On Wed, Dec 05, 2018 at 05:50:23PM +0800, Hongbo Zhang wrote:
> > > +static
> > > +void sbsa_ref_machine_done(Notifier *notifier, void *data)
> > > +{
> > > + VirtMachineState *vms = container_of(notifier, VirtMachineState,
> > > + machine_done);
> > > + ARMCPU *cpu = ARM_CPU(first_cpu);
> > > + struct arm_boot_info *info = &vms->bootinfo;
> > > + AddressSpace *as = arm_boot_address_space(cpu, info);
> > > +
> > > + if (arm_load_dtb(info->dtb_start, info, info->dtb_limit, as) < 0) {
> > > + exit(1);
> > > + }
> > > +}
> >
> > The virt board needs a machine-done notifier because it has
> > to add extra things to the DTB here. You don't, so you don't
> > need one. Don't set bootinfo.skip_dtb_autoload to true, and
> > then the boot.c code will do the arm_load_dtb() call for you.
> >
> After test and check, I think we still need this machine_done callback
> to call arm_load_dtb().
> If only kernel loaded via -kernel but without any firmware, it should
> work to let arm_load_kernel() call the arm_load_dtb(), while in our
> case, we have have firmware loaded but no kernel, so the
> arm_load_kernel() returns before calling arm_load_dtb(), that is,
> firmware runs and there is no chance to call arm_load_dtb(), then we
> get error message and qemu quits.
> Moving arm_load_dtb() to earlier place in arm_load_kernel() cannot fix
> this issue either.
I don't see the value in using -kernel on the SBSA machine.
If it causes complexity, can we drop the functionality?
Regards,
Leif
