Hi Enno,
On Jan 10, 2011, at 9:39 AM, Luebbers, Enno wrote:
> Hi,
>
> first of all, happy new year and all the best wishes for 2011 to all of you.
>
> I was also able to reproduce the sporadic failures of mutex_test_posix. I
> suspect the root of it in the scheduling interactions between the delegate
> threads and the main() thread -- it appears that occasionally the delegate
> thread is not activated in time and thus fails to lock the mutex. This, in
> turn, causes the main() thread to lock the mutex early.
>
> I've played a bit with the thread priorities of main() and the delegate.
> However, even with the delegates having a higher priority than main(), the
> failure still showed up.
I've tried the same, but with changed priorities I could *not* reproduce the
error. (I did 20 tries).
I've added the following lines:
mutex_test_posix.c
37: cyg_handle_t my = cyg_thread_self();
38: cyg_thread_set_priority(my, 10);
ecos-patched/ecos/packages/reconos/current/src/delegate.tmpl
334: cyg_handle_t my = cyg_thread_self();
335: cyg_thread_set_priority(my, 1);
Does this also work for you?
Cheers,
Ariane
> The only way I could prevent the "too early" error from happening was to
> change the test's source could as in the following diff:
>
> diff --git a/tests/automated/mutex/sw/mutex_test_posix.c
> b/tests/automated/mutex
> index 8385469..e066872 100644
> --- a/tests/automated/mutex/sw/mutex_test_posix.c
> +++ b/tests/automated/mutex/sw/mutex_test_posix.c
> @@ -50,7 +50,7 @@ int main( int argc, char *argv[] )
> unsigned long ticks = cyg_current_time();
> //printf("current time = %ld ticks\n",ticks); // XXX remove
> pthread_mutex_unlock(&mutex);
> - while(cyg_current_time() - ticks < 10); // wait for 0.1
> seconds
> + cyg_thread_delay(10);
> pthread_mutex_lock(&mutex);
> ticks = cyg_current_time() - ticks;
> //printf("delta t = %ld ticks\n", ticks); // XXX remove
>
> After this change I wasn't able to reproduce the error, but that doesn't mean
> it's gone. Also, even without the delay(), the first mutex_unlock() should
> cause the delegate to be scheduled if it has a higher priority.
>
> Any ideas what could be causing this?
>
> Best regards
> - Enno
>
>> -----Ursprüngliche Nachricht-----
>> Von: reconos-devel-boun...@lists.reconos.de [mailto:reconos-devel-
>> boun...@lists.reconos.de] Im Auftrag von Ariane Keller
>> Gesendet: Donnerstag, 6. Januar 2011 13:12
>> An: Discussions for ReconOS development
>> Betreff: Re: [reconos-devel] ML605 automated tests
>>
>> Hi all
>>
>> On Dec 14, 2010, at 4:17 PM, Markus wrote:
>>
>>> Hi *,
>>>
>>> I performed the automated tests on the ML605. 64 of 65 tests succeeded,
>> but one test failed (mutex_test_posix).
> [...]
>>> It would be interesting to know, if anyone can reproduce this failure.
>>
>> I finally managed to do the tests. The exec_mutex_posix test also failed in
>> my setup.
>> My statistics are:
>> Success: 6 runs
>> 1 failure: 4 runs
>>
>> failure:
>> mutex lock and release by hwthread: too early
>> (occurrence: once but at different stages)
>>
>> All the other tests were ok.
>>
>> Cheers,
>> Ariane
>>
> _______________________________________________
> Reconos-devel_lists.reconos.de mailing list
> reconos-devel@lists.reconos.de
> https://ml01.ispgateway.de/mailman/listinfo/reconos-devel_lists.reconos.de
_______________________________________________
Reconos-devel_lists.reconos.de mailing list
reconos-devel@lists.reconos.de
https://ml01.ispgateway.de/mailman/listinfo/reconos-devel_lists.reconos.de
