On Thu, Mar 21, 2013 at 12:13:00PM +0200, Michael S. Tsirkin wrote:
> On Thu, Mar 21, 2013 at 06:59:37PM +1030, Rusty Russell wrote:
> > (MST, is this what you were thinking?)
>
> Almost.
>
> Three points:
>
> 1. this is still an offset in BAR so for KVM we are still forced to use
> an IO BAR. I would like an option for hypervisor to simply say "Do IO
> to this fixed address for this VQ". Then virtio can avoid using IO BARs
> completely.
>
> 2. for a real virtio device, offset is only 16 bit, using a 32 bit
> offset in a memory BAR giving each VQ a separate 4K page would allow
> priveledge separation where e.g. RXVQ/TXVQ are passed through to
> hardware but CVQ is handled by the hypervisor.
>
> 3. last thing - (1) applies to ISR reads as well.
>
> So I had in mind a structure like:
>
> struct vq_notify {
> u32 offset;
> u16 data;
> u16 flags;
> }
>
> enum vq_notify_flags {
> VQ_NOTIFY_BAR0,
> VQ_NOTIFY_BAR1,
> VQ_NOTIFY_BAR2,
> VQ_NOTIFY_BAR3,
> VQ_NOTIFY_BAR4,
> VQ_NOTIFY_BAR5,
> VQ_NOTIFY_FIXED_IOPORT,
> }
>
> And then to notify a vq we write a given data at given offset
> or into a given port for VQ_NOTIFY_FIXED_IOPORT.
>
> Only point 1 is really important for me though, I can be
> flexible on the rest of it.
So the minimal change on top of this patch, would be adding a FIXED
option to BIR and reporting data and not just offset for queue_notify
(so it can include device info if we share same address between
devices).
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
