Re: [Xen-devel] [PATCH RFC V6 1/5] kvm hypervisor : Add a hypercall to KVM hypervisor to support pv-ticketlocks

[Raghavendra K T <raghavendra.kt@xxxxxxxxxxxxxxxxxx>]

Thanks Avi, for the review.

On 04/29/2012 06:55 PM, Avi Kivity wrote:
> On 04/23/2012 12:59 PM, Raghavendra K T wrote:
> > From: Srivatsa Vaddagiri<vatsa@xxxxxxxxxxxxxxxxxx>
> >
> > KVM_HC_KICK_CPU allows the calling vcpu to kick another vcpu out of halt state.
> >
> > The presence of these hypercalls is indicated to guest via
> > KVM_FEATURE_PV_UNHALT/KVM_CAP_PV_UNHALT.
> >
> >   #endif
> > diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> > index e216ba0..dad475b 100644
> > --- a/arch/x86/include/asm/kvm_host.h
> > +++ b/arch/x86/include/asm/kvm_host.h
> > @@ -481,6 +481,10 @@ struct kvm_vcpu_arch {
> >                 u64 length;
> >                 u64 status;
> >         } osvw;
> > +       /* pv related host specific info */
> > +       struct {
> > +               int pv_unhalted;
> > +       } pv;
> >   };
> 'bool'.  Or maybe push into vcpu->requests.
Ok. I think you meant
+       struct {
+               bool pv_unhalted;
+       } pv;

and as discussed in old series (V4), cleaner implementation having
vcpu request, would still need a flag to prevent vcpu hang, so back to
having one flag.

> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index 4044ce0..7fc9be6 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -2147,6 +2147,7 @@ int kvm_dev_ioctl_check_extension(long ext)
> >         case KVM_CAP_ASYNC_PF:
> >         case KVM_CAP_GET_TSC_KHZ:
> >         case KVM_CAP_PCI_2_3:
> > +       case KVM_CAP_PV_UNHALT:
> >                 r = 1;
> >                 break;
> >         case KVM_CAP_COALESCED_MMIO:
> Redundant, since we can infer this from KVM_GET_SUPPORTED_CPUID.  But
> please indicate this in the documentation.
Ok. will mention that in documentation added for KVM_CAP_PV_UNHALT.

> > +/*
> > + * kvm_pv_kick_cpu_op:  Kick a vcpu.
> > + *
> > + * @apicid - apicid of vcpu to be kicked.
> > + */
> > +static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
> > +{
> > +       struct kvm_vcpu *vcpu = NULL;
> > +       int i;
> > +
> > +       kvm_for_each_vcpu(i, vcpu, kvm) {
> > +               if (!kvm_apic_present(vcpu))
> > +                       continue;
> > +
> > +               if (kvm_apic_match_dest(vcpu, 0, 0, apicid, 0))
> > +                       break;
> > +       }
> > +       if (vcpu) {
> > +               /*
> > +                * Setting unhalt flag here can result in spurious runnable
> > +                * state when unhalt reset does not happen in vcpu_block.
> > +                * But that is harmless since that should soon result in halt.
> > +                */
> > +               vcpu->arch.pv.pv_unhalted = 1;
> > +               /* We need everybody see unhalt before vcpu unblocks */
> > +               smp_mb();
> smp_wmb().
Done.

> > +               kvm_vcpu_kick(vcpu);
> > +       }
> > +}
> > +
> >
> >   /*
> >    * hypercalls use architecture specific
> > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > index 42b7393..edf56d4 100644
> > --- a/virt/kvm/kvm_main.c
> > +++ b/virt/kvm/kvm_main.c
> > @@ -1500,6 +1500,14 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
> >                 prepare_to_wait(&vcpu->wq,&wait, TASK_INTERRUPTIBLE);
> >
> >                 if (kvm_arch_vcpu_runnable(vcpu)) {
> > +                       /*
> > +                        * This is the only safe place to reset unhalt flag.
> > +                        * otherwise it results in loosing the notification
> > +                        * which eventually can result in vcpu hangs.
> > +                        */
> > +                       kvm_arch_vcpu_reset_pv_unhalted(vcpu);
> > +                       /* preventing reordering should be enough here */
> > +                       barrier();
> >                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
> >                         break;
> >                 }
> Hm, what about reusing KVM_REQ_UNHALT?
Yes, I had experimented this for some time without success.
For e.g. having
make_request(KVM_REQ_UNHALT, vcpu) directly from kick hypercall.

It would still need a flag. (did not get any alternative so far except
the  workaround posted in V4) :(


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel