Re: [PATCH] KVM: Pre-allocate 1 cpumask variable per cpu for both pv tlb and pv ipis

[Thadeu Lima de Souza Cascardo]

On Tue, Feb 04, 2020 at 03:42:04PM +0100, Vitaly Kuznetsov wrote:
> Thadeu Lima de Souza Cascardo <cascardo@xxxxxxxxxxxxx> writes:
> 
> >> > >      /*
> >> > > @@ -624,6 +625,7 @@ static void __init kvm_guest_init(void)
> >> > >          kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
> >> > >          pv_ops.mmu.flush_tlb_others = kvm_flush_tlb_others;
> >> > >          pv_ops.mmu.tlb_remove_table = tlb_remove_table;
> >> > > +        pr_info("KVM setup pv remote TLB flush\n");
> >> > >      }
> >> > >
> >
> > I am more concerned about printing the "KVM setup pv remote TLB flush" message,
> > not only when KVM pv is used, but pv TLB flush is not going to be used, but
> > also when the system is not even paravirtualized.
> 
> Huh? In Wanpeng's patch this print is under
> 
> 	if (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
> 	    !kvm_para_has_hint(KVM_HINTS_REALTIME) &&
> 	    kvm_para_has_feature(KVM_FEATURE_STEAL_TIME))
> 
> and if you mean another patch we descussed before which was adding
>  (!kvm_para_available() || nopv) check than it's still needed. Or,
> alternatively, we can make kvm_para_has_feature() check for that.
> 
> -- 
> Vitaly
> 

Yes, that's what I mean. Though not printing that when allocating the cpumasks
would fix this particular symptom, anyway.

But yes, it doesn't make sense to do all those feature checks when there is no
paravirtualization.

I believe we are in agreement.

Cascardo.