On Fri, 2019-09-27 at 10:11 +0200, Juri Lelli wrote:

Hi Scott,

On 27/07/19 00:56, Scott Wood wrote:

quoted

With the changes to migrate disabling, ->set_cpus_allowed() no longer
gets deferred until migrate_enable().  To avoid releasing the bandwidth
while the task may still be executing on the old CPU, move the
subtraction
to ->migrate_task_rq().

Signed-off-by: Scott Wood <redacted>
---
 kernel/sched/deadline.c | 67 +++++++++++++++++++++++-------------------
-------
 1 file changed, 31 insertions(+), 36 deletions(-)

diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index c18be51f7608..2f18d0cf1b56 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c

@@ -1606,14 +1606,42 @@ static void yield_task_dl(struct rq *rq)
 	return cpu;
 }
 
+static void free_old_cpuset_bw_dl(struct rq *rq, struct task_struct *p)
+{
+	struct root_domain *src_rd = rq->rd;
+
+	/*
+	 * Migrating a SCHED_DEADLINE task between exclusive
+	 * cpusets (different root_domains) entails a bandwidth
+	 * update. We already made space for us in the destination
+	 * domain (see cpuset_can_attach()).
+	 */
+	if (!cpumask_intersects(src_rd->span, p->cpus_ptr)) {
+		struct dl_bw *src_dl_b;
+
+		src_dl_b = dl_bw_of(cpu_of(rq));
+		/*
+		 * We now free resources of the root_domain we are migrating
+		 * off. In the worst case, sched_setattr() may temporary

fail
+		 * until we complete the update.
+		 */
+		raw_spin_lock(&src_dl_b->lock);
+		__dl_sub(src_dl_b, p->dl.dl_bw, dl_bw_cpus(task_cpu(p)));
+		raw_spin_unlock(&src_dl_b->lock);
+	}
+}
+
 static void migrate_task_rq_dl(struct task_struct *p, int new_cpu
__maybe_unused)
 {
 	struct rq *rq;
 
-	if (p->state != TASK_WAKING)
+	rq = task_rq(p);
+
+	if (p->state != TASK_WAKING) {
+		free_old_cpuset_bw_dl(rq, p);

What happens if a DEADLINE task is moved between cpusets while it was
sleeping? Don't we miss removing from the old cpuset if the task gets
migrated on wakeup?

In that case set_task_cpu() is called by ttwu after setting state to
TASK_WAKING.  I guess it could be annoying if the task doesn't wake up for a
long time and therefore doesn't release the bandwidth until then.

-Scott