Re: [PATCH v4 2/2] io_uring: batch getting pcpu references

[Pavel Begunkov <asml.silence@gmail.com>]

On 28/12/2019 14:13, Pavel Begunkov wrote:
> percpu_ref_tryget() has its own overhead. Instead getting a reference
> for each request, grab a bunch once per io_submit_sqes().
> 
> ~5% throughput boost for a "submit and wait 128 nops" benchmark.
> 
> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
> ---
>  fs/io_uring.c | 26 +++++++++++++++++---------
>  1 file changed, 17 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 7fc1158bf9a4..404946080e86 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -1080,9 +1080,6 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
>  	gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
>  	struct io_kiocb *req;
>  
> -	if (!percpu_ref_tryget(&ctx->refs))
> -		return NULL;
> -
>  	if (!state) {
>  		req = kmem_cache_alloc(req_cachep, gfp);
>  		if (unlikely(!req))
> @@ -1141,6 +1138,14 @@ static void io_free_req_many(struct io_ring_ctx *ctx, void **reqs, int *nr)
>  	}
>  }
>  
> +static void __io_req_free_empty(struct io_kiocb *req)

If anybody have better naming (or a better approach at all), I'm all ears.


> +{
> +	if (likely(!io_is_fallback_req(req)))
> +		kmem_cache_free(req_cachep, req);
> +	else
> +		clear_bit_unlock(0, (unsigned long *) req->ctx->fallback_req);
> +}
> +
>  static void __io_free_req(struct io_kiocb *req)
>  {
>  	struct io_ring_ctx *ctx = req->ctx;
> @@ -1162,11 +1167,9 @@ static void __io_free_req(struct io_kiocb *req)
>  			wake_up(&ctx->inflight_wait);
>  		spin_unlock_irqrestore(&ctx->inflight_lock, flags);
>  	}
> -	percpu_ref_put(&ctx->refs);
> -	if (likely(!io_is_fallback_req(req)))
> -		kmem_cache_free(req_cachep, req);
> -	else
> -		clear_bit_unlock(0, (unsigned long *) ctx->fallback_req);
> +
> +	percpu_ref_put(&req->ctx->refs);
> +	__io_req_free_empty(req);
>  }
>  
>  static bool io_link_cancel_timeout(struct io_kiocb *req)
> @@ -4551,6 +4554,9 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  			return -EBUSY;
>  	}
>  
> +	if (!percpu_ref_tryget_many(&ctx->refs, nr))
> +		return -EAGAIN;
> +
>  	if (nr > IO_PLUG_THRESHOLD) {
>  		io_submit_state_start(&state, nr);
>  		statep = &state;
> @@ -4567,7 +4573,7 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  			break;
>  		}
>  		if (!io_get_sqring(ctx, req, &sqe)) {
> -			__io_free_req(req);
> +			__io_req_free_empty(req);
>  			break;
>  		}
>  
> @@ -4598,6 +4604,8 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
>  			break;
>  	}
>  
> +	if (submitted != nr)
> +		percpu_ref_put_many(&ctx->refs, nr - submitted);
>  	if (link)
>  		io_queue_link_head(link);
>  	if (statep)
> 

-- 
Pavel Begunkov