On Mon, Mar 14, 2016 at 07:58:23PM +0100, Jan Kiszka wrote:
> On 2016-03-14 19:52, Marcel Apfelbaum wrote:
> > On 03/12/2016 06:13 PM, Aviv B.D. wrote:
> >> From: "Aviv B.D." <bd.a...@gmail.com <mailto:bd.a...@gmail.com>>
> >>
> >> * Fix bug that prevent qemu from starting up when vIOMMU and VFIO
> >> device are present.
> >> * Advertise Cache Mode capability in iommu cap register.
>
> For the final version: Please keep that feature optional, for the sake
> of emulation accuracy (no modern hw exposes it any more). Maybe turn it
> one once a vfio device is in the scope of the IOMMU?
That would be hard to implement: VFIO supports hotplug and there's no way to
change this on the fly.
I would say
- make the feature an optional flag
- deny adding a VFIO device if the flag is not set
> >> * Register every VFIO device with IOMMU state.
> >> * On page cache invalidation in vIOMMU, check if the domain belong to
> >> VFIO device and mirror the guest requests to host.
> >>
> >> Not working (Yet!):
> >> * Tested only with network interface card (ixgbevf) and
> >> intel_iommu=strict in guest's kernel command line.
> >> * Lock up under high load.
> >> * Errors on guest poweroff.
> >> * High relative latency compare to VFIO without IOMMU.
> >
> > Adding (possibly) interested developers to the thread.
>
> Thanks,
> Jan
>
> --
> Siemens AG, Corporate Technology, CT RDA ITP SES-DE
> Corporate Competence Center Embedded Linux
