On Wed, Jun 07, 2023 at 02:54:22PM +0800, Jason Wang wrote: > On Wed, Jun 7, 2023 at 11:06 AM Parav Pandit <[email protected]> wrote: > > > > > > > > > From: Jason Wang <[email protected]> > > > Sent: Tuesday, June 6, 2023 10:27 PM > > > > > > I can't say I like this. I prefer to do meditation on top of a modern > > > device with > > > some lightweight features like _F_LEAGCY_HEADER. I don't see any > > > advantages > > > of a new legacy ABI over _F_LEGACY_HEADER. I don't want to repeat but I > > > don't > > > get any response here > > > > > It is net only and not generic. Not a big blocker. > > It doesnt stop at _F_LEAGCY_HEADER. > > > > > Anything that prevents you from adding things like > > > VIRTIO_NET_F_LEGACY_HEADER? > > It requires hypervisor to mediate 1.x config region and intercept feature > > negotiation including q config and more. > > For config region, do you mean device configuration space or common > config space? > > I don't see why we need to do mediation on neiter of the above. > Hypervisor just need to prepare > > 1) legacy BAR with legacy config and device configuration space and > 2) modern BAR with modern capabilities (common cfg and device cfg) > > For 2) it could be mapped directly to the modern capabilities. For 1) > hypervisor needs to mediate > > > > > Additionally, 1.x device says the mac address is read only, while legacy is > > RW. > > And that requires yet another feature bit negotiation. > > Hypervisor can trap the legacy device configuration space write and > convert it to cvq commands. > > Thanks
Hypervisor really can't access cvq if guest is doing it too - that requires emulating cvq in software. And *that* is going to be broken for workloads such as confidential computing. > > > > The main design goal is to not mediate 1.x specific region and not modify > > 1.x region definitions. > > Which is why we step forward with the AQ. > > > > > https://lore.kernel.org/virtio- > > > dev/[email protected] > > > mail.com/#t --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
