Re: 2.6.16-rc6-rt1

[Thomas Gleixner <tglx@linutronix.de>]

On Tue, 2006-03-14 at 21:40 +0100, Esben Nielsen wrote:
> The trick is: Maintain the unittest along with the code you are testing.
> Unfortunately a lot of people haven't discovered that yet. As I said
> before, the Linux kernel sould should have a tests/ directory in the main
> directory and a "make tests". For any patch to be accepted, should "make
> tests" should "build". Patches ofcouse include changes to the kernel code
> and the tests/ directory as it is one distribution.  Notice the tests are
> run _without_ running the kernel!
> That is how I do it at work: I have it all in one source repository and
> the "tests" target is the first dependency of "all:" in the makefile.

I did not say that a unittester is bad. It just does not help much when
it only works on your workstation.

> > The deadlock detection is done, when requested. So you _have_ to do it
> > by following the lock chain. When the task goes to sleep, then there is
> > no postmortem. When a futex requests deadlock detection you have to do
>                               --------
> > it in the locking path, as you have to return that information to
> > userspace.
> >
> > http://www.opengroup.org/onlinepubs/009695399/functions/pthread_mutex_lock.html
> >
> The point is that when deadlock detection isn't requested it ought not to
> be forced on the application.

It is not forced. We break out of the loop, when it is not requested.
This is just combined PI / deadlock detection code. And we do the check
in the boosting path anyway to avoid deadlocking there. Given it
works :)

> It happens before getting into that loop

Grmbl, you are right. Fix follows.

	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/