Message-ID: <CAP-5=fXkM52E5VvX9s2CfvF8ckQvcbSm8N_7BCfhW3dyuH7YUg@mail.gmail.com>
Date:   Thu, 3 Nov 2022 09:04:49 -0700
From:   Ian Rogers <irogers@...gle.com>
To:     Arnaldo Carvalho de Melo <acme@...nel.org>
Cc:     Leo Yan <leo.yan@...aro.org>,
        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>,
        linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
        Stephane Eranian <eranian@...gle.com>
Subject: Re: [PATCH v1 6/7] perf trace: 5sec fix libbpf 1.0+ compatibility

On Thu, Nov 3, 2022 at 8:52 AM Arnaldo Carvalho de Melo <acme@...nel.org> wrote:
>
> Em Thu, Nov 03, 2022 at 12:39:59PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Thu, Nov 03, 2022 at 12:36:36PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > [root@...co ~]# perf trace -e /home/acme/git/perf/tools/perf/examples/bpf/5sec.c
> > > /home/acme/git/perf/tools/perf/examples/bpf/5sec.c:42:10: fatal error: 'bpf/bpf_helpers.h' file not found
> > > #include <bpf/bpf_helpers.h>
> > >          ^~~~~~~~~~~~~~~~~~~
> > > 1 error generated.
> > > ERROR:      unable to compile /home/acme/git/perf/tools/perf/examples/bpf/5sec.c
> > > Hint:       Check error message shown above.
> > > Hint:       You can also pre-compile it into .o using:
> > >                     clang -target bpf -O2 -c /home/acme/git/perf/tools/perf/examples/bpf/5sec.c
> > >             with proper -I and -D options.
> > > event syntax error: '/home/acme/git/perf/tools/perf/examples/bpf/5sec.c'
> > >                      \___ Failed to load /home/acme/git/perf/tools/perf/examples/bpf/5sec.c from source: Error when compiling BPF scriptlet
> > >
> > > (add -v to see detail)
> > > Run 'perf list' for a list of valid events
> > >
> > >  Usage: perf trace [<options>] [<command>]
> > >     or: perf trace [<options>] -- <command> [<options>]
> > >     or: perf trace record [<options>] [<command>]
> > >     or: perf trace record [<options>] -- <command> [<options>]
> > >
> > >     -e, --event <event>   event/syscall selector. use 'perf list' to list available events
> > > [root@...co ~]#
> > >
> > > It is not even finding it, in this machine I have libbpf 0.7.0, so there
> > > is a /usr/include/bpf/bpf_helpers.h, but probably that isn't in the
> > > include path set up to build the tools/perf/examples/bpf/ files, perhaps
> > > it should use:
> > >
> > > -Itools/lib/  so that it gets tools/lib/bpf_helpers.h?
> > >
> > > Trying to get this tested...
> >
> > Running with -v:
> >
> > llvm compiling command : /usr/lib64/ccache/clang -D__KERNEL__ -D__NR_CPUS__=8 -DLINUX_VERSION_CODE=0x51310 -g -I/home/acme/lib/perf/include/bpf -nostdinc -I./arch/x86/include -I./arch/x86/include/generated  -I./include -I./arch/x86/include/uapi -I./arch/x86/include/generated/uapi -I./include/uapi -I./include/generated/uapi -include ./include/linux/compiler-version.h -include ./include/linux/kconfig.h  -Wno-unused-value -Wno-pointer-sign -working-directory /lib/modules/5.19.16-200.fc36.x86_64/build -c /home/acme/git/perf/tools/perf/examples/bpf/5sec.c -target bpf  -g -O2 -o -
> > /home/acme/git/perf/tools/perf/examples/bpf/5sec.c:42:10: fatal error: 'bpf/bpf_helpers.h' file not found
> >
> > There is still that -I/home/acme/lib/perf/include/bpf, I'll remove it
> > from the include path and try to replace it with the libbpf path...
>
> Ok, works with the patch below, that needs some more renaming from "perf_" to
> "libbpf_", etc:
>
> [root@...co ~]# perf trace -e /home/acme/git/perf/tools/perf/examples/bpf/5sec.c  sleep 5
>      0.000 sleep/160828 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1474734416, rqtp: 5000000000)
> [root@...co ~]#
>
> Since I have:
>
> [root@...co ~]# cat ~/.perfconfig
> [llvm]
>         dump-obj = true
>         clang-opt = -g
> #
>
> I end up with:
>
> [root@...co ~]# ls -la /home/acme/git/perf/tools/perf/examples/bpf/5sec.o
> -rw-r--r--. 1 root root 3696 Nov  3 12:47 /home/acme/git/perf/tools/perf/examples/bpf/5sec.o
> [root@...co ~]# file /home/acme/git/perf/tools/perf/examples/bpf/5sec.o
> /home/acme/git/perf/tools/perf/examples/bpf/5sec.o: ELF 64-bit LSB relocatable, eBPF, version 1 (SYSV), with debug_info, not stripped
> [root@...co ~]#
>
> and can test with the pre-built .o eBPF bytecode + capped backtrace:
>
> [root@...co ~]# perf trace -e /home/acme/git/perf/tools/perf/examples/bpf/5sec.o/max-stack=6/  sleep 5
>      0.000 sleep/161037 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1474734416, rqtp: 5000000000)
>                                        hrtimer_nanosleep ([kernel.kallsyms])
>                                        common_nsleep ([kernel.kallsyms])
>                                        __x64_sys_clock_nanosleep ([kernel.kallsyms])
>                                        do_syscall_64 ([kernel.kallsyms])
>                                        entry_SYSCALL_64_after_hwframe ([kernel.kallsyms])
>                                        __GI___clock_nanosleep (/usr/lib64/libc.so.6)
> [root@...co ~]#
>
> I'll test the other examples with these changes after I drive Pedro to
> school and get back to the office.

Thanks, I was somewhat coding in the dark with this as I was using a
bpf object file and clearly there's been some attention missed for a
while in these code paths. I couldn't get the hello variant to do
anything with the openat tracing it was set up for, for example, the
change puts the call on the sys_enter raw syscall. I suspect you
remember how these things should be and I'm happy to roll a v2, have
you fix it, etc.

Fwiw, two things I'd like to see further here is somehow the augmented
code to be the default (with a BPF skeleton possibly) and an ability
for perf_event_open tracing to dump the perf_event_attr. I'm unclear
on the advantages of having different augmenters. Perhaps we could
just drop support for different augmenters, use BPF skeletons for the
raw syscall version and keep the BPF logic in parse events for adding
filters - the process of doing augmentation isn't clear to me even
after having produced these changes. There is quite a bit of plumbing
necessary to remove the notion that the BPF object comes from event
parsing and I suspect to achieve this in the current code there is
some overhead that the skeleton could remove - like the empty
sys_enter functions I added.

Anyway, fixing libbpf 1.0+ and removing a header file called bpf.h
were what I was after achieving here :-)

Thanks,
Ian


> - Arnaldo
>
> diff --git a/tools/perf/Makefile.config b/tools/perf/Makefile.config
> index d3d3c13a9f25b55c..067a6e56eeacc9fc 100644
> --- a/tools/perf/Makefile.config
> +++ b/tools/perf/Makefile.config
> @@ -1239,7 +1239,7 @@ includedir = $(abspath $(prefix)/$(includedir_relative))
>  mandir = share/man
>  infodir = share/info
>  perfexecdir = libexec/perf-core
> -perf_include_dir = lib/perf/include
> +perf_include_dir = /usr/include
>  perf_examples_dir = lib/perf/examples
>  sharedir = $(prefix)/share
>  template_dir = share/perf-core/templates
> diff --git a/tools/perf/util/llvm-utils.c b/tools/perf/util/llvm-utils.c
> index 2dc7970074196ca8..a5cac85783d8711f 100644
> --- a/tools/perf/util/llvm-utils.c
> +++ b/tools/perf/util/llvm-utils.c
> @@ -495,7 +495,7 @@ int llvm__compile_bpf(const char *path, void **p_obj_buf,
>
>         snprintf(linux_version_code_str, sizeof(linux_version_code_str),
>                  "0x%x", kernel_version);
> -       if (asprintf(&perf_bpf_include_opts, "-I%s/bpf", perf_include_dir) < 0)
> +       if (asprintf(&perf_bpf_include_opts, "-I%s/", perf_include_dir) < 0)
>                 goto errout;
>         force_set_env("NR_CPUS", nr_cpus_avail_str);
>         force_set_env("LINUX_VERSION_CODE", linux_version_code_str);

Powered by blists - more mailing lists