Re: KCSAN: data-race in find_next_bit / rcu_report_exp_cpu_mult

On Mon, Oct 07, 2019 at 08:33:53PM -0700, Paul E. McKenney wrote:
[...]

quoted

---
 kernel/rcu/tree_exp.h | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)

diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
index af7e7b9c86af..fb51752ac9a6 100644
--- a/kernel/rcu/tree_exp.h
+++ b/kernel/rcu/tree_exp.h

@@ -372,12 +372,10 @@ static void sync_rcu_exp_select_node_cpus(struct work_struct *wp)
 	raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 
 	/* IPI the remaining CPUs for expedited quiescent state. */
-	for_each_leaf_node_cpu_mask(rnp, cpu, rnp->expmask) {
+	for_each_leaf_node_cpu_mask(rnp, cpu, mask_ofl_ipi) {
 		unsigned long mask = leaf_node_cpu_bit(rnp, cpu);
 		struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
 
-		if (!(mask_ofl_ipi & mask))
-			continue;
 retry_ipi:
 		if (rcu_dynticks_in_eqs_since(rdp, rdp->exp_dynticks_snap)) {
 			mask_ofl_test |= mask;

This part I have already on -rcu branch "dev".

quoted

@@ -389,10 +387,10 @@ static void sync_rcu_exp_select_node_cpus(struct work_struct *wp)
 		}
 		ret = smp_call_function_single(cpu, rcu_exp_handler, NULL, 0);
 		put_cpu();
-		if (!ret) {
-			mask_ofl_ipi &= ~mask;
+		/* the CPU responses the IPI, and it will report QS itself */
+		if (!ret)
 			continue;
-		}
+
 		/* Failed, raced with CPU hotplug operation. */
 		raw_spin_lock_irqsave_rcu_node(rnp, flags);
 		if ((rnp->qsmaskinitnext & mask) &&

@@ -403,13 +401,12 @@ static void sync_rcu_exp_select_node_cpus(struct work_struct *wp)
 			schedule_timeout_uninterruptible(1);
 			goto retry_ipi;
 		}
-		/* CPU really is offline, so we can ignore it. */
-		if (!(rnp->expmask & mask))
-			mask_ofl_ipi &= ~mask;
+		/* CPU really is offline, and we need its QS. */
+		if (rnp->expmask & mask)
+			mask_ofl_test |= mask;
 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 	}
 	/* Report quiescent states for those that went offline. */
-	mask_ofl_test |= mask_ofl_ipi;
 	if (mask_ofl_test)
 		rcu_report_exp_cpu_mult(rnp, mask_ofl_test, false);
 }

Would you be willing to port this optimization on top of current -rcu
branch "dev" with an suitably modified commit message?

Sure, will do ;-)

Regards,
Boqun

							Thanx, Paul