Re: [Qemu-devel] [PATCH 3/5] vring: use hostmem's RAM safe api

[Stefan Hajnoczi]

On Fri, Apr 12, 2013 at 12:49:45PM +0800, liu ping fan wrote:
> On Thu, Apr 11, 2013 at 6:15 PM, Stefan Hajnoczi <address@hidden> wrote:
> > On Mon, Apr 01, 2013 at 04:20:32PM +0800, Liu Ping Fan wrote:
> >> @@ -51,7 +50,7 @@ bool vring_setup(Vring *vring, VirtIODevice *vdev, int n)
> >>
> >>  void vring_teardown(Vring *vring)
> >>  {
> >> -    hostmem_finalize(&vring->hostmem);
> >> +    memory_region_unref(vring->vring_mr);
> >>  }
> >
> > dataplane keeps a reference to the vring.  This prevents memory hot
> > unplug while the device is up.  If this is a problem we'll have to
> > reduce the lifespan of the vring mapping.
> >
> hot unplug is rare case, maybe we can ignore the delay of device's finalize.

I thought about the vring more and I think we can keep the current
behavior.  dataplane keeps the vring up when the device is active.

When the guest kernel/device driver resets the device then we unmap the
vring.  It's reasonable that the guest needs to release the virtio
device before hot unplugging memory used for the vring :).

Therefore, in practice this behavior should be fine.

Stefan