Screen turned off after X server exits

[Robert Ancell]

When the X server exits, the screen turns off and doesn't seem to re-enable. The system is still running (i.e. can SSH in).

To reproduce:
1. Log into a session
2. Switch to the greeter (using indicators or running "dm-tool switch-to-greeter" from a terminal")
3. Log back into session

Expected result:
You return to the session

Observed result:
Screen goes black

This appears to be an issue in the Intel drivers, it occurs with the current wily (2:2.99.917+git20150808-0ubuntu2). It does not occur when downgrading to the vivid version (2:2.99.917-1~exp1ubuntu2.2 with some patches to compile in wily).

[Robert Ancell (robert-ancell)]

I am bisecting the problem...

[Robert Ancell (robert-ancell)]

I first noticed this occuring sometime around the end of August.

[Robert Ancell (robert-ancell)]

The issue is in git master and bisecting shows it to be this commit:

commit 7d9a74622e5a936e4860fcef8358619bf59adae8
Author: Chris Wilson <email address hidden>
Date: Mon Jun 15 14:16:34 2015 +0100

    sna: Be robust in handling DPMS failures

    If we fail to turn off an output via DPMS, disable the entire CRTC in
    order to blank the output and save the screeen/power.

    Reported-by: Lukas Hejtmanek <email address hidden>
    References: https://bugs.freedesktop.org/show_bug.cgi?id=90179
    Signed-off-by: Chris Wilson <email address hidden>

diff --git a/src/sna/sna_display.c b/src/sna/sna_display.c
index f58eb8d..efc5fc9 100644
--- a/src/sna/sna_display.c
+++ b/src/sna/sna_display.c
@@ -213,7 +213,7 @@ struct sna_output {
  int panel_vdisplay;

  uint32_t dpms_id;
- int dpms_mode;
+ uint8_t dpms_mode;
  struct backlight backlight;
  int backlight_active_level;

@@ -251,6 +251,7 @@ enum { /* XXX copied from hw/xfree86/modes/xf86Crtc.c */
  OPTION_DEFAULT_MODES,
 };

+static void __sna_output_dpms(xf86OutputPtr output, int dpms, int fixup);
 static void sna_crtc_disable_cursor(struct sna *sna, struct sna_crtc *crtc);

 static bool is_zaphod(ScrnInfoPtr scrn)
@@ -936,7 +937,7 @@ sna_crtc_force_outputs_on(xf86CrtcPtr crtc)
   if (output->crtc != crtc)
    continue;

- output->funcs->dpms(output, DPMSModeOn);
+ __sna_output_dpms(output, DPMSModeOn, false);
  }

 #if XF86_CRTC_VERSION >= 3
@@ -966,7 +967,7 @@ sna_crtc_force_outputs_off(xf86CrtcPtr crtc)
   if (output->crtc != crtc)
    continue;

- output->funcs->dpms(output, DPMSModeOff);
+ __sna_output_dpms(output, DPMSModeOff, false);
  }
 }

@@ -1099,7 +1100,7 @@ sna_crtc_apply(xf86CrtcPtr crtc)
    * and we lose track of the user settings.
    */
   if (output->crtc == NULL)
- output->funcs->dpms(output, DPMSModeOff);
+ __sna_output_dpms(output, DPMSModeOff, false);

   if (output->crtc != crtc)
    continue;
@@ -3580,7 +3581,7 @@ sna_output_destroy(xf86OutputPtr output)
 }

 static void
-sna_output_dpms(xf86OutputPtr output, int dpms)
+__sna_output_dpms(xf86OutputPtr output, int dpms, int fixup)
 {
  struct sna *sna = to_sna(output->scrn);
  struct sna_output *sna_output = output->driver_private;
@@ -3607,8 +3608,9 @@ sna_output_dpms(xf86OutputPtr output, int dpms)
  if (sna_output->backlight.iface && dpms != DPMSModeOn) {
   if (old_dpms == DPMSModeOn) {
    sna_output->backlight_active_level = sna_output_backlight_get(output);
- DBG(("%s: saving current backlight %d\n",
- __FUNCTION__, sna_output->backlight_active_level));
+ DBG(("%s(%s:%d): saving current backlight %d\n",
+ __FUNCTION__, output->name, sna_output->id,
+ sna_output->backlight_active_level));
   }
   sna_output->dpms_mode = dpms;
   sna_output_backlight_off(sna_output);
@@ -3618,18 +3620,31 @@ sna_output_dpms(xf86OutputPtr output, int dpms)
      drmModeConnectorSetProperty(sna->kgem.fd,
      sna_output->id,
      sna_output->dpms_id,
- dpms))
- dpms = old_dpms;
+ dpms)) {
+ DBG(("%s(%s:%d): failed to set DPMS to %d (fixup? %d)\n",
+ __FUNCTION__, output->name, sna_output->id, dpms, fixup));
+ if (fixup) {
+ sna_crtc_disable(output->cr...

[Robert Ancell (robert-ancell)]

https://bugs.freedesktop.org/show_bug.cgi?id=90179 was the reason for this commit.

[Robert Ancell (robert-ancell)]

I've put a patched version of the Intel driver in ppa:robert-ancell/xorg [1]. I think the change might cause a DPMS regression as described in the upstream bug so it probably should get some more testing before uploading.

[1] https://launchpad.net/~robert-ancell/+archive/ubuntu/xorg

[Michał Sawicz (saviq)]

Confirmed, this is the exact problem I'm having here.

[Jean-Baptiste Lallement (jibel)]

I tried the patched version from the PPA and it fixes the problem for me. No more black screen after restarting lightdm

[Robert Ancell (robert-ancell)]

I've uploaded this change to wily. There is a possibility that there might be a regression for some users with DPMS but I think that is less severe than the effect on those who are left with the screen off. We'll look at doing an update if upstream provides a fix.

[Launchpad Janitor (janitor)]

This bug was fixed in the package xserver-xorg-video-intel - 2:2.99.917+git20150808-0ubuntu3

---------------
xserver-xorg-video-intel (2:2.99.917+git20150808-0ubuntu3) wily; urgency=medium

  * debian/patches/revert-dpms-fix.patch:
    - Revert a DPMS change that causes the screen to turn off when returning
      from a greeter (LP: #1501941)

 -- Robert Ancell <email address hidden> Tue, 06 Oct 2015 14:25:00 +1300

[Robert Ancell (robert-ancell)]

On Intel systems when the X server exits the display is turned off and remains off. This is triggered when using LightDM and switching to a greeter (starts a second X server) then authenticating in that greeter and returning to the first X server (the greeter X server is stopped).

The cause seems to be a DPMS change from bug 90179 (revision 7d9a74622e5a936e4860fcef8358619bf59adae8) that causes the driver to turn off the output if DPMS is not working.

Tracked in Launchpad:
https://bugs.launchpad.net/bugs/1501941

[Robert Ancell (robert-ancell)]

We are currently working around this issue in Ubuntu 15.04 by reverting that one commmit.

[Jani-nikula (jani-nikula)]

(In reply to Robert Ancell from comment #1)
> We are currently working around this issue in Ubuntu 15.04 by reverting that
> one commmit.

That one commit being

commit 7d9a74622e5a936e4860fcef8358619bf59adae8
Author: Chris Wilson <email address hidden>
Date: Mon Jun 15 14:16:34 2015 +0100

    sna: Be robust in handling DPMS failures

i.e. not a kernel commit. Apparently Chris' conclusion was that the kernel returns an error on DPMS.

Please attach dmesg with drm.debug=14 module parameter, all the way from boot to reproducing the problem.

[Robert Ancell (robert-ancell)]

I haven't been able to generate a dmesg due to another kernel issue I have where it spams the log. I will generate one when that issue is resolved...

[Ricardo-vega (ricardo-vega)]

This bug is really old, is this problem still persist? Robert are you able now to share your dmesg?