Re: [RFC PATCH RT V4] rwsem: The return of multi-reader PI rwsems

[Steven Rostedt <rostedt@goodmis.org>]

On Fri, 18 Apr 2014 10:19:26 +0200
Ingo Molnar <mingo@kernel.org> wrote:

> 
> * Steven Rostedt <rostedt@goodmis.org> wrote:
> 
> > Changes since v3:
> > 
> > Clark reported that he was seeing a large latency when he added this
> > patch. I tested it out on a 8 logical CPU box, and sure enough I was
> > seeing it too. After spending the day debugging why, I found that I had
> > a bug in rt_mutex_getprio(), where I could do:
> > 
> >   min(task_top_pi_waiter(task)->pi_list_entry.prio, prio)
> > 
> > when there was no "top_pi_waiter", which would give garbage as a
> > result. This would let some tasks have higher priority than they
> > should, and cause other tasks that should have high priority not run.
> 
> Would a sanity check like the one below have helped? (untested and 
> such)

Actually, if I had run this with CONFIG_DEBUG_PI_LIST then this would
have triggered:

#ifdef CONFIG_DEBUG_PI_LIST
# define plist_first_entry(head, type, member)	\
({ \
	WARN_ON(plist_head_empty(head)); \
	container_of(plist_first(head), type, member); \
})
#else
# define plist_first_entry(head, type, member)	\
	container_of(plist_first(head), type, member)
#endif

-- Steve

> 
> Thanks,
> 
> 	Ingo
> 
> ==========>
>  kernel/locking/rtmutex_common.h | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/kernel/locking/rtmutex_common.h b/kernel/locking/rtmutex_common.h
> index 7431a9c..36b1ce8 100644
> --- a/kernel/locking/rtmutex_common.h
> +++ b/kernel/locking/rtmutex_common.h
> @@ -85,6 +85,7 @@ static inline int task_has_pi_waiters(struct task_struct *p)
>  static inline struct rt_mutex_waiter *
>  task_top_pi_waiter(struct task_struct *p)
>  {
> +	WARN_ON_ONCE(!p->pi_waiters_leftmost);
>  	return rb_entry(p->pi_waiters_leftmost, struct rt_mutex_waiter,
>  			pi_tree_entry);
>  }