[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210106133205.617dddd8@kernel.org>
Date: Wed, 6 Jan 2021 13:32:05 +0100
From: Marek BehĂșn <kabel@...nel.org>
To: Andrew Lunn <andrew@...n.ch>
Cc: Sven Auhagen <sven.auhagen@...eatech.de>, netdev@...r.kernel.org,
davem@...emloft.net, Jakub Kicinski <kuba@...nel.org>,
Matteo Croce <mcroce@...rosoft.com>,
Lorenzo Bianconi <lorenzo@...nel.org>
Subject: Re: [PATCH net-next] net: mvpp2: increase MTU limit when XDP
enabled
On Wed, 6 Jan 2021 12:56:08 +0100
Marek BehĂșn <kabel@...nel.org> wrote:
> I also to write a simple NAT masquerading program. I think XDP can
> increase NAT throughput to 2.5gbps as well.
BTW currently if XDP modifies the packet, it has to modify the
checksums accordingly. There is a helper for that even, bpf_csum_diff.
But many drivers can offload csum computation, mvneta and mvpp2 for
example. But for this, somehow the XDP program has to let the driver
know what kind of csum it needs to be computed (L3, L4 TCP/UDP).
This could theoretically be communicated with the driver via metadata
prepended to the packet. But a abstraction is needed, so that every
driver does it in the same way. Maybe someone is already working on
this, I don't know...
Marek
Powered by blists - more mailing lists