Re: [virtio] Groups - Action Item "Create text version of virtio 0.9.5 document" added

[Rusty Russell <rusty@au1.ibm.com>]

Anthony Liguori <aliguori@us.ibm.com> writes:
> Rusty Russell <rusty@au1.ibm.com> writes:
>
>> Rusty Russell <rusty@au1.ibm.com> writes:
>>> -----------------
>>> Action Item Subject: Create text version of virtio 0.9.5 document
>>
>> OK, I've attached this below.  This is exactly the spec as 0.9.5
>> reformatted into text so we can work it.
>>
>> I'm still waiting for virtio to be added to the OASIS issue tracking
>> system, where I will be prompting you all to open an issue for every
>> change we want to consider.
>>
>> The obvious issues I want to open are:
>>
>> o Major rework to make PCI an appendix, and core bus-independent.
>
> We should revisit the use of PCI vendor/device IDs for virtio-pci.
>
> Currently we use the Red Hat/Qumranet vendor ID.  This has led to a
> situation where only Red Hat can produce a signed set of drivers for
> Windows (since they own the vendor ID).  These drivers are released
> under proprietary and discriminatory terms.  As it stands today, it is
> impossible to create a signed driver for Windows that is spec compliant
> unless you are Red Hat.
>
> Can we work with Oasis to get a vendor neutral PCI ID created to avoid
> this problem in the future?

Excellent point.

I'm not sure PCI IDs *can* be vendor neutral?  Beyond RH allowing others
to use theirs as they've done.

Perhaps the spec should not specify them, except to note that RH have
allowed compliant software implementations within their range, perhaps
even requiring all compliant drivers to recognize those.  It's certainly
commonplace on Linux to add new PCI IDs to existing drivers.  Is this
already a solved problem? (Hopeful question!)

Cheers,
Rusty.
PS.  Issue tracking will be up next week; once it is I'll post across
     from this thread.