this seems reasonable.  i wish both SCSI and PCI had 'target'/'card'
abstractions like this, actually.

One question:

How does a 'compound' device present itself to the user?

for simple devices, it seems obvious:

	allow access via device driver if attached, else
	allow "raw" access via udev.

for compound devices what do you do?  It looks from your drawing like
you'll have only one 'udev' for such a case, which makes doing the
right thing harder: it means that you have to have the user do the
sub-"udev" device selection themselves...  That seems ... more complex
than it perhaps should be (and that complexity then has to spill over
into the simple device raw access case, too).


cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.