On 18 October 2018 at 14:19, Philippe Mathieu-Daudé <phi...@redhat.com> wrote:
> On 18/10/2018 15:08, Peter Maydell wrote:
>> This field is the parent-type for the QOM object, so I don't
>> think it makes sense for it to be a union. Any one QOM object
>> should have a single distinct parent type. (There are other
>> examples of "some more or less similar device has ISA and
>> SysBus or PCI and SysBus variations" which should provide
>> some models to copy from.
> So to avoid union, I have to use:
>
> typedef struct PVPanicCommonState {
> MemoryRegion mr;
> uint16_t ioport;
> } PVPanicCommonState;
>
> typedef struct PVPanicISAState {
> /*< private >*/
> ISADevice isadev;
>
> /*< public >*/
> PVPanicCommonState common;
> } PVPanicISAState;
>
> #define PVPANIC_ISA(obj) \
> OBJECT_CHECK(PVPanicISAState, (obj), TYPE_PVPANIC)
>
> typedef struct PVPanicMMIOState {
> /*< private >*/
> SysBusDevice busdev;
>
> /*< public >*/
> PVPanicCommonState common;
> } PVPanicMMIOState;
>
> #define PVPANIC_MMIO(obj) \
> OBJECT_CHECK(PVPanicMMIOState, (obj), TYPE_PVPANIC)
>
Something like that, I think. Basically you have two
distinct types, an ISA PVPanic and a SysBus PVPanic,
which have straightforward inheritance (the ISA PVPanic
is-a ISADevice, the SysBus PVPanic is-a SysBusdevice).
Both have-a PVPanic common state.
See eg hw/display/sm501.c, or hw/net/ne2000{,-isa}.c.
thanks
-- PMM
