| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAP-5=fXxGimJRXKf7bcaPqfjxxGcn1k3CspY_iSjQnpAKs3uFQ@mail.gmail.com>
Date: Fri, 28 Jul 2023 08:45:40 -0700
From: Ian Rogers <irogers@...gle.com>
To: Arnaldo Carvalho de Melo <acme@...nel.org>
Cc: Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Jiri Olsa <jolsa@...nel.org>,
Namhyung Kim <namhyung@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>,
Eduard Zingerman <eddyz87@...il.com>,
Kan Liang <kan.liang@...ux.intel.com>,
Rob Herring <robh@...nel.org>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
bpf@...r.kernel.org, Wang Nan <wangnan0@...wei.com>,
Wang ShaoBo <bobo.shaobowang@...wei.com>,
YueHaibing <yuehaibing@...wei.com>, He Kuang <hekuang@...wei.com>
Subject: Re: [PATCH v1 0/3] Remove BPF arrays from perf event parsing
On Fri, Jul 28, 2023 at 7:29 AM Arnaldo Carvalho de Melo
<acme@...nel.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/perf/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 *args)
> 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 go
> @@ -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", flags: RDONLY|CLOEXEC) = 12
> 95.302 thermald/1231 openat(dfd: CWD, filename: "/sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:2/energy_uj") = 13
> 95.586 thermald/1231 openat(dfd: CWD, filename: "/sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj") = 13
> 95.857 thermald/1231 openat(dfd: CWD, filename: "/sys/class/thermal/thermal_zone2/temp") = 13
> 250.064 systemd-oomd/952 openat(dfd: CWD, filename: "/proc/meminfo", flags: RDONLY|CLOEXEC) = 12
> 500.054 systemd-oomd/952 openat(dfd: CWD, filename: "/proc/meminfo", flags: RDONLY|CLOEXEC) = 12
> 500.472 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 500.669 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.current", flags: RDONLY|CLOEXEC) = 12
> 500.777 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.min", flags: RDONLY|CLOEXEC) = 12
> 500.873 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.low", flags: RDONLY|CLOEXEC) = 12
> 500.968 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.swap.current", flags: RDONLY|CLOEXEC) = 12
> 501.062 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/session.slice/memory.stat", flags: RDONLY|CLOEXEC) = 12
> 501.435 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 501.585 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.current", flags: RDONLY|CLOEXEC) = 12
> 501.676 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.min", flags: RDONLY|CLOEXEC) = 12
> 501.761 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.low", flags: RDONLY|CLOEXEC) = 12
> 501.845 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.swap.current", flags: RDONLY|CLOEXEC) = 12
> 501.929 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/memory.stat", flags: RDONLY|CLOEXEC) = 12
> 502.118 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 502.272 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.current", flags: RDONLY|CLOEXEC) = 12
> 502.367 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.min", flags: RDONLY|CLOEXEC) = 12
> 502.454 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.low", flags: RDONLY|CLOEXEC) = 12
> 502.542 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.swap.current", flags: RDONLY|CLOEXEC) = 12
> 502.628 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-gnome\x2dsession\x2dmanager.slice/memory.stat", flags: RDONLY|CLOEXEC) = 12
> 502.796 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 502.928 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.current", flags: RDONLY|CLOEXEC) = 12
> 503.037 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.min", flags: RDONLY|CLOEXEC) = 12
> 503.162 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.low", flags: RDONLY|CLOEXEC) = 12
> 503.255 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.swap.current", flags: RDONLY|CLOEXEC) = 12
> 503.341 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1000.slice/user@...0.service/app.slice/app-org.gnome.Terminal.slice/memory.stat", flags: RDONLY|CLOEXEC) = 12
> 503.495 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 503.620 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.current", flags: RDONLY|CLOEXEC) = 12
> 503.703 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.min", flags: RDONLY|CLOEXEC) = 12
> 503.782 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.low", flags: RDONLY|CLOEXEC) = 12
> 503.861 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.swap.current", flags: RDONLY|CLOEXEC) = 12
> 503.939 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/system.slice/memory.stat", flags: RDONLY|CLOEXEC) = 12
> 504.079 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1001.slice/user@...1.service/app.slice/memory.pressure", flags: RDONLY|CLOEXEC) = 12
> 504.240 systemd-oomd/952 openat(dfd: CWD, filename: "/sys/fs/cgroup/user.slice/user-1001.slice/user@...1.service/app.slice/memory.current", flags: RDONLY|CLOEXEC) = 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
Powered by blists - more mailing lists