On Tue, 10 May 2022, Thomas Gleixner wrote:
> It does matter whether the synchronization is done via
> __synchronize_hardirq() or via synchronize_irq(). The latter ensured
> that the thread reached the thread function and handled the pending
> wakeup _before_ kthread_stop() become true.
>
> So the fix is required to undo the damage created by 519cc8652b3a.
>
> The synchronize_irq() after free_irq() is a completely different
> problem.
Thank you for the clarification.
I was unsure if I missed something.
Thomas
