[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20191224073239.zcrdaybyj7wj4kvf@kafai-mbp>
Date: Tue, 24 Dec 2019 07:32:43 +0000
From: Martin Lau <kafai@...com>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>
CC: bpf <bpf@...r.kernel.org>, Alexei Starovoitov <ast@...nel.org>,
"Daniel Borkmann" <daniel@...earbox.net>,
David Miller <davem@...emloft.net>,
"Kernel Team" <Kernel-team@...com>,
Networking <netdev@...r.kernel.org>
Subject: Re: [PATCH bpf-next v2 11/11] bpf: Add bpf_dctcp example
On Mon, Dec 23, 2019 at 11:01:55PM -0800, Andrii Nakryiko wrote:
> > > Can all of these types come from vmlinux.h instead of being duplicated here?
> > It can but I prefer leaving it as is in bpf_tcp_helpers.h like another
> > existing test in kfree_skb.c. Without directly using the same struct in
> > vmlinux.h, I think it is a good test for libbpf.
> > That remind me to shuffle the member ordering a little in tcp_congestion_ops
> > here.
>
> Sure no problem. When I looked at this it was a bit discouraging on
> how much types I'd need to duplicate, but surely we don't want to make
> an impression that vmlinux.h is the only way to achieve this.
IMO, it is a very compact set of fields that work for both dctcp and cubic.
Also, duplication is not a concern here. The deviation between
the kernel and bpf_tcp_helpers.h is not a problem with CO-RE.
Powered by blists - more mailing lists