Re: [Xen-devel] Xen on ARM basic questions

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Thu, 2013-11-28 at 12:44 +0000, Stefano Stabellini wrote:
> On Fri, 22 Nov 2013, Vaibhav Bedia wrote:
> > Hi Ian,
> > 
> > On Fri, Nov 22, 2013 at 10:22 AM, Ian Campbell <Ian.Campbell@xxxxxxxxxx> 
> > wrote:
> > > On Fri, 2013-11-22 at 10:05 -0500, Vaibhav Bedia wrote:
> > >> >>  After going through the recent talk from Stefano [1] i got
> > >> >> the impression getting network and SATA requires SWIOTLB support. Is
> > >> >> that correct?
> > >> >
> > >> > The SWIOTLB or an IOMMU/SMMU is necessary for dom0 to drive the 
> > >> > physical
> > >> > devices and provide the virtual interfaces to the guests.
> > >> >
> > >> > Device passthrough will only be supported with an IOMMU/SMMU.
> > >> >
> > >>
> > >> So on SoCs like OMAP5 which don't have an IOMMU device passthrough won't 
> > >> be
> > >> supported? Or it can be done via SWIOTLB if someone works on it?
> > >
> > > I'm not in favour of supporting this. It's a big can of worms wrt
> > > security.
> > >
> > 
> > Ok. I'll leave it to the experts then :)
> 
> Just for clarity: the big can of worms is allowing device passthrough
> without an IOMMU. (Except maybe for some ad-hoc static configurations
> that might be desirable on embedded platform.)
> Using the swiotlb for dom0 is OK.

Yes, sorry, I latched onto the "device passthrough" bit of the question
as being a question about giving devices to domU. swiotlb is fine for
dom0 and OMAP5 is a platform which could be supportable, if you don't
care about giving physical devices to guests.

In fact, we have omap5 platform code already from Chen Baozi. I don't
know which platform he actually has though.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel