On Fri, Sep 14, 2012 at 03:01:55PM -0400, Jason Baron wrote: > On Sat, Sep 15, 2012 at 02:34:18AM +0900, Isaku Yamahata wrote: > > On Fri, Sep 14, 2012 at 11:23:29AM -0400, Jason Baron wrote: > > > On Sat, Sep 15, 2012 at 12:14:01AM +0900, Isaku Yamahata wrote: > > > > On Thu, Sep 13, 2012 at 04:12:34PM -0400, Jason Baron wrote: > > > > > 2) hotplug > > > > > I've added piix acpi style hotplug to ich9. > > > > > > > > What's the point of this? > > > > Its design is ad-hoc and shpc/pcie hotplug are available. > > > > > > The point was to get to feature parity with piix. I'm not sure how hard > > > pcie hotplug is to implement really. I was thinking that we could > > > replace the piix hotplug style with pcie hotplug, once that was > > > implemented... > > > > shpc/pcie hotplug emulators are already available in qemu. > > hw/shpc.c, hw/pcie.c and hw/xio3130_downstream.c. > > > > thanks, > > -- > > yamahata > > > > ok, yes pcie hotplug seems to work :) I guess I got confused b/c I > thought devices would show up on the host bus. So the host bus in q35 > simply doesn't support hotplug? > > Also, the way I'm doing hotplug is by specifying a 'bus=pcie.n' param. > Without the 'bus=pcie.n' param, i get: > > Bus 'pcie.0' does not support hotplugging > > We probably then need to enhance 'device_add' to find a reasonable > default bus for the device on hotplug? > > Thanks, > > -Jason
Maybe though I am not too worried: hotplug is an advanced enough feature. We will also still need a PCI bus in the system so that legacy PCI devices such as virtio can be added. And that bus needs ACPI for hotplug. -- MST
