On 12/01/2023 16:27, David Woodhouse wrote:
On Thu, 2023-01-12 at 16:17 +0000, Paul Durrant wrote:
@@ -309,7 +317,14 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
* gpc1 lock to make lockdep shut up about it.
*/
lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_);
- read_lock(&gpc2->lock);
+ if (atomic) {
+ if (!read_trylock(&gpc2->lock)) {
You could avoid the nesting in this case with:
if (atomic && !read_trylock(&gpc2->lock))
+ read_unlock_irqrestore(&gpc1->lock, flags);
+ return;
+ }
+ } else {
+ read_lock(&gpc2->lock);
+ }
Hm? Wouldn't it take the lock twice then? It'd still take the 'else' branch.
Actually, yes... So much for hoping to make it look prettier.
Paul
