On 09/04, Frederic Weisbecker wrote:
>
> So what happens if, say:
>
>
> CPU 1 CPU 2
> --------------------------------------------------------------
> rcu_read_lock()
> p = rcu_dereference(rq->task)
> if (refcount_inc_not_zero(p->rcu_users)) {
> .....
> release_task() {
> put_task_struct_rcu_user() {
> call_rcu() {
> queue rcu_head
in this particular case call_rcu() won't be called, so
> }
> }
> }
> put_task_struct_rcu_user(); //here rcu_users has been overwritten
rcu_users won't be overwritten.
But nobody should try to increment ->rcu_users,
rcu_read_lock();
p = rcu_dereference(rq->task);
refcount_inc_not_zero(p->rcu_users);
is already wrong because both release_task/last-schedule can happen in
between, before refcount_inc_not_zero().
Oleg.
