[Intel-gfx] [RFC xf86-video-intel] sna: Call dirtyfb for all non-tear-free cases

[Daniel Vetter]

On Tue, Mar 22, 2016 at 09:35:36PM +0000, Zanoni, Paulo R wrote:
> Em Ter, 2016-03-22 às 12:31 +0100, Daniel Vetter escreveu:
> > On Mon, Mar 21, 2016 at 04:26:55PM -0300, Paulo Zanoni wrote:
> > > 
> > > The sna_mode_wants_tear_free() function tries to detect FBC and PSR
> > > based on Kernel module parameters. Currently it fails to detect FBC
> > > due to the default enable_fbc value being -1. While this can easily
> > > be
> > > fixed in the Kernel, I had a conversation with Daniel and he
> > > expressed
> > > unhappiness with that solution, claiming that yet another different
> > > code path just for a feature that should be transparent is not a
> > > good
> > > way to go, and that we should do proper frontbuffer rendering.
> > > 
> > > So with this patch, we'll have the DDX issuing dirtyfb calls even
> > > if
> > > TearFree is not enabled, fixing FBC when i915.enable_fbc=-1.
> > > 
> > > This fixes a bug that happens on SKL with FBC enabled: if you run
> > > lightdm, your login/password won't appear as you type on your
> > > keyboard. You have to move the mouse over the input box for them to
> > > be
> > > displayed.
> > > 
> > > Cc: Chris Wilson <chris at chris-wilson.co.uk>
> > > Signed-off-by: Paulo Zanoni <paulo.r.zanoni at intel.com>
> > I thought we need this anyway to get the kernel to allow fbc, since
> > SNA
> > ends up mmap some of the drm_framebuffer. Even when they're not
> > frontbuffers.
> 
> If we merge patch 2/4, we won't need this one since TearFree will be in
> use, and it seems TearFree doesn't touch frontbuffers, so we'll always
> get the flush calls during page flips.

I'd thought that TearFree would still do rendering with the cpu sometimes,
but only always to the back buffer. So we'd still have cpu mmaps, and
hence the kernel w/a would still potentially kick in and prevent fbc. So
that's not the case then, and we get fbc reliably with the kernel patch,
even without this patch?
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch