Various cleanups to the DRM core initialization and exit handlers:

 - Register chrdev last: Once register_chrdev() returns, open() will
   succeed on the given chrdevs. This is usually not an issue, as no
   chardevs are registered, yet. However, nodes can be created by
   user-space via mknod(2), even though such major/minor combinations are
   unknown to the kernel. Avoid calling into drm_stub_open() in those
   cases.
   Again, drm_stub_open() would just bail out as the inode is unknown,
   but it's really non-obvious if you hack on drm_stub_open().

 - Unify error-paths into just one label. All the error-path helpers can
   be called even though the constructors were not called yet, or failed.
   Hence, just call all cleanups unconditionally.

 - Call into drm_global_release(). This is a no-op, but provides
   debugging helpers in case there're GLOBALS left on module unload. This
   function was unused until now.

 - Use DRM_ERROR() instead of printk(), and also print the error-code on
   failure (even if it is static!).

 - Don't throw away error-codes of register_chrdev()!

 - Don't hardcode -1 as errno. This is just plain wrong.

 - Order exit-handlers in the exact reverse order of initialization
   (except if the order actually matters for syncing-reasons, which is
   not the case here, though).

v2:
 - Call drm_core_exit() directly from the init-error-handler. Requires to
   drop __exit annotation, though.
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20160901124837.680-7-dh.herrmann@gmail.com