Re: [bisected] sh_cmt Division by zero in kernel.

[Arnd Hannemann <arnd@arndnet.de>]

Hi Magnus,

Am 22.07.2010 15:55, schrieb Magnus Damm:
> Hi Arnd,
>
> Thanks for reporting and bisecting.
>
> On Wed, Jul 21, 2010 at 4:25 PM, Arnd Hannemann <arnd@arndnet.de> wrote:
>   
>> Hi,
>>
>> when booting recent genesis-2.6 head on an AP4EVB board a division by zero
>> in kernel is reported:
>> (Looks identical to: http://thread.gmane.org/gmane.linux.ports.sh.devel/8497)
>>
>>
>> [    0.000000]  sh_cmt.10: using deprecated clock lookup
>> [    0.000000]  sh_cmt.10: used for clock events
>> [    0.000000]  sh_cmt.10: used for periodic clock events
>> [    0.000000] Division by zero in kernel.
>> [    0.000000] Backtrace:
>> [    0.000000] [<c0028f08>] (dump_backtrace+0x0/0x110) from [<c0202ecc>] (dump_stack+0x18/0x1c)
>> [    0.000000]  r6:c0298978 r5:cf80e600 r4:cf801460 r3:c02a37ac
>> [    0.000000] [<c0202eb4>] (dump_stack+0x0/0x1c) from [<c002904c>] (__div0+0x18/0x20)
>> [    0.000000] [<c0029034>] (__div0+0x0/0x20) from [<c01177fc> ](Ldiv0+0x8/0x10)
>> [    0.000000] [<c0202444>] (sh_cmt_probe+0x0/0x4e4) from [<c0016f40> ](early_platform_driver_probe+0x1c8/0x264)
>> [    0.000000] [<c0016d78>] (early_platform_driver_probe+0x0/0x264) from[<c000cbb8>] (shmobile_late_time_init+0x24/0x2c)
>> [    0.000000] [<c000cb94>] (shmobile_late_time_init+0x0/0x2c) from[<c0008978>] (start_kernel+0x208/0x2a8)
>> [    0.000000] [<c0008770>] (start_kernel+0x0/0x2a8) from [<40008034>](0x40008034)
>> [    0.000000]  r5:c02bd1f0 r4:10c53c7d
>> [    0.000000]  sh_cmt.10: used as clock source
>>
>> and as a result time is running too fast.
>> For example sleep 60 on bash returns after a few seconds.
>>     
> I suspect 5967d33ce8a030f01a716fc0b25fcb03744a5fda fixes the issue
> with the time running too fast.
>   
Not sure what you mean, in the tested tree with the crash above
5967d33ce8a030f01a716fc0b25fcb03744a5fda was already included.

>   
>> Git bisect identified following commit as bad:
>>
>> commit f4d7c3565c1692c54d9152b52090fe73f0029e37
>> Author: Paul Mundt <lethal@linux-sh.org>
>> Date:   Wed Jun 2 17:10:44 2010 +0900
>>
>>    clocksource: sh_cmt: compute mult and shift before registration
>>       Based on the sh_tmu change in 66f49121ffa41a19c59965b31b046d8368fec3c7
>>    ("clocksource: sh_tmu: compute mult and shift before registration").
>>    The same issues impact the sh_cmt driver, so we take the same approach
>>    here.
>>       Cc: stable@kernel.org
>>    Signed-off-by: Paul Mundt <lethal@linux-sh.org>
>>     
> If you revert that one (and perhaps
> 5967d33ce8a030f01a716fc0b25fcb03744a5fda as well), does it fix your
> crash?
>   

Yes, if I revert both 5967d33ce8a030f01a716fc0b25fcb03744a5fda
and f4d7c3565c1692c54d9152b52090fe73f0029e37, this fixes the
crash and time is running "normal".

I reverted both mainly because the revert of
f4d7c3565c1692c54d9152b52090fe73f0029e37 would not go
through cleanly on top of the current tree.


Best regards,
Arnd