Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: AVIC related warning in enable_irq_window
To:     Maxim Levitsky <mlevitsk@redhat.com>,
        Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>,
        kvm@vger.kernel.org
Cc:     linux-kernel@vger.kernel.org
References: <9ce7bb5c4fb8bcc4ac21103f7534a6edfcbe195d.camel@redhat.com>
 <758b27a8-74c0-087d-d90b-d95faee2f561@redhat.com>
 <c5c32371-4b4e-1382-c616-3830ba46bf85@amd.com>
 <159382e7fdf0f9b50d79e29554842289e92e1ed7.camel@redhat.com>
From:   Paolo Bonzini <pbonzini@redhat.com>
Message-ID: <d22d32de-5d91-662a-bf53-8cfb115dbe8d@redhat.com>
Date:   Mon, 4 May 2020 11:25:31 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.6.0
MIME-Version: 1.0
In-Reply-To: <159382e7fdf0f9b50d79e29554842289e92e1ed7.camel@redhat.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 04/05/20 11:13, Maxim Levitsky wrote:
> On Mon, 2020-05-04 at 15:46 +0700, Suravee Suthikulpanit wrote:
>> Paolo / Maxim,
>>
>> On 5/2/20 11:42 PM, Paolo Bonzini wrote:
>>> On 02/05/20 15:58, Maxim Levitsky wrote:
>>>> The AVIC is disabled by svm_toggle_avic_for_irq_window, which calls
>>>> kvm_request_apicv_update, which broadcasts the KVM_REQ_APICV_UPDATE vcpu request,
>>>> however it doesn't broadcast it to CPU on which now we are running, which seems OK,
>>>> because the code that handles that broadcast runs on each VCPU entry, thus
>>>> when this CPU will enter guest mode it will notice and disable the AVIC.
>>>>
>>>> However later in svm_enable_vintr, there is test 'WARN_ON(kvm_vcpu_apicv_active(&svm->vcpu));'
>>>> which is still true on current CPU because of the above.
>>>
>>> Good point!  We can just remove the WARN_ON I think.  Can you send a patch?
>>
>> Instead, as an alternative to remove the WARN_ON(), would it be better to just explicitly
>> calling kvm_vcpu_update_apicv(vcpu) to update the apicv_active flag right after
>> kvm_request_apicv_update()?
>>
> This should work IMHO, other that the fact kvm_vcpu_update_apicv will be called again,
> when this vcpu is entered since the KVM_REQ_APICV_UPDATE will still be pending on it.
> It shoudn't be a problem, and we can even add a check to do nothing when it is called
> while avic is already in target enable state.

I thought about that but I think it's a bit confusing.  If we want to
keep the WARN_ON, Maxim can add an equivalent one to svm_vcpu_run, which
is even better because the invariant is clearer.

WARN_ON((vmcb->control.int_ctl & (AVIC_ENABLE_MASK | V_IRQ_MASK))
	== (AVIC_ENABLE_MASK | V_IRQ_MASK));

Paolo