry into ent if
323 * it is not NULL. It returns the sub-slot number (0 or 1)
324 * where the desired log entry was found. Negative return values
327 static int btt_log_read(struct arena_info *arena, u32 lane,
328 struct log_entry *ent, int old_flag)
331 int old_ent, ret_ent;
332 struct log_group log;
334 ret = btt_log_group_read(arena, lane, &log);
338 old_ent = btt_log_get_old(arena, &log);
339 if (old_ent < 0 || old_ent > 1) {
340 dev_err(to_dev(arena),
341 "log corruption (%d): lane %d seq [%d, %d]\n",
342 old_ent, lane, log.ent[arena->log_index[0]].seq,
343 log.ent[arena->log_index[1]].seq);
344 /* TODO set error state? */
348 ret_ent = (old_flag ? old_ent : (1 - old_ent));
351 memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE);
357 * This function commits a log entry to media
358 * It does _not_ prepare the freelist entry for the next write
359 * btt_flog_write is the wrapper for updating the freelist elements
361 static int __btt_log_write(struct arena_info *arena, u32 lane,
362 u32 sub, struct log_entry *ent, unsigned long flags)
365 u32 group_slot = arena->log_index[sub];
366 unsigned int log_half = LOG_ENT_SIZE / 2;
370 ns_off = arena->logoff + (lane * LOG_GRP_SIZE) +
371 (group_slot * LOG_ENT_SIZE);
372 /* split the 16B write into atomic, durable halves */
373 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
379 return arena_write_bytes(arena, ns_off, src, log_half, flags);
382 static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
383 struct log_entry *ent)
387 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
391 /* prepare the next free entry */
392 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
393 if (++(arena->freelist[lane].seq) == 4)
394 arena->freelist[lane].seq = 1;
395 if (ent_e_flag(le32_to_cpu(ent->old_map)))
396 arena->freelist[lane].has_err = 1;
397 arena->freelist[lane].block = ent_lba(le32_to_cpu(ent->old_map));
403 * This function initializes the BTT map to the initial state, which is
404 * all-zeroes, and indicates an identity mapping
406 static int btt_map_init(struct arena_info *arena)
411 size_t chunk_size = SZ_2M;
412 size_t mapsize = arena->logoff - arena->mapoff;
414 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
419 * mapoff should always be at least 512B aligned. We rely on that to
420 * make sure rw_bytes does error clearing correctly, so make sure that
423 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
424 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
427 size_t size = min(mapsize, chunk_size);
429 dev_WARN_ONCE(to_dev(arena), size < 512,
430 "chunk size: %#zx is unaligned\n", size);
431 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
447 * This function initializes the BTT log with 'fake' entries pointing
448 * to the initial reserved set of blocks as being free
450 static int btt_log_init(struct arena_info *arena)
452 size_t logsize = arena->info2off - arena->logoff;
453 size_t chunk_size = SZ_4K, offset = 0;
454 struct log_entry ent;
459 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
463 * logoff should always be at least 512B aligned. We rely on that to
464 * make sure rw_bytes does error clearing correctly, so make sure that
467 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
468 "arena->logoff: %#llx is unaligned\n", arena->logoff);
471 size_t size = min(logsize, chunk_size);
473 dev_WARN_ONCE(to_dev(arena), size < 512,
474 "chunk size: %#zx is unaligned\n", size);
475 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
485 for (i = 0; i < arena->nfree; i++) {
486 ent.lba = cpu_to_le32(i);
487 ent.old_map = cpu_to_le32(arena->external_nlba + i);
488 ent.new_map = cpu_to_le32(arena->external_nlba + i);
489 ent.seq = cpu_to_le32(LOG_SEQ_INIT);
490 ret = __btt_log_write(arena, i, 0, &ent, 0);
500 static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
502 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
505 static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
509 if (arena->freelist[lane].has_err) {
510 void *zero_page = page_address(ZERO_PAGE(0));
511 u32 lba = arena->freelist[lane].block;
512 u64 nsoff = to_namespace_offset(arena, lba);
513 unsigned long len = arena->sector_size;
515 mutex_lock(&arena->err_lock);
518 unsigned long chunk = min(len, PAGE_SIZE);
520 ret = arena_write_bytes(arena, nsoff, zero_page,
527 arena->freelist[lane].has_err = 0;
529 mutex_unlock(&arena->err_lock);
534 static int btt_freelist_init(struct arena_info *arena)
537 struct log_entry log_new;
538 u32 i, map_entry, log_oldmap, log_newmap;
540 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
542 if (!arena->freelist)
545 for (i = 0; i < arena->nfree; i++) {
546 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
550 /* old and new map entries with any flags stripped out */
551 log_oldmap = ent_lba(le32_to_cpu(log_new.old_map));
552 log_newmap = ent_lba(le32_to_cpu(log_new.new_map));
554 /* sub points to the next one to be overwritten */
555 arena->freelist[i].sub = 1 - new;
556 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
557 arena->freelist[i].block = log_oldmap;
560 * FIXME: if error clearing fails during init, we want to make
563 if (ent_e_flag(le32_to_cpu(log_new.old_map)) &&
564 !ent_normal(le32_to_cpu(log_new.old_map))) {
565 arena->freelist[i].has_err = 1;
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_oldmap == log_newmap)
576 /* Check if map recovery is needed */
577 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
583 * The map_entry from btt_read_map is stripped of any flag bits,
584 * so use the stripped out versions from the log as well for
585 * testing whether recovery is needed. For restoration, use the
586 * 'raw' version of the log entries as that captured what we
587 * were going to write originally.
589 if ((log_newmap != map_entry) && (log_oldmap == map_entry)) {
591 * Last transaction wrote the flog, but wasn't able
592 * to complete the map write. So fix up the map.
594 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
595 le32_to_cpu(log_new.new_map), 0, 0, 0);
604 static bool ent_is_padding(struct log_entry *ent)
606 return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0)
611 * Detecting valid log indices: We read a log group (see the comments in btt.h
612 * for a description of a 'log_group' and its 'slots'), and iterate over its
613 * four slots. We expect that a padding slot will be all-zeroes, and use this
614 * to detect a padding slot vs. an actual entry.
616 * If a log_group is in the initial state, i.e. hasn't been used since the
617 * creation of this BTT layout, it will have three of the four slots with
618 * zeroes. We skip over these log_groups for the detection of log_index. If
619 * all log_groups are in the initial state (i.e. the BTT has never been
620 * written to), it is safe to assume the 'new format' of log entries in slots
623 static int log_set_indices(struct arena_info *arena)
625 bool idx_set = false, initial_state = true;
626 int ret, log_index[2] = {-1, -1};
627 u32 i, j, next_idx = 0;
628 struct log_group log;
631 for (i = 0; i < arena->nfree; i++) {
632 ret = btt_log_group_read(arena, i, &log);
636 for (j = 0; j < 4; j++) {
638 if (ent_is_padding(&log.ent[j])) {
642 /* Skip if index has been recorded */
643 if ((next_idx == 1) &&
646 /* valid entry, record index */
647 log_index[next_idx] = j;
651 /* two valid entries found */
653 } else if (next_idx > 2) {
654 /* too many valid indices */
659 * once the indices have been set, just verify
660 * that all subsequent log groups are either in
661 * their initial state or follow the same
664 if (j == log_index[0]) {
665 /* entry must be 'valid' */
666 if (ent_is_padding(&log.ent[j]))
668 } else if (j == log_index[1]) {
671 * log_index[1] can be padding if the
672 * lane never got used and it is still
673 * in the initial state (three 'padding'
677 /* entry must be invalid (padding) */
678 if (!ent_is_padding(&log.ent[j]))
684 * If any of the log_groups have more than one valid,
685 * non-padding entry, then the we are no longer in the
689 initial_state = false;
693 if (!initial_state && !idx_set)
697 * If all the entries in the log were in the initial state,
698 * assume new padding scheme
704 * Only allow the known permutations of log/padding indices,
705 * i.e. (0, 1), and (0, 2)
707 if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2)))
708 ; /* known index possibilities */
710 dev_err(to_dev(arena), "Found an unknown padding scheme\n");
714 arena->log_index[0] = log_index[0];
715 arena->log_index[1] = log_index[1];
716 dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]);
717 dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]);
721 static int btt_rtt_init(struct arena_info *arena)
723 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
724 if (arena->rtt == NULL)
730 static int btt_maplocks_init(struct arena_info *arena)
734 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
736 if (!arena->map_locks)
739 for (i = 0; i < arena->nfree; i++)
740 spin_lock_init(&arena->map_locks[i].lock);
745 static struct arena_info *alloc_arena(struct btt *btt, size_t size,
746 size_t start, size_t arena_off)
748 struct arena_info *arena;
749 u64 logsize, mapsize, datasize;
750 u64 available = size;
752 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
755 arena->nd_btt = btt->nd_btt;
756 arena->sector_size = btt->sector_size;
757 mutex_init(&arena->err_lock);
763 arena->external_lba_start = start;
764 arena->external_lbasize = btt->lbasize;
765 arena->internal_lbasize = roundup(arena->external_lbasize,
766 INT_LBASIZE_ALIGNMENT);
767 arena->nfree = BTT_DEFAULT_NFREE;
768 arena->version_major = btt->nd_btt->version_major;
769 arena->version_minor = btt->nd_btt->version_minor;
771 if (available % BTT_PG_SIZE)
772 available -= (available % BTT_PG_SIZE);
774 /* Two pages are reserved for the super block and its copy */
775 available -= 2 * BTT_PG_SIZE;
777 /* The log takes a fixed amount of space based on nfree */
778 logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE);
779 available -= logsize;
781 /* Calculate optimal split between map and data area */
782 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
783 arena->internal_lbasize + MAP_ENT_SIZE);
784 arena->external_nlba = arena->internal_nlba - arena->nfree;
786 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
787 datasize = available - mapsize;
789 /* 'Absolute' values, relative to start of storage space */
790 arena->infooff = arena_off;
791 arena->dataoff = arena->infooff + BTT_PG_SIZE;
792 arena->mapoff = arena->dataoff + datasize;
793 arena->logoff = arena->mapoff + mapsize;
794 arena->info2off = arena->logoff + logsize;
796 /* Default log indices are (0,1) */
797 arena->log_index[0] = 0;
798 arena->log_index[1] = 1;
802 static void free_arenas(struct btt *btt)
804 struct arena_info *arena, *next;
806 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
807 list_del(&arena->list);
809 kfree(arena->map_locks);
810 kfree(arena->freelist);
811 debugfs_remove_recursive(arena->debugfs_dir);
817 * This function reads an existing valid btt superblock and
818 * populates the corresponding arena_info struct
820 static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
823 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
824 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
825 arena->external_nlba = le32_to_cpu(super->external_nlba);
826 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
827 arena->nfree = le32_to_cpu(super->nfree);
828 arena->version_major = le16_to_cpu(super->version_major);
829 arena->version_minor = le16_to_cpu(super->version_minor);
831 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
832 le64_to_cpu(super->nextoff));
833 arena->infooff = arena_off;
834 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
835 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
836 arena->logoff = arena_off + le64_to_cpu(super->logoff);
837 arena->info2off = arena_off + le64_to_cpu(super->info2off);
839 arena->size = (le64_to_cpu(super->nextoff) > 0)
840 ? (le64_to_cpu(super->nextoff))
841 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
843 arena->flags = le32_to_cpu(super->flags);
846 static int discover_arenas(struct btt *btt)
849 struct arena_info *arena;
850 struct btt_sb *super;
851 size_t remaining = btt->rawsize;
856 super = kzalloc(sizeof(*super), GFP_KERNEL);
861 /* Alloc memory for arena */
862 arena = alloc_arena(btt, 0, 0, 0);
868 arena->infooff = cur_off;
869 ret = btt_info_read(arena, super);
873 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
874 if (remaining == btt->rawsize) {
875 btt->init_state = INIT_NOTFOUND;
876 dev_info(to_dev(arena), "No existing arenas\n");
879 dev_err(to_dev(arena),
880 "Found corrupted metadata!\n");
886 arena->external_lba_start = cur_nlba;
887 parse_arena_meta(arena, super, cur_off);
889 ret = log_set_indices(arena);
891 dev_err(to_dev(arena),
892 "Unable to deduce log/padding indices\n");
896 ret = btt_freelist_init(arena);
900 ret = btt_rtt_init(arena);
904 ret = btt_maplocks_init(arena);
908 list_add_tail(&arena->list, &btt->arena_list);
910 remaining -= arena->size;
911 cur_off += arena->size;
912 cur_nlba += arena->external_nlba;
915 if (arena->nextoff == 0)
918 btt->num_arenas = num_arenas;
919 btt->nlba = cur_nlba;
920 btt->init_state = INIT_READY;
933 static int create_arenas(struct btt *btt)
935 size_t remaining = btt->rawsize;
939 struct arena_info *arena;
940 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
942 remaining -= arena_size;
943 if (arena_size < ARENA_MIN_SIZE)
946 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
951 btt->nlba += arena->external_nlba;
952 if (remaining >= ARENA_MIN_SIZE)
953 arena->nextoff = arena->size;
956 cur_off += arena_size;
957 list_add_tail(&arena->list, &btt->arena_list);
964 * This function completes arena initialization by writing
966 * It is only called for an uninitialized arena when a write
967 * to that arena occurs for the first time.
969 static int btt_arena_write_layout(struct arena_info *arena)
973 struct btt_sb *super;
974 struct nd_btt *nd_btt = arena->nd_btt;
975 const uuid_t *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
977 ret = btt_map_init(arena);
981 ret = btt_log_init(arena);
985 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
989 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
990 export_uuid(super->uuid, nd_btt->uuid);
991 export_uuid(super->parent_uuid, parent_uuid);
992 super->flags = cpu_to_le32(arena->flags);
993 super->version_major = cpu_to_le16(arena->version_major);
994 super->version_minor = cpu_to_le16(arena->version_minor);
995 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
996 super->external_nlba = cpu_to_le32(arena->external_nlba);
997 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
998 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
999 super->nfree = cpu_to_le32(arena->nfree);
1000 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
1001 super->nextoff = cpu_to_le64(arena->nextoff);
1003 * Subtract arena->infooff (arena start) so numbers are relative
1006 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
1007 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
1008 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
1009 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
1012 sum = nd_sb_checksum((struct nd_gen_sb *) super);
1013 super->checksum = cpu_to_le64(sum);
1015 ret = btt_info_write(arena, super);
1022 * This function completes the initialization for the BTT namespace
1023 * such that it is ready to accept IOs
1025 static int btt_meta_init(struct btt *btt)
1028 struct arena_info *arena;
1030 mutex_lock(&btt->init_lock);
1031 list_for_each_entry(arena, &btt->arena_list, list) {
1032 ret = btt_arena_write_layout(arena);
1036 ret = btt_freelist_init(arena);
1040 ret = btt_rtt_init(arena);
1044 ret = btt_maplocks_init(arena);
1049 btt->init_state = INIT_READY;
1052 mutex_unlock(&btt->init_lock);
1056 static u32 btt_meta_size(struct btt *btt)
1058 return btt->lbasize - btt->sector_size;
1062 * This function calculates the arena in which the given LBA lies
1063 * by doing a linear walk. This is acceptable since we expect only
1064 * a few arenas. If we have backing devices that get much larger,
1065 * we can construct a balanced binary tree of arenas at init time
1066 * so that this range search becomes faster.
1068 static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
1069 struct arena_info **arena)
1071 struct arena_info *arena_list;
1072 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
1074 list_for_each_entry(arena_list, &btt->arena_list, list) {
1075 if (lba < arena_list->external_nlba) {
1076 *arena = arena_list;
1080 lba -= arena_list->external_nlba;
1087 * The following (lock_map, unlock_map) are mostly just to improve
1088 * readability, since they index into an array of locks
1090 static void lock_map(struct arena_info *arena, u32 premap)
1091 __acquires(&arena->map_locks[idx].lock)
1093 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1095 spin_lock(&arena->map_locks[idx].lock);
1098 static void unlock_map(struct arena_info *arena, u32 premap)
1099 __releases(&arena->map_locks[idx].lock)
1101 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1103 spin_unlock(&arena->map_locks[idx].lock);
1106 static int btt_data_read(struct arena_info *arena, struct page *page,
1107 unsigned int off, u32 lba, u32 len)
1110 u64 nsoff = to_namespace_offset(arena, lba);
1111 void *mem = kmap_atomic(page);
1113 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1119 static int btt_data_write(struct arena_info *arena, u32 lba,
1120 struct page *page, unsigned int off, u32 len)
1123 u64 nsoff = to_namespace_offset(arena, lba);
1124 void *mem = kmap_atomic(page);
1126 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1132 static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1134 void *mem = kmap_atomic(page);
1136 memset(mem + off, 0, len);
1140 #ifdef CONFIG_BLK_DEV_INTEGRITY
1141 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1142 struct arena_info *arena, u32 postmap, int rw)
1144 unsigned int len = btt_meta_size(btt);
1151 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1154 unsigned int cur_len;
1158 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1160 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1161 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1162 * can use those directly
1165 cur_len = min(len, bv.bv_len);
1166 mem = bvec_kmap_local(&bv);
1168 ret = arena_write_bytes(arena, meta_nsoff, mem, cur_len,
1171 ret = arena_read_bytes(arena, meta_nsoff, mem, cur_len,
1179 meta_nsoff += cur_len;
1180 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1187 #else /* CONFIG_BLK_DEV_INTEGRITY */
1188 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1189 struct arena_info *arena, u32 postmap, int rw)
1195 static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1196 struct page *page, unsigned int off, sector_t sector,
1201 struct arena_info *arena = NULL;
1202 u32 lane = 0, premap, postmap;
1207 lane = nd_region_acquire_lane(btt->nd_region);
1209 ret = lba_to_arena(btt, sector, &premap, &arena);
1213 cur_len = min(btt->sector_size, len);
1215 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1221 * We loop to make sure that the post map LBA didn't change
1222 * from under us between writing the RTT and doing the actual
1230 zero_fill_data(page, off, cur_len);
1239 arena->rtt[lane] = RTT_VALID | postmap;
1241 * Barrier to make sure this write is not reordered
1242 * to do the verification map_read before the RTT store
1246 ret = btt_map_read(arena, premap, &new_map, &new_t,
1247 &new_e, NVDIMM_IO_ATOMIC);
1251 if ((postmap == new_map) && (t_flag == new_t) &&
1260 ret = btt_data_read(arena, page, off, postmap, cur_len);
1262 /* Media error - set the e_flag */
1263 if (btt_map_write(arena, premap, postmap, 0, 1, NVDIMM_IO_ATOMIC))
1264 dev_warn_ratelimited(to_dev(arena),
1265 "Error persistently tracking bad blocks at %#x\n",
1271 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1276 arena->rtt[lane] = RTT_INVALID;
1277 nd_region_release_lane(btt->nd_region, lane);
1281 sector += btt->sector_size >> SECTOR_SHIFT;
1287 arena->rtt[lane] = RTT_INVALID;
1289 nd_region_release_lane(btt->nd_region, lane);
1294 * Normally, arena_{read,write}_bytes will take care of the initial offset
1295 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1296 * we need the final, raw namespace offset here
1298 static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1301 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1302 to_namespace_offset(arena, postmap));
1303 sector_t phys_sector = nsoff >> 9;
1305 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1308 static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1309 sector_t sector, struct page *page, unsigned int off,
1313 struct arena_info *arena = NULL;
1314 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1315 struct log_entry log;
1323 lane = nd_region_acquire_lane(btt->nd_region);
1325 ret = lba_to_arena(btt, sector, &premap, &arena);
1328 cur_len = min(btt->sector_size, len);
1330 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1335 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1336 arena->freelist[lane].has_err = 1;
1338 if (mutex_is_locked(&arena->err_lock)
1339 || arena->freelist[lane].has_err) {
1340 nd_region_release_lane(btt->nd_region, lane);
1342 ret = arena_clear_freelist_error(arena, lane);
1346 /* OK to acquire a different lane/free block */
1350 new_postmap = arena->freelist[lane].block;
1352 /* Wait if the new block is being read from */
1353 for (i = 0; i < arena->nfree; i++)
1354 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1358 if (new_postmap >= arena->internal_nlba) {
1363 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1368 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1374 lock_map(arena, premap);
1375 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1379 if (old_postmap >= arena->internal_nlba) {
1384 set_e_flag(old_postmap);
1386 log.lba = cpu_to_le32(premap);
1387 log.old_map = cpu_to_le32(old_postmap);
1388 log.new_map = cpu_to_le32(new_postmap);
1389 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1390 sub = arena->freelist[lane].sub;
1391 ret = btt_flog_write(arena, lane, sub, &log);
1395 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1400 unlock_map(arena, premap);
1401 nd_region_release_lane(btt->nd_region, lane);
1404 ret = arena_clear_freelist_error(arena, lane);
1411 sector += btt->sector_size >> SECTOR_SHIFT;
1417 unlock_map(arena, premap);
1419 nd_region_release_lane(btt->nd_region, lane);
1423 static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1424 struct page *page, unsigned int len, unsigned int off,
1425 unsigned int op, sector_t sector)
1429 if (!op_is_write(op)) {
1430 ret = btt_read_pg(btt, bip, page, off, sector, len);
1431 flush_dcache_page(page);
1433 flush_dcache_page(page);
1434 ret = btt_write_pg(btt, bip, sector, page, off, len);
1440 static void btt_submit_bio(struct bio *bio)
1442 struct bio_integrity_payload *bip = bio_integrity(bio);
1443 struct btt *btt = bio->bi_bdev->bd_disk->private_data;
1444 struct bvec_iter iter;
1445 unsigned long start;
1446 struct bio_vec bvec;
1450 if (!bio_integrity_prep(bio))
1453 do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
1455 start = bio_start_io_acct(bio);
1456 bio_for_each_segment(bvec, bio, iter) {
1457 unsigned int len = bvec.bv_len;
1459 if (len > PAGE_SIZE || len < btt->sector_size ||
1460 len % btt->sector_size) {
1461 dev_err_ratelimited(&btt->nd_btt->dev,
1462 "unaligned bio segment (len: %d)\n", len);
1463 bio->bi_status = BLK_STS_IOERR;
1467 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1468 bio_op(bio), iter.bi_sector);
1470 dev_err(&btt->nd_btt->dev,
1471 "io error in %s sector %lld, len %d,\n",
1472 (op_is_write(bio_op(bio))) ? "WRITE" :
1474 (unsigned long long) iter.bi_sector, len);
1475 bio->bi_status = errno_to_blk_status(err);
1480 bio_end_io_acct(bio, start);
1485 static int btt_rw_page(struct block_device *bdev, sector_t sector,
1486 struct page *page, unsigned int op)
1488 struct btt *btt = bdev->bd_disk->private_data;
1491 rc = btt_do_bvec(btt, NULL, page, thp_size(page), 0, op, sector);
1493 page_endio(page, op_is_write(op), 0);
1499 static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1501 /* some standard values */
1502 geo->heads = 1 << 6;
1503 geo->sectors = 1 << 5;
1504 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1508 static const struct block_device_operations btt_fops = {
1509 .owner = THIS_MODULE,
1510 .submit_bio = btt_submit_bio,
1511 .rw_page = btt_rw_page,
1512 .getgeo = btt_getgeo,
1515 static int btt_blk_init(struct btt *btt)
1517 struct nd_btt *nd_btt = btt->nd_btt;
1518 struct nd_namespace_common *ndns = nd_btt->ndns;
1521 btt->btt_disk = blk_alloc_disk(NUMA_NO_NODE);
1525 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1526 btt->btt_disk->first_minor = 0;
1527 btt->btt_disk->fops = &btt_fops;
1528 btt->btt_disk->private_data = btt;
1530 blk_queue_logical_block_size(btt->btt_disk->queue, btt->sector_size);
1531 blk_queue_max_hw_sectors(btt->btt_disk->queue, UINT_MAX);
1532 blk_queue_flag_set(QUEUE_FLAG_NONROT, btt->btt_disk->queue);
1534 if (btt_meta_size(btt)) {
1535 rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1537 goto out_cleanup_disk;
1540 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1541 rc = device_add_disk(&btt->nd_btt->dev, btt->btt_disk, NULL);
1543 goto out_cleanup_disk;
1545 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1546 nvdimm_check_and_set_ro(btt->btt_disk);
1551 blk_cleanup_disk(btt->btt_disk);
1555 static void btt_blk_cleanup(struct btt *btt)
1557 del_gendisk(btt->btt_disk);
1558 blk_cleanup_disk(btt->btt_disk);
1562 * btt_init - initialize a block translation table for the given device
1563 * @nd_btt: device with BTT geometry and backing device info
1564 * @rawsize: raw size in bytes of the backing device
1565 * @lbasize: lba size of the backing device
1566 * @uuid: A uuid for the backing device - this is stored on media
1567 * @maxlane: maximum number of parallel requests the device can handle
1569 * Initialize a Block Translation Table on a backing device to provide
1570 * single sector power fail atomicity.
1576 * Pointer to a new struct btt on success, NULL on failure.
1578 static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1579 u32 lbasize, uuid_t *uuid,
1580 struct nd_region *nd_region)
1584 struct nd_namespace_io *nsio;
1585 struct device *dev = &nd_btt->dev;
1587 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1591 btt->nd_btt = nd_btt;
1592 btt->rawsize = rawsize;
1593 btt->lbasize = lbasize;
1594 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1595 INIT_LIST_HEAD(&btt->arena_list);
1596 mutex_init(&btt->init_lock);
1597 btt->nd_region = nd_region;
1598 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1599 btt->phys_bb = &nsio->bb;
1601 ret = discover_arenas(btt);
1603 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1607 if (btt->init_state != INIT_READY && nd_region->ro) {
1608 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1609 dev_name(&nd_region->dev));
1611 } else if (btt->init_state != INIT_READY) {
1612 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1613 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1614 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1615 btt->num_arenas, rawsize);
1617 ret = create_arenas(btt);
1619 dev_info(dev, "init: create_arenas: %d\n", ret);
1623 ret = btt_meta_init(btt);
1625 dev_err(dev, "init: error in meta_init: %d\n", ret);
1630 ret = btt_blk_init(btt);
1632 dev_err(dev, "init: error in blk_init: %d\n", ret);
1636 btt_debugfs_init(btt);
1642 * btt_fini - de-initialize a BTT
1643 * @btt: the BTT handle that was generated by btt_init
1645 * De-initialize a Block Translation Table on device removal
1650 static void btt_fini(struct btt *btt)
1653 btt_blk_cleanup(btt);
1655 debugfs_remove_recursive(btt->debugfs_dir);
1659 int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1661 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1662 struct nd_region *nd_region;
1663 struct btt_sb *btt_sb;
1665 size_t size, rawsize;
1668 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1669 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1673 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1677 size = nvdimm_namespace_capacity(ndns);
1678 rc = devm_namespace_enable(&nd_btt->dev, ndns, size);
1683 * If this returns < 0, that is ok as it just means there wasn't
1684 * an existing BTT, and we're creating a new one. We still need to
1685 * call this as we need the version dependent fields in nd_btt to be
1686 * set correctly based on the holder class
1688 nd_btt_version(nd_btt, ndns, btt_sb);
1690 rawsize = size - nd_btt->initial_offset;
1691 if (rawsize < ARENA_MIN_SIZE) {
1692 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1693 dev_name(&ndns->dev),
1694 ARENA_MIN_SIZE + nd_btt->initial_offset);
1697 nd_region = to_nd_region(nd_btt->dev.parent);
1698 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1706 EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1708 int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1710 struct btt *btt = nd_btt->btt;
1717 EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1719 static int __init nd_btt_init(void)
1723 debugfs_root = debugfs_create_dir("btt", NULL);
1724 if (IS_ERR_OR_NULL(debugfs_root))
1730 static void __exit nd_btt_exit(void)
1732 debugfs_remove_recursive(debugfs_root);
1735 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1736 MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1737 MODULE_LICENSE("GPL v2");
1738 module_init(nd_btt_init);
1739 module_exit(nd_btt_exit);