On Tue May 21 12:48:24 2024 +0000, Zhiyi Zhang wrote:
The CI test failures look suspicious, could you take a look? Also, the monitor and vulkan tests timeout at https://testbot.winehq.org/JobDetails.pl?Key=145557
Thanks, there was indeed a deadlock when calling into d3dkmt vulkan, which could then call VulkanInit driver entry, which loads the graphics drivers, call into display devices again and back into d3dkmt pthread_once.
I've fixed this by lazily loading the graphics driver vulkan functions, which are now only needed when surfaces are created, which should only be happening after the display device initialization and when graphics drivers are fully loaded already anyway.