Re: [PATCH] docs: Add '-device intel-iommu' entry

[Peter Xu]

On Wed, Jul 07, 2021 at 11:29:47AM +0200, Eric Auger wrote:
> Hi Peter,

Hi, Eric,

> 
> On 6/11/21 8:55 PM, Peter Xu wrote:
> > The parameters of intel-iommu device are non-trivial to understand.  Add an
> > entry for it so that people can reference to it when using.
> > 
> > There're actually a few more options there, but I hide them explicitly 
> > because
> > they shouldn't be used by normal QEMU users.
> > 
> > Cc: Chao Yang <chayang@redhat.com>
> > Cc: Lei Yang <leiyang@redhat.com>
> > Cc: Jing Zhao <jinzhao@redhat.com>
> > Cc: Jason Wang <jasowang@redhat.com>
> > Cc: Michael S. Tsirkin <mst@redhat.com>
> > Cc: Alex Williamson <alex.williamson@redhat.com>
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> > ---
> >  qemu-options.hx | 32 ++++++++++++++++++++++++++++++++
> >  1 file changed, 32 insertions(+)
> > 
> > diff --git a/qemu-options.hx b/qemu-options.hx
> > index 14258784b3a..4bb04243907 100644
> > --- a/qemu-options.hx
> > +++ b/qemu-options.hx
> > @@ -926,6 +926,38 @@ SRST
> >  
> >  ``-device pci-ipmi-bt,bmc=id``
> >      Like the KCS interface, but defines a BT interface on the PCI bus.
> > +
> > +``-device intel-iommu[,option=...]``
> > +    This is only supported by ``-machine q35``, which will enable Intel 
> > VT-d
> > +    emulation within the guest.  It supports below options:
> > +
> > +    ``intremap=on|off`` (default: auto)
> > +        This enables interrupt remapping feature in the guest.  It's 
> > required
> > +        to enable complete x2apic.  Currently it only supports kvm
> > +        kernel-irqchip modes ``off`` or ``split``.  Full kernel-irqchip is 
> > not
> > +        yet supported.
> maybe explain how the default is chosen (based on kernel-irqhcip mode?)

Will do.

> > +
> > +    ``caching-mode=on|off`` (default: off)
> > +        This enables caching mode for the VT-d emulated device.  When
> > +        caching-mode is enabled, each guest DMA buffer mapping will 
> > generate an
> > +        IOTLB invalidation from the guest IOMMU driver to the vIOMMU 
> > device in
> > +        a synchronous way.  It is required for ``-device vfio-pci`` to work
> > +        with the VT-d device, because host assigned devices requires to 
> > setup
> > +        the DMA mapping on the host before guest DMA starts.
> > +
> > +    ``device-iotlb=on|off`` (default: off)
> > +        This enables device-iotlb capability for the emulated VT-d device. 
> >  So
> > +        far virtio/vhost should be the only real user for this parameter,
> I thought if was needed for vhost. What is the usage for virtio?

I used "virtio/vhost" because vhost is only one type of virtio backend, so
normally they should be used in pair.  Also, the "ats=on" parameter should be
an option to virtio frontend rather than vhost, hence I mentioned both.

> > +        paired with ats=on configured for the device.
> What is not obvious to me is whether there is a kind of diagnosis
> whether device-iotlb and caching mode settings are consistent with
> vfio/vhost usage. Does qemu exit if there is inconsistency?

Caching mode check is done in vtd_machine_done_notify_one().

I think there's no such check for vhost, however iirc vhost should work
with/without ats=on; I did't check deeper.

> > +
> > +    ``aw-bits=39|48`` (default: 39)
> > +        This decides the address width of IOVA address space.  The address
> > +        space has 39 bits width for 3-level IOMMU page tables, and 48 bits 
> > for
> > +        4-level IOMMU page tables.
> > +
> > +    Please also refer to the wiki page for general scenarios of VT-d
> > +    emulation in QEMU: https://wiki.qemu.org/Features/VT-d.
> > +
> >  ERST
> >  
> >  DEF("name", HAS_ARG, QEMU_OPTION_name,
> >
> Besides, it is quite useful.

Thanks for taking a look!

-- 
Peter Xu