Hi, > Looking at it, I suspect there is another bug lurking. > bge_attach() has 14 places where it can return without > reeaching the end.
For some drivers, the progress through the setup is tracked and the detach or failure unwinds the successful steps. Maybe it's worth doing something similar in bge? For example, src/sys/dev/ic/gem.c [1] and specifically the GEM_ATT_* definitions, gem_detach()/gem_partial_detach(). Regards, Julian [1] https://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/ic/gem.c?rev=1.139;content-type=text%2Fx-cvsweb-markup --
