diff mbox

aarch64: advertise the GIC system register interface

Message ID alpine.DEB.2.10.1710311147570.13766@sstabellini-ThinkPad-X260 (mailing list archive)
State New, archived
Headers show

Commit Message

Stefano Stabellini Oct. 31, 2017, 6:51 p.m. UTC 
On Tue, 31 Oct 2017, Peter Maydell wrote:
> On 31 October 2017 at 17:01, Stefano Stabellini <sstabellini@kernel.org> wrote:
> > Fixing QEMU is harder than I expected. Would it be possible to update
> > id_aa64pfr0 at CPU reset time? Like cpu->id_aa64pfr0 |= 0x01000000; ?
> 
> At that point we've already called register_cp_regs_for_features(),
> which is where we read cpu->id_aa64pfr0 when we're creating the
> cpreg. So if you change it after that it's too late. But that
> function is called at CPU realize time, which is before we've
> created the GIC object...

What about something along the lines of

Comments

Peter Maydell Oct. 31, 2017, 7:24 p.m. UTC | #1 
On 31 October 2017 at 18:51, Stefano Stabellini <sstabellini@kernel.org> wrote:
> On Tue, 31 Oct 2017, Peter Maydell wrote:
>> On 31 October 2017 at 17:01, Stefano Stabellini <sstabellini@kernel.org> wrote:
>> > Fixing QEMU is harder than I expected. Would it be possible to update
>> > id_aa64pfr0 at CPU reset time? Like cpu->id_aa64pfr0 |= 0x01000000; ?
>>
>> At that point we've already called register_cp_regs_for_features(),
>> which is where we read cpu->id_aa64pfr0 when we're creating the
>> cpreg. So if you change it after that it's too late. But that
>> function is called at CPU realize time, which is before we've
>> created the GIC object...
>
> What about something along the lines of
>
> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
> index 9e18b41..0851071 100644
> --- a/hw/arm/virt.c
> +++ b/hw/arm/virt.c
> @@ -1401,6 +1400,10 @@ static void machvirt_init(MachineState *machine)
>              object_property_set_link(cpuobj, OBJECT(secure_sysmem),
>                                       "secure-memory", &error_abort);
>          }
> +        if (vms->gic_version == 3) {
> +            ARMCPU *cpu = ARM_CPU(cpuobj);
> +            cpu->id_aa64pfr0 |= 0x01000000;
> +        }
>
>          object_property_set_bool(cpuobj, true, "realized", NULL);
>          object_unref(cpuobj);

Definitely not -- the virt board should not be poking about inside the
internals of the CPU object.

The slightly cleaned up version of this idea is that you give the
CPU an object property for "claim the GICv3 registers exist in the
ID registers" and have virt.c set it. That feels like an ugly
workaround for something we really ought not to have to tell the
board about at all, though.

thanks
-- PMM
diff mbox

Patch

diff --git a/hw/arm/virt.c b/hw/arm/virt.c
index 9e18b41..0851071 100644
--- a/hw/arm/virt.c
+++ b/hw/arm/virt.c
@@ -1401,6 +1400,10 @@  static void machvirt_init(MachineState *machine)
             object_property_set_link(cpuobj, OBJECT(secure_sysmem),
                                      "secure-memory", &error_abort);
         }
+        if (vms->gic_version == 3) {
+            ARMCPU *cpu = ARM_CPU(cpuobj);
+            cpu->id_aa64pfr0 |= 0x01000000;
+        }
 
         object_property_set_bool(cpuobj, true, "realized", NULL);
         object_unref(cpuobj);