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/genhd.h>
15 #include <linux/sizes.h>
16 #include <linux/ndctl.h>
19 #include <linux/backing-dev.h>
23 enum log_ent_request {
28 static struct device *to_dev(struct arena_info *arena)
30 return &arena->nd_btt->dev;
33 static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
35 return offset + nd_btt->initial_offset;
38 static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
39 void *buf, size_t n, unsigned long flags)
41 struct nd_btt *nd_btt = arena->nd_btt;
42 struct nd_namespace_common *ndns = nd_btt->ndns;
44 /* arena offsets may be shifted from the base of the device */
45 offset = adjust_initial_offset(nd_btt, offset);
46 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
49 static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
50 void *buf, size_t n, unsigned long flags)
52 struct nd_btt *nd_btt = arena->nd_btt;
53 struct nd_namespace_common *ndns = nd_btt->ndns;
55 /* arena offsets may be shifted from the base of the device */
56 offset = adjust_initial_offset(nd_btt, offset);
57 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
60 static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
65 * infooff and info2off should always be at least 512B aligned.
66 * We rely on that to make sure rw_bytes does error clearing
67 * correctly, so make sure that is the case.
69 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
70 "arena->infooff: %#llx is unaligned\n", arena->infooff);
71 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
72 "arena->info2off: %#llx is unaligned\n", arena->info2off);
74 ret = arena_write_bytes(arena, arena->info2off, super,
75 sizeof(struct btt_sb), 0);
79 return arena_write_bytes(arena, arena->infooff, super,
80 sizeof(struct btt_sb), 0);
83 static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
85 return arena_read_bytes(arena, arena->infooff, super,
86 sizeof(struct btt_sb), 0);
90 * 'raw' version of btt_map write
92 * mapping is in little-endian
93 * mapping contains 'E' and 'Z' flags as desired
95 static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
98 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
100 if (unlikely(lba >= arena->external_nlba))
101 dev_err_ratelimited(to_dev(arena),
102 "%s: lba %#x out of range (max: %#x)\n",
103 __func__, lba, arena->external_nlba);
104 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
107 static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
108 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
114 * This 'mapping' is supposed to be just the LBA mapping, without
115 * any flags set, so strip the flag bits.
117 mapping = ent_lba(mapping);
119 ze = (z_flag << 1) + e_flag;
123 * We want to set neither of the Z or E flags, and
124 * in the actual layout, this means setting the bit
125 * positions of both to '1' to indicate a 'normal'
128 mapping |= MAP_ENT_NORMAL;
131 mapping |= (1 << MAP_ERR_SHIFT);
134 mapping |= (1 << MAP_TRIM_SHIFT);
138 * The case where Z and E are both sent in as '1' could be
139 * construed as a valid 'normal' case, but we decide not to,
142 dev_err_ratelimited(to_dev(arena),
143 "Invalid use of Z and E flags\n");
147 mapping_le = cpu_to_le32(mapping);
148 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
151 static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
152 int *trim, int *error, unsigned long rwb_flags)
156 u32 raw_mapping, postmap, ze, z_flag, e_flag;
157 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
159 if (unlikely(lba >= arena->external_nlba))
160 dev_err_ratelimited(to_dev(arena),
161 "%s: lba %#x out of range (max: %#x)\n",
162 __func__, lba, arena->external_nlba);
164 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
168 raw_mapping = le32_to_cpu(in);
170 z_flag = ent_z_flag(raw_mapping);
171 e_flag = ent_e_flag(raw_mapping);
172 ze = (z_flag << 1) + e_flag;
173 postmap = ent_lba(raw_mapping);
175 /* Reuse the {z,e}_flag variables for *trim and *error */
181 /* Initial state. Return postmap = premap */
207 static int btt_log_group_read(struct arena_info *arena, u32 lane,
208 struct log_group *log)
210 return arena_read_bytes(arena,
211 arena->logoff + (lane * LOG_GRP_SIZE), log,
215 static struct dentry *debugfs_root;
217 static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
223 /* If for some reason, parent bttN was not created, exit */
227 snprintf(dirname, 32, "arena%d", idx);
228 d = debugfs_create_dir(dirname, parent);
229 if (IS_ERR_OR_NULL(d))
233 debugfs_create_x64("size", S_IRUGO, d, &a->size);
234 debugfs_create_x64("external_lba_start", S_IRUGO, d,
235 &a->external_lba_start);
236 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
237 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
238 &a->internal_lbasize);
239 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
240 debugfs_create_u32("external_lbasize", S_IRUGO, d,
241 &a->external_lbasize);
242 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
243 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
244 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
245 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
246 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
247 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
248 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
249 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
250 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
251 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
252 debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]);
253 debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]);
256 static void btt_debugfs_init(struct btt *btt)
259 struct arena_info *arena;
261 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
263 if (IS_ERR_OR_NULL(btt->debugfs_dir))
266 list_for_each_entry(arena, &btt->arena_list, list) {
267 arena_debugfs_init(arena, btt->debugfs_dir, i);
272 static u32 log_seq(struct log_group *log, int log_idx)
274 return le32_to_cpu(log->ent[log_idx].seq);
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 arena_info *a, struct log_group *log)
288 int idx0 = a->log_index[0];
289 int idx1 = a->log_index[1];
293 * the first ever time this is seen, the entry goes into [0]
294 * the next time, the following logic works out to put this
295 * (next) entry into [1]
297 if (log_seq(log, idx0) == 0) {
298 log->ent[idx0].seq = cpu_to_le32(1);
302 if (log_seq(log, idx0) == log_seq(log, idx1))
304 if (log_seq(log, idx0) + log_seq(log, idx1) > 5)
307 if (log_seq(log, idx0) < log_seq(log, idx1)) {
308 if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1)
313 if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1)
323 * This function copies the desired (old/new) log entry into ent if
324 * it is not NULL. It returns the sub-slot number (0 or 1)
325 * where the desired log entry was found. Negative return values
328 static int btt_log_read(struct arena_info *arena, u32 lane,
329 struct log_entry *ent, int old_flag)
332 int old_ent, ret_ent;
333 struct log_group log;
335 ret = btt_log_group_read(arena, lane, &log);
339 old_ent = btt_log_get_old(arena, &log);
340 if (old_ent < 0 || old_ent > 1) {
341 dev_err(to_dev(arena),
342 "log corruption (%d): lane %d seq [%d, %d]\n",
343 old_ent, lane, log.ent[arena->log_index[0]].seq,
344 log.ent[arena->log_index[1]].seq);
345 /* TODO set error state? */
349 ret_ent = (old_flag ? old_ent : (1 - old_ent));
352 memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE);
358 * This function commits a log entry to media
359 * It does _not_ prepare the freelist entry for the next write
360 * btt_flog_write is the wrapper for updating the freelist elements
362 static int __btt_log_write(struct arena_info *arena, u32 lane,
363 u32 sub, struct log_entry *ent, unsigned long flags)
366 u32 group_slot = arena->log_index[sub];
367 unsigned int log_half = LOG_ENT_SIZE / 2;
371 ns_off = arena->logoff + (lane * LOG_GRP_SIZE) +
372 (group_slot * 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(le32_to_cpu(ent->old_map)))
397 arena->freelist[lane].has_err = 1;
398 arena->freelist[lane].block = ent_lba(le32_to_cpu(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 ent;
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 ent.lba = cpu_to_le32(i);
488 ent.old_map = cpu_to_le32(arena->external_nlba + i);
489 ent.new_map = cpu_to_le32(arena->external_nlba + i);
490 ent.seq = cpu_to_le32(LOG_SEQ_INIT);
491 ret = __btt_log_write(arena, i, 0, &ent, 0);
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)
538 struct log_entry log_new;
539 u32 i, map_entry, log_oldmap, log_newmap;
541 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
543 if (!arena->freelist)
546 for (i = 0; i < arena->nfree; i++) {
547 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
551 /* old and new map entries with any flags stripped out */
552 log_oldmap = ent_lba(le32_to_cpu(log_new.old_map));
553 log_newmap = ent_lba(le32_to_cpu(log_new.new_map));
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 = log_oldmap;
561 * FIXME: if error clearing fails during init, we want to make
564 if (ent_e_flag(le32_to_cpu(log_new.old_map)) &&
565 !ent_normal(le32_to_cpu(log_new.old_map))) {
566 arena->freelist[i].has_err = 1;
567 ret = arena_clear_freelist_error(arena, i);
569 dev_err_ratelimited(to_dev(arena),
570 "Unable to clear known errors\n");
573 /* This implies a newly created or untouched flog entry */
574 if (log_oldmap == log_newmap)
577 /* Check if map recovery is needed */
578 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
584 * The map_entry from btt_read_map is stripped of any flag bits,
585 * so use the stripped out versions from the log as well for
586 * testing whether recovery is needed. For restoration, use the
587 * 'raw' version of the log entries as that captured what we
588 * were going to write originally.
590 if ((log_newmap != map_entry) && (log_oldmap == map_entry)) {
592 * Last transaction wrote the flog, but wasn't able
593 * to complete the map write. So fix up the map.
595 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
596 le32_to_cpu(log_new.new_map), 0, 0, 0);
605 static bool ent_is_padding(struct log_entry *ent)
607 return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0)
612 * Detecting valid log indices: We read a log group (see the comments in btt.h
613 * for a description of a 'log_group' and its 'slots'), and iterate over its
614 * four slots. We expect that a padding slot will be all-zeroes, and use this
615 * to detect a padding slot vs. an actual entry.
617 * If a log_group is in the initial state, i.e. hasn't been used since the
618 * creation of this BTT layout, it will have three of the four slots with
619 * zeroes. We skip over these log_groups for the detection of log_index. If
620 * all log_groups are in the initial state (i.e. the BTT has never been
621 * written to), it is safe to assume the 'new format' of log entries in slots
624 static int log_set_indices(struct arena_info *arena)
626 bool idx_set = false, initial_state = true;
627 int ret, log_index[2] = {-1, -1};
628 u32 i, j, next_idx = 0;
629 struct log_group log;
632 for (i = 0; i < arena->nfree; i++) {
633 ret = btt_log_group_read(arena, i, &log);
637 for (j = 0; j < 4; j++) {
639 if (ent_is_padding(&log.ent[j])) {
643 /* Skip if index has been recorded */
644 if ((next_idx == 1) &&
647 /* valid entry, record index */
648 log_index[next_idx] = j;
652 /* two valid entries found */
654 } else if (next_idx > 2) {
655 /* too many valid indices */
660 * once the indices have been set, just verify
661 * that all subsequent log groups are either in
662 * their initial state or follow the same
665 if (j == log_index[0]) {
666 /* entry must be 'valid' */
667 if (ent_is_padding(&log.ent[j]))
669 } else if (j == log_index[1]) {
672 * log_index[1] can be padding if the
673 * lane never got used and it is still
674 * in the initial state (three 'padding'
678 /* entry must be invalid (padding) */
679 if (!ent_is_padding(&log.ent[j]))
685 * If any of the log_groups have more than one valid,
686 * non-padding entry, then the we are no longer in the
690 initial_state = false;
694 if (!initial_state && !idx_set)
698 * If all the entries in the log were in the initial state,
699 * assume new padding scheme
705 * Only allow the known permutations of log/padding indices,
706 * i.e. (0, 1), and (0, 2)
708 if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2)))
709 ; /* known index possibilities */
711 dev_err(to_dev(arena), "Found an unknown padding scheme\n");
715 arena->log_index[0] = log_index[0];
716 arena->log_index[1] = log_index[1];
717 dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]);
718 dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]);
722 static int btt_rtt_init(struct arena_info *arena)
724 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
725 if (arena->rtt == NULL)
731 static int btt_maplocks_init(struct arena_info *arena)
735 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
737 if (!arena->map_locks)
740 for (i = 0; i < arena->nfree; i++)
741 spin_lock_init(&arena->map_locks[i].lock);
746 static struct arena_info *alloc_arena(struct btt *btt, size_t size,
747 size_t start, size_t arena_off)
749 struct arena_info *arena;
750 u64 logsize, mapsize, datasize;
751 u64 available = size;
753 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
756 arena->nd_btt = btt->nd_btt;
757 arena->sector_size = btt->sector_size;
758 mutex_init(&arena->err_lock);
764 arena->external_lba_start = start;
765 arena->external_lbasize = btt->lbasize;
766 arena->internal_lbasize = roundup(arena->external_lbasize,
767 INT_LBASIZE_ALIGNMENT);
768 arena->nfree = BTT_DEFAULT_NFREE;
769 arena->version_major = btt->nd_btt->version_major;
770 arena->version_minor = btt->nd_btt->version_minor;
772 if (available % BTT_PG_SIZE)
773 available -= (available % BTT_PG_SIZE);
775 /* Two pages are reserved for the super block and its copy */
776 available -= 2 * BTT_PG_SIZE;
778 /* The log takes a fixed amount of space based on nfree */
779 logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE);
780 available -= logsize;
782 /* Calculate optimal split between map and data area */
783 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
784 arena->internal_lbasize + MAP_ENT_SIZE);
785 arena->external_nlba = arena->internal_nlba - arena->nfree;
787 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
788 datasize = available - mapsize;
790 /* 'Absolute' values, relative to start of storage space */
791 arena->infooff = arena_off;
792 arena->dataoff = arena->infooff + BTT_PG_SIZE;
793 arena->mapoff = arena->dataoff + datasize;
794 arena->logoff = arena->mapoff + mapsize;
795 arena->info2off = arena->logoff + logsize;
797 /* Default log indices are (0,1) */
798 arena->log_index[0] = 0;
799 arena->log_index[1] = 1;
803 static void free_arenas(struct btt *btt)
805 struct arena_info *arena, *next;
807 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
808 list_del(&arena->list);
810 kfree(arena->map_locks);
811 kfree(arena->freelist);
812 debugfs_remove_recursive(arena->debugfs_dir);
818 * This function reads an existing valid btt superblock and
819 * populates the corresponding arena_info struct
821 static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
824 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
825 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
826 arena->external_nlba = le32_to_cpu(super->external_nlba);
827 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
828 arena->nfree = le32_to_cpu(super->nfree);
829 arena->version_major = le16_to_cpu(super->version_major);
830 arena->version_minor = le16_to_cpu(super->version_minor);
832 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
833 le64_to_cpu(super->nextoff));
834 arena->infooff = arena_off;
835 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
836 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
837 arena->logoff = arena_off + le64_to_cpu(super->logoff);
838 arena->info2off = arena_off + le64_to_cpu(super->info2off);
840 arena->size = (le64_to_cpu(super->nextoff) > 0)
841 ? (le64_to_cpu(super->nextoff))
842 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
844 arena->flags = le32_to_cpu(super->flags);
847 static int discover_arenas(struct btt *btt)
850 struct arena_info *arena;
851 struct btt_sb *super;
852 size_t remaining = btt->rawsize;
857 super = kzalloc(sizeof(*super), GFP_KERNEL);
862 /* Alloc memory for arena */
863 arena = alloc_arena(btt, 0, 0, 0);
869 arena->infooff = cur_off;
870 ret = btt_info_read(arena, super);
874 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
875 if (remaining == btt->rawsize) {
876 btt->init_state = INIT_NOTFOUND;
877 dev_info(to_dev(arena), "No existing arenas\n");
880 dev_err(to_dev(arena),
881 "Found corrupted metadata!\n");
887 arena->external_lba_start = cur_nlba;
888 parse_arena_meta(arena, super, cur_off);
890 ret = log_set_indices(arena);
892 dev_err(to_dev(arena),
893 "Unable to deduce log/padding indices\n");
897 ret = btt_freelist_init(arena);
901 ret = btt_rtt_init(arena);
905 ret = btt_maplocks_init(arena);
909 list_add_tail(&arena->list, &btt->arena_list);
911 remaining -= arena->size;
912 cur_off += arena->size;
913 cur_nlba += arena->external_nlba;
916 if (arena->nextoff == 0)
919 btt->num_arenas = num_arenas;
920 btt->nlba = cur_nlba;
921 btt->init_state = INIT_READY;
934 static int create_arenas(struct btt *btt)
936 size_t remaining = btt->rawsize;
940 struct arena_info *arena;
941 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
943 remaining -= arena_size;
944 if (arena_size < ARENA_MIN_SIZE)
947 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
952 btt->nlba += arena->external_nlba;
953 if (remaining >= ARENA_MIN_SIZE)
954 arena->nextoff = arena->size;
957 cur_off += arena_size;
958 list_add_tail(&arena->list, &btt->arena_list);
965 * This function completes arena initialization by writing
967 * It is only called for an uninitialized arena when a write
968 * to that arena occurs for the first time.
970 static int btt_arena_write_layout(struct arena_info *arena)
974 struct btt_sb *super;
975 struct nd_btt *nd_btt = arena->nd_btt;
976 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
978 ret = btt_map_init(arena);
982 ret = btt_log_init(arena);
986 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
990 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
991 memcpy(super->uuid, nd_btt->uuid, 16);
992 memcpy(super->parent_uuid, parent_uuid, 16);
993 super->flags = cpu_to_le32(arena->flags);
994 super->version_major = cpu_to_le16(arena->version_major);
995 super->version_minor = cpu_to_le16(arena->version_minor);
996 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
997 super->external_nlba = cpu_to_le32(arena->external_nlba);
998 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
999 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
1000 super->nfree = cpu_to_le32(arena->nfree);
1001 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
1002 super->nextoff = cpu_to_le64(arena->nextoff);
1004 * Subtract arena->infooff (arena start) so numbers are relative
1007 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
1008 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
1009 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
1010 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
1013 sum = nd_sb_checksum((struct nd_gen_sb *) super);
1014 super->checksum = cpu_to_le64(sum);
1016 ret = btt_info_write(arena, super);
1023 * This function completes the initialization for the BTT namespace
1024 * such that it is ready to accept IOs
1026 static int btt_meta_init(struct btt *btt)
1029 struct arena_info *arena;
1031 mutex_lock(&btt->init_lock);
1032 list_for_each_entry(arena, &btt->arena_list, list) {
1033 ret = btt_arena_write_layout(arena);
1037 ret = btt_freelist_init(arena);
1041 ret = btt_rtt_init(arena);
1045 ret = btt_maplocks_init(arena);
1050 btt->init_state = INIT_READY;
1053 mutex_unlock(&btt->init_lock);
1057 static u32 btt_meta_size(struct btt *btt)
1059 return btt->lbasize - btt->sector_size;
1063 * This function calculates the arena in which the given LBA lies
1064 * by doing a linear walk. This is acceptable since we expect only
1065 * a few arenas. If we have backing devices that get much larger,
1066 * we can construct a balanced binary tree of arenas at init time
1067 * so that this range search becomes faster.
1069 static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
1070 struct arena_info **arena)
1072 struct arena_info *arena_list;
1073 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
1075 list_for_each_entry(arena_list, &btt->arena_list, list) {
1076 if (lba < arena_list->external_nlba) {
1077 *arena = arena_list;
1081 lba -= arena_list->external_nlba;
1088 * The following (lock_map, unlock_map) are mostly just to improve
1089 * readability, since they index into an array of locks
1091 static void lock_map(struct arena_info *arena, u32 premap)
1092 __acquires(&arena->map_locks[idx].lock)
1094 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1096 spin_lock(&arena->map_locks[idx].lock);
1099 static void unlock_map(struct arena_info *arena, u32 premap)
1100 __releases(&arena->map_locks[idx].lock)
1102 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1104 spin_unlock(&arena->map_locks[idx].lock);
1107 static int btt_data_read(struct arena_info *arena, struct page *page,
1108 unsigned int off, u32 lba, u32 len)
1111 u64 nsoff = to_namespace_offset(arena, lba);
1112 void *mem = kmap_atomic(page);
1114 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1120 static int btt_data_write(struct arena_info *arena, u32 lba,
1121 struct page *page, unsigned int off, u32 len)
1124 u64 nsoff = to_namespace_offset(arena, lba);
1125 void *mem = kmap_atomic(page);
1127 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1133 static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1135 void *mem = kmap_atomic(page);
1137 memset(mem + off, 0, len);
1141 #ifdef CONFIG_BLK_DEV_INTEGRITY
1142 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1143 struct arena_info *arena, u32 postmap, int rw)
1145 unsigned int len = btt_meta_size(btt);
1152 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1155 unsigned int cur_len;
1159 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1161 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1162 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1163 * can use those directly
1166 cur_len = min(len, bv.bv_len);
1167 mem = kmap_atomic(bv.bv_page);
1169 ret = arena_write_bytes(arena, meta_nsoff,
1170 mem + bv.bv_offset, cur_len,
1173 ret = arena_read_bytes(arena, meta_nsoff,
1174 mem + bv.bv_offset, cur_len,
1182 meta_nsoff += cur_len;
1183 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1190 #else /* CONFIG_BLK_DEV_INTEGRITY */
1191 static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1192 struct arena_info *arena, u32 postmap, int rw)
1198 static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1199 struct page *page, unsigned int off, sector_t sector,
1204 struct arena_info *arena = NULL;
1205 u32 lane = 0, premap, postmap;
1210 lane = nd_region_acquire_lane(btt->nd_region);
1212 ret = lba_to_arena(btt, sector, &premap, &arena);
1216 cur_len = min(btt->sector_size, len);
1218 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1224 * We loop to make sure that the post map LBA didn't change
1225 * from under us between writing the RTT and doing the actual
1233 zero_fill_data(page, off, cur_len);
1242 arena->rtt[lane] = RTT_VALID | postmap;
1244 * Barrier to make sure this write is not reordered
1245 * to do the verification map_read before the RTT store
1249 ret = btt_map_read(arena, premap, &new_map, &new_t,
1250 &new_e, NVDIMM_IO_ATOMIC);
1254 if ((postmap == new_map) && (t_flag == new_t) &&
1263 ret = btt_data_read(arena, page, off, postmap, cur_len);
1265 /* Media error - set the e_flag */
1266 if (btt_map_write(arena, premap, postmap, 0, 1, NVDIMM_IO_ATOMIC))
1267 dev_warn_ratelimited(to_dev(arena),
1268 "Error persistently tracking bad blocks at %#x\n",
1274 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1279 arena->rtt[lane] = RTT_INVALID;
1280 nd_region_release_lane(btt->nd_region, lane);
1284 sector += btt->sector_size >> SECTOR_SHIFT;
1290 arena->rtt[lane] = RTT_INVALID;
1292 nd_region_release_lane(btt->nd_region, lane);
1297 * Normally, arena_{read,write}_bytes will take care of the initial offset
1298 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1299 * we need the final, raw namespace offset here
1301 static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1304 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1305 to_namespace_offset(arena, postmap));
1306 sector_t phys_sector = nsoff >> 9;
1308 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1311 static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1312 sector_t sector, struct page *page, unsigned int off,
1316 struct arena_info *arena = NULL;
1317 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1318 struct log_entry log;
1326 lane = nd_region_acquire_lane(btt->nd_region);
1328 ret = lba_to_arena(btt, sector, &premap, &arena);
1331 cur_len = min(btt->sector_size, len);
1333 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1338 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1339 arena->freelist[lane].has_err = 1;
1341 if (mutex_is_locked(&arena->err_lock)
1342 || arena->freelist[lane].has_err) {
1343 nd_region_release_lane(btt->nd_region, lane);
1345 ret = arena_clear_freelist_error(arena, lane);
1349 /* OK to acquire a different lane/free block */
1353 new_postmap = arena->freelist[lane].block;
1355 /* Wait if the new block is being read from */
1356 for (i = 0; i < arena->nfree; i++)
1357 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1361 if (new_postmap >= arena->internal_nlba) {
1366 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1371 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1377 lock_map(arena, premap);
1378 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1382 if (old_postmap >= arena->internal_nlba) {
1387 set_e_flag(old_postmap);
1389 log.lba = cpu_to_le32(premap);
1390 log.old_map = cpu_to_le32(old_postmap);
1391 log.new_map = cpu_to_le32(new_postmap);
1392 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1393 sub = arena->freelist[lane].sub;
1394 ret = btt_flog_write(arena, lane, sub, &log);
1398 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1403 unlock_map(arena, premap);
1404 nd_region_release_lane(btt->nd_region, lane);
1407 ret = arena_clear_freelist_error(arena, lane);
1414 sector += btt->sector_size >> SECTOR_SHIFT;
1420 unlock_map(arena, premap);
1422 nd_region_release_lane(btt->nd_region, lane);
1426 static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1427 struct page *page, unsigned int len, unsigned int off,
1428 unsigned int op, sector_t sector)
1432 if (!op_is_write(op)) {
1433 ret = btt_read_pg(btt, bip, page, off, sector, len);
1434 flush_dcache_page(page);
1436 flush_dcache_page(page);
1437 ret = btt_write_pg(btt, bip, sector, page, off, len);
1443 static blk_qc_t btt_submit_bio(struct bio *bio)
1445 struct bio_integrity_payload *bip = bio_integrity(bio);
1446 struct btt *btt = bio->bi_bdev->bd_disk->private_data;
1447 struct bvec_iter iter;
1448 unsigned long start;
1449 struct bio_vec bvec;
1453 if (!bio_integrity_prep(bio))
1454 return BLK_QC_T_NONE;
1456 do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
1458 start = bio_start_io_acct(bio);
1459 bio_for_each_segment(bvec, bio, iter) {
1460 unsigned int len = bvec.bv_len;
1462 if (len > PAGE_SIZE || len < btt->sector_size ||
1463 len % btt->sector_size) {
1464 dev_err_ratelimited(&btt->nd_btt->dev,
1465 "unaligned bio segment (len: %d)\n", len);
1466 bio->bi_status = BLK_STS_IOERR;
1470 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1471 bio_op(bio), iter.bi_sector);
1473 dev_err(&btt->nd_btt->dev,
1474 "io error in %s sector %lld, len %d,\n",
1475 (op_is_write(bio_op(bio))) ? "WRITE" :
1477 (unsigned long long) iter.bi_sector, len);
1478 bio->bi_status = errno_to_blk_status(err);
1483 bio_end_io_acct(bio, start);
1486 return BLK_QC_T_NONE;
1489 static int btt_rw_page(struct block_device *bdev, sector_t sector,
1490 struct page *page, unsigned int op)
1492 struct btt *btt = bdev->bd_disk->private_data;
1495 rc = btt_do_bvec(btt, NULL, page, thp_size(page), 0, op, sector);
1497 page_endio(page, op_is_write(op), 0);
1503 static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1505 /* some standard values */
1506 geo->heads = 1 << 6;
1507 geo->sectors = 1 << 5;
1508 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1512 static const struct block_device_operations btt_fops = {
1513 .owner = THIS_MODULE,
1514 .submit_bio = btt_submit_bio,
1515 .rw_page = btt_rw_page,
1516 .getgeo = btt_getgeo,
1519 static int btt_blk_init(struct btt *btt)
1521 struct nd_btt *nd_btt = btt->nd_btt;
1522 struct nd_namespace_common *ndns = nd_btt->ndns;
1524 btt->btt_disk = blk_alloc_disk(NUMA_NO_NODE);
1528 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1529 btt->btt_disk->first_minor = 0;
1530 btt->btt_disk->fops = &btt_fops;
1531 btt->btt_disk->private_data = btt;
1533 blk_queue_logical_block_size(btt->btt_disk->queue, btt->sector_size);
1534 blk_queue_max_hw_sectors(btt->btt_disk->queue, UINT_MAX);
1535 blk_queue_flag_set(QUEUE_FLAG_NONROT, btt->btt_disk->queue);
1537 if (btt_meta_size(btt)) {
1538 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1541 del_gendisk(btt->btt_disk);
1542 blk_cleanup_disk(btt->btt_disk);
1546 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1547 device_add_disk(&btt->nd_btt->dev, btt->btt_disk, NULL);
1548 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1549 nvdimm_check_and_set_ro(btt->btt_disk);
1554 static void btt_blk_cleanup(struct btt *btt)
1556 del_gendisk(btt->btt_disk);
1557 blk_cleanup_disk(btt->btt_disk);
1561 * btt_init - initialize a block translation table for the given device
1562 * @nd_btt: device with BTT geometry and backing device info
1563 * @rawsize: raw size in bytes of the backing device
1564 * @lbasize: lba size of the backing device
1565 * @uuid: A uuid for the backing device - this is stored on media
1566 * @maxlane: maximum number of parallel requests the device can handle
1568 * Initialize a Block Translation Table on a backing device to provide
1569 * single sector power fail atomicity.
1575 * Pointer to a new struct btt on success, NULL on failure.
1577 static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1578 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1582 struct nd_namespace_io *nsio;
1583 struct device *dev = &nd_btt->dev;
1585 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1589 btt->nd_btt = nd_btt;
1590 btt->rawsize = rawsize;
1591 btt->lbasize = lbasize;
1592 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1593 INIT_LIST_HEAD(&btt->arena_list);
1594 mutex_init(&btt->init_lock);
1595 btt->nd_region = nd_region;
1596 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1597 btt->phys_bb = &nsio->bb;
1599 ret = discover_arenas(btt);
1601 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1605 if (btt->init_state != INIT_READY && nd_region->ro) {
1606 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1607 dev_name(&nd_region->dev));
1609 } else if (btt->init_state != INIT_READY) {
1610 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1611 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1612 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1613 btt->num_arenas, rawsize);
1615 ret = create_arenas(btt);
1617 dev_info(dev, "init: create_arenas: %d\n", ret);
1621 ret = btt_meta_init(btt);
1623 dev_err(dev, "init: error in meta_init: %d\n", ret);
1628 ret = btt_blk_init(btt);
1630 dev_err(dev, "init: error in blk_init: %d\n", ret);
1634 btt_debugfs_init(btt);
1640 * btt_fini - de-initialize a BTT
1641 * @btt: the BTT handle that was generated by btt_init
1643 * De-initialize a Block Translation Table on device removal
1648 static void btt_fini(struct btt *btt)
1651 btt_blk_cleanup(btt);
1653 debugfs_remove_recursive(btt->debugfs_dir);
1657 int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1659 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1660 struct nd_region *nd_region;
1661 struct btt_sb *btt_sb;
1663 size_t size, rawsize;
1666 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1667 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1671 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1675 size = nvdimm_namespace_capacity(ndns);
1676 rc = devm_namespace_enable(&nd_btt->dev, ndns, size);
1681 * If this returns < 0, that is ok as it just means there wasn't
1682 * an existing BTT, and we're creating a new one. We still need to
1683 * call this as we need the version dependent fields in nd_btt to be
1684 * set correctly based on the holder class
1686 nd_btt_version(nd_btt, ndns, btt_sb);
1688 rawsize = size - nd_btt->initial_offset;
1689 if (rawsize < ARENA_MIN_SIZE) {
1690 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1691 dev_name(&ndns->dev),
1692 ARENA_MIN_SIZE + nd_btt->initial_offset);
1695 nd_region = to_nd_region(nd_btt->dev.parent);
1696 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1704 EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1706 int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1708 struct btt *btt = nd_btt->btt;
1715 EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1717 static int __init nd_btt_init(void)
1721 debugfs_root = debugfs_create_dir("btt", NULL);
1722 if (IS_ERR_OR_NULL(debugfs_root))
1728 static void __exit nd_btt_exit(void)
1730 debugfs_remove_recursive(debugfs_root);
1733 MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1734 MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1735 MODULE_LICENSE("GPL v2");
1736 module_init(nd_btt_init);
1737 module_exit(nd_btt_exit);