[PATCH 2/2] kvm: arm/arm64: implement kvm_arm_[halt, resume]_guest

[Christoffer Dall]

On Mon, Jul 06, 2015 at 02:49:56PM +0200, Eric Auger wrote:
> On halt, the guest is forced to exit and prevented from being
> re-entered. This is synchronous.
> 
> Those two operations will be needed for IRQ forwarding setting.
> 
> Signed-off-by: Eric Auger <eric.auger at linaro.org>
> 
> ---
> RFC v1 -> v2:
> - add __maybe_unused
> 
> RFC:
> - rename the function and this latter becomes static
> - remove __KVM_HAVE_ARCH_HALT_GUEST
> 
> v4 -> v5: add arm64 support
> - also defines __KVM_HAVE_ARCH_HALT_GUEST for arm64
> - add pause field
> ---
>  arch/arm/include/asm/kvm_host.h   |  3 +++
>  arch/arm/kvm/arm.c                | 35 ++++++++++++++++++++++++++++++++---
>  arch/arm64/include/asm/kvm_host.h |  3 +++
>  3 files changed, 38 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
> index 304004d..899ae27 100644
> --- a/arch/arm/include/asm/kvm_host.h
> +++ b/arch/arm/include/asm/kvm_host.h
> @@ -132,6 +132,9 @@ struct kvm_vcpu_arch {
>  	/* vcpu power-off state */
>  	bool power_off;
>  
> +	/* Don't run the guest */
> +	bool pause;
> +
>  	/* IO related fields */
>  	struct kvm_decode mmio_decode;
>  
> diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
> index 7537e68..46d4ef6 100644
> --- a/arch/arm/kvm/arm.c
> +++ b/arch/arm/kvm/arm.c
> @@ -471,11 +471,39 @@ bool kvm_arch_intc_initialized(struct kvm *kvm)
>  	return vgic_initialized(kvm);
>  }
>  
> +static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused;
> +static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused;
> +
> +static void kvm_arm_halt_guest(struct kvm *kvm)
> +{
> +	int i;
> +	struct kvm_vcpu *vcpu;
> +
> +	kvm_for_each_vcpu(i, vcpu, kvm)
> +		vcpu->arch.pause = true;
> +	force_vm_exit(cpu_all_mask);
> +}
> +
> +static void kvm_arm_resume_guest(struct kvm *kvm)
> +{
> +	int i;
> +	struct kvm_vcpu *vcpu;
> +
> +	kvm_for_each_vcpu(i, vcpu, kvm) {
> +		wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
> +
> +		vcpu->arch.pause = false;
> +		wake_up_interruptible(wq);
> +	}
> +}
> +
> +
>  static void vcpu_pause(struct kvm_vcpu *vcpu)
>  {
>  	wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
>  
> -	wait_event_interruptible(*wq, !vcpu->arch.power_off);
> +	wait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
> +				       (!vcpu->arch.pause)));
>  }
>  
>  static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
> @@ -525,7 +553,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
>  
>  		update_vttbr(vcpu->kvm);
>  
> -		if (vcpu->arch.power_off)
> +		if (vcpu->arch.power_off || vcpu->arch.pause)
>  			vcpu_pause(vcpu);

These two changes really make me feel like kvm_arch_vcpu_runnable()
should be changed to check for both flags and all thes eplaces should
check for runnable instead.

To make the runnable function more robust, perhaps it should check:

1. The vcpu is neither paused nor powered off
2. The vcpu is not in a WFI state (vcpu->arch.wfi_request) OR there are
   pending interrupts (the current check).

Then the logic would fit the name of the function, instead of being a
specific check only called through handle_wfx.

>  
>  		/*
> @@ -551,7 +579,8 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
>  			run->exit_reason = KVM_EXIT_INTR;
>  		}
>  
> -		if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
> +		if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
> +		    vcpu->arch.pause) {

so why do we need to re-check the pause flag, but not the power_off
flag?  That is non-trivial for sure, so if it's correct, deserves a
comment.  Also see my comment on the last patch.

>  			local_irq_enable();
>  			preempt_enable();
>  			kvm_vgic_sync_hwstate(vcpu);
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 009da6b..69e3785 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -125,6 +125,9 @@ struct kvm_vcpu_arch {
>  	/* vcpu power-off state */
>  	bool power_off;
>  
> +	/* Don't run the guest */

Probably need to be more clear about this being an implementation
requirement rather than being guest visible or related to any
architectural concept.

> +	bool pause;
> +
>  	/* IO related fields */
>  	struct kvm_decode mmio_decode;
>  
> -- 
> 1.9.1
> 

Thanks,
-Christoffer