[RFC/PATCH] mm/futex: Fix futex writes on archs with SW tracking of dirty & young

[Shan Hai]

On 07/22/2011 06:59 AM, Andrew Morton wrote:
> On Fri, 22 Jul 2011 08:52:06 +1000
> Benjamin Herrenschmidt<benh at kernel.crashing.org>  wrote:
>
>> On Thu, 2011-07-21 at 15:36 -0700, Andrew Morton wrote:
>>> On Tue, 19 Jul 2011 14:29:22 +1000
>>> Benjamin Herrenschmidt<benh at kernel.crashing.org>  wrote:
>>>
>>>> The futex code currently attempts to write to user memory within
>>>> a pagefault disabled section, and if that fails, tries to fix it
>>>> up using get_user_pages().
>>>>
>>>> This doesn't work on archs where the dirty and young bits are
>>>> maintained by software, since they will gate access permission
>>>> in the TLB, and will not be updated by gup().
>>>>
>>>> In addition, there's an expectation on some archs that a
>>>> spurious write fault triggers a local TLB flush, and that is
>>>> missing from the picture as well.
>>>>
>>>> I decided that adding those "features" to gup() would be too much
>>>> for this already too complex function, and instead added a new
>>>> simpler fixup_user_fault() which is essentially a wrapper around
>>>> handle_mm_fault() which the futex code can call.
>>>>
>>>> Signed-off-by: Benjamin Herrenschmidt<benh at kernel.crashing.org>
>>>> ---
>>>>
>>>> Shan, can you test this ? It might not fix the problem
>>> um, what problem.  There's no description here of the user-visible
>>> effects of the bug hence it's hard to work out what kernel version(s)
>>> should receive this patch.
>> Shan could give you an actual example (it was in the previous thread),
>> but basically, livelock as the kernel keeps trying and trying the
>> in_atomic op and never resolves it.
>>
>>> What kernel version(s) should receive this patch?
>> I haven't dug. Probably anything it applies on as far as we did that
>> trick of atomic + gup() for futex.
> You're not understanding me.
>
> I need a good reason to merge this into 3.0.
>
> The -stable maintainers need even better reasons to merge this into
> earlier kernels.
>
> Please provide those reasons!
>

Summary:
- Encountered a 100% CPU system usage problem on pthread_mutex allocated 
in a
     shared memory region, and the problem occurs only on setting 
PRIORITY_INHERITANCE
     to the pthread_mutex.
- ftrace result reveals that an infinite loop in the futex_lock_pi 
caused high CPU usage.
- The powerpc e500 was affected but the x86 was not.
     I have not tested on other archs so I am not sure whether the other 
archs are attacked
     by the problem.
- Tested it on 2.6.34 and 3.0-rc7, both are affected, earlier versions 
might be affected.

Please refer the threads "[PATCH 0/1] Fixup write permission of TLB on 
powerpc e500 core"
and "[PATCH 1/1] Fixup write permission of TLB on powerpc e500 core" for 
the whole story.
Provided the test case code in the [PATH 0/1].

Thanks
Shan Hai

> (Documentation/stable_kernel_rules.txt, 4th bullet)
>
> (And it's not just me and -stable maintainers.  Distro maintainers will
> also look at this patch and wonder whether they should merge it)