printk: Do not take console lock for console_flush_on_panic()

[ Upstream commit eacb04f ]

Currently console_flush_on_panic() will attempt to acquire the
console lock when flushing the buffer on panic. If it fails to
acquire the lock, it continues anyway because this is the last
chance to get any pending records printed.

The reason why the console lock was attempted at all was to
prevent any other CPUs from acquiring the console lock for
printing while the panic CPU was printing. But as of the
previous commit, non-panic CPUs will no longer attempt to
acquire the console lock in a panic situation. Therefore it is
no longer strictly necessary for a panic CPU to acquire the
console lock.

Avoiding taking the console lock when flushing in panic has
the additional benefit of avoiding possible deadlocks due to
semaphore usage in NMI context (semaphores are not NMI-safe)
and avoiding possible deadlocks if another CPU accesses the
semaphore and is stopped while holding one of the semaphore's
internal spinlocks.

Signed-off-by: John Ogness <john.ogness@linutronix.de>
Reviewed-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Petr Mladek <pmladek@suse.com>
Link: https://lore.kernel.org/r/20230717194607.145135-5-john.ogness@linutronix.de
Signed-off-by: Sasha Levin <sashal@kernel.org>

kernel/printk/printk.c

@@ -3120,14 +3120,24 @@ void console_unblank(void)
  */
 void console_flush_on_panic(enum con_flush_mode mode)
 {
+	bool handover;
+	u64 next_seq;
+
 	/*
-	 * If someone else is holding the console lock, trylock will fail
-	 * and may_schedule may be set.  Ignore and proceed to unlock so
-	 * that messages are flushed out.  As this can be called from any
-	 * context and we don't want to get preempted while flushing,
-	 * ensure may_schedule is cleared.
+	 * Ignore the console lock and flush out the messages. Attempting a
+	 * trylock would not be useful because:
+	 *
+	 *   - if it is contended, it must be ignored anyway
+	 *   - console_lock() and console_trylock() block and fail
+	 *     respectively in panic for non-panic CPUs
+	 *   - semaphores are not NMI-safe
+	 */
+
+	/*
+	 * If another context is holding the console lock,
+	 * @console_may_schedule might be set. Clear it so that
+	 * this context does not call cond_resched() while flushing.
 	 */
-	console_trylock();
 	console_may_schedule = 0;
 
 	if (mode == CONSOLE_REPLAY_ALL) {
@@ -3140,15 +3150,15 @@ void console_flush_on_panic(enum con_flush_mode mode)
 		cookie = console_srcu_read_lock();
 		for_each_console_srcu(c) {
 			/*
-			 * If the above console_trylock() failed, this is an
-			 * unsynchronized assignment. But in that case, the
+			 * This is an unsynchronized assignment, but the
 			 * kernel is in "hope and pray" mode anyway.
 			 */
 			c->seq = seq;
 		}
 		console_srcu_read_unlock(cookie);
 	}
-	console_unlock();
+
+	console_flush_all(false, &next_seq, &handover);
 }
 
 /*