linux-kernel - Re: [PATCH RFC v3 6/9] KVM: rework kvm_vcpu_on

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20170829172439.23fcd6eb.cohuck@redhat.com>
Date:   Tue, 29 Aug 2017 17:24:39 +0200
From:   Cornelia Huck <cohuck@...hat.com>
To:     David Hildenbrand <david@...hat.com>
Cc:     Radim Krčmář <rkrcmar@...hat.com>,
        linux-kernel@...r.kernel.org, kvm@...r.kernel.org,
        linux-mips@...ux-mips.org, kvm-ppc@...r.kernel.org,
        linux-s390@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        Paolo Bonzini <pbonzini@...hat.com>,
        Christoffer Dall <cdall@...aro.org>,
        Marc Zyngier <marc.zyngier@....com>,
        Christian Borntraeger <borntraeger@...ibm.com>,
        James Hogan <james.hogan@...tec.com>,
        Paul Mackerras <paulus@...abs.org>,
        Alexander Graf <agraf@...e.com>
Subject: Re: [PATCH RFC v3 6/9] KVM: rework kvm_vcpu_on_spin loop

On Tue, 22 Aug 2017 16:06:57 +0200
David Hildenbrand <david@...hat.com> wrote:

> On 21.08.2017 22:35, Radim Krčmář wrote:
> > The original code managed to obfuscate a straightforward idea:
> > start iterating from the selected index and reset the index to 0 when
> > reaching the end of online vcpus, then iterate until reaching the index
> > that we started at.
> > 
> > The resulting code is a bit better, IMO.  (Still horrible, though.)  
> 
> I think I prefer dropping this patch and maybe _after_ we have the list
> implementation in place, simply start walking the list from
> last_boosted_vcpu? (store a pointer instead of an index then, of course)
> 
> If I understand correctly, this would then be simply, one walk from
> last_boosted_vcpu until we hit last_boosted_vcpu again.

Yes, doing this change at this point in the series trades an ugly piece
of code for a slightly less ugly one.

> > 
> > Signed-off-by: Radim Krčmář <rkrcmar@...hat.com>
> > ---
> >  include/linux/kvm_host.h | 13 +++++++++++++
> >  virt/kvm/kvm_main.c      | 47 ++++++++++++++++++-----------------------------
> >  2 files changed, 31 insertions(+), 29 deletions(-)
> > 
> > diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> > index abd5cb1feb9e..cfb3c0efdd51 100644
> > --- a/include/linux/kvm_host.h
> > +++ b/include/linux/kvm_host.h
> > @@ -498,6 +498,19 @@ static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
> >  	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
> >  	     idx++)
> >  
> > +#define kvm_for_each_vcpu_from(idx, vcpup, from, kvm) \
> > +	for (idx = from, vcpup = kvm_get_vcpu(kvm, idx); \
> > +	     vcpup; \
> > +	     ({ \
> > +		idx++; \
> > +		if (idx >= atomic_read(&kvm->online_vcpus)) \
> > +			idx = 0; \
> > +		if (idx == from) \
> > +			vcpup = NULL; \
> > +		else \
> > +			vcpup = kvm_get_vcpu(kvm, idx); \
> > +	      }))

The loop below is better after the change, but this macro... it gets at
least a bit better if you push this behind patch 8.

> > +
> >  static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
> >  {
> >  	struct kvm_vcpu *vcpu = NULL;
> > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > index d89261d0d8c6..33a15e176927 100644
> > --- a/virt/kvm/kvm_main.c
> > +++ b/virt/kvm/kvm_main.c
> > @@ -2333,8 +2333,7 @@ void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
> >  	struct kvm_vcpu *vcpu;
> >  	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
> >  	int yielded = 0;
> > -	int try = 3;
> > -	int pass;
> > +	int try = 2;
> >  	int i;
> >  
> >  	kvm_vcpu_set_in_spin_loop(me, true);
> > @@ -2345,34 +2344,24 @@ void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
> >  	 * VCPU is holding the lock that we need and will release it.
> >  	 * We approximate round-robin by starting at the last boosted VCPU.
> >  	 */
> > -	for (pass = 0; pass < 2 && !yielded && try; pass++) {
> > -		kvm_for_each_vcpu(i, vcpu, kvm) {
> > -			if (!pass && i <= last_boosted_vcpu) {
> > -				i = last_boosted_vcpu;
> > -				continue;
> > -			} else if (pass && i > last_boosted_vcpu)
> > -				break;
> > -			if (!ACCESS_ONCE(vcpu->preempted))
> > -				continue;
> > -			if (vcpu == me)
> > -				continue;
> > -			if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
> > -				continue;
> > -			if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
> > -				continue;
> > -			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
> > -				continue;
> > +	kvm_for_each_vcpu_from(i, vcpu, last_boosted_vcpu, kvm) {
> > +		if (!ACCESS_ONCE(vcpu->preempted))
> > +			continue;
> > +		if (vcpu == me)
> > +			continue;
> > +		if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
> > +			continue;
> > +		if (yield_to_kernel_mode && !kvm_arch_vcpu_in_kernel(vcpu))
> > +			continue;
> > +		if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
> > +			continue;
> >  
> > -			yielded = kvm_vcpu_yield_to(vcpu);
> > -			if (yielded > 0) {
> > -				kvm->last_boosted_vcpu = i;
> > -				break;
> > -			} else if (yielded < 0) {
> > -				try--;
> > -				if (!try)
> > -					break;
> > -			}
> > -		}
> > +		yielded = kvm_vcpu_yield_to(vcpu);
> > +		if (yielded > 0) {
> > +			kvm->last_boosted_vcpu = i;
> > +			break;
> > +		} else if (yielded < 0 && !try--)
> > +			break;
> >  	}
> >  	kvm_vcpu_set_in_spin_loop(me, false);
> >  
> >   
> 
>