2024-02-18 19:01:30

by John Ogness

Subject: [PATCH printk v2 25/26] rcu: Mark emergency section in rcu stalls

Mark an emergency section within print_other_cpu_stall(), where
RCU stall information is printed. In this section, the CPU will
not perform console output for the printk() calls. Instead, a
flushing of the console output is triggered when exiting the
emergency section.

Signed-off-by: John Ogness <[email protected]>
---
kernel/rcu/tree_stall.h | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
index ac8e86babe44..efb2be8939a2 100644
--- a/kernel/rcu/tree_stall.h
+++ b/kernel/rcu/tree_stall.h
@@ -9,6 +9,7 @@

#include <linux/kvm_para.h>
#include <linux/rcu_notifier.h>
+#include <linux/console.h>

//////////////////////////////////////////////////////////////////////////////
//
@@ -604,6 +605,8 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
if (rcu_stall_is_suppressed())
return;

+ nbcon_cpu_emergency_enter();
+
/*
* OK, time to rat on our buddy...
* See Documentation/RCU/stallwarn.rst for info on how to debug
@@ -658,6 +661,8 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
panic_on_rcu_stall();

rcu_force_quiescent_state(); /* Kick them all. */
+
+ nbcon_cpu_emergency_exit();
}

static void print_cpu_stall(unsigned long gps)
--
2.39.2

2024-03-01 15:13:40

by Petr Mladek

Subject: Re: [PATCH printk v2 25/26] rcu: Mark emergency section in rcu stalls

On Sun 2024-02-18 20:03:25, John Ogness wrote:
> Mark an emergency section within print_other_cpu_stall(), where
> RCU stall information is printed. In this section, the CPU will
> not perform console output for the printk() calls. Instead, a
> flushing of the console output is triggered when exiting the
> emergency section.
>
> Signed-off-by: John Ogness <[email protected]>

Reviewed-by: Petr Mladek <[email protected]>

I was just curious about one thing. But it seems to work well.

print_other_cpu_stall() print backtraces on other CPUs via NMI.
The other CPUs would not see the emergency context. They would
call defer_console_output() because they are in NMI. As a result:

+ Legacy consoles might be flushed on other CPUs even before
nbcon_cpu_emergency_exit() gets called.

+ nbcon consoles might still be flushed by the printk kthread
until all messages get flushed directly by nbcon_cpu_emergency_exit()

As I wrote. The behavior is corrent. It was just not obvious to me.


> --- a/kernel/rcu/tree_stall.h
> +++ b/kernel/rcu/tree_stall.h
> @@ -9,6 +9,7 @@
>
> #include <linux/kvm_para.h>
> #include <linux/rcu_notifier.h>
> +#include <linux/console.h>
>
> //////////////////////////////////////////////////////////////////////////////
> //
> @@ -604,6 +605,8 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
> if (rcu_stall_is_suppressed())
> return;
>
> + nbcon_cpu_emergency_enter();
> +
> /*
> * OK, time to rat on our buddy...
> * See Documentation/RCU/stallwarn.rst for info on how to debug
> @@ -658,6 +661,8 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
> panic_on_rcu_stall();
>
> rcu_force_quiescent_state(); /* Kick them all. */
> +
> + nbcon_cpu_emergency_exit();
> }
>
> static void print_cpu_stall(unsigned long gps)

Best Regards,
Petr