KVM: x86/xen: Fix lockdep warning on "recursive" gpc locking

In commit 5ec3289 ("KVM: x86/xen: Compatibility fixes for shared runstate
area") we declared it safe to obtain two gfn_to_pfn_cache locks at the same
time:
	/*
	 * The guest's runstate_info is split across two pages and we
	 * need to hold and validate both GPCs simultaneously. We can
	 * declare a lock ordering GPC1 > GPC2 because nothing else
	 * takes them more than one at a time.
	 */

However, we forgot to tell lockdep. Do so, by setting a subclass on the
first lock before taking the second.

Fixes: 5ec3289 ("KVM: x86/xen: Compatibility fixes for shared runstate area")
Suggested-by: Peter Zijlstra <[email protected]>
Signed-off-by: David Woodhouse <[email protected]>
Message-Id: <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

Author: dwmw2

arch/x86/kvm/xen.c

@@ -304,8 +304,10 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 		 * The guest's runstate_info is split across two pages and we
 		 * need to hold and validate both GPCs simultaneously. We can
 		 * declare a lock ordering GPC1 > GPC2 because nothing else
-		 * takes them more than one at a time.
+		 * takes them more than one at a time. Set a subclass on the
+		 * gpc1 lock to make lockdep shut up about it.
 		 */
+		lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_);
 		read_lock(&gpc2->lock);
 
 		if (!kvm_gpc_check(gpc2, user_len2)) {