Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 7 Jul 2022 01:16:38 -0400
From:   "Michael S. Tsirkin" <mst@redhat.com>
To:     Guo Zhi <qtxuning1999@sjtu.edu.cn>
Cc:     jasowang@redhat.com, eperezma@redhat.com,
        virtualization@lists.linux-foundation.org,
        linux-kernel@vger.kernel.org, sgarzare@redhat.com
Subject: Re: [PATCH v2 3/4] virtio_test: use random length scatterlists to
 test descriptor chain
Message-ID: <20220707011015-mutt-send-email-mst@kernel.org>
References: <20220707024409.1869-1-qtxuning1999@sjtu.edu.cn>
 <20220707024409.1869-4-qtxuning1999@sjtu.edu.cn>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220707024409.1869-4-qtxuning1999@sjtu.edu.cn>
Precedence: bulk

On Thu, Jul 07, 2022 at 10:44:08AM +0800, Guo Zhi wrote:
> Prior implementation only use one descriptor for each io event, which
> does't test code of descriptor chain. More importantly, one descriptor
> will not use indirect feature even indirect feature is specified. Use
> random length scatterlists here to test descriptor chain.
> 
> Signed-off-by: Guo Zhi <qtxuning1999@sjtu.edu.cn>
> ---
>  tools/virtio/virtio_test.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/virtio/virtio_test.c b/tools/virtio/virtio_test.c
> index 1ecd64271..363695b33 100644
> --- a/tools/virtio/virtio_test.c
> +++ b/tools/virtio/virtio_test.c
> @@ -20,6 +20,7 @@
>  #include "../../drivers/vhost/test.h"
>  
>  #define RANDOM_BATCH -1
> +#define MAX_SG_FRAGS 8UL
>  #define ALIGN 4096
>  #define RINGSIZE   256
>  #define TEST_BUF_NUM 0x100000
> @@ -172,7 +173,8 @@ static void wait_for_interrupt(struct vdev_info *dev)
>  static void run_test(struct vdev_info *dev, struct vq_info *vq,
>  		     bool delayed, int batch, int reset_n, int bufs)
>  {
> -	struct scatterlist sl;
> +	struct scatterlist sg[MAX_SG_FRAGS];
> +	int sg_size = 0;
>  	long started = 0, completed = 0, next_reset = reset_n;
>  	long completed_before, started_before;
>  	int r, test = 1;
> @@ -197,8 +199,11 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq,
>  
>  			while (started < bufs &&
>  			       (started - completed) < batch) {
> -				sg_init_one(&sl, dev->buf, dev->buf_size);
> -				r = virtqueue_add_outbuf(vq->vq, &sl, 1,
> +				sg_size = random() % (MAX_SG_FRAGS - 1) + 1;
> +				sg_init_table(sg, sg_size);
> +				for (int i = 0; i < sg_size; ++i)
> +					sg_set_buf(&sg[i], dev->buf + i, 0x1);
> +				r = virtqueue_add_outbuf(vq->vq, sg, sg_size,
>  							 dev->buf + started,
>  							 GFP_ATOMIC);
>  				if (unlikely(r != 0)) {

random on data path is pretty expensive.
I would suggest get an array size from user (and maybe a seed?) and
pregenerate some numbers, then reuse.


> -- 
> 2.17.1