On Tue, May 20, 2014 at 08:37:29AM +1000, Benjamin Herrenschmidt wrote:

On Mon, 2014-05-19 at 14:46 +0200, Alexander Graf wrote:

quoted

I don't see the point of VFIO knowing about guest addresses. They are 
not unique across a system and the whole idea that a VFIO device has to 
be owned by a guest is also pretty dubious.

I suppose what you really care about here is just a token for a specific 
device? But why do you need one where we don't have tokens yet?

I think this is going to be needed when doing in-kernel acceleration
of some of the RTAS calls, but yes, Gavin, why do we need that now ?

Thanks for your comments :-)

For now, the newly introduced IOCTL command (VFIO_EEH_INFO) is handled by
VFIO-Container fd, which can't identify IOMMU group (PE) or VFIO-PCI-dev
naturally because one VFIO-container can have multiple IOMMU groups docked
there. I'm going to let VFIO-PCI-dev fd handle the IOCTL command because
VFIO-PCI-dev fd can identify the device and the corresponding IOMMU group
(PE). With that, the address mapping stuff isn't necessary.

Thanks,
Gavin