Re: BUG: scheduling while atomic: swapper/0/0x10000002

["Wouter M. Koolen" <W.M.Koolen-Wijkstra@cwi.nl>]

Hi all,

I noticed this problem has disappeared on 3.2.0.

My code-fu is too limited to figure out how. None of Peter's keywords 
below point me to anything related in the 3.1 -> 3.2 patch.

So instead let me use this channel to thank all involved in fixing this. 
I know you are out there. Your work is much appreciated.

Wouter



On 11/28/2011 01:06 PM, Peter Zijlstra wrote:
> On Mon, 2011-11-28 at 12:14 +0000, Wouter M. Koolen wrote:
>> Dear Paul and others,
>>
>> On vanilla kernel 3.1.3, I got the following during boot.
>>
>> BUG: scheduling while atomic: swapper/0/0x10000002
>> no locks held by swapper/0.
>> Modules linked in:
>> Pid: 0, comm: swapper Not tainted 3.1.3.debug+ #32
>> Call Trace:
>> [<ffffffff814058de>] __schedule_bug+0x60/0x65
>> [<ffffffff8189b85a>] ? pidmap_init+0x84/0xc4
>> [<ffffffff8140a3d9>] __schedule+0x759/0x920
>> [<ffffffff8189b85a>] ? pidmap_init+0x84/0xc4
>> [<ffffffff8103d855>] __cond_resched+0x25/0x40
>> [<ffffffff8140a61d>] _cond_resched+0x2d/0x40
>> [<ffffffff811107df>] kmem_cache_alloc_trace+0x4f/0x1d0
>> [<ffffffff8189b85a>] pidmap_init+0x84/0xc4
>> [<ffffffff8188ab47>] start_kernel+0x339/0x3bc
>> [<ffffffff8188a322>] x86_64_start_reservations+0x132/0x136
>> [<ffffffff8188a416>] x86_64_start_kernel+0xf0/0xf7
>>
>> A little googling revealed that patch [2] "rcu: Avoid having
>> just-onlined CPU resched itself when RCU is idle"
>> is supposed to address this issue. However, booting 3.1.3 with patch [2]
>> leads to three new "BUG: scheduling while atomic: swapper/0/0x10000002"
>> reports every boot.
>>
>> The exact blurb varies a little bit, but all backtraces seem ACPI
>> related. I include three examples below. Some old [4] and new [1,3]
>> similar threads exist, but without resolution as far as I can tell.
>>
>> The machine, a 2008 macbook 4.1, seems to be fine.
>>
>> Is this just noise (produced by overzealous debugging checks) that I
>> should safely ignore? If not, please let me know what I can do to help
>> track this down.
> Bah, looks like d86ee4809d0 ("sched: optimize cond_resched()") is
> broken, what's weird is that it only now shows up.
>
> We reset the preempt_count to 0 at sched_init()->init_idle(), which is
> way before pidmap_init(), loosing the PREEMPT_ACTIVE bit that would
> disable should_resched().
>
>