[PATCH v2 13/15] sched/deadline: Make start_dl_timer callers more robust

From: "Joel Fernandes (Google)" <joel@joelfernandes.org>
To: linux-kernel@vger.kernel.org, Ingo Molnar <mingo@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Juri Lelli <juri.lelli@redhat.com>,
	Vincent Guittot <vincent.guittot@linaro.org>,
	Dietmar Eggemann <dietmar.eggemann@arm.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
	Daniel Bristot de Oliveira <bristot@redhat.com>,
	Valentin Schneider <vschneid@redhat.com>
Cc: Suleiman Souhlal <suleiman@google.com>,
	Youssef Esmat <youssefesmat@google.com>,
	David Vernet <void@manifault.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	"Paul E . McKenney" <paulmck@kernel.org>,
	joseph.salisbury@canonical.com,
	Luca Abeni <luca.abeni@santannapisa.it>,
	Tommaso Cucinotta <tommaso.cucinotta@santannapisa.it>,
	Vineeth Pillai <vineeth@bitbyteword.org>,
	Shuah Khan <skhan@linuxfoundation.org>,
	Phil Auld <pauld@redhat.com>,
	"Joel Fernandes (Google)" <joel@joelfernandes.org>
Subject: [PATCH v2 13/15] sched/deadline: Make start_dl_timer callers more robust
Date: Tue, 12 Mar 2024 21:24:49 -0400	[thread overview]
Message-ID: <20240313012451.1693807-14-joel@joelfernandes.org> (raw)
In-Reply-To: <20240313012451.1693807-1-joel@joelfernandes.org>

For whatever reason, if start_dl_timer() returned 0 during replenish (it
did not start a new timer), then do not marked dl_defer_armed, because
we never really armed.

Further, we need to cancel any old timers,

This is similar to what dl_check_constrained_dl() does.

Add some guardrails for such situations.

Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
---
 kernel/sched/deadline.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index dbba95d364e2..e978e299381c 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -918,7 +918,16 @@ static void replenish_dl_entity(struct sched_dl_entity *dl_se)
 		if (!is_dl_boosted(dl_se)) {
 			dl_se->dl_defer_armed = 1;
 			dl_se->dl_throttled = 1;
-			start_dl_timer(dl_se);
+			if (!start_dl_timer(dl_se)) {
+				/*
+				 * If for whatever reason (delays), if a previous timer was
+				 * queued but not serviced, cancel it.
+				*/
+				hrtimer_try_to_cancel(&dl_se->dl_timer);
+				dl_se->dl_defer_armed = 0;
+				dl_se->dl_throttled = 0;
+				return;
+			}
 		}
 	}
 }
@@ -1465,7 +1474,14 @@ static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64
 		hrtimer_try_to_cancel(&dl_se->dl_timer);
 
 		replenish_dl_new_period(dl_se, dl_se->rq);
-		start_dl_timer(dl_se);
+
+		/*
+		 * Not being able to start the timer seems problematic. If it could not
+		 * be started for whatever reason, we need to "unthrottle" the DL server
+		 * and queue right away. Otherwise nothing might queue it. That's similar
+		 * to what enqueue_dl_entity() does on start_dl_timer==0. For now, just warn.
+		 */
+		WARN_ON_ONCE(!start_dl_timer(dl_se));
 
 		return;
 	}
-- 
2.34.1

