Matthias Drochner <drochner@zel459.zel.kfa-juelich.de> writes:
> aleb@ics.com said:
> > my reading of the kernel sources in dev/ic/vga.c, dev/pci/tga.c, and
> > others is that right now no amount of wishful thinking is going to get
> > a return of WSDISPLAY_TYPE_PCIVGA. 
> 
> You are right. The reason I didn't change it is that the whole
> WSDISPLAYIO_GTYPE is half-done and needs a new interface imho.
> To be more useful, some more information should be provided
> like a more specific device identification (VID/DID in the PCI case)
> and physical attachment information (PCI bus/dev/fcn or even
> mappable memory ranges).

The way i'd intended it:

* there would be an additional ioctl to get more type-specific
data, such as vendor, device, and subsystem IDs for PCI devices.
Dunno if that should be GTYPE-specific, or generic with a large enough
buffer to keep everybody happy.

* mappable memory ranges would be provided by a set of ioctls (which
could also be used to implement a bus_space abstraction in user-land;
i'm not convinced that jason is ever going to put a proposal up about
this... *sigh*).  if he doesn't by the time I get some free time (ha
ha) I will.

* phyical attachment information would _not_ be provided.  If
something thinks it needs to know, i'd say it's wrong.  you don't
configure by b/d/f (or even h/b/d/f), you configure by device node.



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