On Tue, Feb 04, 2025 at 06:50:17PM +0100, David Hildenbrand wrote:
> > /*
> > @@ -595,6 +628,7 @@ static const TypeInfo host_memory_backend_info = {
> > .instance_size = sizeof(HostMemoryBackend),
> > .instance_init = host_memory_backend_init,
> > .instance_post_init = host_memory_backend_post_init,
> > + .instance_finalize = host_memory_backend_finalize,
> > .interfaces = (InterfaceInfo[]) {
> > { TYPE_USER_CREATABLE },
> > { }
> > diff --git a/include/system/hostmem.h b/include/system/hostmem.h
> > index 5c21ca55c0..170849e8a4 100644
> > --- a/include/system/hostmem.h
> > +++ b/include/system/hostmem.h
> > @@ -83,6 +83,7 @@ struct HostMemoryBackend {
> > HostMemPolicy policy;
> > MemoryRegion mr;
> > + RAMBlockNotifier ram_notifier;
> > };
>
> Thinking about Peters comment, it would be a nice improvement to have a
> single global memory-backend notifier that looks up the fitting memory
> backend, instead of having one per memory backend.
Yes, this could also avoid O(N**2).
>
> A per-ramblock notifier might also be possible, but that's a bit
> harder/ackward to configure: e.g., the resize callback is passed to
> memory_region_init_resizeable_ram() right now.
Yes, that can be some fuss on code to be touched up. We could avoid
passing that in when create the ramblock, instead we could allow ramblocks
to opt-in on hooks after ramblock created. Maybe we could move resize()
out too like that.
Either way looks good.
--
Peter Xu
