[PATCH] rcu/tree: Defer setting of jiffies during stall reset

[Joel Fernandes (Google)]

There are instances where rcu_cpu_stall_reset() is called when jiffies
did not get a chance to update for a long time. Before jiffies is
updated, the CPU stall detector can go off triggering false-positives
where a just-started grace period appears to be ages old. In the past,
we disabled stall detection in rcu_cpu_stall_reset() however this got
changed [1]. This is resulting in false-positives in KGDB usecase [2].

Fix this by deferring the update of jiffies to the third run of the FQS
loop. This is more robust, as, even if rcu_cpu_stall_reset() is called
just before jiffies is read, we would end up pushing out the jiffies
read by 3 more FQS loops. Meanwhile the CPU stall detection will be
delayed and we will not get any false positives.

[1] https://lore.kernel.org/all/20210521155624.174524-2-senozhatsky@xxxxxxxxxxxx/
[2] https://lore.kernel.org/all/20230814020045.51950-2-chenhuacai@xxxxxxxxxxx/

Tested with rcutorture.cpu_stall option as well to verify stall behavior
with/without patch.

Reported-by: Huacai Chen <chenhuacai@xxxxxxxxxxx>
Closes: https://lore.kernel.org/all/20230814020045.51950-2-chenhuacai@xxxxxxxxxxx/
Suggested-by: Paul  McKenney <paulmck@xxxxxxxxxx>
Cc: Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx>
Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
---
 kernel/rcu/tree.c       | 11 +++++++++++
 kernel/rcu/tree.h       |  4 ++++
 kernel/rcu/tree_stall.h | 20 ++++++++++++++++++--
 3 files changed, 33 insertions(+), 2 deletions(-)

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 1449cb69a0e0..9273f2318ea1 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -1552,10 +1552,21 @@ static bool rcu_gp_fqs_check_wake(int *gfp)
  */
 static void rcu_gp_fqs(bool first_time)
 {
+	int nr_fqs = READ_ONCE(rcu_state.nr_fqs_jiffies_stall);
 	struct rcu_node *rnp = rcu_get_root();
 
 	WRITE_ONCE(rcu_state.gp_activity, jiffies);
 	WRITE_ONCE(rcu_state.n_force_qs, rcu_state.n_force_qs + 1);
+
+	WARN_ON_ONCE(nr_fqs > 3);
+	if (nr_fqs) {
+		if (nr_fqs == 1) {
+			WRITE_ONCE(rcu_state.jiffies_stall,
+				   jiffies + rcu_jiffies_till_stall_check());
+		}
+		WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, --nr_fqs);
+	}
+
 	if (first_time) {
 		/* Collect dyntick-idle snapshots. */
 		force_qs_rnp(dyntick_save_progress_counter);
diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
index 192536916f9a..e9821a8422db 100644
--- a/kernel/rcu/tree.h
+++ b/kernel/rcu/tree.h
@@ -386,6 +386,10 @@ struct rcu_state {
 						/*  in jiffies. */
 	unsigned long jiffies_stall;		/* Time at which to check */
 						/*  for CPU stalls. */
+	int nr_fqs_jiffies_stall;		/* Number of fqs loops after
+						 * which read jiffies and set
+						 * jiffies_stall. Stall
+						 * warnings disabled if !0. */
 	unsigned long jiffies_resched;		/* Time at which to resched */
 						/*  a reluctant CPU. */
 	unsigned long n_force_qs_gpstart;	/* Snapshot of n_force_qs at */
diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
index b10b8349bb2a..a2fa6b22e248 100644
--- a/kernel/rcu/tree_stall.h
+++ b/kernel/rcu/tree_stall.h
@@ -149,12 +149,17 @@ static void panic_on_rcu_stall(void)
 /**
  * rcu_cpu_stall_reset - restart stall-warning timeout for current grace period
  *
+ * To perform the reset request from the caller, disable stall detection until
+ * 3 fqs loops have passed. This is required to ensure a fresh jiffies is
+ * loaded.  It should be safe to do from the fqs loop as enough timer
+ * interrupts and context switches should have passed.
+ *
  * The caller must disable hard irqs.
  */
 void rcu_cpu_stall_reset(void)
 {
-	WRITE_ONCE(rcu_state.jiffies_stall,
-		   jiffies + rcu_jiffies_till_stall_check());
+	WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 3);
+	WRITE_ONCE(rcu_state.jiffies_stall, ULONG_MAX);
 }
 
 //////////////////////////////////////////////////////////////////////////////
@@ -170,6 +175,7 @@ static void record_gp_stall_check_time(void)
 	WRITE_ONCE(rcu_state.gp_start, j);
 	j1 = rcu_jiffies_till_stall_check();
 	smp_mb(); // ->gp_start before ->jiffies_stall and caller's ->gp_seq.
+	WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 0);
 	WRITE_ONCE(rcu_state.jiffies_stall, j + j1);
 	rcu_state.jiffies_resched = j + j1 / 2;
 	rcu_state.n_force_qs_gpstart = READ_ONCE(rcu_state.n_force_qs);
@@ -725,6 +731,16 @@ static void check_cpu_stall(struct rcu_data *rdp)
 	    !rcu_gp_in_progress())
 		return;
 	rcu_stall_kick_kthreads();
+
+	/*
+	 * Check if it was requested (via rcu_cpu_stall_reset()) that the FQS
+	 * loop has to set jiffies to ensure a non-stale jiffies value. This
+	 * is required to have good jiffies value after coming out of long
+	 * breaks of jiffies updates. Not doing so can cause false positives.
+	 */
+	if (READ_ONCE(rcu_state.nr_fqs_jiffies_stall) > 0)
+		return;
+
 	j = jiffies;
 
 	/*
-- 
2.42.0.rc1.204.g551eb34607-goog