Jens Nerche wrote:
>
> > We are going to try to add a 'pass-through' capability. There are
> > some issues surrounding DMA, PnP, etc that need to be worked out,
> > depending on the device in question. But this is a very useful
> > feature which many people have asked for, so it will likely get
> > a lot of coding attention.
> Hu, DMA is a very hard problem. Once the guest can tell a device to use
> DMA, no host memory is safe anymore. Think, 'pass-through' will work only
> with very simple devices.
If we virtualize the DMA IO from the guest, and emulate the
DMA controller, then it is possible to intelligently pass on
DMA requests to the actual DMA hardware. This of course, has
to be managed with the host so there are no conflicts.
-Kevin
