Re: [PATCH v6 27/43] arm64: rme: support RSI_HOST_CALL

[Steven Price <steven.price@arm.com>]

On 02/02/2025 06:41, Gavin Shan wrote:
> On 12/13/24 1:55 AM, Steven Price wrote:
>> From: Joey Gouly <joey.gouly@arm.com>
>>
>> Forward RSI_HOST_CALLS to KVM's HVC handler.
>>
>> Signed-off-by: Joey Gouly <joey.gouly@arm.com>
>> Signed-off-by: Steven Price <steven.price@arm.com>
>> ---
>> Changes since v4:
>>   * Setting GPRS is now done by kvm_rec_enter() rather than
>>     rec_exit_host_call() (see previous patch - arm64: RME: Handle realm
>>     enter/exit). This fixes a bug where the registers set by user space
>>     were being ignored.
>> ---
>>   arch/arm64/kvm/rme-exit.c | 22 ++++++++++++++++++++++
>>   1 file changed, 22 insertions(+)
>>
>> diff --git a/arch/arm64/kvm/rme-exit.c b/arch/arm64/kvm/rme-exit.c
>> index 8f0f9ab57f28..b2a367474d74 100644
>> --- a/arch/arm64/kvm/rme-exit.c
>> +++ b/arch/arm64/kvm/rme-exit.c
>> @@ -103,6 +103,26 @@ static int rec_exit_ripas_change(struct kvm_vcpu
>> *vcpu)
>>       return 0;
>>   }
>>   +static int rec_exit_host_call(struct kvm_vcpu *vcpu)
>> +{
>> +    int ret, i;
>> +    struct realm_rec *rec = &vcpu->arch.rec;
>> +
>> +    vcpu->stat.hvc_exit_stat++;
>> +
>> +    for (i = 0; i < REC_RUN_GPRS; i++)
>> +        vcpu_set_reg(vcpu, i, rec->run->exit.gprs[i]);
>> +
>> +    ret = kvm_smccc_call_handler(vcpu);
>> +
>> +    if (ret < 0) {
>> +        vcpu_set_reg(vcpu, 0, ~0UL);
>> +        ret = 1;
>> +    }
>> +
>> +    return ret;
>> +}
>> +
> 
> It seems that the return value from kvm_smccc_call() won't be negative.

Well the comment above kvm_psci_call() explains that the return value
can be negative which would be passed through, so there's definitely a
convention that it could be negative. However...

> Besides,
> the host call requests are currently handled by kvm_psci_call(), which
> isn't
> what we want.

Indeed, we shouldn't be getting PSCI calls this way as the RMM needs to
be involved for proper handling of PSCI.

> So I think a new helper is needed and called in> kvm_smccc_call_handler().
> The new helper simply push the error (NOT_SUPPORTED) to x0. Otherwise, a
> unexpected
> return value will be seen by guest.
> 
> handle_rec_exit
>   rec_exit_host_call
>     kvm_smccc_call_handler


I'm not sure I follow here. Are you saying that we should have separate
handling of HOST_CALLs to SMCCC? That's certainly a possibility, but the
expectation is that HOST_CALL is effectively equivalent to a simple
SMC/HVC call in a normal guest. To be honest a "Realm Host Interface" is
something that we're currently lacking a spec for.

Thanks,
Steve

>>   static void update_arch_timer_irq_lines(struct kvm_vcpu *vcpu)
>>   {
>>       struct realm_rec *rec = &vcpu->arch.rec;
>> @@ -164,6 +184,8 @@ int handle_rec_exit(struct kvm_vcpu *vcpu, int
>> rec_run_ret)
>>           return rec_exit_psci(vcpu);
>>       case RMI_EXIT_RIPAS_CHANGE:
>>           return rec_exit_ripas_change(vcpu);
>> +    case RMI_EXIT_HOST_CALL:
>> +        return rec_exit_host_call(vcpu);
>>       }
>>         kvm_pr_unimpl("Unsupported exit reason: %u\n",
> 
> Thanks,
> Gavin
>