Re: [PATCH] driver core: add wait event for deferred probe

[Grant Likely <grant.likely@secretlab.ca>]

On Tue, 12 Feb 2013 10:52:10 +0800, Haojian Zhuang <haojian.zhuang@linaro.org> wrote:
> On 12 February 2013 07:10, Andrew Morton <akpm@linux-foundation.org> wrote:
> > On Sun, 10 Feb 2013 00:57:57 +0800
> > Haojian Zhuang <haojian.zhuang@linaro.org> wrote:
> >
> >> do_initcalls() could call all driver initialization code in kernel_init
> >> thread. It means that probe() function will be also called from that
> >> time. After this, kernel could access console & release __init section
> >> in the same thread.
> >>
> >> But if device probe fails and moves into deferred probe list, a new
> >> thread is created to reinvoke probe. If the device is serial console,
> >> kernel has to open console failure & release __init section before
> >> reinvoking failure. Because there's no synchronization mechanism.
> >> Now add wait event to synchronize after do_initcalls().
> >
> > It sounds like this:
> >
> > static int __ref kernel_init(void *unused)
> > {
> >         kernel_init_freeable();
> >         /* need to finish all async __init code before freeing the memory */
> >         async_synchronize_full();
> >
> > is designed to prevent the problem you describe?
> >
> It can't prevent the problem that I described. Because deferred_probe()
> is introduced recently.
> 
> All synchronization should be finished just after do_initcalls(). Since
> load_default_modules() is also called in the end of kernel_init_freeable(),
> I'm not sure that whether I could remove async_synchronize_full()
> here. So I didn't touch it.
> 
> >> --- a/init/main.c
> >> +++ b/init/main.c
> >> @@ -786,6 +786,7 @@ static void __init do_basic_setup(void)
> >>       do_ctors();
> >>       usermodehelper_enable();
> >>       do_initcalls();
> >> +     wait_for_device_probe();
> >>  }
> >
> > Needs a nice comment here explaining what's going on.
> 
> No problem. I'll add comment here.

Actually, this approach will create new problems. There is no guarantee
that a given device will be able to initialize before exiting
do_basic_setup(). If, for instance, a device depends on a resource
provided by a module, then it will just keep deferring. In that case
you've got a hung kernel.

I think what you really want is the following:

 static int deferred_probe_initcall(void)
 {
	deferred_wq = create_singlethread_workqueue("deferwq");
	if (WARN_ON(!deferred_wq))
		return -ENOMEM;

	driver_deferred_probe_enable = true;
+	deferred_probe_work_func(NULL);
-	driver_deferred_probe_trigger();
	return 0;
 }
 late_initcall(deferred_probe_initcall);

Or something similar. That would guarantee that as many passes as are needed
(which in practical terms only means a couple) for device probing to
settle down before exiting the initcall processing. That should achieve
the effect you desire.

It still masks the __init section issue by making it a lot less likely,
but it does ensure that all of the built-in driver dependency order
issues are processed before continuing on to userspace.

g.

-- 
Grant Likely, B.Sc, P.Eng.
Secret Lab Technologies, Ltd.