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;
Message-ID: <889f2dc5e646992033e0d9b0951d5a42f1907e07.camel@redhat.com>
Subject: Re: [PATCH v4 2/2] gro: optimise redundant parsing of packets
From:   Paolo Abeni <pabeni@redhat.com>
To:     Richard Gobert <richardbgobert@gmail.com>, davem@davemloft.net,
        edumazet@google.com, kuba@kernel.org, dsahern@kernel.org,
        alexanderduyck@fb.com, lucien.xin@gmail.com, lixiaoyan@google.com,
        iwienand@redhat.com, leon@kernel.org, ye.xingchen@zte.com.cn,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Date:   Wed, 22 Mar 2023 10:59:30 +0100
In-Reply-To: <20230320170009.GA27961@debian>
References: <20230320163703.GA27712@debian> <20230320170009.GA27961@debian>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Mon, 2023-03-20 at 18:00 +0100, Richard Gobert wrote:
> Currently the IPv6 extension headers are parsed twice: first in
> ipv6_gro_receive, and then again in ipv6_gro_complete.
>=20
> By using the new ->transport_proto field, and also storing the size of th=
e
> network header, we can avoid parsing extension headers a second time in
> ipv6_gro_complete (which saves multiple memory dereferences and condition=
al
> checks inside ipv6_exthdrs_len for a varying amount of extension headers =
in
> IPv6 packets).
>=20
> The implementation had to handle both inner and outer layers in case of
> encapsulation (as they can't use the same field). I've applied a similar
> optimisation to Ethernet.
>=20
> Performance tests for TCP stream over IPv6 with a varying amount of
> extension headers demonstrate throughput improvement of ~0.7%.

I'm surprised that the improvement is measurable: for large aggregate
packets a single ipv6_exthdrs_len() call is avoided out of tens calls
for the individual pkts. Additionally such figure is comparable to
noise level in my tests.

This adds a couple of additional branches for the common (no extensions
header) case.=20

while patch 1/2 could be useful, patch 2/2 overall looks not worthy to
me.

I suggest to re-post for inclusion only patch 1, unless others have
strong different opinions.

Cheers,

Paolo