linux-kernel - Re: [PATCH v11 023/113] KVM: TDX: allocate/free TDX vcpu structure

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Thu, 2 Mar 2023 16:43:41 -0800
From:   Isaku Yamahata <isaku.yamahata@...il.com>
To:     "Huang, Kai" <kai.huang@...el.com>
Cc:     "isaku.yamahata@...il.com" <isaku.yamahata@...il.com>,
        "Christopherson,, Sean" <seanjc@...gle.com>,
        "Shahar, Sagi" <sagis@...gle.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "Aktas, Erdem" <erdemaktas@...gle.com>,
        "kvm@...r.kernel.org" <kvm@...r.kernel.org>,
        "pbonzini@...hat.com" <pbonzini@...hat.com>,
        "Yamahata, Isaku" <isaku.yamahata@...el.com>,
        "dmatlack@...gle.com" <dmatlack@...gle.com>
Subject: Re: [PATCH v11 023/113] KVM: TDX: allocate/free TDX vcpu structure

On Wed, Mar 01, 2023 at 12:49:06AM +0000,
"Huang, Kai" <kai.huang@...el.com> wrote:

> On Tue, 2023-02-28 at 16:35 -0800, Isaku Yamahata wrote:
> > On Tue, Feb 28, 2023 at 09:49:10PM +0000,
> > "Huang, Kai" <kai.huang@...el.com> wrote:
> > 
> > > On Tue, 2023-02-28 at 12:18 -0800, Isaku Yamahata wrote:
> > > > On Tue, Feb 28, 2023 at 11:52:59AM +0000,
> > > > "Huang, Kai" <kai.huang@...el.com> wrote:
> > > > 
> > > > > On Tue, 2023-02-28 at 03:06 -0800, Isaku Yamahata wrote:
> > > > > > > > +	if (!e)
> > > > > > > > +		return -ENOMEM;
> > > > > > > > +	*e  = (struct kvm_cpuid_entry2) {
> > > > > > > > +		.function = 1,	/* Features for X2APIC */
> > > > > > > > +		.index = 0,
> > > > > > > > +		.eax = 0,
> > > > > > > > +		.ebx = 0,
> > > > > > > > +		.ecx = 1ULL << 21,	/* X2APIC */
> > > > > > > > +		.edx = 0,
> > > > > > > > +	};
> > > > > > > > +	vcpu->arch.cpuid_entries = e;
> > > > > > > > +	vcpu->arch.cpuid_nent = 1;
> > > > > > > 
> > > > > > > As mentioned above, why doing it here? Won't be this be overwritten later in
> > > > > > > KVM_SET_CPUID2?
> > > > > > 
> > > > > > Yes, user space VMM can overwrite cpuid[0x1] and APIC base MSR.  But it
> > > > > > doesn't
> > > > > > matter because it's a bug of user space VMM. user space VMM has to keep the
> > > > > > consistency of cpuid and MSRs.
> > > > > > Because TDX module virtualizes cpuid[0x1].x2apic to fixed 1, KVM value doesn't
> > > > > > matter after vcpu creation.
> > > > > > Because KVM virtualizes APIC base as read only to guest, cpuid[0x1].x2apic
> > > > > > doesn't matter after vcpu creation as long as user space VMM keeps KVM APIC
> > > > > > BASE
> > > > > > value.
> > > > > > 
> > > > > 
> > > > > Contrary, can we depend on userspace VMM to set x2APIC in CPUID, but not do this
> > > > > in KVM?  If userspace doesn't do it, we treat it as userspace's bug.
> > > > > 
> > > > > Plus, userspace anyway needs to set x2APIC in CPUID regardless whether you have
> > > > > done above here, correct?
> > > > > 
> > > > > I don't see the point of doing above in KVM because you are neither enforcing
> > > > > anything in KVM, nor you are reducing effort of userspace.
> > > > 
> > > > Good idea. I can drop cpuid part from tdx_vcpu_create() and apic base part from
> > > > tdx_vcpu_reset(). It needs to modify tdx_has_emulated_msr() to allow user space
> > > > VMM to update APIC BASE MSR.
> > > 
> > > My personal preference would be:
> > > 
> > > 1) In KVM_SET_CPUID2, we do sanity check of CPUIDs provided by userspace, and
> > > return error if not met (i.e X2APIC isn't advertised).  We already have cases
> > > that KVM_SET_CPUID2 can fail, so extending to do TDX-specific check seems
> > > reasonable to me too.
> > 
> > This is moot. The current check does only check maxphys address bit size and
> > specified xfeatures are supported by host.  It's bare minimum for kvm to work.
> > It doesn't try to check consistency.
> > 
> > 
> > > 2) For APIC_BASE, you can just initialize the MSR in tdx_vcpu_reset() and ignore
> > > any update (+pr_warn()?) to MSR_IA32_APIC_BASE.
> > 
> > The x86 common code for KVM_CREATE_VCPU, kvm_arch_vcpu_create(), calls vcpu_create,
> > creates lapic, and calls vcpu_reset(). 
> > 
> > Setting ACPI BASE MSR with X2APIC enabled, checks if cpuid x2apic bit is set.
> > Please notice guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) in kvm_set_apic_base().
> > To work around it, one way is set cpuid artificially in create method as this
> > patch does.  Other way would be to introduce another version of
> > kvm_set_apic_base() that doesn't check cpuid dedicated for this purpose.
> > The third option is to make it user space responsibility to set initial reset
> > value of APIC BASE MSR.
> > 
> > Which option do you prefer?
> > 
> 
> I just recall you have already set all CPUIDs via tdx_td_init().  I would do
> below:
> 
> 1) keep all CPUIDs in tdx_td_init(), and make vcpu->cpuid point to that.
> 2) Ignore KVM_SET_CPUID2 for TDX guest (+ pr_warn(), etc).
> 3) Set TDX-fixed CPU registers/msrs, etc in reset_vcpu().

Finally I come up with the following flow.

- KVM_CREATE_VCPU
  - tdx_vcpu_create()
    no cpuid, no msr operation
- KVM_SET_CPUID2
  user space has to set cpuid...x2apic=1
- KVM_TDX_INIT_VCPU
  tdx_vcpu_ioctl() sets APIC_BASE MSR to x2apic enabled.
  Here if user space VMM doesn't set cpuid properly, it results in error.

ACPI_BASE MSR is read only for both user space VMM and guest TD.
cpuid:
After KVM_TDX_INIT_VCPU, user space VMM can update it by KVM_SET_CPUID2.
KVM doesn't care.  Guest TD see cpuid.x2apic value virtualized by TDX module
while KVM internally doesn't use the value because APIC_BASE won't change.


Thanks,
-- 
Isaku Yamahata <isaku.yamahata@...il.com>