[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20230215034355.481925-1-kuba@kernel.org>
Date: Tue, 14 Feb 2023 19:43:52 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: davem@...emloft.net
Cc: netdev@...r.kernel.org, edumazet@...gle.com, pabeni@...hat.com,
willemb@...gle.com, fw@...len.de, Jakub Kicinski <kuba@...nel.org>
Subject: [PATCH net-next 0/3] net: skbuff: cache one skb_ext for use by GRO
New offload features may require attaching skb_ext to every packet.
PSP support adds fields to the skb in the patches published by Google:
https://github.com/google/psp
Since PSP is designed to be stateless on the device we need to carry
metadata thru the stack and perform validation at the socket layer.
Upstream we'll likely try to use skb_ext to place such metadata.
Lower the cost of allocation of skb_ext by caching a single object
per core.
Obviously the PSP support isn't upstream yet, but I think that this
should also help the TC fallback, and any other skb_ext uses people
may come up with in the meantime.
Jakub Kicinski (3):
net: skb: carve the allocation out of skb_ext_add()
net: skbuff: cache one skb_ext for use by GRO
net: create and use NAPI version of tc_skb_ext_alloc()
.../ethernet/mellanox/mlx5/core/en/rep/tc.c | 2 +-
.../net/ethernet/mellanox/mlx5/core/en_tc.c | 2 +-
include/linux/skbuff.h | 1 +
include/net/pkt_cls.h | 9 ++
net/core/skbuff.c | 131 ++++++++++++++----
5 files changed, 117 insertions(+), 28 deletions(-)
--
2.39.1
Powered by blists - more mailing lists