Re: [PATCH] kvm:vmx:code changes in handle_io() to save some CPU cycles.

[Vitaly Kuznetsov]

Stephen Zhang <starzhangzsd@xxxxxxxxx> writes:

> code changes in handle_io() to save some CPU cycles.
>
> Signed-off-by: Stephen Zhang <starzhangzsd@xxxxxxxxx>
> ---
>  arch/x86/kvm/vmx/vmx.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 47b8357..109bcf64 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -4899,15 +4899,14 @@ static int handle_triple_fault(struct kvm_vcpu *vcpu)
>  static int handle_io(struct kvm_vcpu *vcpu)
>  {
>  	unsigned long exit_qualification;
> -	int size, in, string;
> +	int size, in;
>  	unsigned port;
>  
>  	exit_qualification = vmx_get_exit_qual(vcpu);
> -	string = (exit_qualification & 16) != 0;
>  
>  	++vcpu->stat.io_exits;
>  
> -	if (string)
> +	if (exit_qualification & 16)
>  		return kvm_emulate_instruction(vcpu, 0);
>  
>  	port = exit_qualification >> 16;

I seriously doubt this can save any CPU cycles as compiler will likely
re-order and optimize this check anyway, I don't expect to see any
push/pop operations here. I agree that having local 'string' variable is
likely an overkill, however, the 'exit_qualification & 16' check we have
is hard to read so 'string' here works as a comment :-)

I would've liked the patch in the following shape more:

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index e35552055c07..65b6062ad7bc 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -4898,15 +4898,15 @@ static int handle_triple_fault(struct kvm_vcpu *vcpu)
 static int handle_io(struct kvm_vcpu *vcpu)
 {
        unsigned long exit_qualification;
-       int size, in, string;
+       int size, in;
        unsigned port;
 
        exit_qualification = vmx_get_exit_qual(vcpu);
-       string = (exit_qualification & 16) != 0;
 
        ++vcpu->stat.io_exits;
 
-       if (string)
+       /* String instruction */
+       if (exit_qualification & BIT(4))
                return kvm_emulate_instruction(vcpu, 0);
 
        port = exit_qualification >> 16;

Also, the changelog needs to be re-phrased to state code cleanup and not
CPU cycles optimization.

-- 
Vitaly