On Sat, 24 Nov 2007, Matthias Drochner wrote:

> The memory at sc should be zero initialized by the
> autoconf framework, so we should see a NULL dereference.
> There must be another bug which made that kernel
> did not drop into ddb.

I am not sure, but since sc_dev is NULL, then NULL is passed
to the aprint_error_dev function, but it passes it to device_xname which
thinks it is a valid pointer and returns the offset of dv_xname

so, the invalid memory access would be __offsetof(device_t, dv_xname),
would that be caught?

probably it would be good for aprint_*_dev() and device_*() to contain

	KASSERT(dv != NULL);

?

iain