Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230728001212.457900-1-irogers@google.com> <ZMPQze90mGEdYFtr@kernel.org>
In-Reply-To: <ZMPQze90mGEdYFtr@kernel.org>
From:   Ian Rogers <irogers@google.com>
Date:   Fri, 28 Jul 2023 08:45:40 -0700
Message-ID: <CAP-5=fXxGimJRXKf7bcaPqfjxxGcn1k3CspY_iSjQnpAKs3uFQ@mail.gmail.com>
Subject: Re: [PATCH v1 0/3] Remove BPF arrays from perf event parsing
To:     Arnaldo Carvalho de Melo <acme@kernel.org>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Adrian Hunter <adrian.hunter@intel.com>,
        Eduard Zingerman <eddyz87@gmail.com>,
        Kan Liang <kan.liang@linux.intel.com>,
        Rob Herring <robh@kernel.org>,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
        bpf@vger.kernel.org, Wang Nan <wangnan0@huawei.com>,
        Wang ShaoBo <bobo.shaobowang@huawei.com>,
        YueHaibing <yuehaibing@huawei.com>, He Kuang <hekuang@huawei.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, Jul 28, 2023 at 7:29=E2=80=AFAM Arnaldo Carvalho de Melo
<acme@kernel.org> wrote:
>
> Em Thu, Jul 27, 2023 at 05:12:09PM -0700, Ian Rogers escreveu:
> > Event parsing was recently refactored:
> > https://lore.kernel.org/all/20230502223851.2234828-1-irogers@google.com=
/
> >
> > During these changes I wanted to get coverage of all parts of the
> > parse-events.y and found that I couldn't test the array code.
> >
> > The first patch fixes a BPF testing issue.
> > The 2nd and 3rd patch remove the BPF array event parsing code so that
> > it isn't adding complexity to event parsing.
>
> Thanks, applied and tested that 'perf trace' using the old BPF perf
> integration continues to work modulo the sockaddr collectors, but that
> is a problem with a newer kernel BPF verifier or clang, will
> investigate.
>
> With those commented out:
>
> diff --git a/tools/perf/examples/bpf/augmented_raw_syscalls.c b/tools/per=
f/examples/bpf/augmented_raw_syscalls.c
> index 9a03189d33d3816a..97871e2bd3a47bdf 100644
> --- a/tools/perf/examples/bpf/augmented_raw_syscalls.c
> +++ b/tools/perf/examples/bpf/augmented_raw_syscalls.c
> @@ -176,6 +176,7 @@ int syscall_unaugmented(struct syscall_enter_args *ar=
gs)
>         return 1;
>  }
>
> +#if 0
>  /*
>   * These will be tail_called from SEC("raw_syscalls:sys_enter"), so will=
 find in
>   * augmented_args_tmp what was read by that raw_syscalls:sys_enter and g=
o
> @@ -219,6 +220,7 @@ int sys_enter_sendto(struct syscall_enter_args *args)
>
>         return augmented__output(args, augmented_args, len + socklen);
>  }
> +#endif
>
>  SEC("!syscalls:sys_enter_open")
>  int sys_enter_open(struct syscall_enter_args *args)
>
>
>
>
>   # perf trace -e /home/acme/git/perf-tools-next/tools/perf/examples/bpf/=
augmented_raw_syscalls.c,open*
>      0.000 systemd-oomd/952 openat(dfd: CWD, filename: "/proc/meminfo", f=
lags: RDONLY|CLOEXEC) =3D 12
>     95.302 thermald/1231 openat(dfd: CWD, filename: "/sys/class/powercap/=
intel-rapl/intel-rapl:0/intel-rapl:0:2/energy_uj") =3D 13
>     95.586 thermald/1231 openat(dfd: CWD, filename: "/sys/class/powercap/=
intel-rapl/intel-rapl:0/energy_uj") =3D 13
>     95.857 thermald/1231 openat(dfd: CWD, filename: "/sys/class/thermal/t=
hermal_zone2/temp") =3D 13
>    250.064 systemd-oomd/952 openat(dfd: CWD, filename: "/proc/meminfo", f=
lags: RDONLY|CLOEXEC) =3D 12
>    500.054 systemd-oomd/952 openat(dfd: CWD, filename: "/proc/meminfo", f=
lags: RDONLY|CLOEXEC) =3D 12
>    500.472 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.pressure", =
flags: RDONLY|CLOEXEC) =3D 12
>    500.669 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.current", f=
lags: RDONLY|CLOEXEC) =3D 12
>    500.777 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.min", flags=
: RDONLY|CLOEXEC) =3D 12
>    500.873 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.low", flags=
: RDONLY|CLOEXEC) =3D 12
>    500.968 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.swap.curren=
t", flags: RDONLY|CLOEXEC) =3D 12
>    501.062 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/session.slice/memory.stat", flag=
s: RDONLY|CLOEXEC) =3D 12
>    501.435 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.pressure", flag=
s: RDONLY|CLOEXEC) =3D 12
>    501.585 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.current", flags=
: RDONLY|CLOEXEC) =3D 12
>    501.676 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.min", flags: RD=
ONLY|CLOEXEC) =3D 12
>    501.761 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.low", flags: RD=
ONLY|CLOEXEC) =3D 12
>    501.845 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.swap.current", =
flags: RDONLY|CLOEXEC) =3D 12
>    501.929 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/memory.stat", flags: R=
DONLY|CLOEXEC) =3D 12
>    502.118 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.pressure", flags: RDONLY|CLOEXEC) =3D 12
>    502.272 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.current", flags: RDONLY|CLOEXEC) =3D 12
>    502.367 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.min", flags: RDONLY|CLOEXEC) =3D 12
>    502.454 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.low", flags: RDONLY|CLOEXEC) =3D 12
>    502.542 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.swap.current", flags: RDONLY|CLOEXEC) =3D 12
>    502.628 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-gnome\x2dsession\x=
2dmanager.slice/memory.stat", flags: RDONLY|CLOEXEC) =3D 12
>    502.796 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.pressure", flags: RDONLY|CLOEXEC) =3D 12
>    502.928 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.current", flags: RDONLY|CLOEXEC) =3D 12
>    503.037 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.min", flags: RDONLY|CLOEXEC) =3D 12
>    503.162 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.low", flags: RDONLY|CLOEXEC) =3D 12
>    503.255 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.swap.current", flags: RDONLY|CLOEXEC) =3D 12
>    503.341 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1000.slice/user@1000.service/app.slice/app-org.gnome.Terminal=
.slice/memory.stat", flags: RDONLY|CLOEXEC) =3D 12
>    503.495 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.pressure", flags: RDONLY|CLOEXEC) =3D 12
>    503.620 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.current", flags: RDONLY|CLOEXEC) =3D 12
>    503.703 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.min", flags: RDONLY|CLOEXEC) =3D 12
>    503.782 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.low", flags: RDONLY|CLOEXEC) =3D 12
>    503.861 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.swap.current", flags: RDONLY|CLOEXEC) =3D 12
>    503.939 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/sy=
stem.slice/memory.stat", flags: RDONLY|CLOEXEC) =3D 12
>    504.079 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1001.slice/user@1001.service/app.slice/memory.pressure", flag=
s: RDONLY|CLOEXEC) =3D 12
>    504.240 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/us=
er.slice/user-1001.slice/user@1001.service/app.slice/memory.current", flags=
: RDONLY|CLOEXEC) =3D 12
> ^C

Thanks Arnaldo, it could also be that we need the code and it is my
user error which hasn't found a way of testing it. If the original
Huawei authors could let us know?

I suspect with this code gone, there is more bpf-loader map related
code that can also go as there is no way to use it. I wonder if a
better longer-term strategy is to remove BPF events entirely. My
thoughts on this are:
1) the code had bitrotten a few releases ago, I updated it but the
fact it rotted means I suspect nobody uses it.
2) the map code removed here I think is just redundant and it adds
complexity, so removing it is value add - unless I'm mistaken :-)
Removing more is more value add.
3) the clang/llvm dependencies are problematic in the build, removing
these would I think be a good value add
4) in the event parsing the BPF events have paths with '/' as the path
separator. This matches the modifier separator used by events and so I
think this leads to complexity.
5) the behavior achieved by BPF events can now be achieved with BPF
filters, which are more user friendly:
https://lore.kernel.org/all/20230314234237.3008956-1-namhyung@kernel.org/

There is a dependency with perf trace, but I think perf trace should
use BPF skeletons for augmented syscalls rather than a BPF event.
Using a BPF event in the perf trace feels kind of hacky and I suspect
perf trace users don't know they need to add the event to make
augmentation work.

Thanks,
Ian

> - Arnaldo
>
> > Ian Rogers (3):
> >   perf parse-event: Avoid BPF test segv
> >   perf tools: Revert enable indices setting syntax for BPF map
> >   perf parse-events: Remove array remnants
> >
> >  tools/perf/util/bpf-loader.c   | 101 ---------------------------
> >  tools/perf/util/parse-events.c |  18 +----
> >  tools/perf/util/parse-events.h |  10 ---
> >  tools/perf/util/parse-events.l |  11 ---
> >  tools/perf/util/parse-events.y | 122 ---------------------------------
> >  5 files changed, 2 insertions(+), 260 deletions(-)
> >
> > --
> > 2.41.0.487.g6d72f3e995-goog
> >
>
> --
>
> - Arnaldo