Currently, rcu_gp_cleanup() traces the end of the old grace period after
the old grace period has officially ended. This might make intuitive
sense, but it also makes for confusing event-trace output because the
"end" trace displays not the old but instead the new grace-period number.
This commit therefore traces the end of an old grace period just before
that grace period officially ends.
Reported-by: Aravinda Prasad <aravi...@linux.vnet.ibm.com>
Signed-off-by: Paul E. McKenney <paul...@linux.ibm.com>
---
kernel/rcu/tree.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 429a46fb087a..61bae41b1afe 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -2035,9 +2035,9 @@ static void rcu_gp_cleanup(void)
rnp = rcu_get_root();
raw_spin_lock_irq_rcu_node(rnp); /* GP before ->gp_seq update. */
- /* Declare grace period done. */
- rcu_seq_end(&rcu_state.gp_seq);
+ /* Declare grace period done, trace first to use old GP number. */
trace_rcu_grace_period(rcu_state.name, rcu_state.gp_seq, TPS("end"));
+ rcu_seq_end(&rcu_state.gp_seq);
rcu_state.gp_state = RCU_GP_IDLE;
/* Check for GP requests since above loop. */
rdp = this_cpu_ptr(&rcu_data);
--
2.17.1
