[Intel-gfx] Skylake graphics regression: projector failure with 4.8-rc3

[James Bottomley]

On Wed, 2016-08-31 at 21:51 +0000, Zanoni, Paulo R wrote:
> Hi
> 
> Em Qua, 2016-08-31 às 14:43 -0700, James Bottomley escreveu:
> > On Wed, 2016-08-31 at 11:23 -0700, James Bottomley wrote:
> > > 
> > > On Fri, 2016-08-26 at 09:10 -0400, James Bottomley wrote:
> > > > 
> > > > We seem to have an xrandr regression with skylake now.  What's
> > > > happening is that I can get output on to a projector, but the 
> > > > system is losing video when I change the xrandr sessions (like 
> > > > going from a --above b to a --same-as b).  The main screen goes
> > > > blank, which is basically a reboot situation.  Unfortunately, I
> > > > can't seem to get the logs out of systemd to see if there was a
> > > > dump to dmesg (the system was definitely responding).
> > > > 
> > > > I fell back to 4.6.2 which worked perfectly, so this is
> > > > definitely 
> > > > some sort of regression.  I'll be able to debug more fully when
> > > > I 
> > > > get back home from the Linux Security Summit.
> > > 
> > > I'm home now.  Unfortunately, my monitor isn't as problematic as
> > > the
> > > projector, but by flipping between various modes and separating
> > > and
> > > overlaying the panels with --above and --same-as (xrandr), I can
> > > eventually get it to the point where the main LCD panel goes
> > > black 
> > > and can only be restarted by specifying a different mode.
> > > 
> > > This seems to be associated with these lines in the X
> > > 
> > > [ 14714.389] (EE) intel(0): failed to set mode: Invalid argument
> > > [22]
> > > 
> > > But the curious thing is that even if this fails with the error 
> > > message once, it may succeed a second time, so it looks to be a 
> > > transient error translation problem from the kernel driver.
> > > 
> > > I've attached the full log below.
> > > 
> > > This is only with a VGA output.  I currently don't have a HDMI 
> > > dongle, but I'm in the process of acquiring one.
> > 
> > After more playing around, I'm getting thousands of these in the
> > kernel
> > log (possibly millions: the log wraps very fast):
> > 
> > [23504.873606] [drm:intel_dp_start_link_train [i915]] *ERROR*
> > failed
> > to train DP, aborting
> > 
> > And then finally it gives up with 
> > 
> > [25023.770951] [drm:intel_cpu_fifo_underrun_irq_handler [i915]]
> > *ERROR* CPU pipe B FIFO underrun
> > [25561.926075] [drm:intel_cpu_fifo_underrun_irq_handler [i915]]
> > *ERROR* CPU pipe A FIFO underrun
> > 
> > And the crtc for the VGA output becomes non-responsive to any
> > configuration command.  This requires a reboot and sometimes a UEFI
> > variable reset before it comes back.
> 
> Please see this discussion:
> https://patchwork.freedesktop.org/patch/103237/
> 
> Do you have this patch on your tree? Does the problem go away if you
> revert it?

Yes, I've got it, it went in in 4.8-rc3 according to git:

commit 58e311b09c319183254d9220c50a533e7157c9ab
Author: Matt Roper <matthew.d.roper at intel.com>
Date:   Thu Aug 4 14:08:00 2016 -0700

    drm/i915/gen9: Give one extra block per line for SKL plane WM
calculations

Reverting it causes the secondary display not to sync pretty much at
all.  However, in the flickers I can see, it does work OK and doesn't
now crash switching from --same-as to --above and back

I also still get the logs filling up with the link training errors.

On balance, although the behaviour is different, it's not an
improvement because if I can't sync with the projector, I can't really
use this as a fix.

James