On Wed, Apr 20, 2022 at 08:40:42PM -0500, Bob Pearson wrote: > @@ -217,16 +216,15 @@ int __rxe_cleanup(struct rxe_pool_elem *elem) > struct rxe_pool *pool = elem->pool; > struct xarray *xa = &pool->xa; > static int timeout = RXE_POOL_TIMEOUT; > - unsigned long flags; > int ret, err = 0; > void *xa_ret; > > + WARN_ON(!in_task()); This should be might_sleep() which will show up any held spinlock that Zhu was talking about. There is definately some possible cases like that in the AH mess, RDMA_DESTROY_AH_SLEEPABLE for instance is not handled, and the error unwind during AH ceration, for example. But that is just more of that AH mess that needs fixing, and isn't related to RCU. It needs an atomic version of rxe_cleanup() too. Jason
