Does fglrx simply load built in Intel module when iGPU selected, or more complex?

Azrael84 · 01-27-2014, 07:28 AM

Sorry for multiple posts, but another observation:

When fglrx is in charge and pxpress loads the i915 module, it does so with what appears to be UXA acceleration method, whereas when i915 loads on its own (by which I mean on a system where fglrx has never been installed) it defaults to SNA method.

I would like to try the experiment of setting fglrx to load the i915 module with SNA accel method, but given there is not the usual:

Code:

Section "Device"
   Identifier  "Intel Graphics"
   Driver      "intel"
   Option      "AccelMethod"    "SNA"
EndSection

section of the the xorg.conf file (presumably because of the way pxpress works). How can I set options for the intel driver? For concreteness my xorg.conf looks like:

Code:

Section "ServerLayout"
        Identifier     "Radeon Layout"
        Screen      0  "aticonfig-Screen[0]-0" 0 0
EndSection

Section "Monitor"
        Identifier   "aticonfig-Monitor[0]-0"
        Option      "VendorName" "ATI Proprietary Driver"
        Option      "ModelName" "Generic Autodetecting Monitor"
        Option      "DPMS" "true"
EndSection

Section "Device"
        Identifier  "aticonfig-Device[0]-0"
        Driver      "fglrx"
        BusID       "PCI:6:0:0"
EndSection

Section "Screen"
        Identifier "aticonfig-Screen[0]-0"
        Device     "aticonfig-Device[0]-0"
        Monitor    "aticonfig-Monitor[0]-0"
        DefaultDepth     24
        SubSection "Display"
                Viewport   0 0
                Depth     24
        EndSubSection
EndSection

Trying to add an intel device section (even the most minimal possible) really screws things up, and causes fglrx to segfault, probably because it (pxpress) also loads the intel on its own or some such.
Setting options for the ATI device, in the hope that they would be transfered to when it loaded the i915 did not work.

So I can I set AccelMethod SNA for the intel device, or indeed any other module options for it when fglrx is installed?

Azrael84 · 01-27-2014, 07:44 AM

EUREKA!

Looks like this is the key!

I just booted with an Intel only configured xorg.conf, and set

Code:

Section "Device"
        Identifier      "Intel Graphics"
        Driver          "intel"
        Option          "AccelMethod"  "UXA"
EndSection

the UXA accelmethod, like fglrx seems to utilise, and I get the crash/bug!
Even just on plain old intel.

So I probably need to work out how to use SNA when fglrx loads the i915 now!

Azrael84 · 01-29-2014, 03:13 AM

So if you have read this far congratulations! In the end this is a summary of my conclusions, and apparent fix:

==================================================================================================== ==========================================

On a Haswell system with an Intel HD 4400 as the integrated card in a hybrid system X freezes (moveable mouse cursor but either black screen or frozen image of desktop at time of freeze) often when the user locks the screen (inactivity or otherwise) or closes the lid. The desktop (in 13.10 at least) can be recovered by switching to a tty and back( e.g. ctrl+alt+fn1, followed by ctl+alt+fn7). In 12.04 I needed to restart completely from tty.

This bug occurs because pxpress loads the intel driver with the "AccelMethod" "UXA" option. You can verify this is the cause by using a minimal (intel only) `xorg.conf` (even before fglrx drivers are installed) with just

Code:

   Section "Device"
       Identifier      "Intel Graphics"
       Driver          "intel"
       Option          "AccelMethod"  "UXA" # (mode fglrx uses) I can see the bug
   #   Option          "AccelMethod"  "SNA" #(default) no bug!
   EndSection

and you should reproduce the bug with a few attempts at closing lid/locking screen. Whereas with the change to ` Option "AccelMethod" "SNA"` there is no bug anymore ("SNA" is the default anyway). So this a bug in the intel driver when UXA is selected on a Haswell system it seems.

How to fix

Create (or edit) the DRI config file `~/.drirc` containing

Code:

   <device screen="0" driver="dri2">
       <application name="Default">
               <option name="vblank_mode" value="0"/>
       </application>
   </device>

This turns off "vsync" for the Intel card (the mode that attempts to make your gfx card sync its framerate with that of the display to avoid tearing, rather than racing ahead at its intrinsic higher framerate).

Now even when pxpress loads the intel driver in "UXA" mode, the bug should not be reproducible.

It would be a better solution if it were possible to tell pxpress to load intel driver with SNA mode directly, but I have no idea if this is possible, maybe it is hardcoded into the drivers.