Re: [lkp] [panic] [ BUG: bad unlock balance detected! ]

[Vitaly Kuznetsov <vkuznets@redhat.com>]

Andrew Morton <akpm@linux-foundation.org> writes:

> On Mon, 16 Nov 2015 11:12:50 +0800 kernel test robot <ying.huang@linux.intel.com> wrote:
>
>> FYI, we noticed the below changes on
>> 
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
>> commit 08d78658f393fefaa2e6507ea052c6f8ef4002a2 ("panic: release stale console lock to always get the logbuf printed out")
>> 
>> ...
>>
>> [   38.408623] =====================================
>> [   38.408623] [ BUG: bad unlock balance detected! ]
>> [   38.408627] 4.3.0-08150-g08d7865 #1 Not tainted
>> [   38.408627] -------------------------------------
>> [   38.408631] swapper/0/1 is trying to release lock (console_lock) at:
>> [   38.408639] [<ffffffff811a3cb1>] panic+0x124/0x213
>> [   38.408640] but there are no more locks to release!
>> [   38.408641] 
>> [   38.408641] other info that might help us debug this:
>> [   38.408642] 2 locks held by swapper/0/1:
>> [   38.408651]  #0:  (oom_lock){+.+...}, at: [<ffffffff811add45>] __alloc_pages_slowpath+0x5bd/0x68d
>> [   38.408656]  #1:  (panic_lock){......}, at: [<ffffffff811a3bd9>] panic+0x4c/0x213
>> [   38.408657] 
>> [   38.408657] stack backtrace:
>> [   38.408659] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.3.0-08150-g08d7865 #1
>> [   38.408660] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
>> [   38.408665]  0000000000000000 ffff8800132bba78 ffffffff8173760d ffff8800132b4040
>> [   38.408668]  ffff8800132bbaa0 ffffffff8111328c 00000000ffffffff ffff8800132b4040
>> [   38.408670]  ffffffff8406a080 ffff8800132bbb18 ffffffff8111626c ffffffff811139d1
>> [   38.408671] Call Trace:
>> [   38.408678]  [<ffffffff8173760d>] dump_stack+0x4b/0x63
>> [   38.408682]  [<ffffffff8111328c>] print_unlock_imbalance_bug+0xc7/0xd0
>
> Yup, we can't have this happen.
>
> Can we instead do this?
>
> 	if (is_console_locked())
> 		console_unlock();
>
> It's racy,

Well, at this point all other CPUs are supposed to be stopped by
smp_send_stop() (first by IPIs and then by NMIs).

> but more accurate than what we have now?

Yes, but it won't solve the issue - lockdep is complaining because *we*
didn't take take the lock we're trying to release. I suggest to disable
lockdep here (it is already disabled on OOPS, but if panic() is called
directly it is not -- I didn't test this case and that's what actually
happens on OOM above), please see my "[PATCH] panic: turn off locks
debug before releasing console lock".

-- 
  Vitaly