Re: linux-3.2.1 + KVM: TSC unstable on Core i5-2400S / H67

[Marcelo Tosatti <mtosatti@redhat.com>]

On Tue, Jan 24, 2012 at 12:36:08PM +0100, Johannes Stezenbach wrote:
> Hi,
> 
> I just got the following in dmesg with linux-3.2.1:
> 
> [23379.612251] usb 2-1.2: new high-speed USB device number 6 using ehci_hcd
> [    0.006666] Marking TSC unstable due to KVM discovered backwards TSC
> [23421.511167] Switching to clocksource hpet
> 
> The mainboard is an Asus P8H67-V, BIOS 0806 10/26/2011 with Core i5-2400S.
> The kvm instance was running for a day already, and there was an overnight
> suspend-to-disk cycle in between, but the above happened ~1 hour after resume.
> (I don't think the usb message has anything to do with it, it's just
> the last message before the TSC message.)
> 
> Hm, checking /var/log/kern.* it seems this also happened with
> linux-3.1.* without me noticing.
> 
> Should I be worried?  I thought Core-i5 processors have a stable TSC?

It does.

Can you please apply the following patch, and report back when it
reproduces again.

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 9c912f0b..17d706d 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2242,8 +2242,11 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 		tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
 			     tsc - vcpu->arch.last_guest_tsc;
 
-		if (tsc_delta < 0)
+		if (tsc_delta < 0) {
+			printk(KERN_ERR "tsc = %llx last_guest_tsc = %llx\n",
+				tsc, vcpu->arch.last_guest_tsc);
 			mark_tsc_unstable("KVM discovered backwards TSC");
+		}
 		if (check_tsc_unstable()) {
 			kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
 			vcpu->arch.tsc_catchup = 1;