Re: [PATCH v3 1/3] init / kthread: add module_long_probe_init() and module_long_probe_exit()

On 08/18, Takashi Iwai wrote:

At Mon, 18 Aug 2014 14:22:17 +0200,
Oleg Nesterov wrote:

quoted

On 08/18, Takashi Iwai wrote:

quoted

#define module_long_probe_init(initfn)				\
	static int _long_probe_##initfn(void *arg)		\
	{							\
		module_put_and_exit(initfn());			\
		return 0;					\
	}							\
	static int __init __long_probe_##initfn(void)		\
	{							\
		struct task_struct *__init_thread =		\
			kthread_create(_long_probe_##initfn,	\
				       NULL, #initfn);		\
		if (IS_ERR(__init_thread))			\
			return PTR_ERR(__init_thread);		\
		__module_get(THIS_MODULE);			\
		wake_up_process(__init_thread);			\
		return 0;					\
	}							\
	module_init(__long_probe_##initfn)

... and module_exit() remains identical as the normal version.

Aaaah. This is not true, module_exit() should not call exitfn() if initfn()
fails... So _long_probe_##initfn() needs to save the error code which should
be checked by module_exit().

Oh, right.  So we need a reference in the module exit path in anyway,

We only need to save the error code,

	static int _long_probe_retval;

	static int _long_probe_##initfn(void *arg)
	{
		_long_probe_retval = initfn();
		module_put_and_exit(0); /* noreturn */
	}

	static void __long_probe_##exitfn(void)
	{
		if (!_long_probe_retval)
			exitfn();
	}

and Luis' version might be shorter in the end.

I dont't think that "shorter" does matter in this case. The real difference
is sys_delete_module() behaviour if it is called before initfn() completes.

And, again, I do not really know which version is better.

Oleg.