radeondrm crashes on x86 fixed

[matthew green <mrg%eterna.com.au@localhost>]

hi folks.


i commited some fixes a couple of days ago that fixed the radeondrm
crashes that dyoung's bus_space changes triggered.  totally not his
fault, the code was using uninitialised tags as noted, and turns out
to have been generally broken for these maps anyway...

the fix is to only use bus_space on handles that have a mapping in
some bus space, otherwise using direct accesses is required.

this in turn fixed the radeondrm "writeback" test that has always
failed for me, eg:

        info: [drm] writeback test failed ffffffff deadbeef

now correctly reports:

        info: [drm] writeback test succeeded in 1 usecs


.mrg.