Re: [RFC 20/20] Doc: Add documentation for /dev/iommu

[Jason Gunthorpe]

On Fri, Oct 29, 2021 at 11:15:31AM +1100, David Gibson wrote:

> > +Device must be bound to an iommufd before the attach operation can
> > +be conducted. The binding operation builds the connection between
> > +the devicefd (opened via device-passthrough framework) and IOMMUFD.
> > +IOMMU-protected security context is esbliashed when the binding
> > +operation is completed.
> 
> This can't be quite right.  You can't establish a safe security
> context until all devices in the groun are bound, but you can only
> bind them one at a time.

When any device is bound the entire group is implicitly adopted to
this iommufd and the whole group enters a safe-for-userspace state.

It is symmetrical with the kernel side which is also device focused,
when any struct device is bound to a kernel driver the entire group is
implicitly adopted to kernel mode.

Lu should send a patch series soon that harmonize how this works, it
is a very nice cleanup.

Jason