[Intel-gfx] [PATCH 03/19] drm/i915: add forcewake functions that don't touch runtime PM

[Daniel Vetter]

On Thu, Feb 27, 2014 at 11:48:30AM -0300, Paulo Zanoni wrote:
> Hi
> 
> I'm sorry, I forgot to say. This series is quite old, and I changed it
> a little bit since then (since I found one or two problems), including
> this patch. I think that, to avoid wasting your time reviewing old
> patches, I should resend the new series.
> 
> The problem is that this series should be on top of the 11 patches I
> recently sent (with PC8/runtime PM fixes), so if we could review those
> first, it would be better. We also need to decide the relative order
> of merging your recent series and these patches, because they have
> some conflicts.

Hm, I've merged the first two patches of this series already, hope that
doesn't cause a fuzz ;-)

Now looking closer I'm puzzled by the pm_get/put calls in the forcewake
get/put functions. Imo any place which needs to have a power well up and
runtime (I include runtime pm as the overall power well here) should grab
a runtime pm reference for the entire access, not each register cycle.

The patch which added this was quite old, so probably before we've fixed
the bunch of runtime pm issues around gem batch buffers. And it was part
of a big patch which added get/puts mostly over debugfs files and similar
places. So I wonder whether we really still need this?

I'd much prefer if we could remove it, and if that's not possible fix up
the (hopefully few) places where we currently don't grab a runtime pm ref,
but should.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch