diff options
author | Jens Axboe <axboe@kernel.dk> | 2013-12-31 11:51:02 -0500 |
---|---|---|
committer | Jens Axboe <axboe@kernel.dk> | 2013-12-31 11:51:02 -0500 |
commit | b28bc9b38c52f63f43e3fd875af982f2240a2859 (patch) | |
tree | 76cdb7b52b58f5685993cc15ed81d1c903023358 /drivers/md/bcache | |
parent | 8d30726912cb39c3a3ebde06214d54861f8fdde2 (diff) | |
parent | 802eee95bde72fd0cd0f3a5b2098375a487d1eda (diff) |
Merge tag 'v3.13-rc6' into for-3.14/core
Needed to bring blk-mq uptodate, since changes have been going in
since for-3.14/core was established.
Fixup merge issues related to the immutable biovec changes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Conflicts:
block/blk-flush.c
fs/btrfs/check-integrity.c
fs/btrfs/extent_io.c
fs/btrfs/scrub.c
fs/logfs/dev_bdev.c
Diffstat (limited to 'drivers/md/bcache')
-rw-r--r-- | drivers/md/bcache/alloc.c | 2 | ||||
-rw-r--r-- | drivers/md/bcache/bcache.h | 12 | ||||
-rw-r--r-- | drivers/md/bcache/btree.c | 27 | ||||
-rw-r--r-- | drivers/md/bcache/movinggc.c | 21 | ||||
-rw-r--r-- | drivers/md/bcache/super.c | 2 | ||||
-rw-r--r-- | drivers/md/bcache/sysfs.c | 50 | ||||
-rw-r--r-- | drivers/md/bcache/util.c | 8 | ||||
-rw-r--r-- | drivers/md/bcache/util.h | 2 | ||||
-rw-r--r-- | drivers/md/bcache/writeback.c | 53 |
9 files changed, 111 insertions, 66 deletions
diff --git a/drivers/md/bcache/alloc.c b/drivers/md/bcache/alloc.c index 2b46bf1d7e40..4c9852d92b0a 100644 --- a/drivers/md/bcache/alloc.c +++ b/drivers/md/bcache/alloc.c | |||
@@ -421,9 +421,11 @@ out: | |||
421 | 421 | ||
422 | if (watermark <= WATERMARK_METADATA) { | 422 | if (watermark <= WATERMARK_METADATA) { |
423 | SET_GC_MARK(b, GC_MARK_METADATA); | 423 | SET_GC_MARK(b, GC_MARK_METADATA); |
424 | SET_GC_MOVE(b, 0); | ||
424 | b->prio = BTREE_PRIO; | 425 | b->prio = BTREE_PRIO; |
425 | } else { | 426 | } else { |
426 | SET_GC_MARK(b, GC_MARK_RECLAIMABLE); | 427 | SET_GC_MARK(b, GC_MARK_RECLAIMABLE); |
428 | SET_GC_MOVE(b, 0); | ||
427 | b->prio = INITIAL_PRIO; | 429 | b->prio = INITIAL_PRIO; |
428 | } | 430 | } |
429 | 431 | ||
diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h index 964353c5329d..dbdbca5a9591 100644 --- a/drivers/md/bcache/bcache.h +++ b/drivers/md/bcache/bcache.h | |||
@@ -197,7 +197,7 @@ struct bucket { | |||
197 | uint8_t disk_gen; | 197 | uint8_t disk_gen; |
198 | uint8_t last_gc; /* Most out of date gen in the btree */ | 198 | uint8_t last_gc; /* Most out of date gen in the btree */ |
199 | uint8_t gc_gen; | 199 | uint8_t gc_gen; |
200 | uint16_t gc_mark; | 200 | uint16_t gc_mark; /* Bitfield used by GC. See below for field */ |
201 | }; | 201 | }; |
202 | 202 | ||
203 | /* | 203 | /* |
@@ -209,7 +209,8 @@ BITMASK(GC_MARK, struct bucket, gc_mark, 0, 2); | |||
209 | #define GC_MARK_RECLAIMABLE 0 | 209 | #define GC_MARK_RECLAIMABLE 0 |
210 | #define GC_MARK_DIRTY 1 | 210 | #define GC_MARK_DIRTY 1 |
211 | #define GC_MARK_METADATA 2 | 211 | #define GC_MARK_METADATA 2 |
212 | BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, 14); | 212 | BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, 13); |
213 | BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1); | ||
213 | 214 | ||
214 | #include "journal.h" | 215 | #include "journal.h" |
215 | #include "stats.h" | 216 | #include "stats.h" |
@@ -371,14 +372,14 @@ struct cached_dev { | |||
371 | unsigned char writeback_percent; | 372 | unsigned char writeback_percent; |
372 | unsigned writeback_delay; | 373 | unsigned writeback_delay; |
373 | 374 | ||
374 | int writeback_rate_change; | ||
375 | int64_t writeback_rate_derivative; | ||
376 | uint64_t writeback_rate_target; | 375 | uint64_t writeback_rate_target; |
376 | int64_t writeback_rate_proportional; | ||
377 | int64_t writeback_rate_derivative; | ||
378 | int64_t writeback_rate_change; | ||
377 | 379 | ||
378 | unsigned writeback_rate_update_seconds; | 380 | unsigned writeback_rate_update_seconds; |
379 | unsigned writeback_rate_d_term; | 381 | unsigned writeback_rate_d_term; |
380 | unsigned writeback_rate_p_term_inverse; | 382 | unsigned writeback_rate_p_term_inverse; |
381 | unsigned writeback_rate_d_smooth; | ||
382 | }; | 383 | }; |
383 | 384 | ||
384 | enum alloc_watermarks { | 385 | enum alloc_watermarks { |
@@ -444,7 +445,6 @@ struct cache { | |||
444 | * call prio_write() to keep gens from wrapping. | 445 | * call prio_write() to keep gens from wrapping. |
445 | */ | 446 | */ |
446 | uint8_t need_save_prio; | 447 | uint8_t need_save_prio; |
447 | unsigned gc_move_threshold; | ||
448 | 448 | ||
449 | /* | 449 | /* |
450 | * If nonzero, we know we aren't going to find any buckets to invalidate | 450 | * If nonzero, we know we aren't going to find any buckets to invalidate |
diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c index b62f37925374..946ecd3b048b 100644 --- a/drivers/md/bcache/btree.c +++ b/drivers/md/bcache/btree.c | |||
@@ -1561,6 +1561,28 @@ size_t bch_btree_gc_finish(struct cache_set *c) | |||
1561 | SET_GC_MARK(PTR_BUCKET(c, &c->uuid_bucket, i), | 1561 | SET_GC_MARK(PTR_BUCKET(c, &c->uuid_bucket, i), |
1562 | GC_MARK_METADATA); | 1562 | GC_MARK_METADATA); |
1563 | 1563 | ||
1564 | /* don't reclaim buckets to which writeback keys point */ | ||
1565 | rcu_read_lock(); | ||
1566 | for (i = 0; i < c->nr_uuids; i++) { | ||
1567 | struct bcache_device *d = c->devices[i]; | ||
1568 | struct cached_dev *dc; | ||
1569 | struct keybuf_key *w, *n; | ||
1570 | unsigned j; | ||
1571 | |||
1572 | if (!d || UUID_FLASH_ONLY(&c->uuids[i])) | ||
1573 | continue; | ||
1574 | dc = container_of(d, struct cached_dev, disk); | ||
1575 | |||
1576 | spin_lock(&dc->writeback_keys.lock); | ||
1577 | rbtree_postorder_for_each_entry_safe(w, n, | ||
1578 | &dc->writeback_keys.keys, node) | ||
1579 | for (j = 0; j < KEY_PTRS(&w->key); j++) | ||
1580 | SET_GC_MARK(PTR_BUCKET(c, &w->key, j), | ||
1581 | GC_MARK_DIRTY); | ||
1582 | spin_unlock(&dc->writeback_keys.lock); | ||
1583 | } | ||
1584 | rcu_read_unlock(); | ||
1585 | |||
1564 | for_each_cache(ca, c, i) { | 1586 | for_each_cache(ca, c, i) { |
1565 | uint64_t *i; | 1587 | uint64_t *i; |
1566 | 1588 | ||
@@ -1817,7 +1839,8 @@ static bool fix_overlapping_extents(struct btree *b, struct bkey *insert, | |||
1817 | if (KEY_START(k) > KEY_START(insert) + sectors_found) | 1839 | if (KEY_START(k) > KEY_START(insert) + sectors_found) |
1818 | goto check_failed; | 1840 | goto check_failed; |
1819 | 1841 | ||
1820 | if (KEY_PTRS(replace_key) != KEY_PTRS(k)) | 1842 | if (KEY_PTRS(k) != KEY_PTRS(replace_key) || |
1843 | KEY_DIRTY(k) != KEY_DIRTY(replace_key)) | ||
1821 | goto check_failed; | 1844 | goto check_failed; |
1822 | 1845 | ||
1823 | /* skip past gen */ | 1846 | /* skip past gen */ |
@@ -2217,7 +2240,7 @@ struct btree_insert_op { | |||
2217 | struct bkey *replace_key; | 2240 | struct bkey *replace_key; |
2218 | }; | 2241 | }; |
2219 | 2242 | ||
2220 | int btree_insert_fn(struct btree_op *b_op, struct btree *b) | 2243 | static int btree_insert_fn(struct btree_op *b_op, struct btree *b) |
2221 | { | 2244 | { |
2222 | struct btree_insert_op *op = container_of(b_op, | 2245 | struct btree_insert_op *op = container_of(b_op, |
2223 | struct btree_insert_op, op); | 2246 | struct btree_insert_op, op); |
diff --git a/drivers/md/bcache/movinggc.c b/drivers/md/bcache/movinggc.c index 581f95df8265..052bd24d24b4 100644 --- a/drivers/md/bcache/movinggc.c +++ b/drivers/md/bcache/movinggc.c | |||
@@ -25,10 +25,9 @@ static bool moving_pred(struct keybuf *buf, struct bkey *k) | |||
25 | unsigned i; | 25 | unsigned i; |
26 | 26 | ||
27 | for (i = 0; i < KEY_PTRS(k); i++) { | 27 | for (i = 0; i < KEY_PTRS(k); i++) { |
28 | struct cache *ca = PTR_CACHE(c, k, i); | ||
29 | struct bucket *g = PTR_BUCKET(c, k, i); | 28 | struct bucket *g = PTR_BUCKET(c, k, i); |
30 | 29 | ||
31 | if (GC_SECTORS_USED(g) < ca->gc_move_threshold) | 30 | if (GC_MOVE(g)) |
32 | return true; | 31 | return true; |
33 | } | 32 | } |
34 | 33 | ||
@@ -65,11 +64,16 @@ static void write_moving_finish(struct closure *cl) | |||
65 | 64 | ||
66 | static void read_moving_endio(struct bio *bio, int error) | 65 | static void read_moving_endio(struct bio *bio, int error) |
67 | { | 66 | { |
67 | struct bbio *b = container_of(bio, struct bbio, bio); | ||
68 | struct moving_io *io = container_of(bio->bi_private, | 68 | struct moving_io *io = container_of(bio->bi_private, |
69 | struct moving_io, cl); | 69 | struct moving_io, cl); |
70 | 70 | ||
71 | if (error) | 71 | if (error) |
72 | io->op.error = error; | 72 | io->op.error = error; |
73 | else if (!KEY_DIRTY(&b->key) && | ||
74 | ptr_stale(io->op.c, &b->key, 0)) { | ||
75 | io->op.error = -EINTR; | ||
76 | } | ||
73 | 77 | ||
74 | bch_bbio_endio(io->op.c, bio, error, "reading data to move"); | 78 | bch_bbio_endio(io->op.c, bio, error, "reading data to move"); |
75 | } | 79 | } |
@@ -141,6 +145,11 @@ static void read_moving(struct cache_set *c) | |||
141 | if (!w) | 145 | if (!w) |
142 | break; | 146 | break; |
143 | 147 | ||
148 | if (ptr_stale(c, &w->key, 0)) { | ||
149 | bch_keybuf_del(&c->moving_gc_keys, w); | ||
150 | continue; | ||
151 | } | ||
152 | |||
144 | io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec) | 153 | io = kzalloc(sizeof(struct moving_io) + sizeof(struct bio_vec) |
145 | * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS), | 154 | * DIV_ROUND_UP(KEY_SIZE(&w->key), PAGE_SECTORS), |
146 | GFP_KERNEL); | 155 | GFP_KERNEL); |
@@ -184,7 +193,8 @@ static bool bucket_cmp(struct bucket *l, struct bucket *r) | |||
184 | 193 | ||
185 | static unsigned bucket_heap_top(struct cache *ca) | 194 | static unsigned bucket_heap_top(struct cache *ca) |
186 | { | 195 | { |
187 | return GC_SECTORS_USED(heap_peek(&ca->heap)); | 196 | struct bucket *b; |
197 | return (b = heap_peek(&ca->heap)) ? GC_SECTORS_USED(b) : 0; | ||
188 | } | 198 | } |
189 | 199 | ||
190 | void bch_moving_gc(struct cache_set *c) | 200 | void bch_moving_gc(struct cache_set *c) |
@@ -226,9 +236,8 @@ void bch_moving_gc(struct cache_set *c) | |||
226 | sectors_to_move -= GC_SECTORS_USED(b); | 236 | sectors_to_move -= GC_SECTORS_USED(b); |
227 | } | 237 | } |
228 | 238 | ||
229 | ca->gc_move_threshold = bucket_heap_top(ca); | 239 | while (heap_pop(&ca->heap, b, bucket_cmp)) |
230 | 240 | SET_GC_MOVE(b, 1); | |
231 | pr_debug("threshold %u", ca->gc_move_threshold); | ||
232 | } | 241 | } |
233 | 242 | ||
234 | mutex_unlock(&c->bucket_lock); | 243 | mutex_unlock(&c->bucket_lock); |
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c index 60fb6044b953..93d593f957f6 100644 --- a/drivers/md/bcache/super.c +++ b/drivers/md/bcache/super.c | |||
@@ -1672,7 +1672,7 @@ err: | |||
1672 | static bool can_attach_cache(struct cache *ca, struct cache_set *c) | 1672 | static bool can_attach_cache(struct cache *ca, struct cache_set *c) |
1673 | { | 1673 | { |
1674 | return ca->sb.block_size == c->sb.block_size && | 1674 | return ca->sb.block_size == c->sb.block_size && |
1675 | ca->sb.bucket_size == c->sb.block_size && | 1675 | ca->sb.bucket_size == c->sb.bucket_size && |
1676 | ca->sb.nr_in_set == c->sb.nr_in_set; | 1676 | ca->sb.nr_in_set == c->sb.nr_in_set; |
1677 | } | 1677 | } |
1678 | 1678 | ||
diff --git a/drivers/md/bcache/sysfs.c b/drivers/md/bcache/sysfs.c index 80d4c2bee18a..a1f85612f0b3 100644 --- a/drivers/md/bcache/sysfs.c +++ b/drivers/md/bcache/sysfs.c | |||
@@ -83,7 +83,6 @@ rw_attribute(writeback_rate); | |||
83 | rw_attribute(writeback_rate_update_seconds); | 83 | rw_attribute(writeback_rate_update_seconds); |
84 | rw_attribute(writeback_rate_d_term); | 84 | rw_attribute(writeback_rate_d_term); |
85 | rw_attribute(writeback_rate_p_term_inverse); | 85 | rw_attribute(writeback_rate_p_term_inverse); |
86 | rw_attribute(writeback_rate_d_smooth); | ||
87 | read_attribute(writeback_rate_debug); | 86 | read_attribute(writeback_rate_debug); |
88 | 87 | ||
89 | read_attribute(stripe_size); | 88 | read_attribute(stripe_size); |
@@ -129,31 +128,41 @@ SHOW(__bch_cached_dev) | |||
129 | var_printf(writeback_running, "%i"); | 128 | var_printf(writeback_running, "%i"); |
130 | var_print(writeback_delay); | 129 | var_print(writeback_delay); |
131 | var_print(writeback_percent); | 130 | var_print(writeback_percent); |
132 | sysfs_print(writeback_rate, dc->writeback_rate.rate); | 131 | sysfs_hprint(writeback_rate, dc->writeback_rate.rate << 9); |
133 | 132 | ||
134 | var_print(writeback_rate_update_seconds); | 133 | var_print(writeback_rate_update_seconds); |
135 | var_print(writeback_rate_d_term); | 134 | var_print(writeback_rate_d_term); |
136 | var_print(writeback_rate_p_term_inverse); | 135 | var_print(writeback_rate_p_term_inverse); |
137 | var_print(writeback_rate_d_smooth); | ||
138 | 136 | ||
139 | if (attr == &sysfs_writeback_rate_debug) { | 137 | if (attr == &sysfs_writeback_rate_debug) { |
138 | char rate[20]; | ||
140 | char dirty[20]; | 139 | char dirty[20]; |
141 | char derivative[20]; | ||
142 | char target[20]; | 140 | char target[20]; |
143 | bch_hprint(dirty, | 141 | char proportional[20]; |
144 | bcache_dev_sectors_dirty(&dc->disk) << 9); | 142 | char derivative[20]; |
145 | bch_hprint(derivative, dc->writeback_rate_derivative << 9); | 143 | char change[20]; |
144 | s64 next_io; | ||
145 | |||
146 | bch_hprint(rate, dc->writeback_rate.rate << 9); | ||
147 | bch_hprint(dirty, bcache_dev_sectors_dirty(&dc->disk) << 9); | ||
146 | bch_hprint(target, dc->writeback_rate_target << 9); | 148 | bch_hprint(target, dc->writeback_rate_target << 9); |
149 | bch_hprint(proportional,dc->writeback_rate_proportional << 9); | ||
150 | bch_hprint(derivative, dc->writeback_rate_derivative << 9); | ||
151 | bch_hprint(change, dc->writeback_rate_change << 9); | ||
152 | |||
153 | next_io = div64_s64(dc->writeback_rate.next - local_clock(), | ||
154 | NSEC_PER_MSEC); | ||
147 | 155 | ||
148 | return sprintf(buf, | 156 | return sprintf(buf, |
149 | "rate:\t\t%u\n" | 157 | "rate:\t\t%s/sec\n" |
150 | "change:\t\t%i\n" | ||
151 | "dirty:\t\t%s\n" | 158 | "dirty:\t\t%s\n" |
159 | "target:\t\t%s\n" | ||
160 | "proportional:\t%s\n" | ||
152 | "derivative:\t%s\n" | 161 | "derivative:\t%s\n" |
153 | "target:\t\t%s\n", | 162 | "change:\t\t%s/sec\n" |
154 | dc->writeback_rate.rate, | 163 | "next io:\t%llims\n", |
155 | dc->writeback_rate_change, | 164 | rate, dirty, target, proportional, |
156 | dirty, derivative, target); | 165 | derivative, change, next_io); |
157 | } | 166 | } |
158 | 167 | ||
159 | sysfs_hprint(dirty_data, | 168 | sysfs_hprint(dirty_data, |
@@ -189,6 +198,7 @@ STORE(__cached_dev) | |||
189 | struct kobj_uevent_env *env; | 198 | struct kobj_uevent_env *env; |
190 | 199 | ||
191 | #define d_strtoul(var) sysfs_strtoul(var, dc->var) | 200 | #define d_strtoul(var) sysfs_strtoul(var, dc->var) |
201 | #define d_strtoul_nonzero(var) sysfs_strtoul_clamp(var, dc->var, 1, INT_MAX) | ||
192 | #define d_strtoi_h(var) sysfs_hatoi(var, dc->var) | 202 | #define d_strtoi_h(var) sysfs_hatoi(var, dc->var) |
193 | 203 | ||
194 | sysfs_strtoul(data_csum, dc->disk.data_csum); | 204 | sysfs_strtoul(data_csum, dc->disk.data_csum); |
@@ -197,16 +207,15 @@ STORE(__cached_dev) | |||
197 | d_strtoul(writeback_metadata); | 207 | d_strtoul(writeback_metadata); |
198 | d_strtoul(writeback_running); | 208 | d_strtoul(writeback_running); |
199 | d_strtoul(writeback_delay); | 209 | d_strtoul(writeback_delay); |
200 | sysfs_strtoul_clamp(writeback_rate, | 210 | |
201 | dc->writeback_rate.rate, 1, 1000000); | ||
202 | sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40); | 211 | sysfs_strtoul_clamp(writeback_percent, dc->writeback_percent, 0, 40); |
203 | 212 | ||
204 | d_strtoul(writeback_rate_update_seconds); | 213 | sysfs_strtoul_clamp(writeback_rate, |
214 | dc->writeback_rate.rate, 1, INT_MAX); | ||
215 | |||
216 | d_strtoul_nonzero(writeback_rate_update_seconds); | ||
205 | d_strtoul(writeback_rate_d_term); | 217 | d_strtoul(writeback_rate_d_term); |
206 | d_strtoul(writeback_rate_p_term_inverse); | 218 | d_strtoul_nonzero(writeback_rate_p_term_inverse); |
207 | sysfs_strtoul_clamp(writeback_rate_p_term_inverse, | ||
208 | dc->writeback_rate_p_term_inverse, 1, INT_MAX); | ||
209 | d_strtoul(writeback_rate_d_smooth); | ||
210 | 219 | ||
211 | d_strtoi_h(sequential_cutoff); | 220 | d_strtoi_h(sequential_cutoff); |
212 | d_strtoi_h(readahead); | 221 | d_strtoi_h(readahead); |
@@ -313,7 +322,6 @@ static struct attribute *bch_cached_dev_files[] = { | |||
313 | &sysfs_writeback_rate_update_seconds, | 322 | &sysfs_writeback_rate_update_seconds, |
314 | &sysfs_writeback_rate_d_term, | 323 | &sysfs_writeback_rate_d_term, |
315 | &sysfs_writeback_rate_p_term_inverse, | 324 | &sysfs_writeback_rate_p_term_inverse, |
316 | &sysfs_writeback_rate_d_smooth, | ||
317 | &sysfs_writeback_rate_debug, | 325 | &sysfs_writeback_rate_debug, |
318 | &sysfs_dirty_data, | 326 | &sysfs_dirty_data, |
319 | &sysfs_stripe_size, | 327 | &sysfs_stripe_size, |
diff --git a/drivers/md/bcache/util.c b/drivers/md/bcache/util.c index c57621e49dc0..db3ae4c2b223 100644 --- a/drivers/md/bcache/util.c +++ b/drivers/md/bcache/util.c | |||
@@ -209,7 +209,13 @@ uint64_t bch_next_delay(struct bch_ratelimit *d, uint64_t done) | |||
209 | { | 209 | { |
210 | uint64_t now = local_clock(); | 210 | uint64_t now = local_clock(); |
211 | 211 | ||
212 | d->next += div_u64(done, d->rate); | 212 | d->next += div_u64(done * NSEC_PER_SEC, d->rate); |
213 | |||
214 | if (time_before64(now + NSEC_PER_SEC, d->next)) | ||
215 | d->next = now + NSEC_PER_SEC; | ||
216 | |||
217 | if (time_after64(now - NSEC_PER_SEC * 2, d->next)) | ||
218 | d->next = now - NSEC_PER_SEC * 2; | ||
213 | 219 | ||
214 | return time_after64(d->next, now) | 220 | return time_after64(d->next, now) |
215 | ? div_u64(d->next - now, NSEC_PER_SEC / HZ) | 221 | ? div_u64(d->next - now, NSEC_PER_SEC / HZ) |
diff --git a/drivers/md/bcache/util.h b/drivers/md/bcache/util.h index 362c4b3f8b4a..1030c6020e98 100644 --- a/drivers/md/bcache/util.h +++ b/drivers/md/bcache/util.h | |||
@@ -110,7 +110,7 @@ do { \ | |||
110 | _r; \ | 110 | _r; \ |
111 | }) | 111 | }) |
112 | 112 | ||
113 | #define heap_peek(h) ((h)->size ? (h)->data[0] : NULL) | 113 | #define heap_peek(h) ((h)->used ? (h)->data[0] : NULL) |
114 | 114 | ||
115 | #define heap_full(h) ((h)->used == (h)->size) | 115 | #define heap_full(h) ((h)->used == (h)->size) |
116 | 116 | ||
diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c index 04657e93f4fd..f4300e4c0114 100644 --- a/drivers/md/bcache/writeback.c +++ b/drivers/md/bcache/writeback.c | |||
@@ -30,38 +30,40 @@ static void __update_writeback_rate(struct cached_dev *dc) | |||
30 | 30 | ||
31 | /* PD controller */ | 31 | /* PD controller */ |
32 | 32 | ||
33 | int change = 0; | ||
34 | int64_t error; | ||
35 | int64_t dirty = bcache_dev_sectors_dirty(&dc->disk); | 33 | int64_t dirty = bcache_dev_sectors_dirty(&dc->disk); |
36 | int64_t derivative = dirty - dc->disk.sectors_dirty_last; | 34 | int64_t derivative = dirty - dc->disk.sectors_dirty_last; |
35 | int64_t proportional = dirty - target; | ||
36 | int64_t change; | ||
37 | 37 | ||
38 | dc->disk.sectors_dirty_last = dirty; | 38 | dc->disk.sectors_dirty_last = dirty; |
39 | 39 | ||
40 | derivative *= dc->writeback_rate_d_term; | 40 | /* Scale to sectors per second */ |
41 | derivative = clamp(derivative, -dirty, dirty); | ||
42 | 41 | ||
43 | derivative = ewma_add(dc->disk.sectors_dirty_derivative, derivative, | 42 | proportional *= dc->writeback_rate_update_seconds; |
44 | dc->writeback_rate_d_smooth, 0); | 43 | proportional = div_s64(proportional, dc->writeback_rate_p_term_inverse); |
45 | 44 | ||
46 | /* Avoid divide by zero */ | 45 | derivative = div_s64(derivative, dc->writeback_rate_update_seconds); |
47 | if (!target) | ||
48 | goto out; | ||
49 | 46 | ||
50 | error = div64_s64((dirty + derivative - target) << 8, target); | 47 | derivative = ewma_add(dc->disk.sectors_dirty_derivative, derivative, |
48 | (dc->writeback_rate_d_term / | ||
49 | dc->writeback_rate_update_seconds) ?: 1, 0); | ||
50 | |||
51 | derivative *= dc->writeback_rate_d_term; | ||
52 | derivative = div_s64(derivative, dc->writeback_rate_p_term_inverse); | ||
51 | 53 | ||
52 | change = div_s64((dc->writeback_rate.rate * error) >> 8, | 54 | change = proportional + derivative; |
53 | dc->writeback_rate_p_term_inverse); | ||
54 | 55 | ||
55 | /* Don't increase writeback rate if the device isn't keeping up */ | 56 | /* Don't increase writeback rate if the device isn't keeping up */ |
56 | if (change > 0 && | 57 | if (change > 0 && |
57 | time_after64(local_clock(), | 58 | time_after64(local_clock(), |
58 | dc->writeback_rate.next + 10 * NSEC_PER_MSEC)) | 59 | dc->writeback_rate.next + NSEC_PER_MSEC)) |
59 | change = 0; | 60 | change = 0; |
60 | 61 | ||
61 | dc->writeback_rate.rate = | 62 | dc->writeback_rate.rate = |
62 | clamp_t(int64_t, dc->writeback_rate.rate + change, | 63 | clamp_t(int64_t, (int64_t) dc->writeback_rate.rate + change, |
63 | 1, NSEC_PER_MSEC); | 64 | 1, NSEC_PER_MSEC); |
64 | out: | 65 | |
66 | dc->writeback_rate_proportional = proportional; | ||
65 | dc->writeback_rate_derivative = derivative; | 67 | dc->writeback_rate_derivative = derivative; |
66 | dc->writeback_rate_change = change; | 68 | dc->writeback_rate_change = change; |
67 | dc->writeback_rate_target = target; | 69 | dc->writeback_rate_target = target; |
@@ -87,15 +89,11 @@ static void update_writeback_rate(struct work_struct *work) | |||
87 | 89 | ||
88 | static unsigned writeback_delay(struct cached_dev *dc, unsigned sectors) | 90 | static unsigned writeback_delay(struct cached_dev *dc, unsigned sectors) |
89 | { | 91 | { |
90 | uint64_t ret; | ||
91 | |||
92 | if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) || | 92 | if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) || |
93 | !dc->writeback_percent) | 93 | !dc->writeback_percent) |
94 | return 0; | 94 | return 0; |
95 | 95 | ||
96 | ret = bch_next_delay(&dc->writeback_rate, sectors * 10000000ULL); | 96 | return bch_next_delay(&dc->writeback_rate, sectors); |
97 | |||
98 | return min_t(uint64_t, ret, HZ); | ||
99 | } | 97 | } |
100 | 98 | ||
101 | struct dirty_io { | 99 | struct dirty_io { |
@@ -241,7 +239,7 @@ static void read_dirty(struct cached_dev *dc) | |||
241 | if (KEY_START(&w->key) != dc->last_read || | 239 | if (KEY_START(&w->key) != dc->last_read || |
242 | jiffies_to_msecs(delay) > 50) | 240 | jiffies_to_msecs(delay) > 50) |
243 | while (!kthread_should_stop() && delay) | 241 | while (!kthread_should_stop() && delay) |
244 | delay = schedule_timeout_interruptible(delay); | 242 | delay = schedule_timeout_uninterruptible(delay); |
245 | 243 | ||
246 | dc->last_read = KEY_OFFSET(&w->key); | 244 | dc->last_read = KEY_OFFSET(&w->key); |
247 | 245 | ||
@@ -438,7 +436,7 @@ static int bch_writeback_thread(void *arg) | |||
438 | while (delay && | 436 | while (delay && |
439 | !kthread_should_stop() && | 437 | !kthread_should_stop() && |
440 | !test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) | 438 | !test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags)) |
441 | delay = schedule_timeout_interruptible(delay); | 439 | delay = schedule_timeout_uninterruptible(delay); |
442 | } | 440 | } |
443 | } | 441 | } |
444 | 442 | ||
@@ -476,6 +474,8 @@ void bch_sectors_dirty_init(struct cached_dev *dc) | |||
476 | 474 | ||
477 | bch_btree_map_keys(&op.op, dc->disk.c, &KEY(op.inode, 0, 0), | 475 | bch_btree_map_keys(&op.op, dc->disk.c, &KEY(op.inode, 0, 0), |
478 | sectors_dirty_init_fn, 0); | 476 | sectors_dirty_init_fn, 0); |
477 | |||
478 | dc->disk.sectors_dirty_last = bcache_dev_sectors_dirty(&dc->disk); | ||
479 | } | 479 | } |
480 | 480 | ||
481 | int bch_cached_dev_writeback_init(struct cached_dev *dc) | 481 | int bch_cached_dev_writeback_init(struct cached_dev *dc) |
@@ -490,18 +490,15 @@ int bch_cached_dev_writeback_init(struct cached_dev *dc) | |||
490 | dc->writeback_delay = 30; | 490 | dc->writeback_delay = 30; |
491 | dc->writeback_rate.rate = 1024; | 491 | dc->writeback_rate.rate = 1024; |
492 | 492 | ||
493 | dc->writeback_rate_update_seconds = 30; | 493 | dc->writeback_rate_update_seconds = 5; |
494 | dc->writeback_rate_d_term = 16; | 494 | dc->writeback_rate_d_term = 30; |
495 | dc->writeback_rate_p_term_inverse = 64; | 495 | dc->writeback_rate_p_term_inverse = 6000; |
496 | dc->writeback_rate_d_smooth = 8; | ||
497 | 496 | ||
498 | dc->writeback_thread = kthread_create(bch_writeback_thread, dc, | 497 | dc->writeback_thread = kthread_create(bch_writeback_thread, dc, |
499 | "bcache_writeback"); | 498 | "bcache_writeback"); |
500 | if (IS_ERR(dc->writeback_thread)) | 499 | if (IS_ERR(dc->writeback_thread)) |
501 | return PTR_ERR(dc->writeback_thread); | 500 | return PTR_ERR(dc->writeback_thread); |
502 | 501 | ||
503 | set_task_state(dc->writeback_thread, TASK_INTERRUPTIBLE); | ||
504 | |||
505 | INIT_DELAYED_WORK(&dc->writeback_rate_update, update_writeback_rate); | 502 | INIT_DELAYED_WORK(&dc->writeback_rate_update, update_writeback_rate); |
506 | schedule_delayed_work(&dc->writeback_rate_update, | 503 | schedule_delayed_work(&dc->writeback_rate_update, |
507 | dc->writeback_rate_update_seconds * HZ); | 504 | dc->writeback_rate_update_seconds * HZ); |