On Fri, Nov 30, 2012 at 12:14 PM, Will Deacon [off-list ref] wrote:

quoted

On Fri, Nov 30, 2012 at 04:47:40PM +0000, Christoffer Dall wrote:

quoted

On Fri, Nov 30, 2012 at 10:15 AM, Will Deacon [off-list ref] wrote:

quoted

At this point, VM1 is running and VM0:VCPU1 is running. VM0:VCPU0 is not
running because physical CPU0 is handling an interrupt. The problem is that
when VCPU0 *is* resumed, it will update the VMID of VM0 and could be
scheduled in parallel with VCPU1 but with a different VMID.

How do you avoid this in the current code?

I don't. Nice catch. Please apply your interesting brain to the following fix:)

I'm far too sober to look at your patch right now, but I'll think about it
over the weekend [I can't break it at a quick glance] :)

In the meantime, can you think about whether the TLB operations need to run
on every CPU please?

they don't we can invalidate the TLB and the icache using the inner
shareability domain. Here's a patch:

diff --git a/arch/arm/include/asm/kvm_host.h b/arch/arm/include/asm/kvm_host.h
index ad1390f..df1b753 100644
--- a/arch/arm/include/asm/kvm_host.h
+++ b/arch/arm/include/asm/kvm_host.h

@@ -146,6 +146,7 @@ struct kvm_one_reg;
 int kvm_arm_get_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
 int kvm_arm_set_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
 u64 kvm_call_hyp(void *hypfn, ...);
+void force_vm_exit(const cpumask_t *mask);

 #define KVM_ARCH_WANT_MMU_NOTIFIER
 struct kvm;

diff --git a/arch/arm/kvm/arm.c b/arch/arm/kvm/arm.c
index c4f631e..674592e 100644
--- a/arch/arm/kvm/arm.c
+++ b/arch/arm/kvm/arm.c

@@ -405,9 +405,14 @@ int kvm_arch_vcpu_in_guest_mode(struct kvm_vcpu *v)
 	return v->mode == IN_GUEST_MODE;
 }

-static void reset_vm_context(void *info)
+/* Just ensure a guest exit from a particular CPU */
+static void exit_vm_noop(void *info)
 {
-	kvm_call_hyp(__kvm_flush_vm_context);
+}
+
+void force_vm_exit(const cpumask_t *mask)
+{
+	smp_call_function_many(mask, exit_vm_noop, NULL, true);
 }