[interfacekit] Re: BScreen (again)

On Mon, 28 Jul 2003 18:10:34 +0200 CEST "Axel Dörfler" <axeld@pinc-
software.de> wrote:
> "Ingo Weinhold" <bonefish@xxxxxxxxxxxxxxx> wrote:
> > BPrivateScreen seems to have a sem_id member, which is acquired in 
> > WaitForRetrace() -- if not yet initialized, the method gets the 
> > sem_id 
> > from the app server. I suspect, all threads of all applications 
> > wait 
> > on 
> > the same semaphore for the vertical blank. I wonder a bit, how this 
> > works, but probably there is something like an undocumented 
> > release_sem_etc() flag that releases the semaphore exactly as often 
> > as 
> > necessary to unblock all waiting threads.
> 
> What about this:
> 
> int32 *sWaitForRetraceWaiters = locationInSomeSharedMemory;
> sem_id sRetraceSemaphore;
> 
> status_t
> WaitForRetrace()
> {
>       atomic_add(sWaitForRetraceWaiters, 1);
>       return acquire_sem(sRetraceSemaphore);
> }
> 
> 
> and in the driver:
> 
> int32 waiters = atomic_and(sWaitForRetraceWaiters, 0);
> release_sem_etc(sRetraceSemaphore, waiters);
> 
> ?
> Not tested, but it should work safely AFAICT.

Yep, looks good in principle, but it may have problems, since 
acquire_sem() is actually an acquire_sem_etc() with timeout. Which will 
cause the counter and the semaphore counter to get out of sync. As in 
BLocker I don't think it is possible to fix this (not without a further 
semaphore at least).

CU, Ingo

Other related posts: