[PATCH 0/3] xf86: xf86Crtc.c compat_output patches

[vdb at picaros.org]

There is a lurking bug in xserver/hw/xfree86/modes/xf86Crtc.c 
about the use of compat_output.  It will show once users start 
to disconnect crtcs and monitors from outputs.  

This stable code is probably unfamiliar to most on the list.  Might 
someone be willing to try these patches ?

A summary is here below, a detailed report is in the bugzilla entry.  
http://bugs.freedesktop.org/show_bug.cgi?id=65210

--------------------------------

There is a lurking bug in xf86Crtc.c about the use of compat_output.  
The bug is twofold: a use-before-set at initial configuration and a 
desynchronization of scrn->{monitor, modes} when (re)connecting a 
monitor.  

The bug surfaced due to commit 37d956e3ac9513b74078882dff489f9b0a7a5a28 
into xserver/hw/xfree86/modes/xf86Crtc.c which presets

    config->compat_output = -1;

This parameter is referenced during initial screen configuration for: 

1. scrn->monitor         set by xf86SetDDCproperties(scrn, edid_mon),
2. config->compat_output set by SetCompatOutput(config),
3. scrn->modes           set by xf86SetScrnInfoModes(scrn).

The use-before-set for case 1 results from the video driver callback:

xf86InitialConfiguration() calls
xf86ProbeOutputModes() which calls into the video driver via 
(*output->funcs->get_modes)() which callbacks 
xf86OutputSetEDID() where 
xf86CompatOutput() dereferences
config->output[config->compat_output]

with compat_output == -1.  Since at this initial stage the compat 
output can't be known a NULL return from xf86Crtc.h 
xf86CompatOutput() is a logical solution. 

Also, xf86SetDDCproperties() must be called after each 
SetCompatOutput() compat_output change for scrn->monitor and 
scrn->modes consistency.

The first two patches try to stay as close as possible to the code.  
The third patch guarantees a compat_output in the range of 
[ -1 .. num_output-1 ] by reorganizing the if-then-else fallback 
logic.