[Qemu-devel] [RFC] target-arm: do not change CNTVOFF_EL2

[Qemu-devel] [RFC] target-arm: do not change CNTVOFF_EL2

[hw.claudio]

From: Claudio Fontana <claudio.fontana@huawei.com>

after a vm is created, the cntvoff_el2 is already calculated
for us by kvm, do not attempt to change it.

Signed-off-by: Claudio Fontana <claudio.fontana@huawei.com>
---
 target-arm/kvm.c | 5 +++++
 1 file changed, 5 insertions(+)

This is just an RFC to show where QEMU is causing the change
of CNTVOFF_EL2. This fixes the issue for me, but maybe this
is not the right way to do it?

Thanks,

Claudio

diff --git a/target-arm/kvm.c b/target-arm/kvm.c
index 548bfd7..c5ad580 100644
--- a/target-arm/kvm.c
+++ b/target-arm/kvm.c
@@ -421,6 +421,11 @@ bool write_list_to_kvmstate(ARMCPU *cpu)
         uint32_t v32;
         int ret;
 
+	if (regidx == KVM_REG_ARM_TIMER_CNT) {
+            /* do not cause a change of CNTVOFF_EL2. */
+            continue;
+	}
+
         r.id = regidx;
         switch (regidx & KVM_REG_SIZE_MASK) {
         case KVM_REG_SIZE_U32:
-- 
1.8.5.3

Re: [Qemu-devel] [RFC] target-arm: do not change CNTVOFF_EL2

[Peter Maydell]

On 9 July 2015 at 15:38,  <hw.claudio@gmail.com> wrote:
> From: Claudio Fontana <claudio.fontana@huawei.com>
>
> after a vm is created, the cntvoff_el2 is already calculated
> for us by kvm, do not attempt to change it.
>
> Signed-off-by: Claudio Fontana <claudio.fontana@huawei.com>
> ---
>  target-arm/kvm.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> This is just an RFC to show where QEMU is causing the change
> of CNTVOFF_EL2. This fixes the issue for me, but maybe this
> is not the right way to do it?

Yeah, pretty sure it isn't -- see the discussion in
the kvmarm thread. We do want to fix this on the
QEMU side, but not like this...

-- PMM