Re: [RFC][PATCH] fixup pi_state in futex_requeue on lock steal

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

Ingo Molnar wrote:
> * Darren Hart <dvhltc@us.ibm.com> wrote:
> 
>> So, I think we're fine with respect to the pi_state ownership!  In 
>> fact I finally managed to catch the lock steal in the requeue loop 
>> in my tracing, and everything worked fine.  Going to go rerun a 
>> bunch more tests and see if I hit any other issues, if I do, I 
>> suspect they are unrelated to this.
>>
>> Thanks for the help in thinking this through.
> 
> i've got these queued up:
> 
>  00235fe: futex: Update woken requeued futex_q lock_ptr
>  1bbf208: rtmutex: Avoid deadlock in rt_mutex_start_proxy_lock()
> 
> should i drop them?

My apologies for the churn on these Ingo.  My comments above only apply 
to this RFC thread, the other patches are needed.  You should include 
the following patches:

tip/core/urgent
===============
rtmutex: Avoid deadlock in rt_mutex_start_proxy_lock()
	1bbf20835c4e088667a090ce6523a0f70b62dc76

[PATCH] futex: Update futex_q lock_ptr on requeue proxy lock (resend)
	from Aug 7, 2009
	The one you committed is older, I resent it on Aug 7 with an
	improved patch description, commentary, and DEBUG_PI_LIST
	ifdefs.  Please drop 00235fe25eba6d3a13f3349b2e3a2d94b699a414
	and pull in the new one.

[PATCH V2] futex: Fix handling of bad requeue syscall pairing
	from Aug 7, 2009


tip/rt/something
================
[PATCH 2/2][RT] Avoid deadlock in rt_mutex_start_proxy_lock()
	from Aug 5, 2009
	This one uses the new atomic_spinlock calls for the RT tree.  I
	suspect you may instead choose to make that one line change
	yourself.

Thanks,

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