Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <24b74f5bd8810c7f79777ed6898baeaf47bfe3e3.camel@redhat.com>
Subject: Re: [PATCH v2 08/12] KVM: SVM: Update AVIC settings when changing
 APIC mode
From:   Maxim Levitsky <mlevitsk@redhat.com>
To:     Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>,
        linux-kernel@vger.kernel.org, kvm@vger.kernel.org
Cc:     pbonzini@redhat.com, seanjc@google.com, joro@8bytes.org,
        jon.grimm@amd.com, wei.huang2@amd.com, terry.bowman@amd.com
Date:   Mon, 02 May 2022 20:13:19 +0300
In-Reply-To: <9307c734-3473-0bdc-57be-c39e96bca4d8@amd.com>
References: <20220412115822.14351-1-suravee.suthikulpanit@amd.com>
         <20220412115822.14351-9-suravee.suthikulpanit@amd.com>
         <abb93e2d73b7ada6cbabcd3ebbf7b38e4701ec57.camel@redhat.com>
         <9307c734-3473-0bdc-57be-c39e96bca4d8@amd.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.36.5 (3.36.5-2.fc32) 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Mon, 2022-05-02 at 21:07 +0700, Suravee Suthikulpanit wrote:
> Maxim, Sean
> 
> On 4/18/22 7:55 PM, Maxim Levitsky wrote:
> > On Tue, 2022-04-12 at 06:58 -0500, Suravee Suthikulpanit wrote:
> > > When APIC mode is updated (e.g. disabled, xAPIC, or x2APIC),
> > > KVM needs to call kvm_vcpu_update_apicv() to update AVIC settings
> > > accordingly.
> > > 
> > > Signed-off-by: Suravee Suthikulpanit<suravee.suthikulpanit@amd.com>
> > > ---
> > >   arch/x86/kvm/svm/avic.c | 15 +++++++++++++++
> > >   arch/x86/kvm/svm/svm.c  |  1 +
> > >   2 files changed, 16 insertions(+)
> > > 
> > > diff --git a/arch/x86/kvm/svm/avic.c b/arch/x86/kvm/svm/avic.c
> > > index 22ee1098e2a5..01392b8364f4 100644
> > > --- a/arch/x86/kvm/svm/avic.c
> > > +++ b/arch/x86/kvm/svm/avic.c
> > > @@ -616,6 +616,21 @@ void avic_apicv_post_state_restore(struct kvm_vcpu *vcpu)
> > >   	avic_handle_ldr_update(vcpu);
> > >   }
> > >   
> > > +void avic_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
> > > +{
> > > +	struct vcpu_svm *svm = to_svm(vcpu);
> > > +
> > > +	if (!lapic_in_kernel(vcpu) || (avic_mode == AVIC_MODE_NONE))
> > > +		return;
> > > +
> > > +	if (kvm_get_apic_mode(vcpu) == LAPIC_MODE_INVALID) {
> > > +		WARN_ONCE(true, "Invalid local APIC state (vcpu_id=%d)", vcpu->vcpu_id);
> > > +		return;
> > > +	}
> > > +
> > > +	kvm_vcpu_update_apicv(&svm->vcpu);
> > I think it makes sense to call avic_refresh_apicv_exec_ctrl directly here.
> >   
> > I am not sure that kvm_vcpu_update_apicv will even call it
> > because it has an optimization of doing nothing when inhibition status
> > didn't change.
> >   
> >   
> > Another semi-related note:
> >   
> > the current way the x2avic msrs are configured creates slight performance
> > problem for nesting:
> >   
> > The problem is that when entering a nested guest, AVIC on the current vCPU
> > is inhibited, but this is done only so that this vCPU*peers*  don't
> > try to use AVIC to send IPIs to it, so there is no need to update vmcb01
> > msr interception bitmap, and vmcb02 should have all these msrs intercepted always.
> > Same with returning to host.
> > 
> > It also should be checked that during nested entry, at least vmcb01 msr bitmap
> > is updated - TL;DR - please check that x2avic works when there is a nested guest running.
> 
> In the kvm/queue branch, I found a regression on nested SVM guest, where L2 guest cannot
> launch. The bad commit is:
> 
> commit a4cfff3f0f8c07f1f7873a82bdeb3995807dac8c (bisect)
> Merge: 42dcbe7d8bac 8d5678a76689
> Author: Paolo Bonzini <pbonzini@redhat.com>
> Date:   Fri Apr 8 12:43:40 2022 -0400
> 
>      Merge branch 'kvm-older-features' into HEAD
> 
>      Merge branch for features that did not make it into 5.18:
> 
>      * New ioctls to get/set TSC frequency for a whole VM
> 
>      * Allow userspace to opt out of hypercall patching
> 
>      Nested virtualization improvements for AMD:
> 
>      * Support for "nested nested" optimizations (nested vVMLOAD/VMSAVE,
>        nested vGIF)
> 
>      * Allow AVIC to co-exist with a nested guest running
> 
>      * Fixes for LBR virtualizations when a nested guest is running,
>        and nested LBR virtualization support
> 
>      * PAUSE filtering for nested hypervisors
> 
>      Guest support:
> 
>      * Decoupling of vcpu_is_preempted from PV spinlocks
> 
>      Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> 
> I am still working on the bisect into the merge commits.
> 
> Regards,
> Suravee
> 

What happens when the guest can't launch? It sure works for me for kvm/queue
from yesterday.

I'll test again tomorrow.


Best regards,
	Maxim Levitsky