Re: new (common) WARNING message?

[Ingo Molnar <mingo@elte.hu>]

* Stefan Richter <stefanr@s5r6.in-berlin.de> wrote:

> Mike Travis wrote:
> > I've been seeing these on and off for a while now.  They don't
> > seem to affect anything but perhaps I should track them down?
> > 
> > Do you know what it means?  (I'm asking you because evidently
> > debugobjects.c is from you?  I read the good explanation in the
> > changelog of 3ac7fe5a, but it didn't clue me into what's going
> > on. ;-)
> > 
> > Thanks,
> > Mike
> > --
> > 
> > [ 1294.973592] ODEBUG: object is on stack, but not annotated
> > [ 1294.974067] ------------[ cut here ]------------
> > [ 1294.974067] WARNING: at /mdata/lwork/polaris3/travis/workareas/ingo-latest/linux-2.6-for-ingo/lib/debugobjects.c:253 __debug_object_init+0x27c/0x2f0()
> > [ 1294.974067] Hardware name: X7DBT
> > [ 1294.974067] Pid: 4150, comm: offline-test Tainted: G        W  2.6.29-rc1-4k-defconfig.01151634-00255-gc99dbbe-dirty #89
> > [ 1294.974067] Call Trace:
> > [ 1294.974067]  [<ffffffff81047e3f>] warn_slowpath+0xd3/0xf2
> > [ 1294.974067]  [<ffffffff8106c64f>] ? trace_hardirqs_off+0xd/0xf
> > [ 1294.974067]  [<ffffffff815ca6ac>] ? _spin_unlock_irqrestore+0x3d/0x4c
> > [ 1294.974067]  [<ffffffff8125bebe>] ? __debug_check_no_obj_freed+0xf6/0x14c
> > [ 1294.974067]  [<ffffffff81048d3c>] ? printk+0x6c/0x6e
> > [ 1294.974067]  [<ffffffff8125c449>] ? __debug_object_init+0xf9/0x2f0
> > [ 1294.974067]  [<ffffffff8125b3ae>] ? _raw_spin_unlock+0xc7/0xf6
> > [ 1294.974067]  [<ffffffff8125c5cc>] __debug_object_init+0x27c/0x2f0
> > [ 1294.974067]  [<ffffffff8125c66d>] debug_object_init+0x14/0x17
> > [ 1294.974067]  [<ffffffff81052320>] init_timer+0x1d/0x2a
> > [ 1294.974067]  [<ffffffff81025077>] hpet_cpuhp_notify+0xad/0x124
> > [ 1294.974067]  [<ffffffff81025233>] ? hpet_work+0x0/0x21a
> > [ 1294.974067]  [<ffffffff81062057>] notifier_call_chain+0x33/0x5b
> > [ 1294.974067]  [<ffffffff810620ad>] __raw_notifier_call_chain+0xe/0x10
> > [ 1294.974067]  [<ffffffff810620c3>] raw_notifier_call_chain+0x14/0x16
> > [ 1294.974067]  [<ffffffff815c5190>] _cpu_up+0x115/0x155
> > [ 1294.974067]  [<ffffffff815c527c>] cpu_up+0x65/0x73
> > [ 1294.974067]  [<ffffffff815ac21d>] store_online+0x52/0x7b
> > [ 1294.974067]  [<ffffffff812ff775>] sysdev_store+0x20/0x22
> > [ 1294.974067]  [<ffffffff81134f6b>] sysfs_write_file+0xe9/0x11e
> > [ 1294.974067]  [<ffffffff810e0da5>] vfs_write+0xea/0x148
> > [ 1294.974067]  [<ffffffff810e0ec7>] sys_write+0x4c/0x72
> > [ 1294.974067]  [<ffffffff8100c6cb>] system_call_fastpath+0x16/0x1b
> > [ 1294.974067] ---[ end trace 4eaa2a86a8e2da24 ]---
> > --
> 
> The warning tells us that arch/x86/kernel/hpet.c::hpet_cpuhp_notify()
> allocates a timer on the stack.  In particular, it allocates a struct
> hpet_work_struct on the stack which incorporates a timer.
> 
> Now, a timer instance on a stack would be wrong if the function which
> defined that timer exits too early.  But in case of hpet_cpuhp_notify(),
> this is OK since
>   - it waits for the work be done,
>   - the work does not rearm itself.
> IOW the work and thus the timer won't be accessed anymore when
> hpet_cpuhp_notify() returns.
> 
> So, hpet_cpuhp_notify() apparently needs a variant of INIT_DELAYED_WORK
> which uses init_timer_on_stack instead of init_timer, to say "trust me,
> I know what I'm doing".

that's already fixed in tip/timers/urgent and queued up for Linus:

  6d612b0: locking, hpet: annotate false positive warning

	Ingo