Bug 21877

Created attachment 26115 [details]
Xorg.0.log

Created attachment 26116 [details]
xrandr --prop --verbose

Also props of DVI-D_1 aren't nice. Backlight is much too big, and _HDMI is 0 even I set it to 1:
$ grep HDMI /etc/X11/xorg.conf
  Option       "HDMI" "DVI-D_1"

Changing _HDMI works fine:
$ xrandr --output PANEL --set _HDMI 1
$ xrandr --prop
PANEL connected 1600x900+0+0 (normal left inverted right x axis y axis) 423mm x 238mm
        _HDMI: 1 (0x00000001)   range:  (0,1)
        _Coherent: 150136792 (0x08f2e7d8)       range:  (0,1)
        _Backlight: 150136792 (0x08f2e7d8)      range:  (0,255)


But changing _Backlight doesn't work:
$ xrandr --output PANEL --set _Backlight 100
$ xrandr --prop
PANEL connected 1600x900+0+0 (normal left inverted right x axis y axis) 423mm x 238mm
        _HDMI: 1 (0x00000001)   range:  (0,1)
        _Coherent: 150136792 (0x08f2e7d8)       range:  (0,1)
        _Backlight: 150136792 (0x08f2e7d8)      range:  (0,255)


Next thing is that changing _HDMI for DVI-D_1 crashes my X:
$ xrandr --output DVI-D_1 --set _HDMI 1

Created attachment 26117 [details]
Xorg.0.log.old after crash

Hm, I'm suspecting some piece of code. Started my debugging with PANEL.


PANEL is driven with:
(--) RADEONHD(0): Attaching Output UNIPHY_KLDSKP_LVTMA to Connector PANEL


My PANEL's Output->Property points to DigPropertyControl.


Here goes some forwardtrace:
1) rhdRROutputGetProperty(PANEL Output, atom_Backlight);
2) DigPropertyControl(PANEL Output, rhdPropertyGet, RHD_OUTPUT_BACKLIGHT, *val);
3) LVDSTransmitterPropertyControl(PANEL Output, rhdPropertyGet, RHD_OUTPUT_BACKLIGHT, *val);

Now:
LVDSTransmitterPropertyControl detects (Private->BlLevel < 0), returns FALSE;
DigPropertyControl doesn't check returned vales, returns TRUE

Created attachment 26120 [details] [review]
Fix for DigPropertyControl to return real result

This fixes DigPropertyControl, now it returns FALSE correctly. Still there is something wrong as Output->Private should be something working (now it doesn't have any my PANELS's properties: backlight, hdmi, coherent).

Reporting unexisting properites was fixed in commit efa0b5475517e4252c5a97389ee5612ed7da453f

Returning 0 for _HDMI every time was fixed in commit b8533a98448f8dc2d67137e02a180430def17e0b

The one left bug is that using:
xrandr --output DVI-D_1 --set _HDMI 0
crashes X.

Created attachment 26602 [details]
Patch that exposes source of crash

Created attachment 26603 [details]
verbose Xorg.0.log with exposing patch applied

What happens is that calling:
xrandr --output DVI-D_1 --set _HDMI 0
executes rhdRROutputSetProperty(...)

-rhdRROutputSetProperty(xf86OutputPtr, atom_HdmiProperty, RRPropertyValuePtr)
--DigPropertyControl(..., rhdPropertySet, RHD_OUTPUT_HDMI, &val)
---TMDSTransmitterPropertyControl(..., rhdPropertySet, RHD_OUTPUT_HDMI, &val)
--DigPropertyControl(..., rhdPropertyCommit, RHD_OUTPUT_HDMI, NULL)
---TMDSTransmitterPropertyControl(..., rhdPropertyCommit, RHD_OUTPUT_HDMI, NULL)

The last TMDSTransmitterPropertyControl executes:
> Output->Mode(Output, Private->Mode);
but Private->Mode is not set. A little deeper ATOMTransmitterSet tries to use Mode->SynthClock and that ends with crash.

Thanks for debugging. I'll think of something.

Fixed with git commit 4b4a439ef35af340b9c8e48e3760e375b07f873c