Re: [PATCH 2/2] hw/char/serial: Allow migration of the I/O serial device

[Peter Maydell]

On Tue, 7 Jul 2020 at 11:40, Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 03/07/20 20:58, Philippe Mathieu-Daudé wrote:
> > The serial device mapped on the I/O bus hold a migratable
> > SerialState. Keep the same version range from SerialState:
> >
> >  837 const VMStateDescription vmstate_serial = {
> >  838     .name = "serial",
> >  839     .version_id = 3,
> >  840     .minimum_version_id = 2,
> >
> > Fixes: 10315a7089 ("serial: make SerialIO a sysbus device")
> > Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> > ---
> >  hw/char/serial.c | 12 +++++++++++-
> >  1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/hw/char/serial.c b/hw/char/serial.c
> > index 9eebcb27e7..c167b584fb 100644
> > --- a/hw/char/serial.c
> > +++ b/hw/char/serial.c
> > @@ -1002,12 +1002,22 @@ static void serial_io_realize(DeviceState *dev, 
> > Error **errp)
> >      sysbus_init_irq(SYS_BUS_DEVICE(sio), &s->irq);
> >  }
> >
> > +static const VMStateDescription vmstate_serial_io = {
> > +    .name = "serial",
> > +    .version_id = 3,
> > +    .minimum_version_id = 2,
> > +    .fields = (VMStateField[]) {
> > +        VMSTATE_STRUCT(serial, SerialIO, 0, vmstate_serial, SerialState),
> > +        VMSTATE_END_OF_LIST()
> > +    }
> > +};
> > +
> >  static void serial_io_class_init(ObjectClass *klass, void* data)
> >  {
> >      DeviceClass *dc = DEVICE_CLASS(klass);
> >
> >      dc->realize = serial_io_realize;
> > -    /* No dc->vmsd: class has no migratable state */
> > +    dc->vmsd = &vmstate_serial_io;
> >  }
> >
> >  static void serial_io_instance_init(Object *o)
> >
>
> Is there any difference between SerialMM and SerialIO at this point?

SerialIO insists on 1-byte wide accesses; SerialMM allows the
creator of the device to specify the spacing between registers
and the endianness. So I suppose SerialIO is kind of a subset of
SerialMM.

It looks like the only user of TYPE_SERIAL_IO is hw/mips/mipssim.c,
Adding the migration state here would be a forwards migration
compat break anyway, so I think we could just change that
machine to use TYPE_SERIAL_MM and then delete TYPE_SERIAL_IO ?

thanks
-- PMM