From: Huaisheng Ye <[email protected]>
This patch set is a collection about some patches for dm-writecache.
Patches 0001-0003 have got Acked-by or Signed-off-by from the author and
dm maintainer.
Patches 0004 is still under review. Any comments and suggestion are welcome.
Huaisheng Ye (4):
dm writecache: remove needless dereferences in
__writecache_writeback_pmem()
dm writecache: add unlikely for returned value of rb_next/prev
dm writecache: remove unused member page_offset in writeback_struct
dm writecache: avoid unnecessary lookups in writecache_find_entry
drivers/md/dm-writecache.c | 30 ++++++++++++++++++------------
1 file changed, 18 insertions(+), 12 deletions(-)
--
1.8.3.1
From: Huaisheng Ye <[email protected]>
bio is already available so there is no need to access it in terms of
the wb pointer.
Signed-off-by: Huaisheng Ye <[email protected]>
Signed-off-by: Mike Snitzer <[email protected]>
---
drivers/md/dm-writecache.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/md/dm-writecache.c b/drivers/md/dm-writecache.c
index f782287..5b4d1c1 100644
--- a/drivers/md/dm-writecache.c
+++ b/drivers/md/dm-writecache.c
@@ -1478,9 +1478,9 @@ static void __writecache_writeback_pmem(struct dm_writecache *wc, struct writeba
bio = bio_alloc_bioset(GFP_NOIO, max_pages, &wc->bio_set);
wb = container_of(bio, struct writeback_struct, bio);
wb->wc = wc;
- wb->bio.bi_end_io = writecache_writeback_endio;
- bio_set_dev(&wb->bio, wc->dev->bdev);
- wb->bio.bi_iter.bi_sector = read_original_sector(wc, e);
+ bio->bi_end_io = writecache_writeback_endio;
+ bio_set_dev(bio, wc->dev->bdev);
+ bio->bi_iter.bi_sector = read_original_sector(wc, e);
wb->page_offset = PAGE_SIZE;
if (max_pages <= WB_LIST_INLINE ||
unlikely(!(wb->wc_list = kmalloc_array(max_pages, sizeof(struct wc_entry *),
@@ -1507,12 +1507,12 @@ static void __writecache_writeback_pmem(struct dm_writecache *wc, struct writeba
wb->wc_list[wb->wc_list_n++] = f;
e = f;
}
- bio_set_op_attrs(&wb->bio, REQ_OP_WRITE, WC_MODE_FUA(wc) * REQ_FUA);
+ bio_set_op_attrs(bio, REQ_OP_WRITE, WC_MODE_FUA(wc) * REQ_FUA);
if (writecache_has_error(wc)) {
bio->bi_status = BLK_STS_IOERR;
- bio_endio(&wb->bio);
+ bio_endio(bio);
} else {
- submit_bio(&wb->bio);
+ submit_bio(bio);
}
__writeback_throttle(wc, wbl);
--
1.8.3.1
From: Huaisheng Ye <[email protected]>
In functions writecache_discard() and writecache_find_entry() there is a
high probablity that the pointer of structure rb_node won't equal NULL.
Add unlikely for the pointer node NULL.
Signed-off-by: Huaisheng Ye <[email protected]>
Acked-by: Mikulas Patocka <[email protected]>
Signed-off-by: Mike Snitzer <[email protected]>
---
drivers/md/dm-writecache.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/md/dm-writecache.c b/drivers/md/dm-writecache.c
index 5b4d1c1..cfbbfbc 100644
--- a/drivers/md/dm-writecache.c
+++ b/drivers/md/dm-writecache.c
@@ -571,7 +571,7 @@ static struct wc_entry *writecache_find_entry(struct dm_writecache *wc,
node = rb_prev(&e->rb_node);
else
node = rb_next(&e->rb_node);
- if (!node)
+ if (unlikely(!node))
return e;
e2 = container_of(node, struct wc_entry, rb_node);
if (read_original_sector(wc, e2) != block)
@@ -804,7 +804,7 @@ static void writecache_discard(struct dm_writecache *wc, sector_t start, sector_
writecache_free_entry(wc, e);
}
- if (!node)
+ if (unlikely(!node))
break;
e = container_of(node, struct wc_entry, rb_node);
--
1.8.3.1
From: Huaisheng Ye <[email protected]>
The stucture member page_offset in writeback_struct never has been
used actually. Remove it.
Signed-off-by: Huaisheng Ye <[email protected]>
Acked-by: Mikulas Patocka <[email protected]>
---
drivers/md/dm-writecache.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/drivers/md/dm-writecache.c b/drivers/md/dm-writecache.c
index cfbbfbc..ddf1732 100644
--- a/drivers/md/dm-writecache.c
+++ b/drivers/md/dm-writecache.c
@@ -190,7 +190,6 @@ struct writeback_struct {
struct dm_writecache *wc;
struct wc_entry **wc_list;
unsigned wc_list_n;
- unsigned page_offset;
struct page *page;
struct wc_entry *wc_list_inline[WB_LIST_INLINE];
struct bio bio;
@@ -1481,7 +1480,6 @@ static void __writecache_writeback_pmem(struct dm_writecache *wc, struct writeba
bio->bi_end_io = writecache_writeback_endio;
bio_set_dev(bio, wc->dev->bdev);
bio->bi_iter.bi_sector = read_original_sector(wc, e);
- wb->page_offset = PAGE_SIZE;
if (max_pages <= WB_LIST_INLINE ||
unlikely(!(wb->wc_list = kmalloc_array(max_pages, sizeof(struct wc_entry *),
GFP_NOIO | __GFP_NORETRY |
--
1.8.3.1
From: Huaisheng Ye <[email protected]>
Only when entry has been found, that would only be necessary to check the
lowest or highest seq-count.
Add local variable "found" in writecache_find_entry, if no entry has been
found, it is meaningless that having a useless rb_prev or rb_next.
This patch is not designed for fixing logical error. That is used for
optimizing the behavior of writecache_find_entry.
Give an example to illustrate the point below.
Suppose that is the case, here is a normal READ bio comes to writecache_map.
And because of bio's direction is READ, writecache_find_entry would be called
with flags WFE_RETURN_FOLLOWING.
Now there are two scenarios,
1. writecache_find_entry successfully get an existing entry by searching
rb_tree, we could call it HIT. Then the first 'while' will be finished by
'break'. Next it will move to second 'while' loop, because of the flags
hasn't been marked as WFE_LOWEST_SEQ. writecache_find_entry will try to
return an entry with HIGHEST_SEQ, if there are other entries which has same
original_sector in rb_tree.
For this situation, the current code is okay to deal with that.
2. writecache_find_entry couldn't get an existing entry from rb_tree, we
could call it MISS. Because of same flags WFE_RETURN_FOLLOWING,
writecache_find_entry will get other entry, which's original_sector will
slightly larger than input parameter block, with big probability.
For this scenario, function writecache_find_entry doesn't need to enter
second 'while' loop. But current code would still try to check there were
other entry with same original_sector.
So the additional rb_next or rb_prev is unnecessary by this case, also the
code doesn't need to compare the original_sector of 'e2' with parameter
'block'.
This patch is designed to optimize the second case. so it could skip the
second 'while' loop when the block is missed from rb_tree.
Signed-off-by: Huaisheng Ye <[email protected]>
---
drivers/md/dm-writecache.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/drivers/md/dm-writecache.c b/drivers/md/dm-writecache.c
index ddf1732..047ae09 100644
--- a/drivers/md/dm-writecache.c
+++ b/drivers/md/dm-writecache.c
@@ -537,14 +537,18 @@ static struct wc_entry *writecache_find_entry(struct dm_writecache *wc,
{
struct wc_entry *e;
struct rb_node *node = wc->tree.rb_node;
+ bool found = false;
if (unlikely(!node))
return NULL;
while (1) {
e = container_of(node, struct wc_entry, rb_node);
- if (read_original_sector(wc, e) == block)
+ if (read_original_sector(wc, e) == block) {
+ found = true;
break;
+ }
+
node = (read_original_sector(wc, e) >= block ?
e->rb_node.rb_left : e->rb_node.rb_right);
if (unlikely(!node)) {
@@ -564,7 +568,8 @@ static struct wc_entry *writecache_find_entry(struct dm_writecache *wc,
}
}
- while (1) {
+ /* only need to check lowest or highest seq-count when entry has been found */
+ while (found) {
struct wc_entry *e2;
if (flags & WFE_LOWEST_SEQ)
node = rb_prev(&e->rb_node);
@@ -577,6 +582,9 @@ static struct wc_entry *writecache_find_entry(struct dm_writecache *wc,
return e;
e = e2;
}
+
+ /* no entry has been found, return the following entry */
+ return e;
}
static void writecache_insert_entry(struct dm_writecache *wc, struct wc_entry *ins)
--
1.8.3.1