Re: [PATCH printk v3 14/15] printk: extend console_lock for proper kthread support

[John Ogness <john.ogness@linutronix.de>]

On 2022-04-22, Petr Mladek <pmladek@suse.com> wrote:
> IMHO, it is actually a generic problem of the complex locking scheme
> when there are too many combinations of the protected data.

Sure. We are in a delicate situation of continuing to support the old
locking scheme while transitioning to a new one.

> In the current state, the problem seems to be only with CON_ENABLED
> flag but there might be other hidden races in the future.
>
> IMHO, it would be much easier when there are the following rules:
>
>    + console_lock() blocks taking con->lock
>    + con->lock blocks taking console_lock()
>    + Different con->lock might be taken in parallel
>
> The result would be:
>
>    + global variables need to be guarded by the big console_lock()
>    + con->lock should be enough to guard per-console variables
>    + the big console_lock() would serialize also access to
>      per-console variables.

It looks like you are talking about nested locking. This was my original
idea but I had problems relating to kthread stopping. However, the code
has changed a lot since then and now when I look at it, it does not look
like it would be a problem. Getting rid of CON_THD_BLOCKED would greatly
simplify the relationship between console_lock and kthreads.

For this we would need the console list to become a list_head so that it
is doubly linked (in order to unlock in reverse order). That probably
would be a good idea anyway. It is a bit bizarre that printk implements
its own linked list.

> Of course, it is not that simple. I am not 100% that we could
> even achieve this.

It just might be that simple. I will explore it again.

> Anyway, I think about the following wrapper:
>
> void single_console_lock(struct console *con)
> {
> 	for (;;) {
> 		error = wait_event_interruptible(log_wait,
> 					con->flags & CON_THB_BLOCKED);
>
> 		if (error)
> 			continue;
>
> 		mutex_lock(&con->lock);
>
> 		if (!con->flags & CON_THB_BLOCKED)
> 			break;
>
> 		mutex_unlock(&con->lock);
> 	}
> }
>
> void single_console_unlock(struct console *con)
> {
> 	mutex_unlock(&con->lock);
> }
>
> We should use it everywhere instead of the simple mutex_lock(con->lock)
> and mutex_lock(con->lock). And we could remove mutex_lock()/unlock()
> from code called under the big console_lock().

Hmmm. Waiting on @log_wait is not correct. A @log_wait wakeup with the
kthread already in the blocked state is unusual. There would need to be
a per-console waitqueue for when the kthread unlocks its mutex.

Maybe something like:

void single_console_lock(struct console *con)
{
	for (;;) {
		error = wait_event_interruptible(con->lock_wait,
				!(con->flags & CON_THB_BLOCKED));
		if (error)
			continue;

		mutex_lock(&con->lock);

		if (!(con->flags & CON_THB_BLOCKED))
			break;

		mutex_unlock(&con->lock);
	}
}

And in printk_kthread_func(), after the kthread unlocks its con->lock,
it calls:

if (wq_has_sleeper(&con->lock_wait))
	wake_up_interruptible_all(&con->lock_wait);

But single_console_lock() would not be allowed to be called under
console_lock(), so I don't see how it is useful. con->flags is always
modified under @console_sem to make sure the console does not disappear.

Anyway, I will first look into the nested locking solution. That seems
more promising to me and it would go a long way to simplify the locking
hierarchy.

John