Re: [PATCH v2 16/16] ARM64: XEN: Initialize Xen specific UEFI runtime services

[Shannon Zhao]

On 2016/1/19 21:03, Mark Rutland wrote:
> On Tue, Jan 19, 2016 at 12:03:59PM +0000, Stefano Stabellini wrote:
> > On Mon, 18 Jan 2016, Mark Rutland wrote:
> > > On Mon, Jan 18, 2016 at 05:45:24PM +0000, Stefano Stabellini wrote:
> > > > On Mon, 18 Jan 2016, Mark Rutland wrote:
> > > > > On Fri, Jan 15, 2016 at 02:55:29PM +0800, Shannon Zhao wrote:
> > > > > > +void __init xen_efi_runtime_setup(void)
> > > > > > +{
> > > > > > +	efi.get_time                 = xen_efi_get_time;
> > > > > > +	efi.set_time                 = xen_efi_set_time;
> > > > > > +	efi.get_wakeup_time          = xen_efi_get_wakeup_time;
> > > > > > +	efi.set_wakeup_time          = xen_efi_set_wakeup_time;
> > > > > > +	efi.get_variable             = xen_efi_get_variable;
> > > > > > +	efi.get_next_variable        = xen_efi_get_next_variable;
> > > > > > +	efi.set_variable             = xen_efi_set_variable;
> > > > > > +	efi.query_variable_info      = xen_efi_query_variable_info;
> > > > > > +	efi.update_capsule           = xen_efi_update_capsule;
> > > > > > +	efi.query_capsule_caps       = xen_efi_query_capsule_caps;
> > > > > > +	efi.get_next_high_mono_count = xen_efi_get_next_high_mono_count;
> > > > > > +	efi.reset_system             = NULL;
> > > > > > +}
> > > > >
> > > > > How do capsules work in the absence of an EFI system reset?
> > > >
> > > > Actually I don't think that capsules are available in Xen on ARM64 yet,
> > > > see "TODO - disabled until implemented on ARM" in
> > > > xen/common/efi/runtime.c.
> > > >
> > > > FYI system reset is available, but it is provided via a different
> > > > mechanism (HYPERVISOR_sched_op(xen_restart...)
> > >
> > > Will that trigger Xen to do the right thing to trigger capsule updates
> > > when implemented in Xen? Or do we need a xen_efi_reset_system?
> >
> > On ARM, to reboot the hardware, Xen calls the native PSCI system_reset
> > method. On x86, Xen calls efi_reset_system on EFI systems, and has
> > several fall backs if that doesn't work as expected (see
> > xen/arch/x86/shutdown.c:machine_restart).
> >
> > But on a second look it doesn't look like that the capsule hypercalls
> > are implemented correctly even on x86 (there is an "XXX fall through for
> > now" comment in the code). I guess they are not available on Xen at all
> > unfortunately.
>
> That is incredibly unfortunate. It effectively renders the firmware
> non-updateable when using Xen.
>
> > > Does that override PSCI?
> >
> > It does not, HYPERVISOR_sched_op(xen_restart,) is in addition to it. It
> > ends up calling the same function within Xen as PSCI system_reset.
>
> I meant within Dom0.
>
> Presumably Dom0 calls HYPERVISOR_sched_op(xen_restart,), and doesn't
> ever call PSCI SYSTEM_RESET?
I think executing reset in Dom0 will reset not only Dom0 but also the 
Xen hypervisor, right?

> > > In machine_restart we try efi_reboot first specifically to allow for
> > > capsule updates. Similarly drivers/firmware/efi/reboot.c registers
> > > efi_power_off late in order to override anything else, though that's
> > > best-effort at present.
> >
> > That's very interesting. I think that Xen on ARM should follow what
> > Linux does and what Xen already does on x86 and try efi_reset_system
> > first on efi systems.
>
> I would agree.
>
> Thanks,
> Mark.

--
Shannon