Some thread might be destroying a window, calling destroy_client_window from release_gl_drawable, holding the GL context mutex and trying to enter the win_data mutex.
At the same time, another thread might be moving its window, calling sync_gl_drawable from X11DRV_WindowPosChanged, holding the win_data mutex and trying to enter the GL context mutex.
The deadlock was present before already, although less frequently triggered as sync_gl_drawable was done conditionally if the client window has been moved.
This triggers now more frequently in the dxgi:dxgi tests.