multiple tasks and timer setting, starting

[Paolo Mantegazza]

Herman Bruyninckx wrote:
> On Fri, 16 Jan 2004, Matteo Martegani wrote:
> 
> 
>>    I'm coding a simple test case where different tasks are running on 
>>the same machine. Any task can start the timer and set the mode, if the 
>>timer isn't running yet (i.e. first task running starts the timer). The 
>>obvious way to proceed is something in the lines of:
>>
>>        if (!rt_is_hard_timer_running()){
>>                rt_set_oneshot_mode();
>>                start_rt_timer(0);
>>        }
>>
> 
> [...]
> The code above is prone to race conditions: what if it is called from
> two separate tasks, and one task is preempted at the check of the
> conditional?
> 
> The generic design for this sort of "unique but shared" resources is
> that tasks do not do the actions themselves, but ask the "resource
> manager" to do it for them. The resource manager does the locking and
> synchronization of the access to the resource internally, guaranteeing
> race free execution.
> I think there are still many places within RTAI (and Linux!) where
> this "centralized resource management" design pattern should be
> introduced...  (At least more explicitly than is done now.)
> Note that the "resource manager" need not be complex!
> 

Well I think he had to do that as a way to avoid crashing an already 
running hard timer. The race in this case is theoritically possible 
UP/SMP but is difficult to have preemption in UP without the hard timer 
running. In SMP it could but it is very very unlikely, so it will happen 
:-). In any case it must be taken into account in a rewrite.

The problem I see is the one of braking any existing application 
purposely issuing start_rt_timer, in a safe way and place, to reset it 
and start anything anew.

If nobody says anything more I'll have a look, taking both suggestions 
into account.

Paolo.