[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAATStaOY+JJ7mXWRwfXWuR_imhwsv=PecKSWiuZUetHTcz3EXg@mail.gmail.com>
Date: Mon, 25 May 2020 11:43:00 +1000
From: "Anand K. Mistry" <amistry@...gle.com>
To: Andi Kleen <ak@...ux.intel.com>
Cc: linux-perf-users@...r.kernel.org,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Ingo Molnar <mingo@...hat.com>, Jiri Olsa <jolsa@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Namhyung Kim <namhyung@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3] perf record: Use an eventfd to wakeup when done
On Sat, 23 May 2020 at 23:35, Andi Kleen <ak@...ux.intel.com> wrote:
>
> Anand K Mistry <amistry@...gle.com> writes:
> > }
> >
> > + done_fd = eventfd(0, EFD_NONBLOCK);
>
> This will make perf depend on a recent glibc or other library
> that implements eventfd. Wouldn't surprise me if some kind
> of build time check is needed for this to pass all of Arnaldo's
> built tests.
Looks like Arnaldo made that change when merging:
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/commit/?h=perf/core&id=e9db221d37f91409040cf7f3fbed08b44e055ae9
This makes me curious. How old a kernel should modern tools support?
>From the man page, eventfd was added in 2.6.22 (and eventfd2 in
2.6.27), which was 2007 (or 2008 for eventfd2) and glibc-2.8 which was
2008. I understand the kernel's policy of never breaking userspace,
but what about userspace tools?
>
>
> -Andi
--
Anand K. Mistry
Software Engineer
Google Australia
Powered by blists - more mailing lists