[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <55d391d1-da32-5748-ddda-272fb3e2c105@gmail.com>
Date: Wed, 2 May 2018 12:37:48 +0900
From: Toshiaki Makita <toshiaki.makita1@...il.com>
To: Jesper Dangaard Brouer <brouer@...hat.com>
Cc: Toshiaki Makita <makita.toshiaki@....ntt.co.jp>,
netdev@...r.kernel.org
Subject: Re: [PATCH RFC 4/9] veth: Use NAPI for XDP
On 18/05/01 (火) 17:43, Jesper Dangaard Brouer wrote:
> On Tue, 1 May 2018 17:02:34 +0900
> Toshiaki Makita <makita.toshiaki@....ntt.co.jp> wrote:
>
>> On 2018/05/01 16:50, Jesper Dangaard Brouer wrote:
>>> On Tue, 24 Apr 2018 23:39:18 +0900
>>> Toshiaki Makita <toshiaki.makita1@...il.com> wrote:
>>>
>>>> +static int veth_xdp_enqueue(struct veth_priv *priv, void *ptr)
>>>> +{
>>>> + if (unlikely(ptr_ring_produce(&priv->xdp_ring, ptr)))
>>>> + return -ENOSPC;
>>>> +
>>>> + return 0;
>>>> +}
>>>
>>> Here we have a lock per (enqueued) packet. I'm working on changing the
>>> ndo_xdp_xmit API to allow bulking. And the tun driver have exact same
>>> issue/need.
>>
>> Probably I should have noted in commitlog, but this per-packet lock is
>> removed in patch 9.
>> I'm curious about if any change is needed by your new API.
>
> Again, I'm just moving this into the generic code, to avoid having to
> implement this for every driver.
>
> Plus, with CONFIG_RETPOLINE we have the advantage of only calling
> ndo_xdp_xmit once (indirect function pointer call).
Nice. Once it becomes available I'll use it instead.
Toshiaki Makita
Powered by blists - more mailing lists