[OE-core] lttng-modules may trigger deadlock in the timekeeping subsystem with Linux <= 3.10.13, <= 3.11.2

[Woegerer, Paul]

On 10/15/2013 12:50 AM, Bruce Ashfield wrote:
> On Mon, Oct 14, 2013 at 12:32 PM, Tom Zanussi
> <tom.zanussi at linux.intel.com> wrote:
>> On Mon, 2013-10-14 at 17:54 +0200, Woegerer, Paul wrote:
>>> Hi,
>>>
>>> looking at the following commits shows that the current version of
>>> lttng-modules 2.3.x can trigger a deadlock in the timekeeping subsystem
>>> of Linux.
>>>
>>> http://git.lttng.org/?p=lttng-modules.git;a=commit;h=fc8216ae9ec5d18172d8227d179475e7cc1fb45c
>>> http://git.lttng.org/?p=lttng-modules.git;a=commit;h=9998f5216f4641a79e158135c4c1658dcc6cd2d8
>>> http://git.lttng.org/?p=lttng-modules.git;a=commit;h=e14bf96416c39675a5f785b032d1c5279020b93d
>>>
>>> This is caused by a bug in the kernel that was introduced in 3.10. The
>>> problem is already fixed in Linux 3.10.14, 3.11.3 and 3.12. See:
>>> http://marc.info/?l=linux-kernel&m=137894350929349
>>>
>>> lttng-modules 2.3.2 contains preprocessor-code that makes sure that
>>> building of lttng-modules fails for the unlucky combinations of kernel
>>> and lttng-modules. The problem is when we update our lttng-modules 2.3.0
>>> to 2.3.2 then we would have lttng-modules failing to build for our qemu*
>>> MACHINEs.
>>>
>>> Any recommendations on how we should address this ?
>>>
>> Probably just get Bruce to update linux-yocto-3.10 to 3.10.14, or pull
>> the specific commit that fixes it in.
> Precisely. And I already have all the -stable updates queued, it just needs
> to be 1.5.1, since I stopped sending updates to 3.10 for 1.5 proper a few
> weeks ago.

Bruce, as soon as you have updated linux-yocto-3.10 to 3.10.14 (or
later) I will send the patch that updates lttng-modules from 2.3.0 to
2.3.2. This will ensure we will have the checks from 2.3.2 in, so that
users of other bsps will get a compile time error if they are tring to
build a fatal combination of lttng-modules and kernel.

Thanks,
Paul

>
> This isn't severe enough to warrant any special handling outside of getting
> the -stable update into 1.5.1.
>
> Cheers,
>
> Bruce
>
>> Tom
>>
>>> Many thanks,
>>> Paul
>>>
>>
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core at lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-core
>
>


-- 
Paul Woegerer, SW Development Engineer
Sourcery Analyzer <http://go.mentor.com/sourceryanalyzer>
Mentor Graphics, Embedded Software Division