Re: Problem with kvm on -tip

[Luis Henriques <henrix@sapo.pt>]

Hi,

On Sat, Apr 11, 2009 at 03:08:55PM +0300, Avi Kivity wrote:
>
> This might be fixed by the attached patch.

I confirm that the patch you sent removes the warnings but does it really solve
the issue? (Sorry, I really do not know this code so I might be saying something
really stupid.)

What I understand from your patch is that the only portion of code that needs
protection is the __get_cpu_var().  If this is true then a patch like the one
below would do a better job.  But I am not sure that nothing else needs
protection since the code immediately following the preempt_enable (in your
patch) is an invocation to local_irq_save()...

---
 arch/x86/kvm/x86.c |   10 +++++++---
 1 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 8ca100a..cf918b5 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -626,13 +626,17 @@ static void kvm_write_guest_time(struct kvm_vcpu *v)
 	unsigned long flags;
 	struct kvm_vcpu_arch *vcpu = &v->arch;
 	void *shared_kaddr;
+	uint32_t tsc_khz;
 
 	if ((!vcpu->time_page))
 		return;
 
-	if (unlikely(vcpu->hv_clock_tsc_khz != __get_cpu_var(cpu_tsc_khz))) {
-		kvm_set_time_scale(__get_cpu_var(cpu_tsc_khz), &vcpu->hv_clock);
-		vcpu->hv_clock_tsc_khz = __get_cpu_var(cpu_tsc_khz);
+	preempt_disable();
+	tsc_khz = __get_cpu_var(cpu_tsc_khz);
+	preempt_enable();
+	if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
+		kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
+		vcpu->hv_clock_tsc_khz = tsc_khz;
 	}

 	/* Keep irq disabled to prevent changes to the clock */
-- 
1.6.2.1