[PATCH] sched: Fix stop_one_cpu_nowait() vs hotplug

[Peter Zijlstra]

On Tue, Oct 10, 2023 at 04:57:47PM +0200, Peter Zijlstra wrote:
> On Tue, Oct 10, 2023 at 02:40:22PM +0000, Kuyo Chang (張建文) wrote:

> > It is running good so far(more than a week)on hotplug/set affinity
> > stress test. I will keep it testing and report back if it happens
> > again.
> 
> OK, I suppose I should look at writing a coherent Changelog for this
> then...

Something like the below... ?

---
Subject: sched: Fix stop_one_cpu_nowait() vs hotplug
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Tue Oct 10 20:57:39 CEST 2023

Kuyo reported sporadic failures on a sched_setaffinity() vs CPU
hotplug stress-test -- notably affine_move_task() remains stuck in
wait_for_completion(), leading to a hung-task detector warning.

Specifically, it was reported that stop_one_cpu_nowait(.fn =
migration_cpu_stop) returns false -- this stopper is responsible for
the matching complete().

The race scenario is:

	CPU0					CPU1

					// doing _cpu_down()

  __set_cpus_allowed_ptr()
    task_rq_lock();
					takedown_cpu()
					  stop_machine_cpuslocked(take_cpu_down..)
	
					<PREEMPT: cpu_stopper_thread()
					  MULTI_STOP_PREPARE
					  ...
    __set_cpus_allowed_ptr_locked()
      affine_move_task()
        task_rq_unlock();

  <PREEMPT: cpu_stopper_thread()\> 
    ack_state()
					  MULTI_STOP_RUN
					    take_cpu_down()
					      __cpu_disable();
					      stop_machine_park();
						stopper->enabled = false;
					 />
   />
	stop_one_cpu_nowait(.fn = migration_cpu_stop);
          if (stopper->enabled) // false!!!

	
That is, by doing stop_one_cpu_nowait() after dropping rq-lock, the
stopper thread gets a chance to preempt and allows the cpu-down for
the target CPU to complete.

OTOH, since stop_one_cpu_nowait() / cpu_stop_queue_work() needs to
issue a wakeup, it must not be ran under the scheduler locks.

Solve this apparent contradiction by keeping preemption disabled over
the unlock + queue_stopper combination:

	preempt_disable();
	task_rq_unlock(...);
	if (!stop_pending)
	  stop_one_cpu_nowait(...)
	preempt_enable();

This respects the lock ordering contraints while still avoiding the
above race. That is, if we find the CPU is online under rq-lock, the
targeted stop_one_cpu_nowait() must succeed.

Apply this pattern to all similar stop_one_cpu_nowait() invocations.

Fixes: 6d337eab041d ("sched: Fix migrate_disable() vs set_cpus_allowed_ptr()")
Reported-by: "Kuyo Chang (張建文)" <Kuyo.Chang@xxxxxxxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Tested-by: "Kuyo Chang (張建文)" <Kuyo.Chang@xxxxxxxxxxxx>
---
 kernel/sched/core.c     |   10 ++++++++--
 kernel/sched/deadline.c |    2 ++
 kernel/sched/fair.c     |    4 +++-
 3 files changed, 13 insertions(+), 3 deletions(-)

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2645,9 +2645,11 @@ static int migration_cpu_stop(void *data
 		 * it.
 		 */
 		WARN_ON_ONCE(!pending->stop_pending);
+		preempt_disable();
 		task_rq_unlock(rq, p, &rf);
 		stop_one_cpu_nowait(task_cpu(p), migration_cpu_stop,
 				    &pending->arg, &pending->stop_work);
+		preempt_enable();
 		return 0;
 	}
 out:
@@ -2967,12 +2969,13 @@ static int affine_move_task(struct rq *r
 			complete = true;
 		}
 
+		preempt_disable();
 		task_rq_unlock(rq, p, rf);
-
 		if (push_task) {
 			stop_one_cpu_nowait(rq->cpu, push_cpu_stop,
 					    p, &rq->push_work);
 		}
+		preempt_enable();
 
 		if (complete)
 			complete_all(&pending->done);
@@ -3038,12 +3041,13 @@ static int affine_move_task(struct rq *r
 		if (flags & SCA_MIGRATE_ENABLE)
 			p->migration_flags &= ~MDF_PUSH;
 
+		preempt_disable();
 		task_rq_unlock(rq, p, rf);
-
 		if (!stop_pending) {
 			stop_one_cpu_nowait(cpu_of(rq), migration_cpu_stop,
 					    &pending->arg, &pending->stop_work);
 		}
+		preempt_enable();
 
 		if (flags & SCA_MIGRATE_ENABLE)
 			return 0;
@@ -9459,9 +9463,11 @@ static void balance_push(struct rq *rq)
 	 * Temporarily drop rq->lock such that we can wake-up the stop task.
 	 * Both preemption and IRQs are still disabled.
 	 */
+	preempt_disable();
 	raw_spin_rq_unlock(rq);
 	stop_one_cpu_nowait(rq->cpu, __balance_push_cpu_stop, push_task,
 			    this_cpu_ptr(&push_work));
+	preempt_enable();
 	/*
 	 * At this point need_resched() is true and we'll take the loop in
 	 * schedule(). The next pick is obviously going to be the stop task
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -2420,9 +2420,11 @@ static void pull_dl_task(struct rq *this
 		double_unlock_balance(this_rq, src_rq);
 
 		if (push_task) {
+			preempt_disable();
 			raw_spin_rq_unlock(this_rq);
 			stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
 					    push_task, &src_rq->push_work);
+			preempt_enable();
 			raw_spin_rq_lock(this_rq);
 		}
 	}
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -11299,13 +11299,15 @@ static int load_balance(int this_cpu, st
 				busiest->push_cpu = this_cpu;
 				active_balance = 1;
 			}
-			raw_spin_rq_unlock_irqrestore(busiest, flags);
 
+			preempt_disable();
+			raw_spin_rq_unlock_irqrestore(busiest, flags);
 			if (active_balance) {
 				stop_one_cpu_nowait(cpu_of(busiest),
 					active_load_balance_cpu_stop, busiest,
 					&busiest->active_balance_work);
 			}
+			preempt_enable();
 		}
 	} else {
 		sd->nr_balance_failed = 0;