Re: [PATCH] virtio_net: reduce raw_smp_processor_id() calling in virtnet_xdp_get_sq

Re: [PATCH] virtio_net: reduce raw_smp_processor_id() calling in virtnet_xdp_get_sq

[Michael S. Tsirkin]

On Thu, Aug 26, 2021 at 04:21:35PM +0800, Li RongQing wrote:
> smp_processor_id()/raw* will be called once each when not
> more queues in virtnet_xdp_get_sq() which is called in
> non-preemptible context, so it's safe to call the function
> smp_processor_id() once.
> 
> Signed-off-by: Li RongQing <lirongqing@baidu.com>

commit log should probably explain why it's a good idea
to replace raw_smp_processor_id with smp_processor_id
in the case of curr_queue_pairs <= nr_cpu_ids.

> ---
>  drivers/net/virtio_net.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 2e42210a6503..2a7b368c1da2 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -528,19 +528,20 @@ static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
>   * functions to perfectly solve these three problems at the same time.
>   */
>  #define virtnet_xdp_get_sq(vi) ({                                       \
> +	int cpu = smp_processor_id();                                   \
>  	struct netdev_queue *txq;                                       \
>  	typeof(vi) v = (vi);                                            \
>  	unsigned int qp;                                                \
>  									\
>  	if (v->curr_queue_pairs > nr_cpu_ids) {                         \
>  		qp = v->curr_queue_pairs - v->xdp_queue_pairs;          \
> -		qp += smp_processor_id();                               \
> +		qp += cpu;                                              \
>  		txq = netdev_get_tx_queue(v->dev, qp);                  \
>  		__netif_tx_acquire(txq);                                \
>  	} else {                                                        \
> -		qp = smp_processor_id() % v->curr_queue_pairs;          \
> +		qp = cpu % v->curr_queue_pairs;                         \
>  		txq = netdev_get_tx_queue(v->dev, qp);                  \
> -		__netif_tx_lock(txq, raw_smp_processor_id());           \
> +		__netif_tx_lock(txq, cpu);                              \
>  	}                                                               \
>  	v->sq + qp;                                                     \
>  })
> -- 
> 2.33.0.69.gc420321.dirty
> 
> 

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: 答复: [PATCH] virtio_net: reduce raw_smp_processor_id() calling in virtnet_xdp_get_sq

[Michael S. Tsirkin]

On Tue, Aug 31, 2021 at 02:09:36AM +0000, Li,Rongqing wrote:
> > -----邮件原件-----
> > 发件人: Michael S. Tsirkin <mst@redhat.com>
> > 发送时间: 2021年8月31日 5:10
> > 收件人: Li,Rongqing <lirongqing@baidu.com>
> > 抄送: netdev@vger.kernel.org; bpf@vger.kernel.org;
> > virtualization@lists.linux-foundation.org
> > 主题: Re: [PATCH] virtio_net: reduce raw_smp_processor_id() calling in
> > virtnet_xdp_get_sq
> > 
> > On Thu, Aug 26, 2021 at 04:21:35PM +0800, Li RongQing wrote:
> > > smp_processor_id()/raw* will be called once each when not more queues
> > > in virtnet_xdp_get_sq() which is called in non-preemptible context, so
> > > it's safe to call the function
> > > smp_processor_id() once.
> > >
> > > Signed-off-by: Li RongQing <lirongqing@baidu.com>
> > 
> > commit log should probably explain why it's a good idea to replace
> > raw_smp_processor_id with smp_processor_id in the case of curr_queue_pairs
> > <= nr_cpu_ids.
> > 
> 
> 
> I change it as below, is it ok?
> 
>     virtio_net: reduce raw_smp_processor_id() calling in virtnet_xdp_get_sq

shorter:

virtio_net: s/raw_smp_processor_id/smp_processor_id/ in virtnet_xdp_get_sq


> 
>     smp_processor_id() and raw_smp_processor_id() are called once
>     each in virtnet_xdp_get_sq(), when curr_queue_pairs <= nr_cpu_ids,
>     should be merged

I'd just drop this part.

> 
>     virtnet_xdp_get_sq() is called in non-preemptible context, so
>     it's safe to call the function smp_processor_id(), and keep
>     smp_processor_id(), and remove the calling of raw_smp_processor_id(),
>     avoid the wrong use virtnet_xdp_get_sq to preemptible context
>     in the future

s/avoid.*/this way we'll get a warning if this is ever called in a preemptible context/


> 
> -Li

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization