On 31/08/2018 14:52, Auger Eric wrote: > Do we agree here we can get rid of the struct device * parameter? That's fine by me, in my opinion the bind operation should only be on the domain, like map/unmap. For the invalidation however, I think we need to keep the device as an optional parameter, because the guest may want to invalidate the ATC (DTLB) of a single device. Thanks, Jean