On 7 May 2018 at 15:30, Gerd Hoffmann <kra...@redhat.com> wrote: > The only one which passes global_vmstate == false is "-device > secondary-vga", and that does *not* register vga ioports (the > registers can be accessed via mmio pci bar though).
Looks like virtio-vga does as well. > So, right now global_vmstate == "registers vga ioports", and > I don't expect that to change. If we want that to remain true I think it's more likely to continue to hold if we (a) document it and (b) give the parameter a name that closer matches its function. But since vga_common_init() doesn't actually do anything related to registering the ioports, it's kind of odd. My assumption reading the code was "this is purely for migration back compat and if we wrote a new VGA device model we should have it pass 'false', whether it has IO ports or not. thanks -- PMM
