Re: [PATCH v5 7/8] hisi_acc_vfio_pci: Add support for VFIO live migration

[Jason Gunthorpe <jgg@nvidia.com>]

On Wed, Feb 23, 2022 at 09:34:43AM -0700, Alex Williamson wrote:
> On Tue, 22 Feb 2022 20:52:51 -0400
> Jason Gunthorpe <jgg@nvidia.com> wrote:
> 
> > On Mon, Feb 21, 2022 at 11:40:42AM +0000, Shameer Kolothum wrote:
> > 
> > > +	/*
> > > +	 * ACC VF dev BAR2 region consists of both functional register space
> > > +	 * and migration control register space. For migration to work, we
> > > +	 * need access to both. Hence, we map the entire BAR2 region here.
> > > +	 * But from a security point of view, we restrict access to the
> > > +	 * migration control space from Guest(Please see mmap/ioctl/read/write
> > > +	 * override functions).
> > > +	 *
> > > +	 * Also the HiSilicon ACC VF devices supported by this driver on
> > > +	 * HiSilicon hardware platforms are integrated end point devices
> > > +	 * and has no capability to perform PCIe P2P.  
> > 
> > If that is the case why not implement the RUNNING_P2P as well as a
> > NOP?
> > 
> > Alex expressed concerned about proliferation of non-P2P devices as it
> > complicates qemu to support mixes
> 
> I read the above as more of a statement about isolation, ie. grouping.
> Given that all DMA from the device is translated by the IOMMU, how is
> it possible that a device can entirely lack p2p support, or even know
> that the target address post-translation is to a peer device rather
> than system memory.  If this is the case, it sounds like a restriction
> of the SMMU not supporting translations that reflect back to the I/O
> bus rather than a feature of the device itself.  Thanks,

This is an interesting point..

Arguably if P2P addresses are invalid in an IOPTE then
pci_p2pdma_distance() should fail and we shouldn't have installed them
into the iommu in the first place.

Jason