On 06/07/2018 18:25, Michael S. Tsirkin wrote: > On Fri, Jul 06, 2018 at 05:55:23PM +0200, Paolo Bonzini wrote: >> On 06/07/2018 03:51, Peter Xu wrote: >>> >>> A question about memory region auto destruction (which might not >>> related to this patch): I see that we have object_property_add_child() >>> in memory_region_do_init() to achieve the auto destruction but only if >>> the "name" of memory region is specified. Could we just do that >>> unconditionally (though we might of course need to generate some of >>> the names), or is there a reason not to do so? >> >> I'm not sure actually if there are still regions without a name... >> >> Paolo > > Answer to Peter's question would be a yes then? > > With all the autodestruct I'm unsure when is calling vmstate_unregister_ram > appropriate. > Is it necessary to invoke that from pci any longer? >
I think vmstate_unregister_ram is not necessary at all. This patch, or Alex's suggestion, are smaller changes in that direction---more suitable as we're closer to the release. Paolo