[Intel-gfx] BUG: sleeping function called from invalid context on 3.10.10-rt7

Steven Rostedt rostedt at goodmis.org
Wed Sep 25 07:18:03 PDT 2013

On Wed, 25 Sep 2013 10:49:36 +0300
Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:

> The preempt_disable/enable is not needed. The spinlock serves the same
> purpose already.

As stated, that was only for the -rt patch, as spin_lock_irqsave does
not disable preemption nor does it even disable interrupts.

But I agree, the added preempt_disable() should be sent to us to keep
in the -rt patch itself. We really appreciate that you are thinking
about us :-)  But something like this will just confuse the mainline
folks. Having a "preempt_disable_rt()" would make a lot more sense
(which exists in the -rt patch).

> As far as ktime_get(), I've used it from spinlocked/irq disabled sections
> and so far haven't seen it do bad things. But would be nice to get some
> official statement to that effect.

It's just a read seqlock. It may do a few loops to get the correct
time, but it's fine to have in a preempt/irq disabled section.

-- Steve

More information about the dri-devel mailing list