Re: [PATCH] perf record: Skip side-band event setup if HAVE_LIBBPF_SUPPORT is not set

[Arnaldo Carvalho de Melo]

Em Fri, Aug 07, 2020 at 09:16:29AM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Fri, Aug 07, 2020 at 09:09:56AM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Fri, Aug 07, 2020 at 09:08:24AM -0300, Arnaldo Carvalho de Melo escreveu:
> > > Em Thu, Aug 06, 2020 at 09:43:57PM +0200, Jiri Olsa escreveu:
> > > > On Wed, Aug 05, 2020 at 10:29:37AM +0800, Jin Yao wrote:
> > > > > We received an error report that perf-record caused 'Segmentation fault'
> > > > > on a newly system (e.g. on the new installed ubuntu).
> > > > > 
> > > > >  (gdb) backtrace
> > > > >  #0  __read_once_size (size=4, res=<synthetic pointer>, p=0x14) at /root/0-jinyao/acme/tools/include/linux/compiler.h:139
> > > > >  #1  atomic_read (v=0x14) at /root/0-jinyao/acme/tools/include/asm/../../arch/x86/include/asm/atomic.h:28
> > > > >  #2  refcount_read (r=0x14) at /root/0-jinyao/acme/tools/include/linux/refcount.h:65
> > > > >  #3  perf_mmap__read_init (map=map@entry=0x0) at mmap.c:177
> > > > >  #4  0x0000561ce5c0de39 in perf_evlist__poll_thread (arg=0x561ce68584d0) at util/sideband_evlist.c:62
> > > > >  #5  0x00007fad78491609 in start_thread (arg=<optimized out>) at pthread_create.c:477
> > > > >  #6  0x00007fad7823c103 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
> > > > 
> > > > hum, I recall discussing the same issue,
> > > > I thought it was already fixed :-\ in any case:
> > > > 
> > > > Acked-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> > > 
> > > I added this:
> > > 
> > > Fixes: 23cbb41c939a ("perf record: Move side band evlist setup to separate routine")
> > > 
> > > To help this fixe go back to a few stable kernels. Technically the
> > > problem was introduced much earlier, when the side band thread was
> > > added, but then this would require more cherry-picking of patches and
> > > since we have a workaround, i.e. build with libbpf, I think this is
> > > enough.
> > 
> > I backtrack on that, as the sideband event is not just for BPF... The
> > switch-output-event code uses it as well, so it can't be dependent on
> > LIBBPF being built...
> > 
> > I'll see what I can do
> 
> So this is the change I made, the side band thread is may have been
> already created by 
> 
> [acme@quaco ~]$ perf record -h switch-output-event
> 
>  Usage: perf record [<options>] [<command>]
>     or: perf record [<options>] -- <command> [<options>]
> 
>         --switch-output-event <switch output event>
>                           switch output event selector. use 'perf list' to list available events
> 
> [acme@quaco ~]$
> 
> I'm doing some extra checking now on your report, and the patch below
> has skews because it clashed with the clockid patches by Jiri so I had
> to resolve its merge.
 
Sorry, I resent your patch, doh, here is the end result, with my change,
its ok to call perf_evlist__start_sb_thread() with a NULL evlist, it'll
just return 0.

- Arnald


commit b13536a7e93680625094beb18cdce4ae47a3dbfb
Author: Jin Yao <yao.jin@xxxxxxxxxxxxxxx>
Date:   Wed Aug 5 10:29:37 2020 +0800

    perf record: Skip side-band event setup if HAVE_LIBBPF_SUPPORT is not set
    
    We received an error report that perf-record caused 'Segmentation fault'
    on a newly system (e.g. on the new installed ubuntu).
    
      (gdb) backtrace
      #0  __read_once_size (size=4, res=<synthetic pointer>, p=0x14) at /root/0-jinyao/acme/tools/include/linux/compiler.h:139
      #1  atomic_read (v=0x14) at /root/0-jinyao/acme/tools/include/asm/../../arch/x86/include/asm/atomic.h:28
      #2  refcount_read (r=0x14) at /root/0-jinyao/acme/tools/include/linux/refcount.h:65
      #3  perf_mmap__read_init (map=map@entry=0x0) at mmap.c:177
      #4  0x0000561ce5c0de39 in perf_evlist__poll_thread (arg=0x561ce68584d0) at util/sideband_evlist.c:62
      #5  0x00007fad78491609 in start_thread (arg=<optimized out>) at pthread_create.c:477
      #6  0x00007fad7823c103 in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
    
    The root cause is, evlist__add_bpf_sb_event() just returns 0 if
    HAVE_LIBBPF_SUPPORT is not defined (inline function path). So it will
    not create a valid evsel for side-band event.
    
    But perf-record still creates BPF side band thread to process the
    side-band event, then the error happpens.
    
    We can reproduce this issue by removing the libelf-dev. e.g.
    1. apt-get remove libelf-dev
    2. perf record -a -- sleep 1
    
      root@test:~# ./perf record -a -- sleep 1
      perf: Segmentation fault
      Obtained 6 stack frames.
      ./perf(+0x28eee8) [0x5562d6ef6ee8]
      /lib/x86_64-linux-gnu/libc.so.6(+0x46210) [0x7fbfdc65f210]
      ./perf(+0x342e74) [0x5562d6faae74]
      ./perf(+0x257e39) [0x5562d6ebfe39]
      /lib/x86_64-linux-gnu/libpthread.so.0(+0x9609) [0x7fbfdc990609]
      /lib/x86_64-linux-gnu/libc.so.6(clone+0x43) [0x7fbfdc73b103]
      Segmentation fault (core dumped)
    
    To fix this issue,
    
    1. We either install the missing libraries to let HAVE_LIBBPF_SUPPORT
       be defined.
       e.g. apt-get install libelf-dev and install other related libraries.
    
    2. Use this patch to skip the side-band event setup if HAVE_LIBBPF_SUPPORT
       is not set.
    
    Committer notes:
    
    The side band thread is not used just with BPF, it is also used with
    --switch-output-event, so narrow the ifdef to the BPF specific part.
    
    Fixes: 23cbb41c939a ("perf record: Move side band evlist setup to separate routine")
    Signed-off-by: Jin Yao <yao.jin@xxxxxxxxxxxxxxx>
    Acked-by: Jiri Olsa <jolsa@xxxxxxxxxx>
    Cc: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
    Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx>
    Cc: Jin Yao <yao.jin@xxxxxxxxx>
    Cc: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
    Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
    Link: http://lore.kernel.org/lkml/20200805022937.29184-1-yao.jin@xxxxxxxxxxxxxxx
    Signed-off-by: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 7d97c8e9f7f9..f91352f847c0 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1531,7 +1531,7 @@ static int record__setup_sb_evlist(struct record *rec)
 		evlist__set_cb(rec->sb_evlist, record__process_signal_event, rec);
 		rec->thread_id = pthread_self();
 	}
-
+#ifdef HAVE_LIBBPF_SUPPORT
 	if (!opts->no_bpf_event) {
 		if (rec->sb_evlist == NULL) {
 			rec->sb_evlist = evlist__new();
@@ -1547,7 +1547,7 @@ static int record__setup_sb_evlist(struct record *rec)
 			return -1;
 		}
 	}
-
+#endif
 	if (perf_evlist__start_sb_thread(rec->sb_evlist, &rec->opts.target)) {
 		pr_debug("Couldn't start the BPF side band thread:\nBPF programs starting from now on won't be annotatable\n");
 		opts->no_bpf_event = true;