kvm-guest-drivers-windows: viogpudo /virtio-vga broken in recent Windows builds
In recent Windows versions, like Windows 11 IoT 24H2, when using the virtio-vga graphics device, the screen turns black when installing the viogpudo driver and stays black until the next reboot.
Using the std VGA device everything works.
The problem even exists with the new beta drivers, version 0.1.258.
About this issue
- Original URL
- State: open
- Created 2 years ago
- Comments: 56 (17 by maintainers)
I’ve tried debugging this issue some more. Following https://learn.microsoft.com/en-us/windows-hardware/manufacture/desktop/winpe-debug-apps?view=windows-11 I was able to create a WinPE boot ISO with debugging and test signing enabled. This allows me to attach windbg in WinPE. Using
drvloadallows me to load a test build of the driver.This is the log output when loading the driver in WinPE:
The log stops there. In a normal working windows install, a bunch of log entries continue. The next lines should be something like:
This says to me that Windows isn’t deciding to enumerate modes as per https://learn.microsoft.com/en-us/windows-hardware/drivers/display/enumerating-cofunctional-vidpn-source-and-target-modes
The docs say “from time to time” Windows will enumerate modes but doesn’t specify the exact conditions. From testing it seems like in normal boot and upon driver install, it will happen after the new display miniport driver gets loaded. However, in WinPE, that doesn’t happen. It would be nice if anyone with direct contact with Microsoft can ask them about this and if the driver has to do anything extra to get dxgkrnl to call IsSupportedVidPn
I don’t know if this is related but there was another issue that has been bothering me for a long time (as in years) which is that the viogpudo driver doesn’t inhibit the Microsoft Basic Display on arm64. On x86_64 builds, when viogpudo gets loaded, Microsoft Basic Display gets disabled and only one display shows up in settings. However, on arm64, when viogpudo gets loaded, there are two displays because the boot display (Microsoft Basic Display attached to the UEFI initialized framebuffer) isn’t disabled. I’ve tried all sorts of ways to try to get Windows to hide Microsoft Basic Display but in the end I used a registry hack to manually disable it.
What’s the point of this story? I think it might be related because it’s possible that this strange behaviour that I only observed on arm64 is now showing up on x86_64 as well because if you try to render to the boot display, you will get a black screen. I’m not a windows debugging expert, but can someone check if it goes to a black screen how many display devices does windows think is there and which one it thinks is the primary video device?
Issue still present in v0.1.266
@kroese
Dehan shared with me all the scripts that he uses. I’m trying to reproduce and investigate this issue.
Hi @kroese, I could reproduce this issue on
scenario: