Re: [PATCH v13 34/48] arm64: RMI: support RSI_HOST_CALL

[Suzuki K Poulose <suzuki.poulose@arm.com>]

On 18/03/2026 15:53, Steven Price wrote:
> From: Joey Gouly <joey.gouly@arm.com>
> 
> Forward RSI_HOST_CALLS to KVM's HVC handler.

Minor nit: Please could we add a line or two, explaining what 
RSI_HOST_CALL is ? e.g.:

Realm's can talk to the hypervisor using the RSI_HOST_CALL, which
the RMM forwards to the KVM. Handle them as regular hypercalls.

Suzuki

> 
> Signed-off-by: Joey Gouly <joey.gouly@arm.com>
> Signed-off-by: Steven Price <steven.price@arm.com>
> Reviewed-by: Gavin Shan <gshan@redhat.com>
> ---
> Changes since v7:
>   * Avoid turning a negative return from kvm_smccc_call_handler() into a
>     error response to the guest. Instead propogate the error back to user
>     space.
> 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/rmi-exit.c | 15 +++++++++++++++
>   1 file changed, 15 insertions(+)
> 
> diff --git a/arch/arm64/kvm/rmi-exit.c b/arch/arm64/kvm/rmi-exit.c
> index b4843f094615..7eff6967530c 100644
> --- a/arch/arm64/kvm/rmi-exit.c
> +++ b/arch/arm64/kvm/rmi-exit.c
> @@ -116,6 +116,19 @@ static int rec_exit_ripas_change(struct kvm_vcpu *vcpu)
>   	return -EFAULT;
>   }
>   
> +static int rec_exit_host_call(struct kvm_vcpu *vcpu)
> +{
> +	int 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]);
> +
> +	return kvm_smccc_call_handler(vcpu);
> +}
> +
>   static void update_arch_timer_irq_lines(struct kvm_vcpu *vcpu)
>   {
>   	struct realm_rec *rec = &vcpu->arch.rec;
> @@ -183,6 +196,8 @@ 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);
>   	}

Probably we should move the RMI_EXIT_HOST_CALL case addition in 
kvm_rec_pre_enter() to this hunk to keep all in one place ?

Otherwise, looks good to me.

Suzuki


>   
>   	kvm_pr_unimpl("Unsupported exit reason: %u\n",