[RFC V2] virtio: Add platform specific DMA API translation for virito devices

[Michael S. Tsirkin]

On Mon, Jun 04, 2018 at 07:48:54PM +1000, Benjamin Herrenschmidt wrote:
> On Mon, 2018-06-04 at 18:57 +1000, David Gibson wrote:
> > 
> > > - First qemu doesn't know that the guest will switch to "secure mode"
> > > in advance. There is no difference between a normal and a secure
> > > partition until the partition does the magic UV call to "enter secure
> > > mode" and qemu doesn't see any of it. So who can set the flag here ?
> > 
> > This seems weird to me.  As a rule HV calls should go through qemu -
> > or be allowed to go directly to KVM *by* qemu.
> 
> It's not an HV call, it's a UV call, qemu won't see it, qemu isn't
> trusted. Now the UV *will* reflect that to the HV via some synthetized
> HV calls, and we *could* have those do a pass by qemu, however, so far,
> our entire design doesn't rely on *any* qemu knowledge whatsoever and
> it would be sad to add it just for that purpose.

It's a temporary work-around. I think that the long-term fix is to
support per-device quirks and have the DMA API DTRT for virtio.

> Additionally, this is rather orthogonal, see my other email, the
> problem we are trying to solve is *not* a qemu problem and it doesn't
> make sense to leak that into qemu.
> 
> >   We generally reserve
> > the latter for hot path things.  Since this isn't a hot path, having
> > the call handled directly by the kernel seems wrong.
> >
> > Unless a "UV call" is something different I don't know about.
> 
> Yes, a UV call goes to the Ultravisor, not the Hypervisor. The
> Hypervisor isn't trusted.
> 
> > > - Second, when using VIRTIO_F_IOMMU_PLATFORM, we also make qemu (or
> > > vhost) go through the emulated MMIO for every access to the guest,
> > > which adds additional overhead.
> > 
> Ben.