On 2/28/25 16:36, Keith Busch wrote:

On Fri, Feb 28, 2025 at 07:29:45AM -0800, Sean Christopherson wrote:

quoted

On Fri, Feb 28, 2025, Keith Busch wrote:

quoted

On Fri, Feb 28, 2025 at 06:32:47AM -0800, Sean Christopherson wrote:

quoted

quoted

@@ -35,10 +35,12 @@ static inline int call_once(struct once *once, int (*cb)(struct once *))
                 return 0;
  
          guard(mutex)(&once->lock);
-        WARN_ON(atomic_read(&once->state) == ONCE_RUNNING);
-        if (atomic_read(&once->state) != ONCE_NOT_STARTED)
+        if (WARN_ON(atomic_read(&once->state) == ONCE_RUNNING))
                  return -EINVAL;
  
+        if (atomic_read(&once->state) == ONCE_COMPLETED)
+                return 0;
+
          atomic_set(&once->state, ONCE_RUNNING);
         r = cb(once);
         if (r)

Possible suggestion since it seems odd to do an atomic_read twice on the
same value.

Yeah, good call.  At the risk of getting too cute, how about this?

Sure, that also looks good to me.

Just to overthink it a bit more, I'm changing "if (r)" to "if (r < 0)". 
Not because it's particularly useful to return a meaningful nonzero 
value on the first initialization, but more because 0+ for success and 
-errno for failure is a more common.

Queued with this change, thanks.

(Keith, I haven't forgotten about AVX by the way).

Paolo