With tiger b28, VolanoMark and VolanoTest hang when running with -client -Xprof -XX:+UseSerialGC. The hang happened from just a few minutes to a couple of hours.

Test machine: j2se-bigapps.west ( please contact me if you need to access the machine )
# uname -a
SunOS j2se-bigapps 5.9 Generic_112233-09 sun4u sparc SUNW,Sun-Fire
The machine has 20 cpus * 1050MHZ, 98GB RAM.

The hang is reproducible on big5server.sfbay ( SunOS big5server 5.8 Generic_108528-13 sun4u sparc SUNW,Ultra-Enterprise ) the machine
has 8 cpu * 400MHZ.

With b26, the hang is not reproducible ( test ran for 20 hours ).

###@###.### 2003-11-14

Karen Kinnear has examined the pstack of the client process. Here is her
finding:

"Looks like we have a deadlock.

In the client, t@8 (periodic task thread - which does the profiling) is
waiting for
all other threads to complete suspending so it can get the profiling
snapshot.
It loops through all threads calling thread->java_suspend. At the moment
it is examining t@197 (JavaThread 0x0c0046e8). java_suspend acquires
the SR_lock for t@197, and then calls is_ext_suspend_completed to
wait for the suspension to be done.

While that was happening another thread requested a VM operation and the
VMThread (t@2) asked us to safepoint.

The periodic task thread in is_ext_suspend_completed, where
being a well-behaved thread it carefully blocks itself on the Threads_lock.
Unfortunately it still owns the SR_lock of the thread it is examining. "

current thread: t@8
  [1] ___lwp_mutex_lock(0x35ff8, 0x74, 0xff055e14, 0xff315884, 0xff148574, 0x35f
d0), at 0xff31e840
  [2] Mutex::lock(0x35fa0, 0xfec681c4, 0x6c00, 0x6e48, 0xff135e70, 0x227c98), at
 0xfec6da5c
  [3] JavaThread::is_ext_suspend_completed(0xc0046e8, 0x0, 0xf927fb3c, 0xdeac, 0
xff3, 0x55), at 0xff0511e4
  [4] JavaThread::java_suspend(0xc0050b0, 0x6cb0, 0xc0046e8, 0x35fa0, 0x6c00, 0x
ff118000), at 0xff052a3c
  [5] FlatProfiler::record_thread_ticks(0x195dd4d0, 0x0, 0x195dd81c, 0xff1324ec,
 0xff118000, 0xff12b79c), at 0xfee73aa8
  [6] PeriodicTask::real_time_tick(0xa, 0x6cd8, 0x0, 0xff135bf0, 0xff140cf0, 0xf
f118000), at 0xfed510bc
  [7] WatcherThread::run(0x227c98, 0xa, 0xff140cf8, 0xff135dc4, 0xff118000, 0x0)
, at 0xfed50a3c
  [8] _start(0x227c98, 0x4c48, 0x6400, 0x66f8, 0x4c00, 0xff118000), at 0xfecdb00
4

current thread: t@2
  [1] ___lwp_mutex_lock(0xc005108, 0x72, 0x0, 0xff3158ac, 0xff315884, 0xff148574
), at 0xff31e840
  [2] Mutex::lock_without_safepoint_check(0xc0050b0, 0xc0050e0, 0x4b03cc, 0x1160
b8, 0xff3, 0x7f9a81), at 0xfec67cd4
  [3] ThreadSafepointState::examine_state_of_thread(0xc1062c0, 0x0, 0x0, 0x80000
000, 0x80000000, 0xc0050b0), at 0xfed73ea0
  [4] SafepointSynchronize::begin(0x35f10, 0xc3, 0x0, 0xc0046e8, 0x1, 0xff134c98
), at 0xfed73c84
  [5] VMThread::run(0xff132e8c, 0xff13d73c, 0xff13d738, 0xff13d728, 0x0, 0x5c00)
, at 0xfecdb6e0
  [6] _start(0x1160b8, 0x4c48, 0x6400, 0x66f8, 0x4c00, 0xff118000), at 0xfecdb00
4

###@###.### 2003-11-14