[Intel-gfx] [PATCH 5/7] drm/i915: Make sure we invalidate frontbuffer on fbcon.

[Daniel Vetter]

On Wed, Mar 04, 2015 at 03:05:11PM -0800, Rodrigo Vivi wrote:
> On Wed, Mar 4, 2015 at 6:30 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
> > On Tue, Mar 03, 2015 at 08:03:13PM +0000, Vivi, Rodrigo wrote:
> >> On Tue, 2015-03-03 at 09:28 +0100, Daniel Vetter wrote:
> >> > On Mon, Mar 02, 2015 at 06:35:26PM +0000, Vivi, Rodrigo wrote:
> >> > > On Mon, 2015-03-02 at 18:59 +0100, Daniel Vetter wrote:
> >> > > > On Fri, Feb 27, 2015 at 08:26:05PM -0500, Rodrigo Vivi wrote:
> >> > > > > There are some cases like suspend/resume or dpms off/on sequences
> >> > > > > that can flush frontbuffer bits. In these cases features that relies
> >> > > > > on frontbuffer tracking can start working and user can stop getting
> >> > > > > screen updates on fbcon having impression the system is frozen.
> >> > > > >
> >> > > > > So, let's make sure on fbcon write operation we also invalidate
> >> > > > > frontbuffer bits so we will be on the safest side with fbcon.
> >> > > >
> >> > > > This is just a bandaid since you can always just directly access the
> >> > > > fbdev framebuffer. We really need to figure out why we have frontbuffer
> >> > > > bit flushes after we've invalidated them for fbcon and catch them all.
> >> > >
> >> > > yeah, an ugly bandaid... Just to make PSR a bit more reliable without
> >> > > breaking fbcon environment when it gets enabled by default.
> >> > >
> >> > > The issue is that on the logs I see:
> >> > >
> >> > > 1.fbdev_blank dpms off
> >> > > 2. disable planes
> >> > > 3. flush frontbuffer bits
> >> > > --- blank stage ---
> >> > > 4. fbdev_blank dpms on
> >> >
> >> > so fbdev_blank returns _before_ the below enable_planes/frontbuf_flush?
> >> > Can you please attach full backtraces for steps 5&6?
> >>
> >> [  156.665517] [drm:intel_fbdev_set_par] PSR FBDEV modeset
> >> [  759.232969] [drm:intel_fbdev_blank] PSR FBDEV blank normal
> >> [  759.232987] [drm:intel_crtc_disable_planes] PSR FBDEV crtc disable
> >> planes flush fb bits
> >> [  897.313095] [drm:intel_fbdev_blank] PSR FBDEV unblank
> >> [  897.313112] [drm:intel_crtc_control] PSR FBDEV crtc enable planes
> >> [  897.527818] [drm:haswell_crtc_enable] PSR FBDEV crtc enable planes
> >> [  897.542925] [drm:intel_crtc_enable_planes] PSR FBDEV crtc enable
> >> planes flush fb bits
> >
> > I didn't mean the drm.debug log but the full backtrace for every time we
> > flush psr fb bits. I want to know who's the top-level function which
> > eventualy leads to the fb flush. I.e. something like
> >
> >         WARN_ON(frontbuffer_bits);
> >
> > in intel_psr_flush (after we've filtered out any already set bits and
> > other stuff that doesn't apply ofc).
> 
> I'm not sure if I understood your request...
> This [  897.542925] [drm:intel_crtc_enable_planes] PSR FBDEV crtc
> enable planes flush fb bits
> is the point where frontbuffer bits will be flushed calling psr_flush
> and psr gets back to life after the fbdev unblank.

I know, but I want the stacktrace for that point in time. The easiest way
to get that is a WARN_ON. If this would be userspace you could just use
the gdb backtrace command.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch