Re: [PATCH net-next v5 12/15] virtio-net: support AF_XDP zc tx

From: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
To: Jason Wang <jasowang@redhat.com>
Cc: "Song Liu" <songliubraving@fb.com>,
	"Martin KaFai Lau" <kafai@fb.com>,
	"Jesper Dangaard Brouer" <hawk@kernel.org>,
	"Daniel Borkmann" <daniel@iogearbox.net>,
	"Michael S. Tsirkin" <mst@redhat.com>,
	"Yonghong Song" <yhs@fb.com>,
	"John Fastabend" <john.fastabend@gmail.com>,
	"Alexei Starovoitov" <ast@kernel.org>,
	"Andrii Nakryiko" <andrii@kernel.org>,
	netdev@vger.kernel.org, "Björn Töpel" <bjorn@kernel.org>,
	"dust.li" <dust.li@linux.alibaba.com>,
	"Jonathan Lemon" <jonathan.lemon@gmail.com>,
	"KP Singh" <kpsingh@kernel.org>,
	"Jakub Kicinski" <kuba@kernel.org>,
	bpf@vger.kernel.org, virtualization@lists.linux-foundation.org,
	"David S. Miller" <davem@davemloft.net>,
	"Magnus Karlsson" <magnus.karlsson@intel.com>
Subject: Re: [PATCH net-next v5 12/15] virtio-net: support AF_XDP zc tx
Date: Wed, 16 Jun 2021 20:57:45 +0800	[thread overview]
Message-ID: <1623848265.175296-1-xuanzhuo@linux.alibaba.com> (raw)
In-Reply-To: <f81775f8-7df9-5ca0-0bd2-99c86786fe78@redhat.com>

On Wed, 16 Jun 2021 20:51:41 +0800, Jason Wang <jasowang@redhat.com> wrote:
>
> 在 2021/6/16 下午6:19, Xuan Zhuo 写道:
> >>> + * In this way, even if xsk has been unbundled with rq/sq, or a new xsk and
> >>> + * rq/sq  are bound, and a new virtnet_xsk_ctx_head is created. It will not
> >>> + * affect the old virtnet_xsk_ctx to be recycled. And free all head and ctx when
> >>> + * ref is 0.
> >> This looks complicated and it will increase the footprint. Consider the
> >> performance penalty and the complexity, I would suggest to use reset
> >> instead.
> >>
> >> Then we don't need to introduce such context.
> > I don't like this either. It is best if we can reset the queue, but then,
> > according to my understanding, the backend should also be supported
> > synchronously, so if you don't update the backend synchronously, you can't use
> > xsk.
>
>
> Yes, actually, vhost-net support per vq suspending. The problem is that
> we're lacking a proper API at virtio level.
>
> Virtio-pci has queue_enable but it forbids writing zero to that.
>
>
> >
> > I don’t think resetting the entire dev is a good solution. If you want to bind
> > xsk to 10 queues, you may have to reset the entire device 10 times. I don’t
> > think this is a good way. But the current spec does not support reset single
> > queue, so I chose the current solution.
> >
> > Jason, what do you think we are going to do? Realize the reset function of a
> > single queue?
>
>
> Yes, it's the best way. Do you want to work on that?

Of course, I am very willing to continue this work. Although users must upgrade
the backend to use virtio-net + xsk in the future, this makes the situation a
bit troublesome.

I will complete the kernel modification as soon as possible, but I am not
familiar with the process of submitting the spec patch. Can you give me some
guidance and where should I send the spec patch.

Thanks.

>
> We can start from the spec patch, and introduce it as basic facility and
> implement it in the PCI transport first.
>
> Thanks
>
>
> >
> > Looking forward to your reply!!!
> >
> > Thanks
> >
>
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization