2022-06-16 13:57:55

by Zqiang

Subject: [PATCH] rcu: Fix rcu_read_unlock_strict() strict QS reporting

When running a kerenl with CONFIG_PREEMPT=n and
CONFIG_RCU_STRICT_GRACE_PERIOD=y, the QS state will be reported
directly after exiting the last level of RCU critical section and
in non irqs-disable context, but maybe the CPU's rcu_data
structure's ->cpu_no_qs.b.norm is not cleared, as a result the
rcu_report_qs_rdp() will exit early, and not report QS state.

This commit will clear CPU's rcu_data structure's ->cpu_no_qs.b.norm
before invoke rcu_report_qs_rdp().

Signed-off-by: Zqiang <[email protected]>
---
kernel/rcu/tree_plugin.h | 1 +
1 file changed, 1 insertion(+)

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index b2c01919b92c..dc78726b993f 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -824,6 +824,7 @@ void rcu_read_unlock_strict(void)
if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread)
return;
rdp = this_cpu_ptr(&rcu_data);
+ rdp->cpu_no_qs.b.norm = false;
rcu_report_qs_rdp(rdp);
udelay(rcu_unlock_delay);
}
--
2.25.1

2022-06-17 00:20:22

by Paul E. McKenney

Subject: Re: [PATCH] rcu: Fix rcu_read_unlock_strict() strict QS reporting

On Thu, Jun 16, 2022 at 09:53:47PM +0800, Zqiang wrote:
> When running a kerenl with CONFIG_PREEMPT=n and
> CONFIG_RCU_STRICT_GRACE_PERIOD=y, the QS state will be reported
> directly after exiting the last level of RCU critical section and
> in non irqs-disable context, but maybe the CPU's rcu_data
> structure's ->cpu_no_qs.b.norm is not cleared, as a result the
> rcu_report_qs_rdp() will exit early, and not report QS state.
>
> This commit will clear CPU's rcu_data structure's ->cpu_no_qs.b.norm
> before invoke rcu_report_qs_rdp().
>
> Signed-off-by: Zqiang <[email protected]>

Good point, thank you! Queued for review and testing with the usual
wordsmithing shown below. As always, please check.

Thanx, Paul

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

commit cb58562318cf1fa7ad7c2c4c8d8e847c7942df66
Author: Zqiang <[email protected]>
Date: Thu Jun 16 21:53:47 2022 +0800

rcu: Fix rcu_read_unlock_strict() strict QS reporting

Kernels built with CONFIG_PREEMPT=n and CONFIG_RCU_STRICT_GRACE_PERIOD=y
report the quiescent state directly from the outermost rcu_read_unlock().
However, the current CPU's rcu_data structure's ->cpu_no_qs.b.norm
might still be set, in which case rcu_report_qs_rdp() will exit early,
thus failing to report quiescent state.

This commit therefore causes rcu_read_unlock_strict() to clear
CPU's rcu_data structure's ->cpu_no_qs.b.norm field before invoking
rcu_report_qs_rdp().

Signed-off-by: Zqiang <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index b2c01919b92c0..dc78726b993fd 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -824,6 +824,7 @@ void rcu_read_unlock_strict(void)
if (irqs_disabled() || preempt_count() || !rcu_state.gp_kthread)
return;
rdp = this_cpu_ptr(&rcu_data);
+ rdp->cpu_no_qs.b.norm = false;
rcu_report_qs_rdp(rdp);
udelay(rcu_unlock_delay);
}