Re: i915 regression on 3.6-rc1: lid blanks screen

[Takashi Iwai <tiwai@suse.de>]

At Fri, 10 Aug 2012 11:53:51 -0700 (PDT),
Hugh Dickins wrote:
> 
> On Fri, 10 Aug 2012, Takashi Iwai wrote:
> > At Fri, 10 Aug 2012 14:35:13 +0200,
> > Daniel Vetter wrote:
> > > 
> > > On Fri, Aug 10, 2012 at 1:59 PM, Takashi Iwai <tiwai@suse.de> wrote:
> > > > At Mon, 6 Aug 2012 11:25:30 -0700 (PDT),
> > > > Hugh Dickins wrote:
> > > >>
> > > >> On Mon, 6 Aug 2012, Daniel Vetter wrote:
> > > >> > On Mon, Aug 6, 2012 at 6:21 AM, Hugh Dickins <hughd@google.com> wrote:
> > > >> > > On Sun, 5 Aug 2012, Takashi Iwai wrote:
> > > >> > >> At Sat, 4 Aug 2012 10:01:13 -0700 (PDT),
> > > >> > >> Hugh Dickins wrote:
> > > >> > >> >
> > > >> > >> > Sorry to report that with 3.6-rc1, closing and opening the lid on
> > > >> > >> > this ThinkPad T420s leaves the screen blank, and I have to reboot.
> > > >> > >> >
> > > >> > >> > Bisection led to this commit, and reverting indeed gets my screen back:
> > > >> > >> >
> > > >> > >> > commit 520c41cf2fa029d1e8b923ac2026f96664f17c4b
> > > >> > >> > Author: Daniel Vetter <daniel.vetter@ffwll.ch>
> > > >> > >> > Date:   Wed Jul 11 16:27:52 2012 +0200
> > > >> > >> >
> > > >> > >> >     drm/i915/lvds: ditch ->prepare special case
> > > >> > > ...
> > > >> > >>
> > > >> > >> Hm, it's surprising.
> > > >> > >>
> > > >> > >> Could you check whether the counter-part intel_lvds_enable() is
> > > >> > >> called?  If the prepare callback affects, it must be from the mode
> > > >> > >> setting (drm_crtc_helper_set_mode()).
> > > >> > >
> > > >> > > Yes, I put a dump_stack() in both, and intel_lvds_enable() gets called
> > > >> > > about 0.28 seconds after the intel_lvds_disable() when I lift the lid;
> > > >> > > but with no video display until I revert that commit.
> > > >> >
> > > >> > Can you please boot with drm.debug=0xe added to your kernel cmdline,
> > > >> > reproduce the issue (with the two dump_stack calls added) and then
> > > >> > attach the full dmesg?
> > > >>
> > > >> Collected, I'll send it to you both privately in a moment.
> > > >>
> > > >> >
> > > >> > Also a few other things to try: What happens if you do a modeset on
> > > >> > the LVDS while it's still working, e.g.
> > > >>
> > > >> In the dmesg, I've only gone to runlevel 3, simply working on the
> > > >> console without startx.  For these xrandrs to work, I did startx
> > > >> and used the graphics screen.
> > > >
> > > > OK, now I can see the problem here, too.  The key is that it happens
> > > > only on Linux console, not on X.  That's why no one else reported.
> > > > I guess the problem can be seen on many laptops with LVDS on PCH.
> 
> A correction there: for me it was happening both on X and on console;
> but once I found that it happened even on the simple console, I mostly
> stuck to bisecting and testing on that.

Maybe depending on the desktop setup.  When a power-management stuff
is running, it may readjust the backlight level upon lid open, then it
fixes the issue casually.


> > > >
> > > > Looking at intel_reg_dumper output, BLC_PWM_CPU_CTL is 0 while other
> > > > registers are set correctly.  This seems coming from the rewrite of
> > > > backlight control code by commit
> > > >   24ded204: drm/i915: properly enable the blc controller on the right pipe
> > > > and
> > > >   a4f32fc3: drm/i915: don't forget the PCH backlight registers
> > > >
> > > > While the latter fixes the regression by the former commit, it still
> > > > doesn't cover this regression.
> > > >
> > > > I don't know the exact hardware behavior, but it looks like that
> > > > resetting BLC_PWM_PCH_CTL2 and BLC_PWM_PCH_CTL1 clears the
> > > > BLC_PWM_CPU_CTL (oh what confusing reg names).
> > > >
> > > > FWIW, the commit 520c41cf you mentioned is no direct cause.  This
> > > > patch works fine on the top of 3.5 kernel.  But it's like a bad drug,
> > > > the combination of this and other two commits break things.
> > > >
> > > > The patch below is my quick fix.  It worked on an HP laptop.
> > > > Hugh, could you give it a try?
> > > 
> > > Hm, this sounds eerily familiar to the backlight bug you've recently fixed in
> > > 
> > > commit 6db65cbb941f9d433659bdad02b307f6d94465df
> > > Author: Takashi Iwai <tiwai@suse.de>
> > > Date:   Thu Jun 21 15:30:41 2012 +0200
> > > 
> > >     drm/i915: Fix eDP blank screen after S3 resume on HP desktops
> > > 
> > > Have you checked other code-paths for such issues?
> > 
> > Not yet, I'd leave such a joy rather to you guys :)
> > 
> > > The resume code
> > > seems to follow this order already ... also, when you submit this
> > > patch, can you please add a small comment to explain the ordering
> > > constraint, like in the resume register restore function?
> > 
> > Sure, I'll add a comment and resubmit once when I hear it really fixes
> > on Hugh's machine, too.
> 
> Indeed, your patch really fixes it on my machine: many thanks!

OK, I'll resend the patch with more comments.


thanks,

Takashi