kernel thread exit race

[Nikita Danilov <Nikita@Namesys.COM>]

Hello,

what is the politically correct way to exit from a kernel thread daemon
without module unload races?

Currently most kernel threads exit with something like

	wake_up(&daemon_done_wait_queue);
	return 0;

(or complete() in stead of wake_up()). Problem is that thread waiting
for daemon shutdown can start running on another CPU while daemon is
still executing and unload module, in particular unmapping page with
daemon code.

Reiserfs used to do something like

    /*
     * BKL will be released in do_exit()
     */
    lock_kernel();
	wake_up(&daemon_done_wait_queue);
	return 0;

and wait for daemon completion under BKL so that when waiter resumes,
daemon is definitely not executing module code. This looks like a hack,
though. Is there a better solution?

Nikita.