Re: 2.6.16-rt10

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 4 Apr 2006, Simon Derr wrote:

>
> But I must be severely misunderstanding something.
>
> What I understood is that in preemptible sections preempt_count() is
> zero, and in non preemptible sections it is >0.
>
> If preempt_count() is 1, then preempt_enable_no_resched() will decrement
> it and issue a warning. This is what happens in disabled_fph_fault().
>
> Where am I wrong ?

You're not.

There's nothing unstable about it.  The problem is that you didn't
schedule when you could have.  With Linux, this really isn't an issue,
but with the -rt kernel we concentrate on low latency, and by not
calling schedule when preempt count goes to zero and the need_resched
flag may be set, you may be delaying a high priority process
unnecessarily, for longer than needed.  This in the -rt kernel _is_ a
bug.

So for a stability point of view, that missed schedule wont crash the
kernel, but it might cause an xrun in JACK.

Now, sometimes a call to preempt_enable_no_resched is called just before
schedule is called, this is done so you don't have a double schedule.
For this, it is ok to call *_no_resched, but you need to flag that this
is ok by calling __preempt_enable_no_resched directly.

-- Steve
-
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/