[PATCH] nmi_backtrace: use printk_safe_flush_on_panic when triggering backtraces

[Hoeun Ryu]

From: Hoeun Ryu <hoeun.ryu@xxxxxxx>

 Use printk_safe_flush_on_panic() in nmi_trigger_cpumask_backtrace().
nmi_trigger_cpumask_backtrace() can be called in NMI context. For example the
function is called in watchdog_overflow_callback() if the flag of hardlockup
backtrace (sysctl_hardlockup_all_cpu_backtrace) is true and
watchdog_overflow_callback() function is called in NMI context on some
architectures.
 printk_safe_flush() eventually tries to lock logbuf_lock in vprintk_emit() but the
logbuf_lock can be already locked in preempted contexts (task or irq in this case)
or by other CPUs and it may cause deadlocks.
 By using printk_safe_flush_on_panic() instead of printk_safe_flush() in
in nmi_trigger_cpumask_backtrace(), the backtrace triggering CPU can access the lock
safely as the lock is reinitialized before calling printk_safe_flush().
 The messages in logbuf can be corrupted if the NMI context preempts other contexts
or ignores the lock held by other CPUs writing the buffer.

Signed-off-by: Hoeun Ryu <hoeun.ryu@xxxxxxx>
---
 lib/nmi_backtrace.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/lib/nmi_backtrace.c b/lib/nmi_backtrace.c
index 61a6b5a..0ace3c9 100644
--- a/lib/nmi_backtrace.c
+++ b/lib/nmi_backtrace.c
@@ -78,8 +78,11 @@ void nmi_trigger_cpumask_backtrace(const cpumask_t *mask,
 	/*
 	 * Force flush any remote buffers that might be stuck in IRQ context
 	 * and therefore could not run their irq_work.
+	 * Call nmi-safe version of printk_safe_flush() because this function can be
+	 * called in NMI context like watchdog_overflow_callback() if
+	 * sysctl_hardlockup_all_cpu_backtrace is true.
 	 */
-	printk_safe_flush();
+	printk_safe_flush_on_panic();
 
 	clear_bit_unlock(0, &backtrace_flag);
 	put_cpu();
-- 
2.1.4