Re: [Qemu-devel] Re: [PATCH 2 of 5] add can_dma/post_dma for direct IO

[Gerd Hoffmann <kraxel@redhat.com>]

Avi Kivity wrote:
> Ian Jackson wrote:
>> I think there is continued value in being able to emulate a guest
>> whose physical address space exceeds the virtual address space in the
>> host.  The whole assumption that all guest accessible RAM is mapped at
>> once, contiguously, is I think wrong.
> 
> If Xen has a problem with 64-bit hosts, we can try to
> accommodate it,

Yes, this is the whole point.  Flesh out the DMA api in a way that Xen
can use it.  Large memory guests are only one reason why Xen needs
map/unmap.  The other one is grant table support.

> but to have 32-bit qemu/tcg or qemu/kvm support large
> address spaces is pointless IMO.

You don't have to support that.

I think the DMA api should look like this in the end:

struct qemu_dma_ops {
     map(...);
     xfer(...);
     unmap(...);
};

The qemu/xen implementation would actually map/unmap.

For qemu/tcg and qemu/kvm which has all guest memory permanently mapped
map() would be a guest-physical -> host-virtual address translation and
unmap() would be a nop (or refcount--).

cheers,
  Gerd