Re: [Qemu-devel] [PATCH 27/28] sysbus: apic: ioapic: convert to QEMU Object Model

[Anthony Liguori]

On 01/25/2012 08:23 AM, Jan Kiszka wrote:
> On 2012-01-25 15:00, Anthony Liguori wrote:
> > On 01/25/2012 02:37 AM, Jan Kiszka wrote:
> > > On 2012-01-25 00:03, Anthony Liguori wrote:
> > > > They're exactly the same size (16 lines).  If you embed TypeInfo into
> > > > DeviceTypeInfo, and introduce a Device specific type registration
> > > > function, then you could do:
> > > >
> > > > static DeviceTypeInfo my_device_type_info = {
> > > >       .type.name = TYPE_MY_DEVICE,
> > > >       .type.parent = TYPE_PARENT_DEVICE,
> > > >       .reset = my_device_reset,
> > >
> > > And if you introduce some
> > >
> > > #define TYPE_UNIMPLEMENTED (void *)&dummy_variable
> >
> > (void *) isn't compatible with integers or function pointers (at least
> > not in a portable way).

I don't see how they can be compatible since on some platforms sizeof(void 
(*)(void)) != sizeof(void *).

> The former is true, the latter not. However, if you use pointers in the
> TypeInfo structs consistently (const int * etc.), this issue should be
> solvable as well.

How would you handle PCIDeviceInfo?

> > I'm not being dismissive, I've already spent a lot of time trying to
> > make it work and have convinced myself that it isn't workable.
> >
> > If you can show a mechanism that works, I don't mind scripting a mass
> > conversion.  I'm also fairly confident that a workable solution is going
> > to be better than what we have.
>
> Well, I'm surely not happy having to go through all of the devices
> again, even if scripted (scripts tend to neglect some coding style rules
> e.g.). But if this change is in a hurry and my concern is the only one,
> then let it be.

I wouldn't be happy with it either, but in the absence of a concrete counter 
proposal and given that I've spent a fair amount of time trying to figure this 
out, I strongly believe we won't need to make the change.

I'm happy to be proven wrong though.

Regards,

Anthony Liguori

> Jan