Re: [PATCH bpf-next 3/3] libbpf: auto-generate list of BPF helper definitions

From: Alexei Starovoitov <ast@fb.com>
To: Andrii Nakryiko <andrii.nakryiko@gmail.com>
Cc: Andrii Nakryiko <andriin@fb.com>,
	"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"daniel@iogearbox.net" <daniel@iogearbox.net>,
	Kernel Team <Kernel-team@fb.com>,
	Quentin Monnet <quentin.monnet@netronome.com>
Subject: Re: [PATCH bpf-next 3/3] libbpf: auto-generate list of BPF helper definitions
Date: Sun, 6 Oct 2019 01:01:39 +0000	[thread overview]
Message-ID: <c621f33f-4c17-f96f-e2f5-625d452f3df3@fb.com> (raw)
In-Reply-To: <CAEf4BzZdhBTovTfv+Ar0En__RmuP6Lr=RWF2ix3uo9hZ84oHcg@mail.gmail.com>

On 10/5/19 11:24 AM, Andrii Nakryiko wrote:
> On Sat, Oct 5, 2019 at 10:10 AM Alexei Starovoitov <ast@fb.com> wrote:
>>
>> On 10/5/19 12:59 AM, Andrii Nakryiko wrote:
>>> Get rid of list of BPF helpers in bpf_helpers.h (irony...) and
>>> auto-generate it into bpf_helpers_defs.h, which is now included from
>>> bpf_helpers.h.
>>>
>>> Suggested-by: Alexei Starovoitov<ast@fb.com>
>>> Signed-off-by: Andrii Nakryiko<andriin@fb.com>
>>> ---
>>>    tools/lib/bpf/Makefile           |    8 +-
>>>    tools/lib/bpf/bpf_helpers.h      |  264 +--
>>>    tools/lib/bpf/bpf_helpers_defs.h | 2677 ++++++++++++++++++++++++++++++
>>>    3 files changed, 2685 insertions(+), 264 deletions(-)
>>>    create mode 100644 tools/lib/bpf/bpf_helpers_defs.h
>>
>> Approach looks good to me.
>> imo that's better than messing with macros.
>>
>> Using bpf_helpers_doc.py as part of build will help man pages too.
>> I think we were sloppy documenting helpers, since only Quentin
>> was running that script regularly.
> 
> Yep, I agree, I had to fix few things, as well as (char *) vs (void *)
> vs (__u8 *) differences were causing some extra warnings.
> Please check the list of type translations whether they make sense.

yes. These conversions make sense.
Only one sk_buff->__sk_buff won't work for too long.
Once my btf vmlinux stuff lands both structs will be possible.

>>
>> Only question is what is the reason to commit generated .h into git?
> 
> So originally I didn't want to depend on system UAPI headers during
> Github build. But now I recalled that we do have latest UAPI synced
> into Github's include/ subdir, so that's not an obstacle really. We'll
> just need to re-license bpf_helpers_doc.py (I don't think Quentin will
> mind) and start syncing it to Github (not a big deal at all).

github sync script can run make, grab generated .h,
and check it in into github.

> 
> There is still a benefit in having it checked in: easy to spot if
> script does something wrong and double-check the changes (after
> initial big commit, of course).
> 
> If you think that's not reason enough, let me know and I can drop it in v2.

I don't remember too many cases when generated files were committed.
My preference to keep it out of kernel git.
github/libbpf is a pure mirror. That .h can be committed there.
The whole thing is automatic, so we don't need to move .py
into github too.