On Wed, Oct 08, 2003 at 08:46:06PM +0200, Frank van der Linden wrote:
> On Wed, Oct 08, 2003 at 06:41:37PM +0000, Charles M. Hannum wrote:
> > 2) A controller+drive older than that will read all-1s.  Therefore it
> > will fail the 0 test and continue polling.  This is completely harmless.
> > Because the drive also is too old to support the slave register shadowing,
> > this test would not work for it anyway even if it read the main status
> > register.
> 
> Is that guaranteed? This is fine on i386, but there are architectures
> out there, like macppc, that will fault if you read a non-existent I/O
> address.

Others will return the previous contents of the data bus [1].

Although I suspect that in this case you are reading an address that
the controller would have aliased to something else, rather than
not have been decoded.

Has anyone seen a vmebus IDE disk controller (or whatever it used to
be called) - vmebus is the obvious bus that would fault transfers.

	David

[1] subject to its natural capacitance.

-- 
David Laight: david@l8s.co.uk