tty/sysrq: Convert show_lock to raw_spinlock_t
authorJulien Grall <>
Wed, 13 Mar 2019 11:40:34 +0000 (11:40 +0000)
committerGreg Kroah-Hartman <>
Wed, 27 Mar 2019 16:24:55 +0000 (01:24 +0900)
Systems which don't provide arch_trigger_cpumask_backtrace() will
invoke showacpu() from a smp_call_function() function which is invoked
with disabled interrupts even on -RT systems.

The function acquires the show_lock lock which only purpose is to
ensure that the CPUs don't print simultaneously. Otherwise the
output would clash and it would be hard to tell the output from CPUx
apart from CPUy.

On -RT the spin_lock() can not be acquired from this context. A
raw_spin_lock() is required. It will introduce the system's latency
by performing the sysrq request and other CPUs will block on the lock
until the request is done. This is okay because the user asked for a
backtrace of all active CPUs and under "normal circumstances in
production" this path should not be triggered.

Signed-off-by: Julien Grall <>
[ commit description]
Signed-off-by: Sebastian Andrzej Siewior <>
Acked-by: Sebastian Andrzej Siewior <>
Signed-off-by: Greg Kroah-Hartman <>

index fa0ce7d..59e82e6 100644 (file)
@@ -208,7 +208,7 @@ static struct sysrq_key_op sysrq_showlocks_op = {
 #ifdef CONFIG_SMP
-static DEFINE_SPINLOCK(show_lock);
+static DEFINE_RAW_SPINLOCK(show_lock);
 static void showacpu(void *dummy)
@@ -218,10 +218,10 @@ static void showacpu(void *dummy)
        if (idle_cpu(smp_processor_id()))
-       spin_lock_irqsave(&show_lock, flags);
+       raw_spin_lock_irqsave(&show_lock, flags);
        pr_info("CPU%d:\n", smp_processor_id());
        show_stack(NULL, NULL);
-       spin_unlock_irqrestore(&show_lock, flags);
+       raw_spin_unlock_irqrestore(&show_lock, flags);
 static void sysrq_showregs_othercpus(struct work_struct *dummy)