On Tue, 2 Apr 2019 at 09:46, Like Xu <like...@linux.intel.com> wrote:
>
> On 2019/4/2 7:38, Eduardo Habkost wrote:
> > On Mon, Apr 01, 2019 at 10:56:30AM +0800, Like Xu wrote:
> >> On 2019/3/29 17:27, Alex Bennée wrote:
> > [...]
> >>>> @@ -1713,6 +1717,9 @@ static void cortex_a9_initfn(Object *obj)
> >>>> #ifndef CONFIG_USER_ONLY
> >>>> static uint64_t a15_l2ctlr_read(CPUARMState *env, const ARMCPRegInfo
> >>>> *ri)
> >>>> {
> >>>> + MachineState *ms = MACHINE(qdev_get_machine());
> >>>
> >>> How expensive is qdev_get_machine? This could potentially be a
> >>> performance issue if this register is read a lot.
> >>
> >> It may not bother us cause the MachineState object would be feteched from
> >> qemu QOM at its first requested time and save it as static variable for
> >> incoming requests.
> >
> > We already have a current_machine variable declared in
> > include/hw/boards.h. We normally avoid using it and use
> > MACHINE(qdev_get_machine()) instead, but I never understood why.
> >
> Which one do you prefer to access smp machine properties?
My suggestion would be that we use qdev_get_machine(). I think
it would be nice to make the remaining dozen or so uses of
the global current_machine outside vl.c use qdev_get_machine()
instead, and then make current_machine local to vl.c instead
of global.
thanks
-- PMM
