> They may do it in backgroud, while doing others tasks. Also, A bios with
> all devices turned to "auto" will take longer than with unexistent devices
> turned off.

My entire BIOS POST sequence takes less than 31s, so I sincerely doubt
that's the case.

> It may be possible to speed up ghost detection via a command:
> - we may have a ghost only if we have the same signature for device 0 and
>   device 1 after a reset
> - if we may have a ghost, issue IDENTIFY or ATAPI_IDENTIFY as appropriate.
>   After that, we should have either BSY set, or DRQ set, or an aborted
>   command. If the status register remains all 0, then this isn't a real
>   device.

Hm, that might work.  The flowchart in the spec seems to indicate that
DRQ should be asserted before BSY is cleared, but is that guaranteed?

> Note that we have to IDENTIFY both device, as in a slave-only config the
> controller may fake a master.

No, the slave will never fake the master registers, because it can't
know whether or not a master is present.  (The master only knows the
presence of the slave via DASP and PDIAG, which are one-way.)