Re: [PATCH] perf tools: session: avoid infinite loop

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Fri, Sep 18, 2015 at 02:37:09PM +0100, Mark Rutland escreveu:
> > > So it looks like I shouldn't have any synthesized events. Have I missed
> > > anything?
> > 
> > Yes, you are right.  But you are not getting the COMM and MMAP events from
> > the exec which means you are killing perf before it execs the workload.
> 
> Oh, I see.
> 
> > Perf writes through a pipe to its forked child to do the exec, so
> > to reproduce it you just need to put everything on the same cpu and play around
> > with the sleep number
> > 
> > 	taskset -c 0 tools/perf/perf record -- sleep 1 & sleep 0.005 ; kill -2 $!
> > 
> > reproduces the problem for me.  Of course, since the workload doesn't get exec'ed
> > it doesn't matter if it is bogus i.e.
> > 
> > 	taskset -c 0 tools/perf/perf record -- sdfgsdgdg & sleep 0.005 ; kill -2 $!
> > 
> > also reproduces the problem.
> 
> Thanks for confirming!

I tried what you suggested, but couldn't reproduce it, I only managed
when I applied the patch below _and_ executed;

 # echo 1 > /proc/sys/kernel/kptr_restrict

To be on the same page as Mark. Then, yes, I get stuck in:

#0  0x00007f746a1d8aba in mmap64 () at ../sysdeps/unix/syscall-template.S:81
#1  0x00000000004be66e in __perf_session__process_events (file_size=232, data_size=<optimized out>, data_offset=<optimized out>, session=0x1ba8590)
    at util/session.c:1604
#2  perf_session__process_events (session=0x1ba8590) at util/session.c:1685
#3  0x000000000042d97b in process_buildids (rec=0x869ac0 <record>) at builtin-record.c:364
#4  __cmd_record (rec=0x869ac0 <record>, argv=<optimized out>, argc=<optimized out>) at builtin-record.c:734
#5  cmd_record (argc=<optimized out>, argv=<optimized out>, prefix=<optimized out>) at builtin-record.c:1199
#6  0x0000000000479123 in run_builtin (p=p@entry=0x874a90 <commands+144>, argc=argc@entry=3, argv=argv@entry=0x7ffd92eb5ee0) at perf.c:370
#7  0x0000000000420aba in handle_internal_command (argv=0x7ffd92eb5ee0, argc=3) at perf.c:429
#8  run_argv (argv=0x7ffd92eb5c70, argcp=0x7ffd92eb5c7c) at perf.c:473
#9  main (argc=3, argv=0x7ffd92eb5ee0) at perf.c:588
(gdb) 

After applying your patch it works, had to tweak the commit log to avoid
starting lines with ---, breaks git scripts, also added a commiter log,
check it, patch is below, after the one I used to not process any
samples.

- Arnaldo

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 142eeb341b29..b32bb9814b35 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -645,7 +645,7 @@ static int __cmd_record(struct record *rec, int argc, const char **argv)
 	}
 
 	auxtrace_snapshot_enabled = 1;
-	for (;;) {
+	for (;child_finished=1,0;) {
 		int hits = rec->samples;
 
 		if (record__mmap_read_all(rec) < 0) {

----

commit dd486ec4aa33cfca2fd912ef501d49909005de79
Author: Mark Rutland <mark.rutland@arm.com>
Date:   Wed Sep 16 18:18:49 2015 +0100

    perf record: Avoid infinite loop at buildid processing with no samples
    
    If a session contains no events, we can get stuck in an infinite loop in
    __perf_session__process_events, with a non-zero file_size and data_offset, but
    a zero data_size.
    
    In this case, we can mmap the entirety of the file (consisting of the file and
    attribute headers), and fetch_mmaped_event will correctly refuse to read any
    (unmapped and non-existent) event headers. This causes
    __perf_session__process_events to unmap the file and retry with the exact same
    parameters, getting stuck in an infinite loop.
    
    This has been observed to result in an exit-time hang when counting
    rare/unschedulable events with perf record, and can be triggered artificially
    with the script below:
    
      ----
      #!/bin/sh
      printf "REPRO: launching perf\n";
      ./perf record -e software/config=9/ sleep 1 &
      PERF_PID=$!;
      sleep 0.002;
      kill -2 $PERF_PID;
      printf "REPRO: waiting for perf (%d) to exit...\n" "$PERF_PID";
      wait $PERF_PID;
      printf "REPRO: perf exited\n";
      ----
    
    To avoid this, have __perf_session__process_events bail out early when
    the file has no data (i.e. it has no events).
    
    Commiter note:
    
    I only managed to reproduce this when setting
    /proc/sys/kernel/kptr_restrict to '1' and changing the code to
    purposefully not process any samples and no synthesized samples, i.e.
    kptr_restrict prevents 'record' from synthesizing the kernel mmaps for
    vmlinux + modules and since it is a workload started from perf, we don't
    synthesize mmap/comm records for existing threads.
    
    Adrian Hunter managed to reproduce it in his environment tho.
    
    Reported-by: Arnaldo Carvalho de Melo <acme@kernel.org>
    Signed-off-by: Mark Rutland <mark.rutland@arm.com>
    Cc: Adrian Hunter <adrian.hunter@intel.com>
    Cc: Jiri Olsa <jolsa@redhat.com>
    Cc: Peter Zijlstra <peterz@infradead.org>
    Link: http://lkml.kernel.org/r/1442423929-12253-1-git-send-email-mark.rutland@arm.com
    Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
index 8a4537ee9bc3..fc3f7c922f99 100644
--- a/tools/perf/util/session.c
+++ b/tools/perf/util/session.c
@@ -1580,7 +1580,10 @@ static int __perf_session__process_events(struct perf_session *session,
 	file_offset = page_offset;
 	head = data_offset - page_offset;
 
-	if (data_size && (data_offset + data_size < file_size))
+	if (data_size == 0)
+		goto out;
+
+	if (data_offset + data_size < file_size)
 		file_size = data_offset + data_size;
 
 	ui_progress__init(&prog, file_size, "Processing events...");