[PATCH v9 07/12] ARM/ARM64: KVM: Emulate PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET

[Marc Zyngier]

On Tue, Apr 15 2014 at 12:26:06 pm BST, Anup Patel <anup at brainfault.org> wrote:
> On Tue, Apr 15, 2014 at 4:04 PM, Marc Zyngier <marc.zyngier at arm.com> wrote:
>> On Tue, Apr 15 2014 at  7:14:10 am BST, Anup Patel <anup.patel at linaro.org> wrote:
>>> The PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET functions are system-level
>>> functions hence cannot be fully emulated by in-kernel PSCI emulation code.
>>>
>>> To tackle this, we forward PSCI v0.2 SYSTEM_OFF and SYSTEM_RESET function
>>> calls from vcpu to user space (i.e. QEMU or KVMTOOL) via kvm_run structure
>>> using KVM_EXIT_SYSTEM_EVENT exit reasons.
>>>
>>> Signed-off-by: Anup Patel <anup.patel at linaro.org>
>>> Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar at linaro.org>
>>> Reviewed-by: Christoffer Dall <christoffer.dall at linaro.org>
>>> ---
>>>  arch/arm/kvm/psci.c |   32 +++++++++++++++++++++++++++++---
>>>  1 file changed, 29 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/arm/kvm/psci.c b/arch/arm/kvm/psci.c
>>> index 14e6fa6..b964aa4 100644
>>> --- a/arch/arm/kvm/psci.c
>>> +++ b/arch/arm/kvm/psci.c
>>> @@ -85,6 +85,23 @@ static unsigned long kvm_psci_vcpu_on(struct kvm_vcpu *source_vcpu)
>>>       return PSCI_RET_SUCCESS;
>>>  }
>>>
>>> +static inline void kvm_prepare_system_event(struct kvm_vcpu *vcpu, u32 type)
>>
>> Loose the "inline". This is not performance critical, and the compiler
>> does a pretty good job doing that for you.
>
> OK, I will drop the "inline" attribute.
>
>>
>>> +{
>>> +     memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
>>> +     vcpu->run->system_event.type = type;
>>> +     vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
>>> +}
>>> +
>>> +static void kvm_psci_system_off(struct kvm_vcpu *vcpu)
>>> +{
>>> +     kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_SHUTDOWN);
>>> +}
>>> +
>>> +static void kvm_psci_system_reset(struct kvm_vcpu *vcpu)
>>> +{
>>> +     kvm_prepare_system_event(vcpu, KVM_SYSTEM_EVENT_RESET);
>>> +}
>>> +
>>>  int kvm_psci_version(struct kvm_vcpu *vcpu)
>>>  {
>>>       if (test_bit(KVM_ARM_VCPU_PSCI_0_2, vcpu->arch.features))
>>> @@ -95,6 +112,7 @@ int kvm_psci_version(struct kvm_vcpu *vcpu)
>>>
>>>  static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
>>>  {
>>> +     int ret = 1;
>>>       unsigned long psci_fn = *vcpu_reg(vcpu, 0) & ~((u32) 0);
>>>       unsigned long val;
>>>
>>> @@ -114,13 +132,21 @@ static int kvm_psci_0_2_call(struct kvm_vcpu *vcpu)
>>>       case PSCI_0_2_FN64_CPU_ON:
>>>               val = kvm_psci_vcpu_on(vcpu);
>>>               break;
>>> +     case PSCI_0_2_FN_SYSTEM_OFF:
>>> +             kvm_psci_system_off(vcpu);
>>> +             val = PSCI_RET_SUCCESS;
>>> +             ret = 0;
>>> +             break;
>>> +     case PSCI_0_2_FN_SYSTEM_RESET:
>>> +             kvm_psci_system_reset(vcpu);
>>> +             val = PSCI_RET_SUCCESS;
>>> +             ret = 0;
>>> +             break;
>>
>> What is the significance of setting val to PSCI_RET_SUCCESS here? We're
>> exiting to userspace, so surely only the platform emulation can set
>> that. Am I missing something?
>
> Actually, return value is undefined for SYSTEM_OFF and SYSTEM_RESET
> because these functions are not expected to return. Currently, we are updating
> r0 (or x0) for all PSCI functions.
>
> Are you suggesting that we update r0 (or x0) only when there are no error
> (i.e. ret == 0) ?

What I'm suggesting is that the only valid return value should be
indicative of a failure. If you're coming back to the guest after a
SYSTEM_OFF or a SYSTEM_RESET, it means you've failed to perform the
operation.

So userspace either should report the failure (by putting it in x0), and
there is no need to pre-load x0 with SUCCESS, or you start by storing
FAILURE in x0, before exiting to userspace.

	M.
-- 
Jazz is not dead. It just smells funny.