diff options
Diffstat (limited to 'results/classifier/accel-gemma3:12b/kvm/1107')
| -rw-r--r-- | results/classifier/accel-gemma3:12b/kvm/1107 | 25 |
1 files changed, 25 insertions, 0 deletions
diff --git a/results/classifier/accel-gemma3:12b/kvm/1107 b/results/classifier/accel-gemma3:12b/kvm/1107 new file mode 100644 index 000000000..357381aba --- /dev/null +++ b/results/classifier/accel-gemma3:12b/kvm/1107 @@ -0,0 +1,25 @@ + +Virtual monitor heads are not "connected" until viewed in a front end +Description of problem: +When you attach a virtual GPU to a guest, qemu appears to only "attach" a virtual monitor to an output port when that virtual display is +viewed using the GUI. For example, when you boot using the above command line, there will be four displays in ```/sys/class/drm/``` on the guest, +```card0-Virtual-1``` through to ```card0-Virtual-4```. In each of these directories, there is an "enabled" file, which contains either +"enabled" or "disabled". These contain "disabled" until you switch tab/view to look at it using the GUI, at which point they change to "enabled". + +This causes a problem for us because Weston will not initialise displays that do not have a monitor attached, meaning the system we are trying +to boot fails because not all the Weston display surfaces are available. + +There does not appear to be a command line option to force virtual monitors to be attached to virtual displays immediately. Looking through the +Gtk user interface code (and the other front ends) there does not appear to be a call into the qemu core that requests the connection of a virtual +monitor to the virtual displays - my guess is that qemu only connects a monitor when a render request first happens (or similar), but I have not followed the code paths deeper than the source files in ```QEMU/ui/```. + +I also tried using the ```screengrab``` command to screenshot each head, but this does not need sufficient to cause the display to be marked +enabled in the guest. + +While we could possibly automate the GUI using some external tool, we ultimately need to run this in a CI environment using +```egl-headless``` or similar. +Steps to reproduce: +1. Launch qemu with virtio-gpu-gl setting max_outputs > 1 +2. On guest, ```cat /sys/drm/class/card0-Virtual-2``` - it reads "disabled" +3. On host, switch the view to look at the second display ("virtio-gpu-gl-pci.1") +4. On guest, ```cat /sys/drm/class/card0-Virtual-2``` - it now reads "enabled" |