Re: PI patch against 2.6.16-rt9

[Thomas Gleixner <tglx@linutronix.de>]

On Wed, 2006-03-29 at 00:34 +0100, Esben Nielsen wrote:
> > Your method is tempting, but I do not see how it works out right now
> 
> It works for PI.

Well, works and effective are two things. In the worst case it
introduces scheduler floods.

> It might give false positives for deadlock detection even
> without signals involved. But that might be solved by simply checking
> again.

Which is even more broken. Rechecking is less deterministic as the
global lock fall back solution.

> If it is stored on a task when they blocked on a lock it
> could be seen if they had released and reobtained the task since the last
> traversal.

-ENOPARSE

> If I should choose between a 100% certain deadlock detection and
> rescheduling while doing PI I would choose that latter as that gives a
> deterministic RT system. Are there at all applications depending on
> deadlock detection or is it only for debug perposes anyway?

No, userspace can request deadlock checking and we have to return
-EDEADLK in that case.

[EDEADLK]
        A deadlock condition was detected or the current thread already
        owns the mutex.
        
Returning false positives might break well designed applications and
prevent real deadlock detection.

Btw, your get/put_task proposal adds two atomic ops. Atomic ops are
implicit memory barriers and therefor you add two extra slow downs into
the non conflict case.

	tglx


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/