Re: [PATCH] devtmpfs: Add missing lockdep annotation

[Helge Deller]

* Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>:
> On Fri, Aug 11, 2023 at 06:02:29PM +0200, Helge Deller wrote:
> > Add the lockdep annotation to the setup_done completer to avoid this
> > kernel warning:
> >
> >  Backtrace:
> >   [<000000004030bcd0>] show_stack+0x74/0xb0
> >   [<000000004146c63c>] dump_stack_lvl+0x104/0x180
> >   [<000000004146c6ec>] dump_stack+0x34/0x48
> >   [<000000004040e5b4>] register_lock_class+0xd24/0xd30
> >   [<000000004040c21c>] __lock_acquire.isra.0+0xb4/0xac8
> >   [<000000004040cd60>] lock_acquire+0x130/0x298
> >   [<000000004147095c>] _raw_spin_lock_irq+0x60/0xb8
> >   [<0000000041474a4c>] wait_for_completion+0xa0/0x2d0
> >   [<000000004015d9f4>] devtmpfs_init+0x1e0/0x2b8
> >   [<000000004015d0e4>] driver_init+0x68/0x1b8
> >   [<0000000040102b68>] kernel_init_freeable+0x4ac/0x7f0
> >   [<000000004146df68>] kernel_init+0x64/0x3a8
> >   [<0000000040302020>] ret_from_kernel_thread+0x20/0x28
> >
> > Signed-off-by: Helge Deller <deller@xxxxxx>
> >
> > diff --git a/drivers/base/devtmpfs.c b/drivers/base/devtmpfs.c
> > index b848764ef018..f98d58b0225c 100644
> > --- a/drivers/base/devtmpfs.c
> > +++ b/drivers/base/devtmpfs.c
> > @@ -462,6 +462,7 @@ int __init devtmpfs_init(void)
> >  		return err;
> >  	}
> >
> > +	init_completion(&setup_done);
> >  	thread = kthread_run(devtmpfsd, &err, "kdevtmpfs");
> >  	if (!IS_ERR(thread)) {
> >  		wait_for_completion(&setup_done);
>
> What changed to required this now?  What commit id does this fix?  Why
> doesn't the declaration of DECLARE_COMPLETION() initialize this properly
> for us already?

You're right!

The problem is not the missing init_completion() call.

That part of the kernel warning indicates the real problem:
 INFO: trying to register non-static key.

Lockdep uses static_obj() to check if the given lock is in "static" memory,
aka inside a valid memory region: from &_stext to &_end.

The "setup_done" completer is used at bootup only, and thus located in
the __initdata section. And on parisc the __initdata section is outside of
the &_stext to &_end range, so the static_obj() check fails.

The patch below fixes it.
I'll send a proper patch to the mailing list in a moment.

Thanks!
Helge


diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
index 111607d91489..aa99245d8e12 100644
--- a/kernel/locking/lockdep.c
+++ b/kernel/locking/lockdep.c
@@ -855,6 +855,15 @@ static int static_obj(const void *obj)
 	if (is_kernel_percpu_address(addr))
 		return 1;

+	/*
+	 * in initdata section and used during bootup only?
+	 * NOTE: On some platforms the initdata section is
+	 * outside of the _stext ... _end range.
+	 */
+	if (system_state < SYSTEM_FREEING_INITMEM &&
+		init_section_contains((void*)addr, 1))
+		return 1;
+
 	/*
 	 * module static or percpu var?
 	 */