Re: [RFC PATCH 09/11] kvm: simplify processor compat check

[Paolo Bonzini <pbonzini@redhat.com>]

Il 27/09/2013 15:13, Aneesh Kumar K.V ha scritto:
> Alexander Graf <agraf@suse.de> writes:
> 
>> On 27.09.2013, at 12:03, Aneesh Kumar K.V wrote:
>>
>>> From: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
>>
>> Missing patch description.
>>
>>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
>>
>> I fail to see how this really simplifies things, but at the end of the
>> day it's Gleb's and Paolo's call.
> 
> will do. It avoid calling 
> 
> 	for_each_online_cpu(cpu) {
> 		smp_call_function_single() 
> 
> on multiple architecture.

I agree with Alex.

The current code is not specially awesome; having
kvm_arch_check_processor_compat take an int* disguised as a void* is a
bit ugly indeed.

However, the API makes sense and tells you that it is being passed as a
callback (to smp_call_function_single in this case).

You are making the API more complicated to use on the arch layer
(because arch maintainers now have to think "do I need to check this on
all online CPUs?") and making the "leaf" POWER code less legible because
it still has the weird void()(void *) calling convention.

If anything, you could change kvm_arch_check_processor_compat to return
an int and accept no argument, and introduce a wrapper that kvm_init
passes to smp_call_function_single.

Paolo

> We also want to make the smp call function a callback of opaque. Hence
> this should be made arch specific. 
> 
> int kvm_arch_check_processor_compat(void *opaque)
> {
> 	int r,cpu;
> 	struct kvmppc_ops *kvm_ops = (struct kvmppc_ops *)opaque;
> 	for_each_online_cpu(cpu) {
> 		smp_call_function_single(cpu,
> 					 kvm_ops->check_processor_compat,
> 					 &r, 1);
> 		if (r < 0)
> 			break;
> 	}
> 	return r;
> }
> 
> against
> 
> -	for_each_online_cpu(cpu) {
> -		smp_call_function_single(cpu,
> -				kvm_arch_check_processor_compat,
> -				&r, 1);
> -		if (r < 0)
> -			goto out_free_1;
> -	}
> +
> +	r = kvm_arch_check_processor_compat(opaque);
> +	if (r < 0)
> +		goto out_free_1;
> 
> 
> 
>>
>> Which brings me to the next issue: You forgot to CC kvm@vger on your
>> patch set. Gleb and Paolo don't read kvm-ppc@vger. And they shouldn't
>> have to. Every kvm patch that you want review on or that should get
>> applied needs to be sent to kvm@vger. If you want to tag it as PPC
>> specific patch, do so by CC'ing kvm-ppc@vger.
> 
> Will do in the next update
> 
> -aneesh
>