The current theory of the root cause of the nv_107/108 GUI install
failures is that the miniroot is missing libraries that the Xorg
server is linked against - since Xorg is linked -z lazyload, it gets
far enough to initialize the screen, and then calls into libdbus,
which it can't find, and gets the fatal relocation error.
If it had crashed due to a bug in the server, our SIGBUS/SIGSEGV
signal handler would have been triggered to restore the devices
back to sane state before exiting, but in this case it just looked
like the server was hung because the screen was left in the
initialized state, so we were looking down the wrong path for causes
for a while.
Is there any way to register an error handler to clean up like this
for runtime linker errors like failure to find a lazy loaded library?
Or should we just make sure to not lazyload the libraries we know we'll
always need at startup?
--
-Alan Coopersmith- alan.coopersmith at sun.com
Sun Microsystems, Inc. - X Window System Engineering
