Re: [PATCH 10/16] freezer: fix set_freezable[_with_signal]() race

[Oleg Nesterov <oleg@redhat.com>]

On 08/19, Tejun Heo wrote:
>
> A kthread doing set_freezable*() may race with on-going PM freeze and
> the freezer might think all tasks are frozen while the new freezable
> kthread is marrily proceeding to execute code paths which aren't
> supposed to be executing during PM freeze.

Yes,

> +bool __set_freezable(bool with_signal)
> +{
> +	might_sleep();
> +
> +	/*
> +	 * Modify flags while holding freezer_lock.  This ensures the
> +	 * freezer notices that we aren't frozen yet or the freezing
> +	 * condition is visible to try_to_freeze() below.
> +	 */
> +	spin_lock_irq(&freezer_lock);
> +	current->flags &= ~PF_NOFREEZE;
> +	if (with_signal)
> +		current->flags &= ~PF_FREEZER_NOSIG;
> +	spin_unlock_irq(&freezer_lock);
> +
> +	return try_to_freeze();
> +}

You know, I was really, really puzzled by this change. Because it
"obviously can fix nothing", and try_to_freeze() makes no sense
(ignoring the very unlikely case when TIF_FREEZING was set right
after we drop freezer_lock).

But I guess this works along with 14/16 which removes TIF_FREEZING.

Oleg.