Subject: Re: -current bha driver attaches PCI device multiple times (as ISA)
To: Jonathan Stone <jonathan@DSG.Stanford.EDU>
From: Justin T. Gibbs <gibbs@freefall.freebsd.org>
List: current-users
Date: 10/08/1996 08:03:52
>Thank you for pointing that out; but, as it happens, I had that in an
>editor buffer at the time I sent the message. I assumed we didn't
>need explicit advice to look at other souce.
Well, you didn't mention it, so I didn't know you knew this was an option.
>It does, of course, leave open the issue of whether it's
>architecturally preferable for the NetBSD Buslogic attach functions
>to
>
> a) autodisable the ISA ports on a PCI buslogic adaptor
> (when the ports were not disabled by the user via the
> SCSI BIOS) , making the releveant io ports available for
> other, conflicting devices; or
>
> b) leave a PCI/EISAthe BusLogic card which is
> BIOS-configured to respond to ISA primary or secondary
> addresses in that state, whilst _not_ configuring the
> alreayd-configured PCI devices as an ISA device too.
>
>(or both).
Well, since the ISA compatibility address is nothing more than a hack to
allow the new boards that are now fully PCI compliant to work with old
drivers, I think its best to disable the compatibility address. According
to severy problem reports on this issue in FreeBSD, even touching the
adapter to determine is ID string from the ISA address port once the device
has been properly configured in the upper address causes the card to go out
to lunch. It would seem really strange to me as a user if my ethernet card
at the ISA compatibility address of the Buslogic card worked fine under
Win95 (which will never use that address for the Buslogic adapter), but may
not under NetBSD because it doesn't use the far more preferable PCI BIOS
selected address for the card.
>This could make a difference to, for instance, user-space direct
>access to the relevant io ports. I concede that's a far-fechted
>example. It's not clear (to me) whether it's better for the kerrnel
>to second-guess the user by disabling access to a PCI card via the
>standard ISA ioports.
Why would using the upper address restrict this? It's still a 16bit
addressed I/O port.
--
Justin T. Gibbs
===========================================
FreeBSD: Turning PCs into workstations
===========================================