static struct rcu_head **rcu_tasks_cbs_tail = &rcu_tasks_cbs_head;
static DEFINE_RAW_SPINLOCK(rcu_tasks_cbs_lock);
+/* Track exiting tasks in order to allow them to be waited for. */
+DEFINE_SRCU(tasks_rcu_exit_srcu);
+
+/* Control stall timeouts. Disable with <= 0, otherwise jiffies till stall. */
+static int rcu_task_stall_timeout __read_mostly = HZ * 60 * 3;
+module_param(rcu_task_stall_timeout, int, 0644);
+
/* Post an RCU-tasks callback. */
void call_rcu_tasks(struct rcu_head *rhp, void (*func)(struct rcu_head *rhp))
{
}
rcu_read_unlock();
+ /*
+ * Wait for tasks that are in the process of exiting.
+ * This does only part of the job, ensuring that all
+ * tasks that were previously exiting reach the point
+ * where they have disabled preemption, allowing the
+ * later synchronize_sched() to finish the job.
+ */
+ synchronize_srcu(&tasks_rcu_exit_srcu);
+
/*
* Each pass through the following loop scans the list
* of holdout tasks, removing any that are no longer
* ->rcu_tasks_holdout accesses to be within the grace
* period, avoiding the need for memory barriers for
* ->rcu_tasks_holdout accesses.
+ *
+ * In addition, this synchronize_sched() waits for exiting
+ * tasks to complete their final preempt_disable() region
+ * of execution, cleaning up after the synchronize_srcu()
+ * above.
*/
synchronize_sched();