Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 14 Mar 2022 12:09:42 +0800
From:   Chao Gao <chao.gao@intel.com>
To:     Maxim Levitsky <mlevitsk@redhat.com>
Cc:     Zeng Guang <guang.zeng@intel.com>,
        Sean Christopherson <seanjc@google.com>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>,
        "kvm@vger.kernel.org" <kvm@vger.kernel.org>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        "Luck, Tony" <tony.luck@intel.com>,
        Kan Liang <kan.liang@linux.intel.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Kim Phillips <kim.phillips@amd.com>,
        Jarkko Sakkinen <jarkko@kernel.org>,
        Jethro Beekman <jethro@fortanix.com>,
        "Huang, Kai" <kai.huang@intel.com>,
        "x86@kernel.org" <x86@kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "Hu, Robert" <robert.hu@intel.com>
Subject: Re: [PATCH v6 6/9] KVM: x86: lapic: don't allow to change APIC ID
 unconditionally
Message-ID: <20220314040941.GA18296@gao-cwp>
References: <Yifg4bea6zYEz1BK@google.com>
 <20220309052013.GA2915@gao-cwp>
 <YihCtvDps/qJ2TOW@google.com>
 <6dc7cff15812864ed14b5c014769488d80ce7f49.camel@redhat.com>
 <YirPkr5efyylrD0x@google.com>
 <29c76393-4884-94a8-f224-08d313b73f71@intel.com>
 <01586c518de0c72ff3997d32654b8fa6e7df257d.camel@redhat.com>
 <2900660d947a878e583ebedf60e7332e74a1af5f.camel@redhat.com>
 <20220313135335.GA18405@gao-cwp>
 <fbf929e0793a6b4df59ec9d95a018d1f6737db35.camel@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <fbf929e0793a6b4df59ec9d95a018d1f6737db35.camel@redhat.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

>> > > This won't work with nested AVIC - we can't just inhibit a nested guest using its own AVIC,
>> > > because migration happens.
>> > 
>> > I mean because host decided to change its apic id, which it can in theory do any time,
>> > even after the nested guest has started. Seriously, the only reason guest has to change apic id,
>> > is to try to exploit some security hole.
>> 
>> Hi
>> 
>> Thanks for the information.  
>> 
>> IIUC, you mean KVM applies APICv inhibition only to L1 VM, leaving APICv
>> enabled for L2 VM. Shouldn't KVM disable APICv for L2 VM in this case?
>> It looks like a generic issue in dynamically toggling APICv scheme,
>> e.g., qemu can set KVM_GUESTDBG_BLOCKIRQ after nested guest has started.
>> 
>
>That is the problem - you can't disable it for L2, unless you are willing to emulate it in software.
>Or in other words, when nested guest uses a hardware feature, you can't at some point say to it:
>sorry buddy - hardware feature disappeared.

Agreed. I missed this.

>
>It is *currently* not a problem for APICv because it doesn't do IPI virtualization,
>and even with these patches, it doesn't do this for nesting.
>It does become when you allow nested guest to use this which I did in the nested AVIC code.
>
>
>and writable apic ids do pose a large problem, since nested AVIC, will target L1's apic ids,
>and when they can change under you without any notice, and even worse be duplicate,
>it is just nightmare.

OK. So the problem of disabling APICv is if we choose to disable APICv instead
of making APIC ID read-only, although it can work perfectly for VMX IPIv, it
effectively makes future cleanup to AVIC difficult/impossible because nested
AVIC is practically to implement without assuming APIC IDs of L1 is immutable.

Sean & Maxim

How about go back to use a module parameter to opt in to read-only APIC ID.
Although migration in some cases may fail but it shouldn't be a big issue as
migration VMs from a KVM with nested=on to a KVM with nested=off may also fail.