[PATCH 0/2] optimise bvec/bio iteration

[PATCH 0/2] optimise bvec/bio iteration

[Pavel Begunkov]

This adds simpler versions of bvec_iter_advance() and bio_advance_iter()
(i.e. *_single()), that are faster but work with the restriction that
@bytes shouldn't be more than available in the current bvec segment.

That covers most of bvec/bio iteration/foreach, that are massively
inlined, and thus also nicely shrinks binary.

Others non core-block users might be updated on case by case basis
(if applicable) after the change is merged.

Pavel Begunkov (2):
  block: optimise for_each_bvec() advance
  bio: optimise bvec iteration

 block/bio.c          |  4 ++--
 include/linux/bio.h  | 17 +++++++++++++++--
 include/linux/bvec.h | 20 +++++++++++++++-----
 3 files changed, 32 insertions(+), 9 deletions(-)

-- 
2.24.0

[PATCH 1/2] block: optimise for_each_bvec() advance

[Pavel Begunkov]

Because of how for_each_bvec() works it never advances across multiple
entries at a time, so bvec_iter_advance() is an overkill. Add
specialised bvec_iter_advance_single() that is faster. It also handles
zero-len bvecs, so can kill bvec_iter_skip_zero_bvec().

   text    data     bss     dec     hex filename
before:
  23977     805       0   24782    60ce lib/iov_iter.o
before, bvec_iter_advance() w/o WARN_ONCE()
  22886     600       0   23486    5bbe ./lib/iov_iter.o
after:
  21862     600       0   22462    57be lib/iov_iter.o

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 include/linux/bvec.h | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/include/linux/bvec.h b/include/linux/bvec.h
index 2efec10bf792..ff832e698efb 100644
--- a/include/linux/bvec.h
+++ b/include/linux/bvec.h
@@ -121,18 +121,28 @@ static inline bool bvec_iter_advance(const struct bio_vec *bv,
 	return true;
 }
 
-static inline void bvec_iter_skip_zero_bvec(struct bvec_iter *iter)
+/*
+ * A simpler version of bvec_iter_advance(), @bytes should not span
+ * across multiple bvec entries, i.e. bytes <= bv[i->bi_idx].bv_len
+ */
+static inline void bvec_iter_advance_single(const struct bio_vec *bv,
+				struct bvec_iter *iter, unsigned int bytes)
 {
-	iter->bi_bvec_done = 0;
-	iter->bi_idx++;
+	unsigned int done = iter->bi_bvec_done + bytes;
+
+	if (done == bv[iter->bi_idx].bv_len) {
+		done = 0;
+		iter->bi_idx++;
+	}
+	iter->bi_bvec_done = done;
+	iter->bi_size -= bytes;
 }
 
 #define for_each_bvec(bvl, bio_vec, iter, start)			\
 	for (iter = (start);						\
 	     (iter).bi_size &&						\
 		((bvl = bvec_iter_bvec((bio_vec), (iter))), 1);	\
-	     (bvl).bv_len ? (void)bvec_iter_advance((bio_vec), &(iter),	\
-		     (bvl).bv_len) : bvec_iter_skip_zero_bvec(&(iter)))
+	     bvec_iter_advance_single((bio_vec), &(iter), (bvl).bv_len))
 
 /* for iterating one bio from start to end */
 #define BVEC_ITER_ALL_INIT (struct bvec_iter)				\
-- 
2.24.0

[PATCH 2/2] bio: optimise bvec iteration

[Pavel Begunkov]

__bio_for_each_bvec(), __bio_for_each_segment() and bio_copy_data_iter()
fall under conditions of bvec_iter_advance_single(), which is a faster
and slimmer version of bvec_iter_advance(). Add
bio_advance_iter_single() and convert them.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 block/bio.c         |  4 ++--
 include/linux/bio.h | 17 +++++++++++++++--
 2 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/block/bio.c b/block/bio.c
index fa01bef35bb1..8e718920457a 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -1212,8 +1212,8 @@ void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
 
 		flush_dcache_page(dst_bv.bv_page);
 
-		bio_advance_iter(src, src_iter, bytes);
-		bio_advance_iter(dst, dst_iter, bytes);
+		bio_advance_iter_single(src, src_iter, bytes);
+		bio_advance_iter_single(dst, dst_iter, bytes);
 	}
 }
 EXPORT_SYMBOL(bio_copy_data_iter);
diff --git a/include/linux/bio.h b/include/linux/bio.h
index c6d765382926..d55d53c49ae4 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -148,11 +148,24 @@ static inline void bio_advance_iter(const struct bio *bio,
 		/* TODO: It is reasonable to complete bio with error here. */
 }
 
+/* @bytes should be less or equal to bvec[i->bi_idx].bv_len */
+static inline void bio_advance_iter_single(const struct bio *bio,
+					   struct bvec_iter *iter,
+					   unsigned int bytes)
+{
+	iter->bi_sector += bytes >> 9;
+
+	if (bio_no_advance_iter(bio))
+		iter->bi_size -= bytes;
+	else
+		bvec_iter_advance_single(bio->bi_io_vec, iter, bytes);
+}
+
 #define __bio_for_each_segment(bvl, bio, iter, start)			\
 	for (iter = (start);						\
 	     (iter).bi_size &&						\
 		((bvl = bio_iter_iovec((bio), (iter))), 1);		\
-	     bio_advance_iter((bio), &(iter), (bvl).bv_len))
+	     bio_advance_iter_single((bio), &(iter), (bvl).bv_len))
 
 #define bio_for_each_segment(bvl, bio, iter)				\
 	__bio_for_each_segment(bvl, bio, iter, (bio)->bi_iter)
@@ -161,7 +174,7 @@ static inline void bio_advance_iter(const struct bio *bio,
 	for (iter = (start);						\
 	     (iter).bi_size &&						\
 		((bvl = mp_bvec_iter_bvec((bio)->bi_io_vec, (iter))), 1); \
-	     bio_advance_iter((bio), &(iter), (bvl).bv_len))
+	     bio_advance_iter_single((bio), &(iter), (bvl).bv_len))
 
 /* iterate over multi-page bvec */
 #define bio_for_each_bvec(bvl, bio, iter)			\
-- 
2.24.0

Re: [PATCH 1/2] block: optimise for_each_bvec() advance

[Christoph Hellwig]

On Tue, Nov 24, 2020 at 05:58:12PM +0000, Pavel Begunkov wrote:
> Because of how for_each_bvec() works it never advances across multiple
> entries at a time, so bvec_iter_advance() is an overkill. Add
> specialised bvec_iter_advance_single() that is faster. It also handles
> zero-len bvecs, so can kill bvec_iter_skip_zero_bvec().
> 
>    text    data     bss     dec     hex filename
> before:
>   23977     805       0   24782    60ce lib/iov_iter.o
> before, bvec_iter_advance() w/o WARN_ONCE()
>   22886     600       0   23486    5bbe ./lib/iov_iter.o
> after:
>   21862     600       0   22462    57be lib/iov_iter.o
> 
> Reviewed-by: Ming Lei <ming.lei@redhat.com>
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH 2/2] bio: optimise bvec iteration

[Christoph Hellwig]

On Tue, Nov 24, 2020 at 05:58:13PM +0000, Pavel Begunkov wrote:
> __bio_for_each_bvec(), __bio_for_each_segment() and bio_copy_data_iter()
> fall under conditions of bvec_iter_advance_single(), which is a faster
> and slimmer version of bvec_iter_advance(). Add
> bio_advance_iter_single() and convert them.

Are you sure about bio_advance_iter()?  That API looks like it might
not always be limited to a single segment, and might at least need a
WARN_ON_ONCE to make sure it is not abused.

Re: [PATCH 2/2] bio: optimise bvec iteration

[Pavel Begunkov]

On 26/11/2020 10:02, Christoph Hellwig wrote:
> On Tue, Nov 24, 2020 at 05:58:13PM +0000, Pavel Begunkov wrote:
>> __bio_for_each_bvec(), __bio_for_each_segment() and bio_copy_data_iter()
>> fall under conditions of bvec_iter_advance_single(), which is a faster
>> and slimmer version of bvec_iter_advance(). Add
>> bio_advance_iter_single() and convert them.
> 
> Are you sure about bio_advance_iter()?  That API looks like it might

Both those listed bio_for_each*() pass bvl.bv_len, which is truncated to
current segment by bio_iter_iovec() (i.e. bvec_iter_bvec) or
mp_bvec_iter_bvec().

And just to note that I didn't change bio_advance_iter() but added a
new function.

There is always space for stupid mistakes, but I'm sure. What makes you
to think opposite? I may have missed it.

> not always be limited to a single segment, and might at least need a
> WARN_ON_ONCE to make sure it is not abused.

I thought twice about converting other places as you commented before,
and it looks saner to not do that exactly for that reason. I prefer
to leave *_single() versions for rare but high impact cases like
for_each()s.
And as it's contained I decided to not add overhead on WARN_ONCE(),
e.g. with inlining and w/o string dedup it grows .data section much.

-- 
Pavel Begunkov

Re: [PATCH 2/2] bio: optimise bvec iteration

[Pavel Begunkov]

On 26/11/2020 12:32, Pavel Begunkov wrote:
> On 26/11/2020 10:02, Christoph Hellwig wrote:
>> On Tue, Nov 24, 2020 at 05:58:13PM +0000, Pavel Begunkov wrote:
>>> __bio_for_each_bvec(), __bio_for_each_segment() and bio_copy_data_iter()
>>> fall under conditions of bvec_iter_advance_single(), which is a faster
>>> and slimmer version of bvec_iter_advance(). Add
>>> bio_advance_iter_single() and convert them.
>>
>> Are you sure about bio_advance_iter()?  That API looks like it might
> 
> Both those listed bio_for_each*() pass bvl.bv_len, which is truncated to
> current segment by bio_iter_iovec() (i.e. bvec_iter_bvec) or
> mp_bvec_iter_bvec().
> 
> And just to note that I didn't change bio_advance_iter() but added a
> new function.

> There is always space for stupid mistakes, but I'm sure. What makes you
> to think opposite? I may have missed it.

Christoph, any doubts left?

>> not always be limited to a single segment, and might at least need a
>> WARN_ON_ONCE to make sure it is not abused.
> 
> I thought twice about converting other places as you commented before,
> and it looks saner to not do that exactly for that reason. I prefer
> to leave *_single() versions for rare but high impact cases like
> for_each()s.
> And as it's contained I decided to not add overhead on WARN_ONCE(),
> e.g. with inlining and w/o string dedup it grows .data section much.
> 

-- 
Pavel Begunkov

Re: [PATCH 2/2] bio: optimise bvec iteration

[Christoph Hellwig]

On Tue, Nov 24, 2020 at 05:58:13PM +0000, Pavel Begunkov wrote:
> __bio_for_each_bvec(), __bio_for_each_segment() and bio_copy_data_iter()
> fall under conditions of bvec_iter_advance_single(), which is a faster
> and slimmer version of bvec_iter_advance(). Add
> bio_advance_iter_single() and convert them.
> 
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH 0/2] optimise bvec/bio iteration

[Jens Axboe]

On 11/24/20 10:58 AM, Pavel Begunkov wrote:
> This adds simpler versions of bvec_iter_advance() and bio_advance_iter()
> (i.e. *_single()), that are faster but work with the restriction that
> @bytes shouldn't be more than available in the current bvec segment.
> 
> That covers most of bvec/bio iteration/foreach, that are massively
> inlined, and thus also nicely shrinks binary.
> 
> Others non core-block users might be updated on case by case basis
> (if applicable) after the change is merged.

Applied, thanks.

-- 
Jens Axboe