The valgrind indication is a false positive in this case, is seems there is a race of memory read and free of the libaio and aio kernel threads which had not timer-d out yet during exit()
reducing the aio_idle_time timeout to mitigate
```
man aio_init
...
aio_idle_time
This field specifies the amount of time in seconds that a worker thread
should wait for further requests before terminating, after having
completed a previous request. The
default value is 1.
...
```
Fixes: https://tracker.ceph.com/issues/64835 Signed-off-by: Mark Kogan <mkogan@ibm.com>