[Bug dyninst/14637] New: using dyninst, probing pthread mutex markers can hang the process

[Bug dyninst/14637] New: using dyninst, probing pthread mutex markers can hang the process

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14637

             Bug #: 14637
           Summary: using dyninst, probing pthread mutex markers can hang
                    the process
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: dyninst
        AssignedTo: systemtap@sourceware.org
        ReportedBy: dsmith@redhat.com
    Classification: Unclassified


When using --runtime=dyninst, probing the pthread mutex markers can hang the
process.

The test script I'm using is a modified version of the test from
systemtap.exelib/pthreadprobes.exp.  I've modified that testcase to use
statistics as counters.

What I believe is going on is that the systemtap stat code uses pthread mutexes
internally, Since that test puts probes on the mutex markers, systemtap is
getting "internal" probe hits (possibly causing the program to infinitely
loop).

We may have to somehow ignore those hits or blacklist those markers and pthread
library routines.

(I've got a patch I'm testing patch that changes the mutexes to rwlocks, but
that just lessens the problem a bit.)

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug dyninst/14637] using dyninst, probing pthread mutex markers can hang the process

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14637

--- Comment #1 from David Smith <dsmith at redhat dot com> 2012-09-27 20:22:59 UTC ---
Created attachment 6658
  --> http://sourceware.org/bugzilla/attachment.cgi?id=6658
test script

To duplicate this problem, run the following command from the testsuite
directory (using the test program from
testsuite/systemtap.exelib/pthreadprobes.c):

# stap -v --runtime=dyninst ./pthreadprobes.stp -c ./pthreadprobes
/lib/libpthread-2.16.20

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug dyninst/14637] using dyninst, probing pthread mutex markers can hang the process

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14637

--- Comment #2 from David Smith <dsmith at redhat dot com> 2012-09-27 20:58:33 UTC ---
One thing I forgot, is that I also had the dyninst 'big_dumb_lock' turned off
in runtime/dyninst/runtime.h

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug dyninst/14637] using dyninst, probing pthread mutex markers can hang the process

[dsmith at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=14637

David Smith <dsmith at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |WORKSFORME

--- Comment #3 from David Smith <dsmith at redhat dot com> 2012-11-27 22:29:59 UTC ---
With the latest stapdyn multithreading work (commit 1d0e697, see bug #14571),
this problem no longer occurs. (This might be the result of the stat code using
less pthread mutexes internally.)

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.