[PATCH] iov_iter: optimise bvec iov_iter_advance()

[PATCH] iov_iter: optimise bvec iov_iter_advance()

[Pavel Begunkov]

iov_iter_advance() is heavily used, but implemented through generic
iteration. As bvecs have a specifically crafted advance() function, i.e.
bvec_iter_advance(), which is faster and slimmer, use it instead.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 lib/iov_iter.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/lib/iov_iter.c b/lib/iov_iter.c
index 1635111c5bd2..7fbdd9ef3ff0 100644
--- a/lib/iov_iter.c
+++ b/lib/iov_iter.c
@@ -1077,6 +1077,20 @@ void iov_iter_advance(struct iov_iter *i, size_t size)
 		i->count -= size;
 		return;
 	}
+	if (iov_iter_is_bvec(i)) {
+		struct bvec_iter bi;
+
+		bi.bi_size = i->count;
+		bi.bi_bvec_done = i->iov_offset;
+		bi.bi_idx = 0;
+		bvec_iter_advance(i->bvec, &bi, size);
+
+		i->bvec += bi.bi_idx;
+		i->nr_segs -= bi.bi_idx;
+		i->count = bi.bi_size;
+		i->iov_offset = bi.bi_bvec_done;
+		return;
+	}
 	iterate_and_advance(i, size, v, 0, 0, 0)
 }
 EXPORT_SYMBOL(iov_iter_advance);
-- 
2.24.0

Re: [PATCH] iov_iter: optimise bvec iov_iter_advance()

[Christoph Hellwig]

> @@ -1077,6 +1077,20 @@ void iov_iter_advance(struct iov_iter *i, size_t size)
>  		i->count -= size;
>  		return;
>  	}
> +	if (iov_iter_is_bvec(i)) {
> +		struct bvec_iter bi;
> +
> +		bi.bi_size = i->count;
> +		bi.bi_bvec_done = i->iov_offset;
> +		bi.bi_idx = 0;
> +		bvec_iter_advance(i->bvec, &bi, size);
> +
> +		i->bvec += bi.bi_idx;
> +		i->nr_segs -= bi.bi_idx;
> +		i->count = bi.bi_size;
> +		i->iov_offset = bi.bi_bvec_done;
> +		return;
> +	}
>  	iterate_and_advance(i, size, v, 0, 0, 0)

I like the idea, but whu not avoid the on-stack bvec_iter and just
open code this entirely using a new helper?  E.g.

static void bio_iov_iter_advance(struct iov_iter *i, size_t bytes)
{
	unsigned int cnt;

	i->count -= bytes;

	bytes += i->iov_offset;
	for (cnt = 0; bytes && bytes >= i->bvec[cnt].bv_len; cnt++)
		bytes -= i->bvec[cnt].bv_len;
	i->iov_offset = bytes;

	i->bvec += cnt;
	i->nr_segs -= cnt;
}

Re: [PATCH] iov_iter: optimise bvec iov_iter_advance()

[Pavel Begunkov]

On 03/12/2020 09:14, Christoph Hellwig wrote:
>> @@ -1077,6 +1077,20 @@ void iov_iter_advance(struct iov_iter *i, size_t size)
>>  		i->count -= size;
>>  		return;
>>  	}
>> +	if (iov_iter_is_bvec(i)) {
>> +		struct bvec_iter bi;
>> +
>> +		bi.bi_size = i->count;
>> +		bi.bi_bvec_done = i->iov_offset;
>> +		bi.bi_idx = 0;
>> +		bvec_iter_advance(i->bvec, &bi, size);
>> +
>> +		i->bvec += bi.bi_idx;
>> +		i->nr_segs -= bi.bi_idx;
>> +		i->count = bi.bi_size;
>> +		i->iov_offset = bi.bi_bvec_done;
>> +		return;
>> +	}
>>  	iterate_and_advance(i, size, v, 0, 0, 0)
> 
> I like the idea, but whu not avoid the on-stack bvec_iter and just
> open code this entirely using a new helper?  E.g.

It's inlined and the on-stack iter is completely optimised out. Frankly,
I'd rather not open-code bvec_iter_advance(), at least for this chunk to
be findable from bvec.h, e.g. grep bvec_iter and bvec_for_each. Though,
I don't like myself that preamble/postamble.

> 
> static void bio_iov_iter_advance(struct iov_iter *i, size_t bytes)
> {
> 	unsigned int cnt;
> 
> 	i->count -= bytes;
> 
> 	bytes += i->iov_offset;
> 	for (cnt = 0; bytes && bytes >= i->bvec[cnt].bv_len; cnt++)
> 		bytes -= i->bvec[cnt].bv_len;
> 	i->iov_offset = bytes;
> 
> 	i->bvec += cnt;
> 	i->nr_segs -= cnt;
> }
> 

-- 
Pavel Begunkov

Re: [PATCH] iov_iter: optimise bvec iov_iter_advance()

[Christoph Hellwig]

On Thu, Dec 03, 2020 at 11:48:56AM +0000, Pavel Begunkov wrote:
> It's inlined and the on-stack iter is completely optimised out. Frankly,
> I'd rather not open-code bvec_iter_advance(), at least for this chunk to
> be findable from bvec.h, e.g. grep bvec_iter and bvec_for_each. Though,
> I don't like myself that preamble/postamble.

Ok.  I still think we at least want a separate helper instead of
bloating the main iov_iter_advance.

Re: [PATCH] iov_iter: optimise bvec iov_iter_advance()

[Pavel Begunkov]

On 09/12/2020 08:33, Christoph Hellwig wrote:
> On Thu, Dec 03, 2020 at 11:48:56AM +0000, Pavel Begunkov wrote:
>> It's inlined and the on-stack iter is completely optimised out. Frankly,
>> I'd rather not open-code bvec_iter_advance(), at least for this chunk to
>> be findable from bvec.h, e.g. grep bvec_iter and bvec_for_each. Though,
>> I don't like myself that preamble/postamble.
> 
> Ok.  I still think we at least want a separate helper instead of
> bloating the main iov_iter_advance.

I agree on that, just haven't got back to it yet.

-- 
Pavel Begunkov