Re: [PATCH 00/11] hw/arm: Define machines as generic QOM types

[Pierrick Bouvier]

On 4/18/25 11:48, BALATON Zoltan wrote:
> On Fri, 18 Apr 2025, Philippe Mathieu-Daudé wrote:
> > On 18/4/25 18:33, Pierrick Bouvier wrote:
> > > On 4/18/25 01:53, BALATON Zoltan wrote:
> > > > On Fri, 18 Apr 2025, Philippe Mathieu-Daudé wrote:
> > > > > While DEFINE_MACHINE() is a succinct macro, it doesn't
> > > > > allow registering QOM interfaces to the defined machine.
> > > > > Convert to the generic DEFINE_TYPES() in preparation to
> > > > > register interfaces.
> > > > >
> > > > > Philippe Mathieu-Daudé (11):
> > > > >    hw/core/null-machine: Define machine as generic QOM type
> > > > >    hw/arm/bananapi: Define machine as generic QOM type
> > > > >    hw/arm/cubieboard: Define machine as generic QOM type
> > > > >    hw/arm/digic: Define machine as generic QOM type
> > > > >    hw/arm/imx: Define machines as generic QOM types
> > > > >    hw/arm/integratorcp: Define machine as generic QOM type
> > > > >    hw/arm/kzm: Define machine as generic QOM type
> > > > >    hw/arm/msf2: Define machine as generic QOM type
> > > > >    hw/arm/musicpal: Define machine as generic QOM type
> > > > >    hw/arm/orangepi: Define machine as generic QOM type
> > > > >    hw/arm/stm32: Define machines as generic QOM types
> > > > >
> > > > > hw/arm/bananapi_m2u.c      | 13 +++++++++++--
> > > > > hw/arm/cubieboard.c        | 13 +++++++++++--
> > > > > hw/arm/digic_boards.c      | 14 ++++++++++++--
> > > > > hw/arm/imx25_pdk.c         | 14 ++++++++++++--
> > > > > hw/arm/imx8mp-evk.c        | 15 +++++++++++++--
> > > > > hw/arm/integratorcp.c      | 16 +++++++++++++---
> > > > > hw/arm/kzm.c               | 14 ++++++++++++--
> > > > > hw/arm/mcimx6ul-evk.c      | 15 +++++++++++++--
> > > > > hw/arm/mcimx7d-sabre.c     | 15 +++++++++++++--
> > > > > hw/arm/msf2-som.c          | 13 +++++++++++--
> > > > > hw/arm/musicpal.c          | 16 +++++++++++++---
> > > > > hw/arm/netduino2.c         | 13 +++++++++++--
> > > > > hw/arm/netduinoplus2.c     | 13 +++++++++++--
> > > > > hw/arm/olimex-stm32-h405.c | 13 +++++++++++--
> > > > > hw/arm/orangepi.c          | 13 +++++++++++--
> > > > > hw/arm/sabrelite.c         | 14 ++++++++++++--
> > > > > hw/arm/stm32vldiscovery.c  | 13 +++++++++++--
> > > > > hw/core/null-machine.c     | 14 ++++++++++++--
> > > > > 18 files changed, 213 insertions(+), 38 deletions(-)
> > > >
> > > > This is much longer and exposing boiler plate code. Is it possible instead
> > > > to change DEFINE_MACHINE or add another similar macro that allows
> > > > specifying more details such as class state type and interfaces like we
> > > > already have for OBJECT_DEFINE macros to keep the boiler plate code hidden
> > > > and not bring it back?
> > >
> > > We can eventually modify DEFINE_MACHINES, to take an additional interfaces
> > > parameter, and replace all call sites, with an empty list for all boards
> > > out of hw/arm.
> > >
> > > As long as we avoid something like:
> > > DEFINE_MACHINES_WITH_INTERFACE_1(...)
> > > DEFINE_MACHINES_WITH_INTERFACE_2(...)
> > > DEFINE_MACHINES_WITH_INTERFACE_3(...)
> > > I'm ok with keeping the macro.
> > >
> > > Would that work for you folks?
> >
> > But then we'll want DEFINE_PPC32_MACHINE() ->
> > DEFINE_MACHINES_WITH_INTERFACE_1() etc...
> >
> > We want to eventually use declarative file to structure most of the
> > machine boiler plate code. Maybe being momentarily verbose is
> > acceptable...
>
> Moments in QEMU can last years... I was thinking about similar to
> OBJECT_DEFINE_SIMPLE_TYPE_WITH_INTERFACES. Would something like that be
> possible for DEFINE_MACHINE too?

We can just add the list of interfaces the machine is implementing as an 
additional parameter to DEFINE_MACHINE, and this solves the problem 
without adding any boilerplate.

Ultimately, we'll need to tag all machines to mention targets supported, 
so this information has to be written somewhere.

> Regards,
> BALATON Zoltan