Date:   Mon, 29 Nov 2021 19:20:10 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Paolo Bonzini <pbonzini@redhat.com>
Cc:     isaku.yamahata@intel.com, Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H . Peter Anvin" <hpa@zytor.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>, erdemaktas@google.com,
        Connor Kuehl <ckuehl@redhat.com>, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org, isaku.yamahata@gmail.com,
        Sean Christopherson <sean.j.christopherson@intel.com>
Subject: Re: [RFC PATCH v3 28/59] KVM: x86: Check for pending APICv interrupt
 in kvm_vcpu_has_events()
Message-ID: <YaUn6pqZHrw4Z8zn@google.com>
References: <cover.1637799475.git.isaku.yamahata@intel.com>
 <9852ad79d1078088743a57008226c869b0316da1.1637799475.git.isaku.yamahata@intel.com>
 <4708e92c-373b-a07f-c80c-fe194ca706df@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4708e92c-373b-a07f-c80c-fe194ca706df@redhat.com>
Precedence: bulk

On Thu, Nov 25, 2021, Paolo Bonzini wrote:
> On 11/25/21 01:20, isaku.yamahata@intel.com wrote:
> > From: Sean Christopherson<sean.j.christopherson@intel.com>
> > 
> > Return true for kvm_vcpu_has_events() if the vCPU has a pending APICv
> > interrupt to support TDX's usage of APICv.  Unlike VMX, TDX doesn't have
> > access to vmcs.GUEST_INTR_STATUS and so can't emulate posted interrupts,
> > i.e. needs to generate a posted interrupt and more importantly can't
> > manually move requested interrupts into the vIRR (which it also doesn't
> > have access to).
> 
> Does this mean it is impossible to disable APICv on TDX?  If so, please add
> a WARN.

Yes, APICv is forced.

Rereading this patch, checking only for a pending posted interrupt isn't correct,
a pending interrupt that's below the PPR shouldn't be considered a wake event.

A much better approach would be to have vt_sync_pir_to_irr() redirect to a TDX
implementation to read the PIR but not update the vIRR, that way common x86 doesn't
need to be touched.  Hopefully that can be done in a race-free way.