Re: [PATCH 2.6.27-rc5 incremental re-resubmit] Fix itimer/many thread hang.

[Ingo Molnar <mingo@elte.hu>]

* Frank Mayhar <fmayhar@google.com> wrote:

>  /*
>   * Return any ns on the sched_clock that have not yet been banked in
>   * @p in case that task is currently running.
> - *
> - * Called with task_rq_lock() held on @rq.
>   */
> -static unsigned long long task_delta_exec(struct task_struct *p, struct rq *rq)
> +unsigned long long task_delta_exec(struct task_struct *p)
>  {
> +	struct rq *rq;
> +	unsigned long flags;
> +	u64 ns = 0;
> +
>  	if (task_current(rq, p)) {
>  		u64 delta_exec;

hmmm ... where do we get 'rq' from?

in v3 you did this:

-       rq = task_rq_lock(p, &flags);

which removed the deadlock but left us with a random uninitialized rq 
variable ...

the right solution for the bug would have been to unlock it. 
Miraculously we didnt actually crash anywhere visibly, found it by 
reviewing the code. I thought this code gets excercised quite 
frequently. The commit below fixes it.

Could you please functionality-test latest tip/master:

  http://people.redhat.com/mingo/tip.git/README

with your testcase that excercises these codepaths heavily?

Thanks,

	Ingo