Re: [Qemu-devel] [Xen-devel] [PATCH] Citrix PV Bus device [V3]

[Anthony Liguori <anthony@codemonkey.ws>]

Ian Campbell <Ian.Campbell@citrix.com> writes:

> On Wed, 2013-07-03 at 09:34 +0100, Paul Durrant wrote:
>> Already did that  :-)
>> 
>> > There are also sub-vendor and sub-device IDs but I don't think they are
>> > so useful for us (AFAIK they are intended to allow the board
>> > manufacturer to "subclass" the IDs baked into the ASIC).
>> > 
>> 
>> I'm always hazy about what those mean. I thought the idea was that a
>> vendor may collect together many devices, potentially from different
>> vendors, into a single chip/board and they should use common subsystem
>> vendor/device info for all those devices to indicate that they were
>> all part of that larger unit - but maybe I'm completely wrong.
>
> I figured it was so the board vendor could add "value" in their drivers
> by taking advantage of the higher precedence given to the binding to the
> sub-ids by OSs.

It's essentially an OEM id.  Often times it's an EEPROM setting on real
hardware.  A different subsystem vendor/device does not indicate a
different programming interface.

We set it to a vendor/device ID reserved for QEMU by default.  It's best
to keep it the QEMU ID to identify it as a device implemented by QEMU.

It's mighty handy as it lets software figure out that it's a Cirrus VGA
card emulated by QEMU (not real hardware).  In fact, I believe the
kernel KMS driver for Cirrus only binds to the QEMU vendor/device ID
since that's the only thing reasonably testable these days.

Regards,

Anthony Liguori