dump_stack: Do not get cpu_sync for panic CPU

dump_stack() is called in panic(). If for some reason another CPU
is holding the printk_cpu_sync and is unable to release it, the
panic CPU will be unable to continue and print the stacktrace.

Since non-panic CPUs are not allowed to store new printk messages
anyway, there is no need to synchronize the stacktrace output in
a panic situation.

For the panic CPU, do not get the printk_cpu_sync because it is
not needed and avoids a potential deadlock scenario in panic().

Link: https://lore.kernel.org/lkml/ZcIGKU8sxti38Kok@alley
Signed-off-by: John Ogness <john.ogness@linutronix.de>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Link: https://lore.kernel.org/r/20240207134103.1357162-15-john.ogness@linutronix.de
Signed-off-by: Petr Mladek <pmladek@suse.com>

1 files changed, 13 insertions, 3 deletions

diff --git a/lib/dump_stack.c b/lib/dump_stack.c
index 83471e81501a..222c6d6c8281 100644
--- a/lib/dump_stack.c
+++ b/lib/dump_stack.c
@@ -96,15 +96,25 @@ static void __dump_stack(const char *log_lvl)
  */
 asmlinkage __visible void dump_stack_lvl(const char *log_lvl)
 {
+	bool in_panic = this_cpu_in_panic();
 	unsigned long flags;
 
 	/*
 	 * Permit this cpu to perform nested stack dumps while serialising
-	 * against other CPUs
+	 * against other CPUs, unless this CPU is in panic.
+	 *
+	 * When in panic, non-panic CPUs are not permitted to store new
+	 * printk messages so there is no need to synchronize the output.
+	 * This avoids potential deadlock in panic() if another CPU is
+	 * holding and unable to release the printk_cpu_sync.
 	 */
-	printk_cpu_sync_get_irqsave(flags);
+	if (!in_panic)
+		printk_cpu_sync_get_irqsave(flags);
+
 	__dump_stack(log_lvl);
-	printk_cpu_sync_put_irqrestore(flags);
+
+	if (!in_panic)
+		printk_cpu_sync_put_irqrestore(flags);
 }
 EXPORT_SYMBOL(dump_stack_lvl);