Re: [PATCH bpf-next v7 0/7] libbpf: Implement BTFGen

From: "Mauricio Vásquez Bernal" <mauricio@kinvolk.io>
To: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Networking <netdev@vger.kernel.org>, bpf <bpf@vger.kernel.org>,
	Alexei Starovoitov <ast@kernel.org>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Andrii Nakryiko <andrii@kernel.org>,
	Quentin Monnet <quentin@isovalent.com>,
	Rafael David Tinoco <rafaeldtinoco@gmail.com>,
	Lorenzo Fontana <lorenzo.fontana@elastic.co>,
	Leonardo Di Donato <leonardo.didonato@elastic.co>
Subject: Re: [PATCH bpf-next v7 0/7] libbpf: Implement BTFGen
Date: Thu, 17 Feb 2022 17:07:15 -0500	[thread overview]
Message-ID: <CAHap4zuD4ei9XT-+L0tjah_nG0n1o+wAkdV_HMBM23SErg8CWA@mail.gmail.com> (raw)
In-Reply-To: <CAEf4BzbxcoP8hoHM_1+QX4Nx=F0NPkc-CXDq=H_JkQfc9PAzLQ@mail.gmail.com>

>
> Fixed up few things I pointed out in respective patches. Applied to
> bpf-next. Great work, congrats!

Thanks a lot for all your patience and helpful reviews!

>
> It would be great as a next step to add this as (probably optional at
> first) step for libbpf-tools in BCC repo, so that those CO-RE-based
> tools can be used much more widely than today.

I like this idea. It'll also help us to understand and improve the way
to ship those files within the application.

> How much work that
> would be, do you think?

Probably the most difficult part is to embed the generated files into
the executable. I think generating a header file with the BTF info for
each tool and some helpers to extract it at runtime according to the
kernel version should work.

> And how slow would it be to download all those
> BTFs and run min_core_btf on all of them?

The whole thing takes like 5 minutes on my system (AMD Ryzen 7 3700X
with 60mbps connection), given that almost 3 minutes are spent
downloading the files I'd say that with a fast connection and some
performance improvements (multicore?) it could take around 2~3
minutes.

Let me think better about this integration and will be back with some ideas.