[Intel-gfx] [PATCH 00/16] drm/i915: 830M/ns201 fixes again

[Daniel Vetter]

On Fri, Aug 15, 2014 at 10:57:21AM +0300, Ville Syrjälä wrote:
> On Fri, Aug 15, 2014 at 01:21:52AM +0300, ville.syrjala at linux.intel.com wrote:
> > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > 
> > Thomas asked me to repost my 830/ns2501 patches. So here they are. I added
> > a few more patches (trickle feed and unused ring init) to fix some post-resume
> > issues. The primary plane rmw elimination patches and some locking/load detect
> > fixes already got merged.
> > 
> > Apart from these we still lack the minimum watermark check. I guess
> > we could just take the patch Thomas posted [1]. Doesn't look like a
> > more advanced solution is coming any time soon. Though the commit
> > message of that patch needs work and it lacks a s-o-b.
> > 
> > The VGACNTR patch might not be necessary any longer since Daniel's
> > vga/dummycon stuff. I don't recall if I tested without it, but my
> > gut feeling is that it's no longer needed. But I included the patch
> > here anyway.
> > 
> > [1] http://patchwork.freedesktop.org/patch/27318/
> 
> Here's the branch (includes my earlier watermark hack):
> 
> git://gitorious.org/vsyrjala/linux.git alm_fixes11

Ok, I've merged all the patches here except the dvo 2x mode thing. And imo
we can also pull in the burst size hack since imo that's what we want -
afaik the hw has some bits to select the burst size between 4, 8, ...
entries, and atm we always set it to 8. And Bspec explicitly says that
setting the fifo to something smaller than burst size will lead to tears.

So the patch with a pimped commit message + bit a better comment is good
to go. Originally I wanted to implement "proper" burst size selection, but
that ran afoul of the general gen2/3 wm mess.

Aside: We might want to give intel_calculate_wm some i9xx prefix, since
it's very much not generic.

Thanks, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch