Re: [PATCH 5/6] futex: unlock before returning -EFAULT

[Darren Hart <dvhltc@us.ibm.com>]

Peter Zijlstra wrote:
> On Thu, 2009-03-12 at 11:47 +0100, Thomas Gleixner wrote:
>> On Thu, 12 Mar 2009, Peter Zijlstra wrote:
>>
>>> On Thu, 2009-03-12 at 00:56 -0700, Darren Hart wrote:
>>>> futex_lock_pi can potentially return -EFAULT with the rt_mutex held.  This
>>>> seems like the wrong thing to do as userspace should assume -EFAULT means the
>>>> lock was not taken.  Even if it could figure this out, we'd be leaving the
>>>> pi_state->owner in an inconsistent state.  This patch unlocks the rt_mutex
>>>> prior to returning -EFAULT to userspace.
>>> lockdep would complain, one is not to leave the kernel with locks held.
>> That would break pi futexes in bits and pieces.
>>
>>      T1 takes F1
>>      T2 blocks on F1
>>      	-> T2 sets up rt_mutex and locks it for T1
>> 	   T2 blocks on rt_mutex and boosts T1
>>
>>      T1 calls a non futex syscall
>>      T1 returns from syscall with the rt_mutex still locked
>>
>> Thanks,
> 
> Oh right, raw rt_mutex stuff isn't lockdep annotated, and you use the
> robust futex infrastructure to ensure stuff gets unlocked when holder
> dies. That should work out.
> 

OK, are there any other concerns with this patch?

-- 
Darren Hart
IBM Linux Technology Center
Real-Time Linux Team