[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200225173748.7429cd8c@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Date: Tue, 25 Feb 2020 17:37:48 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: David Ahern <dsahern@...nel.org>
Cc: netdev@...r.kernel.org, davem@...emloft.net,
David Ahern <dahern@...italocean.com>,
Jason Wang <jasowang@...hat.com>,
"Michael S . Tsirkin" <mst@...hat.com>
Subject: Re: [PATCH RFC net-next] virtio_net: Relax queue requirement for
using XDP
On Tue, 25 Feb 2020 17:57:44 -0700 David Ahern wrote:
> From: David Ahern <dahern@...italocean.com>
>
> virtio_net currently requires extra queues to install an XDP program,
> with the rule being twice as many queues as vcpus. From a host
> perspective this means the VM needs to have 2*vcpus vhost threads
> for each guest NIC for which XDP is to be allowed. For example, a
> 16 vcpu VM with 2 tap devices needs 64 vhost threads.
>
> The extra queues are only needed in case an XDP program wants to
> return XDP_TX. XDP_PASS, XDP_DROP and XDP_REDIRECT do not need
> additional queues. Relax the queue requirement and allow XDP
> functionality based on resources. If an XDP program is loaded and
> there are insufficient queues, then return a warning to the user
> and if a program returns XDP_TX just drop the packet. This allows
> the use of the rest of the XDP functionality to work without
> putting an unreasonable burden on the host.
>
> Cc: Jason Wang <jasowang@...hat.com>
> Cc: Michael S. Tsirkin <mst@...hat.com>
> Signed-off-by: David Ahern <dahern@...italocean.com>
The plan^W hope in my head was that Magnus will expose more info about
NIC queues. Then we can introduce a "manual queue setup" mode for XDP,
where users is responsible for making sure that the cores they care
about have TX/REDIRECT queues (with or without an XDP program attached).
Right now the XDP queues are entirely invisible.
That's just FWIW, in absence of actual code - regardless if it's an
obvious or pointless idea - it may not have much weight in this
discussion..
Powered by blists - more mailing lists