Subrata Modak wrote:
>
>
> Hi,
>> We have seen that prio_preempt testcase that is part of realtime tests
>> occasionally hangs. This can be easily recreated on a 8-cpu system,
>> but can be recreated on a 4-cpu system as well when run for a number
>> of iterations.
>>
>> The problem occurs because the first worker thread sometimes fails to
>> call cond_wait on the condvar through which the master thread signals
>> it to start. Since the first thread needs to start the chain of
>> signalling from then on, all other threads just sit in a cond_wait
>> without ever being woken up.
>>
>> The fix is to have a barrier to ensure that the first worker thread
>> doesn't miss the cond_signal.
>>
>> I have tested this patch by running 10s of thousands of iterations of
>> the testcase. Without the patch I can recreate the problem fairly
>> easily.
>>
>> Thanks,
>> Sripathi.
>>
>> Signed-off-by: Sripathi Kodi <[EMAIL PROTECTED]>
Acked-by: Darren Hart <[EMAIL PROTECTED]>
>>
>
> Over to the Real Time Linux Community for comments :-)
>
> Regards--
> Subrata
>> ---
>> ltp-full-20080229/testcases/realtime/func/prio-preempt/prio-preempt.c.org
>>
>> 2008-03-12 22:10:05.000000000 +0530
>> +++
>> ltp-full-20080229/testcases/realtime/func/prio-preempt/prio-preempt.c
>> 2008-03-12 22:14:02.000000000 +0530
>> @@ -90,6 +90,8 @@ static int t_after_wait[NUM_WORKERS];
>> static int ret = 0;
>> static int run_jvmsim=0;
>>
>> +pthread_barrier_t barrier;
>> +
>> void usage(void)
>> {
>> rt_help();
>> @@ -184,6 +186,8 @@ void *worker_thread(void* arg)
>>
>> /* block */
>> rc = pthread_mutex_lock(&mutex[tid]);
>> + if (tid == 0)
>> + pthread_barrier_wait(&barrier);
>> rc = pthread_cond_wait(&cond[tid], &mutex[tid]);
>> rc = pthread_mutex_unlock(&mutex[tid]);
>>
>> @@ -234,6 +238,8 @@ void *master_thread(void* arg)
>> {
>> int i, pri_boost;
>>
>> + pthread_barrier_init(&barrier, NULL, 2);
>> +
>> /* start interrupter thread */
>> if (int_threads) {
>> pri_boost = 90;
>> @@ -270,6 +276,9 @@ void *master_thread(void* arg)
>> while (threads_running < NUM_WORKERS)
>> usleep(100);
>>
>> + /* Ensure the first worker has called cond_wait */
>> + pthread_barrier_wait(&barrier);
>> +
>> printf("Signaling first thread\n");
>> pthread_mutex_lock(&mutex[0]);
>> pthread_cond_signal(&cond[0]);
>>
>>
>
>
--
Darren Hart
IBM Linux Technology Center
Real-Time Linux Team
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Ltp-list mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list
