On Mon, Jun 08, 2015 at 05:55:10PM +0200, Paolo Bonzini wrote: > On 08/06/2015 17:51, Eduardo Habkost wrote: > > > > Having a numa_set_memory_region_node_id(MemoryRegion *mr, int node) API > > would probably be better, and make the discussion about pc_dimm.addr > > moot. But it would require changing > > memory_region_allocate_system_memory() to avoid > > allocate_system_memory_nonnuma() even in the !have_memdevs case. > > This in turn may have complications due to migration. We can do it later.
I agree we can do it later. But what kind of complications do you see? Is the choice of MemoryRegion layout really supposed to affect migration? (Is it just about the single vmstate_register_ram_global() call at allocate_system_memory_nonnuma()?) -- Eduardo
