Re: [PATCH] watchdog: Fix lockdep warning

[Helge Deller]

On 8/15/23 13:54, Petr Mladek wrote:
> Adding workqueue and lockdep people into Cc.
>
> On Fri 2023-08-11 19:11:46, Helge Deller wrote:
> > Fully initialize detector_work work struct to avoid this kernel warning
> > when lockdep is enabled:
> >
> >   =====================================
> >   WARNING: bad unlock balance detected!
> >   6.5.0-rc5+ #687 Not tainted
> >   -------------------------------------
> >   swapper/0/1 is trying to release lock (detector_work) at:
> >   [<000000004037e554>] __flush_work+0x60/0x658
> >   but there are no more locks to release!
> >
> >   other info that might help us debug this:
> >   no locks held by swapper/0/1.
> >
> >   stack backtrace:
> >   CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.5.0-rc5+ #687
> >   Hardware name: 9000/785/C3700
> >   Backtrace:
> >    [<0000000041455d5c>] print_unlock_imbalance_bug.part.0+0x20c/0x230
> >    [<000000004040d5e8>] lock_release+0x2e8/0x3f8
> >    [<000000004037e5cc>] __flush_work+0xd8/0x658
> >    [<000000004037eb7c>] flush_work+0x30/0x60
> >    [<000000004011f140>] lockup_detector_check+0x54/0x128
> >    [<0000000040306430>] do_one_initcall+0x9c/0x408
> >    [<0000000040102d44>] kernel_init_freeable+0x688/0x7f0
> >    [<000000004146df68>] kernel_init+0x64/0x3a8
> >    [<0000000040302020>] ret_from_kernel_thread+0x20/0x28
> >
> > Signed-off-by: Helge Deller <deller@xxxxxx>
> >
> > ---
> >
> > diff --git a/kernel/watchdog.c b/kernel/watchdog.c
> > index be38276a365f..eab0dfcfa3f9 100644
> > --- a/kernel/watchdog.c
> > +++ b/kernel/watchdog.c
> > @@ -1022,5 +1022,6 @@ void __init lockup_detector_init(void)
> >   	else
> >   		allow_lockup_detector_init_retry = true;
> >
> > +	INIT_WORK(&detector_work, lockup_detector_delay_init);
> >   	lockup_detector_setup();
> >   }
>
> Strange. The work is initialized when declared:
>
> static struct work_struct detector_work __initdata =
> 		__WORK_INITIALIZER(detector_work, lockup_detector_delay_init);
>
> , which should initialize the lockdep map as well. I would expect
> this is enough.

Right. It's enough...

> And I do not see this on x86_64 with CONFIG_PROVE_LOCKING=y
> on today's Linus' tree. And the flush_work() is always called
> from the lockup_detector_check() late init call.
>
> IMHO, it must be a bug somewhere else. Otherwise, many other
> users of DECLARE_WORK() or __WORK_INITIALIZER() would have
> the same problem.

Right. I should have replied earlier to this patch.
Andrew already dropped this patch here from his patch collection.

It turns out the problem is in function static_obj() from
lockdep, which doesn't recognize the __initdata section
for all platforms.

Please look at this patch on the LKML instead, which
fixes the real problem:
[PATCH v2] lockdep: Fix static memory detection even more

Thanks!
Helge