sched/deadline: Check bandwidth overflow earlier for hotplug

Currently we check for bandwidth overflow potentially due to hotplug
operations at the end of sched_cpu_deactivate(), after the cpu going
offline has already been removed from scheduling, active_mask, etc.
This can create issues for DEADLINE tasks, as there is a substantial
race window between the start of sched_cpu_deactivate() and the moment
we possibly decide to roll-back the operation if dl_bw_deactivate()
returns failure in cpuset_cpu_inactive(). An example is a throttled
task that sees its replenishment timer firing while the cpu it was
previously running on is considered offline, but before
dl_bw_deactivate() had a chance to say no and roll-back happened.

Fix this by directly calling dl_bw_deactivate() first thing in
sched_cpu_deactivate() and do the required calculation in the former
function considering the cpu passed as an argument as offline already.

By doing so we also simplify sched_cpu_deactivate(), as there is no need
anymore for any kind of roll-back if we fail early.

Signed-off-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Phil Auld <pauld@redhat.com>
Tested-by: Waiman Long <longman@redhat.com>
Link: https://lore.kernel.org/r/Zzc1DfPhbvqDDIJR@jlelli-thinkpadt14gen4.remote.csb

kernel/sched/core.c

@@ -8182,19 +8182,14 @@ static void cpuset_cpu_active(void)
 	cpuset_update_active_cpus();
 }
 
-static int cpuset_cpu_inactive(unsigned int cpu)
+static void cpuset_cpu_inactive(unsigned int cpu)
 {
 	if (!cpuhp_tasks_frozen) {
-		int ret = dl_bw_deactivate(cpu);
-
-		if (ret)
-			return ret;
 		cpuset_update_active_cpus();
 	} else {
 		num_cpus_frozen++;
 		partition_sched_domains(1, NULL, NULL);
 	}
-	return 0;
 }
 
 static inline void sched_smt_present_inc(int cpu)
@@ -8256,6 +8251,11 @@ int sched_cpu_deactivate(unsigned int cpu)
 	struct rq *rq = cpu_rq(cpu);
 	int ret;
 
+	ret = dl_bw_deactivate(cpu);
+
+	if (ret)
+		return ret;
+
 	/*
 	 * Remove CPU from nohz.idle_cpus_mask to prevent participating in
 	 * load balancing when not active
@@ -8301,15 +8301,7 @@ int sched_cpu_deactivate(unsigned int cpu)
 		return 0;
 
 	sched_update_numa(cpu, false);
-	ret = cpuset_cpu_inactive(cpu);
+	cpuset_cpu_inactive(cpu);
-	if (ret) {
-		sched_smt_present_inc(cpu);
-		sched_set_rq_online(rq, cpu);
-		balance_push_set(cpu, false);
-		set_cpu_active(cpu, true);
-		sched_update_numa(cpu, true);
-		return ret;
-	}
 	sched_domains_numa_masks_clear(cpu);
 	return 0;
 }

kernel/sched/deadline.c

@@ -3495,6 +3495,13 @@ static int dl_bw_manage(enum dl_bw_request req, int cpu, u64 dl_bw)
 		}
 		break;
 	case dl_bw_req_deactivate:
+		/*
+		 * cpu is not off yet, but we need to do the math by
+		 * considering it off already (i.e., what would happen if we
+		 * turn cpu off?).
+		 */
+		cap -= arch_scale_cpu_capacity(cpu);
+
 		/*
 		 * cpu is going offline and NORMAL tasks will be moved away
 		 * from it. We can thus discount dl_server bandwidth
@@ -3512,9 +3519,10 @@ static int dl_bw_manage(enum dl_bw_request req, int cpu, u64 dl_bw)
 		if (dl_b->total_bw - fair_server_bw > 0) {
 			/*
 			 * Leaving at least one CPU for DEADLINE tasks seems a
-			 * wise thing to do.
+			 * wise thing to do. As said above, cpu is not offline
+			 * yet, so account for that.
 			 */
-			if (dl_bw_cpus(cpu))
+			if (dl_bw_cpus(cpu) - 1)
 				overflow = __dl_overflow(dl_b, cap, fair_server_bw, 0);
 			else
 				overflow = 1;