Re: [PATCH 3/4] KVM-S390: Less function calls in kvm_s390_import_bp_data() after error detection

[Cornelia Huck]

On Mon, 22 Aug 2016 23:17:26 +0200
SF Markus Elfring <elfring@xxxxxxxxxxxxxxxxxxxxx> wrote:

> >>>> @@ -273,10 +273,12 @@ int kvm_s390_import_bp_data(struct kvm_vcpu *vcpu,
> >>>>  	vcpu->arch.guestdbg.nr_hw_wp = nr_wp;
> >>>>  	vcpu->arch.guestdbg.hw_wp_info = wp_info;
> >>>>  	return 0;
> >>>> -error:
> >>>> -	kfree(bp_data);
> >>>> -	kfree(wp_info);
> >>>> +free_bp_info:
> >>>>  	kfree(bp_info);
> >>>> +free_wp_info:
> >>>> +	kfree(wp_info);
> >>>> +free_bp_data:
> >>>> +	kfree(bp_data);
> >>>>  	return ret;
> >>>>  }
> >>>>
> >>>
> >>> This replaces a perfectly fine fallthrough
> >>
> >> The usage of a single goto label like "error" seems to be convenient.
> >> But how do these habits fit to the current Linux coding style convention?
> >>
> >>
> >>> with some horrible labels.
> >>
> >> Do they explain better which processing steps should be performed
> >> for an efficient exception handling in this function implementation?
> > 
> > *sigh*
> > 
> > It's _exception handling_. It does not need to be "efficient",
> 
> I imagine that run time situations could evolve where software efficiency
> will also matter for this purpose.

*major sigh*

We can start to optimize error handling that should never run after we
fixed every other performance problem that we have. Not earlier.

> 
> 
> > it needs to be easily parsable by humans.
> 
> I guess that we have got different preferences for this detail.

And I'm maintainer for this code.

> 
> 
> > If in doubt, the compiler will be _much_ better at optimizing
> > that kind of stuff anyway.
> 
> Which compiler (or optimizer) implementation is capable to restructure
> the jump targets for you automatically in the way I propose here?

No, please stop right here. NACK. EOD.