Re: [PATCH v2 1/3] KVM: x86: implement KVM_{GET|SET}_TSC_STATE

From: Oliver Upton <oupton@google.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>,
	Maxim Levitsky <mlevitsk@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Marcelo Tosatti <mtosatti@redhat.com>,
	kvm list <kvm@vger.kernel.org>, "H. Peter Anvin" <hpa@zytor.com>,
	Jonathan Corbet <corbet@lwn.net>,
	Jim Mattson <jmattson@google.com>,
	Wanpeng Li <wanpengli@tencent.com>,
	"open list:KERNEL SELFTEST FRAMEWORK" 
	<linux-kselftest@vger.kernel.org>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Sean Christopherson <sean.j.christopherson@intel.com>,
	open list <linux-kernel@vger.kernel.org>,
	Ingo Molnar <mingo@redhat.com>,
	"maintainer:X86 ARCHITECTURE (32-BIT AND 64-BIT)"
	<x86@kernel.org>, Joerg Roedel <joro@8bytes.org>,
	Borislav Petkov <bp@alien8.de>, Shuah Khan <shuah@kernel.org>,
	Andrew Jones <drjones@redhat.com>,
	"open list:DOCUMENTATION" <linux-doc@vger.kernel.org>
Subject: Re: [PATCH v2 1/3] KVM: x86: implement KVM_{GET|SET}_TSC_STATE
Date: Thu, 10 Dec 2020 12:13:12 -0600	[thread overview]
Message-ID: <CAOQ_QsggH=RaPbiOTGjDviKUCa0r4YgJjLsP7ghUGbcVtr2YJQ@mail.gmail.com> (raw)
In-Reply-To: <eb0cbfaa-251a-810b-3c12-4ee63d082bc8@redhat.com>

On Thu, Dec 10, 2020 at 12:05 PM Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> On 10/12/20 18:59, Oliver Upton wrote:
> > However, I don't believe we can assume the guest's TSCs to be synchronized,
> > even if sane guests will never touch them. In this case, I think a per-vCPU
> > ioctl is still warranted, allowing userspace to get at the guest CPU adjust
> > component of Thomas' equation below (paraphrased):
> >
> >          TSC guest CPU = host tsc base + guest base offset + guest CPU adjust
>
> Right now that would be:
>
> - KVM_GET_TSC_STATE (vm) returns host tsc base + guest base offset (plus
> the associated time)
>
> - KVM_GET_MSR *without* KVM_X86_QUIRK_TSC_HOST_ACCESS for guest CPU adjust
>
> and the corresponding SET ioctls.  What am *I* missing?
>
> > Alternatively, a write from userspace to the guest's IA32_TSC_ADJUST with
> > KVM_X86_QUIRK_TSC_HOST_ACCESS could have the same effect, but that seems to be
> > problematic for a couple reasons. First, depending on the guest's CPUID the
> > TSC_ADJUST MSR may not even be available, meaning that the guest could've used
> > IA32_TSC to adjust the TSC (eww).
>
> Indeed, the host should always be able to read/write IA32_TSC and
> IA32_TSC_ADJUST.

So long as it is guaranteed that guest manipulations of IA32_TSC are
reflected in IA32_TSC_ADJUST even if it isn't in the guest's CPUID,
then this seems OK. I think having clear documentation on this subject
is also necessary, as we're going to rely on the combination of
KVM_{GET,SET}_TSC_STATE, disabling KVM_X86_QUIRK_TSC_HOST_ACCESS, and
userspace reading/writing a possibly hidden MSR to pull this off
right.

--
Thanks,
Oliver

> Thanks,
>
> Paolo
>
> > Second, userspace replaying writes to IA32_TSC
> > (in the case IA32_TSC_ADJUST doesn't exist for the guest) seems_very_
> > unlikely to work given all the magic handling that KVM does for
> > writes to it.
> >
> > Is this roughly where we are or have I entirely missed the mark?:-)
>