[PATCH tip/core/rcu 0/3] RCU stall-warning updates for v5.8

[PATCH tip/core/rcu 0/3] RCU stall-warning updates for v5.8

[Paul E. McKenney]

Hello!

This series provides a few updates to RCU CPU stall-warning messages.

1.	Use data_race() for RCU expedited CPU stall-warning prints.

2.	When GP kthread is starved, tag idle threads as false positives.
	After all, it is not the idle threads' fault that the GP
	kthread hasn't reported their extended quiescent states!

3.	Remove self-stack-trace when all quiescent states seen.

							Thanx, Paul

------------------------------------------------------------------------

 tree_exp.h   |    8 ++++----
 tree_stall.h |   26 +++++++++++++++++++-------
 2 files changed, 23 insertions(+), 11 deletions(-)

[PATCH tip/core/rcu 1/3] rcu: Use data_race() for RCU expedited CPU stall-warning prints

[paulmck]

From: "Paul E. McKenney" <paulmck@kernel.org>

Although the accesses used to determine whether or not an expedited
stall should be printed are an integral part of the concurrency algorithm
governing use of the corresponding variables, the values that are simply
printed are ancillary.  As such, it is best to use data_race() for these
accesses in order to provide the greatest latitude in the use of KCSAN
for the other accesses that are an integral part of the algorithm.  This
commit therefore changes the relevant uses of READ_ONCE() to data_race().

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
---
 kernel/rcu/tree_exp.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
index 1a617b9..e1a7986 100644
--- a/kernel/rcu/tree_exp.h
+++ b/kernel/rcu/tree_exp.h
@@ -542,8 +542,8 @@ static void synchronize_rcu_expedited_wait(void)
 		}
 		pr_cont(" } %lu jiffies s: %lu root: %#lx/%c\n",
 			jiffies - jiffies_start, rcu_state.expedited_sequence,
-			READ_ONCE(rnp_root->expmask),
-			".T"[!!rnp_root->exp_tasks]);
+			data_race(rnp_root->expmask),
+			".T"[!!data_race(rnp_root->exp_tasks)]);
 		if (ndetected) {
 			pr_err("blocking rcu_node structures:");
 			rcu_for_each_node_breadth_first(rnp) {
@@ -553,8 +553,8 @@ static void synchronize_rcu_expedited_wait(void)
 					continue;
 				pr_cont(" l=%u:%d-%d:%#lx/%c",
 					rnp->level, rnp->grplo, rnp->grphi,
-					READ_ONCE(rnp->expmask),
-					".T"[!!rnp->exp_tasks]);
+					data_race(rnp->expmask),
+					".T"[!!data_race(rnp->exp_tasks)]);
 			}
 			pr_cont("\n");
 		}
-- 
2.9.5

[PATCH tip/core/rcu 2/3] rcu: When GP kthread is starved, tag idle threads as false positives

[paulmck]

From: "Paul E. McKenney" <paulmck@kernel.org>

If the grace-period kthread is starved, idle threads' extended quiescent
states are not reported.  These idle threads thus wrongly appear to
be blocking the current grace period.  This commit therefore tags such
idle threads as probable false positives when the grace-period kthread
is being starved.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
---
 kernel/rcu/tree_stall.h | 24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)

diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
index 119ed6a..4da41a6 100644
--- a/kernel/rcu/tree_stall.h
+++ b/kernel/rcu/tree_stall.h
@@ -299,6 +299,16 @@ static const char *gp_state_getname(short gs)
 	return gp_state_names[gs];
 }
 
+/* Is the RCU grace-period kthread being starved of CPU time? */
+static bool rcu_is_gp_kthread_starving(unsigned long *jp)
+{
+	unsigned long j = jiffies - READ_ONCE(rcu_state.gp_activity);
+
+	if (jp)
+		*jp = j;
+	return j > 2 * HZ;
+}
+
 /*
  * Print out diagnostic information for the specified stalled CPU.
  *
@@ -313,6 +323,7 @@ static const char *gp_state_getname(short gs)
 static void print_cpu_stall_info(int cpu)
 {
 	unsigned long delta;
+	bool falsepositive;
 	char fast_no_hz[72];
 	struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
 	char *ticks_title;
@@ -333,7 +344,9 @@ static void print_cpu_stall_info(int cpu)
 	}
 	print_cpu_stall_fast_no_hz(fast_no_hz, cpu);
 	delta = rcu_seq_ctr(rdp->mynode->gp_seq - rdp->rcu_iw_gp_seq);
-	pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%#lx softirq=%u/%u fqs=%ld %s\n",
+	falsepositive = rcu_is_gp_kthread_starving(NULL) &&
+			rcu_dynticks_in_eqs(rcu_dynticks_snap(rdp));
+	pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%ld/%#lx softirq=%u/%u fqs=%ld %s%s\n",
 	       cpu,
 	       "O."[!!cpu_online(cpu)],
 	       "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)],
@@ -345,8 +358,9 @@ static void print_cpu_stall_info(int cpu)
 	       rcu_dynticks_snap(rdp) & 0xfff,
 	       rdp->dynticks_nesting, rdp->dynticks_nmi_nesting,
 	       rdp->softirq_snap, kstat_softirqs_cpu(RCU_SOFTIRQ, cpu),
-	       READ_ONCE(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
-	       fast_no_hz);
+	       data_race(rcu_state.n_force_qs) - rcu_state.n_force_qs_gpstart,
+	       fast_no_hz,
+	       falsepositive ? " (false positive?)" : "");
 }
 
 /* Complain about starvation of grace-period kthread.  */
@@ -355,8 +369,7 @@ static void rcu_check_gp_kthread_starvation(void)
 	struct task_struct *gpk = rcu_state.gp_kthread;
 	unsigned long j;
 
-	j = jiffies - READ_ONCE(rcu_state.gp_activity);
-	if (j > 2 * HZ) {
+	if (rcu_is_gp_kthread_starving(&j)) {
 		pr_err("%s kthread starved for %ld jiffies! g%ld f%#x %s(%d) ->state=%#lx ->cpu=%d\n",
 		       rcu_state.name, j,
 		       (long)rcu_seq_current(&rcu_state.gp_seq),
@@ -364,6 +377,7 @@ static void rcu_check_gp_kthread_starvation(void)
 		       gp_state_getname(rcu_state.gp_state), rcu_state.gp_state,
 		       gpk ? gpk->state : ~0, gpk ? task_cpu(gpk) : -1);
 		if (gpk) {
+			pr_err("\tUnless %s kthread gets sufficient CPU time, OOM is now expected behavior.\n", rcu_state.name);
 			pr_err("RCU grace-period kthread stack dump:\n");
 			sched_show_task(gpk);
 			wake_up_process(gpk);
-- 
2.9.5

[PATCH tip/core/rcu 3/3] rcu: Remove self-stack-trace when all quiescent states seen

[paulmck]

From: "Paul E. McKenney" <paulmck@kernel.org>

When all quiescent states have been seen, it is normally the grace-period
kthread that is in trouble.  Although the existing stack trace from
the current CPU might possibly provide useful information, experience
indicates that there is too much noise for this to be worthwhile.

This commit therefore removes this stack trace from the output.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
---
 kernel/rcu/tree_stall.h | 2 --
 1 file changed, 2 deletions(-)

diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
index 4da41a6..535762b 100644
--- a/kernel/rcu/tree_stall.h
+++ b/kernel/rcu/tree_stall.h
@@ -440,8 +440,6 @@ static void print_other_cpu_stall(unsigned long gp_seq)
 			       rcu_state.name, j - gpa, j, gpa,
 			       READ_ONCE(jiffies_till_next_fqs),
 			       rcu_get_root()->qsmask);
-			/* In this case, the current CPU might be at fault. */
-			sched_show_task(current);
 		}
 	}
 	/* Rewrite if needed in case of slow consoles. */
-- 
2.9.5