Re: [Xenomai] Issue with cobalt_monitor_wait()

[Philippe Gerum <rpm@xenomai.org>]

On 07/08/2014 06:10 PM, Matthias Schneider wrote:
> ----- Original Message -----
>
>> From: Philippe Gerum <rpm@xenomai.org>
>> To: Matthias Schneider <ma30002000@yahoo.de>; "xenomai@xenomai.org" <xenomai@xenomai.org>
>> Cc:
>> Sent: Sunday, July 6, 2014 11:15 PM
>> Subject: Re: [Xenomai] Issue with cobalt_monitor_wait()
>>
>> On 07/06/2014 10:16 PM, Matthias Schneider wrote:
>>
>> [snip]
>>
>>>   On thing I do not understand is:
>>>
>>>   in kernel cobalt_monitor_wait(), the synch object is unlocked via
>>>   xnsynch_release(). What happens if this synchobj was locked via
>>>   mon->gate.fastlock ? Shouldnt that also be released?
>>>
>>
>> xnsynch_release() handles fastlocks as well.
>>
>>
>>>   What other reason could there be if the synch object was released
>>>   via xnsynch_release, xnsynch_acquire was interrupted for
>>>   xnsynch_release to block?
>>>
>>
>> Since the issue seems to be easily reproducible, could you send a
>> self-contained piece of code illustrating it?
>>
>> Also, please mention if you are seeing this issue only when running your
>> app over GDB, or if it currently happens without any debugger attached.
>>
>> TIA,
>
>
> It seems I have not described the problematic scenario completely -
>
> there were two other threads that call called syncobj_lock()
> / cobalt_monitor_enter() at about the same time. (Actually there
> are three concurrent on the queue that is being tested, two receive
> operation and one send operation). I am pretty sure that the issue is
> extremely timing dependent.
>
> Anyway, the testcase would be
>
> queue_test_receive_peek_multiple_tasks()
>

I could not reproduce the issue yet, but could you check if this patch 
has any influence on this bug? TIA,

diff --git a/kernel/cobalt/posix/syscall.c b/kernel/cobalt/posix/syscall.c
index d921d81..3856794 100644
--- a/kernel/cobalt/posix/syscall.c
+++ b/kernel/cobalt/posix/syscall.c
@@ -156,7 +156,7 @@ static struct xnsyscall cobalt_syscalls[] = {
  	SKINCALL_DEF(sc_cobalt_monitor_enter, cobalt_monitor_enter, primary),
  	SKINCALL_DEF(sc_cobalt_monitor_wait, cobalt_monitor_wait, 
nonrestartable),
  	SKINCALL_DEF(sc_cobalt_monitor_sync, cobalt_monitor_sync, 
nonrestartable),
-	SKINCALL_DEF(sc_cobalt_monitor_exit, cobalt_monitor_exit, primary),
+	SKINCALL_DEF(sc_cobalt_monitor_exit, cobalt_monitor_exit, nonrestartable),
  	SKINCALL_DEF(sc_cobalt_event_init, cobalt_event_init, current),
  	SKINCALL_DEF(sc_cobalt_event_destroy, cobalt_event_destroy, current),
  	SKINCALL_DEF(sc_cobalt_event_wait, cobalt_event_wait, primary),
diff --git a/lib/cobalt/internal.c b/lib/cobalt/internal.c
index e0d990d..6c1331d 100644
--- a/lib/cobalt/internal.c
+++ b/lib/cobalt/internal.c
@@ -230,6 +230,7 @@ int cobalt_monitor_exit(cobalt_monitor_t *mon)
  	struct cobalt_monitor_data *datp;
  	unsigned long status;
  	xnhandle_t cur;
+	int ret;

  	__sync_synchronize();

@@ -246,9 +247,13 @@ int cobalt_monitor_exit(cobalt_monitor_t *mon)
  	if (xnsynch_fast_release(&datp->owner, cur))
  		return 0;
  syscall:
-	return XENOMAI_SKINCALL1(__cobalt_muxid,
-				 sc_cobalt_monitor_exit,
-				 mon);
+	do
+		ret = XENOMAI_SKINCALL1(__cobalt_muxid,
+					sc_cobalt_monitor_exit,
+					mon);
+	while (ret == -EINTR);
+
+	return ret;
  }

  int cobalt_monitor_wait(cobalt_monitor_t *mon, int event,

-- 
Philippe.