On 28 May 2018 at 07:11, Philippe Mathieu-Daudé <f4...@amsat.org> wrote: > The create_unimplemented_device() function is very useful to > register unimplemented devices to the SysBus 'absolute' address. > > Some devices are modelled as container of memory subregions, > the subregions being mmio-mapped relatively to the container > base address. > > With these container devices, the current create_unimplemented_device() > does not work. > Add a function to remove the current limitation, and allow > containerized devices to use the helpful UnimplementedDevice. > > Signed-off-by: Philippe Mathieu-Daudé <f4...@amsat.org> > --- > I don't like the function name much, I first tried with > create_unimplemented_device_mr() :( Would it be cleaner renaming current > create_unimplemented_device() -> create_unimplemented_sysbus_device()? > > Also, is this useful to have this code inlined?
My experience has been that devices that want to map the unimplemented device into some other container probably also want to do in-place initialization rather than using qdev_create() -- eg hw/arm/iotkit.c which directly creates TYPE_UNIMPLEMENTED_DEVICEs which are embedded in its state struct. It was always the intention that you can create a TYPE_UNIMPLEMENTED_DEVICE directly; create_unimplemented_device() is just a convenience wrapper. I'm not opposed to having another convenience wrapper here; but I'm not sure this is the right API for one (ie if it would be used in very many places). thanks -- PMM
