[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <4e7f40fc-114c-4786-86f7-532dce6cb04c@kernel.org>
Date: Fri, 5 Dec 2025 10:28:03 +0000
From: Quentin Monnet <qmo@...nel.org>
To: Ian Rogers <irogers@...gle.com>, Namhyung Kim <namhyung@...nel.org>,
Arnaldo Carvalho de Melo <acme@...nel.org>
Cc: bpf@...r.kernel.org, James Clark <james.clark@...aro.org>,
Jiri Olsa <jolsa@...nel.org>, Adrian Hunter <adrian.hunter@...el.com>,
Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...nel.org>,
LKML <linux-kernel@...r.kernel.org>, linux-perf-users@...r.kernel.org,
KP Singh <kpsingh@...nel.org>
Subject: Re: [PATCH 1/2] tools/build: Add a feature test for libopenssl
2025-12-04 22:27 UTC-0800 ~ Ian Rogers <irogers@...gle.com>
> On Thu, Dec 4, 2025 at 3:52 PM Namhyung Kim <namhyung@...nel.org> wrote:
>>
>> On Thu, Dec 04, 2025 at 01:16:54PM -0300, Arnaldo Carvalho de Melo wrote:
>>> On Wed, Dec 03, 2025 at 04:34:56PM -0800, Ian Rogers wrote:
>>>> On Wed, Dec 3, 2025 at 3:29 PM Namhyung Kim <namhyung@...nel.org> wrote:
>>>>>
>>>>> It's used by bpftool and the kernel build. Let's add a feature test so
>>>>> that perf can decide what to do based on the availability.
>>>>
>>>> It seems strange to add a feature test that bpftool is missing and
>>>> then use it only in the perf build. The signing of bpf programs isn't
>>>
>>> It is strange indeed, I agree that since we don't use BPF signing at
>>> this point in the perf BPf skels, then we could just bootstrap a bpftool
>>> without such feature and continue building the existing features.
>>>
>>> Adding the bpftool maintainer to the CC list, Quentin?
>>
>> I've already talked to Quentin and they want libopenssl as a
>> requirement.
>>
>> https://lore.kernel.org/linux-perf-users/e44f70bf-8f50-4a4b-97b8-eaf988aabced@kernel.org/
>
> You can have libopenssl as a requirement and have a bootstrap bpftool
> that doesn't require it, as the bootstrap version only provides
> minimal features typically to just build bpftool. You can also have
> libopenssl as a requirement and have a feature test that fails in the
> bpftool build saying you are missing a requirement. Having the perf
> build detect that a feature for the bpftool dependency is missing is
> fine as we can then recommend installing bpftool or the missing
> dependency, but doing this without bpftool also doing something just
> seems inconsistent.
>
> Thanks,
> Ian
>From bpftool's perspective, it doesn't really make sense to skip the
OpenSSL dependency for the bootstrap version, given that we want to ship
the main binary with the signing feature: so you could build a bootstrap
version without signing, but you won't be able to use it to build the
final binary because, well, you miss a required dependency.
This being said, if it really makes it easier for you to build perf, I'd
be open to adjusting the bootstrap version, as long as it doesn't affect
the final bpftool build. It might lead to further headaches if someone
needs to sign the BPF programs when building perf in the future though.
I'm also OK with adding a dependency check with a simple build error for
bpftool, although we don't currently do it for other required
dependencies in bpftool.
Quentin
Powered by blists - more mailing lists