Re: [ckrm-tech] Re: [patch 02/38] CKRM e18: Processor Delay Accounting

[Shailabh Nagar <nagar@watson.ibm.com>]

Ingo Molnar wrote:
> * Ingo Molnar <mingo@elte.hu> wrote:
> 
> 
>>* Gerrit Huizenga <gh@us.ibm.com> wrote:
>>
>>
>>>+#ifdef CONFIG_DELAY_ACCT
>>>+int task_running_sys(struct task_struct *p)
>>>+{
>>>+	return task_is_running(p);
>>>+}
>>>+EXPORT_SYMBOL_GPL(task_running_sys);
>>>+#endif
>>
>>why is this function defined, and why is it exported?

The wrapping of the macro and export of the function was
to allow its use by a module (crbce).

> 
> this:
> 
> +#define task_is_running(p)     (this_rq() == task_rq(p))
> 
> is totally bogus. What you are checking is not whether 'the task is 
> running', but it is a complex way of doing p->thread_info->cpu == 
> smp_processor_id(). This, combined with:
> 
> +               if (pdata == NULL)
> +                       /* some wierdo race condition .. simply ignore */
> +                       continue;
> +               if (thread->state == TASK_RUNNING) {
> +                       if (task_running_sys(thread)) {
> +                               atomic_inc((atomic_t *) &
> +                                          (PSAMPLE(pdata)->cpu_running));
> +                               run++;
> +                       } else {
> +                               atomic_inc((atomic_t *) &
> +                                          (PSAMPLE(pdata)->cpu_waiting));
> +                               wait++;
> +                       }
> +               }
> 
> yields completely incorrect code, and bogus data. If your goal is to 
> sample executing-on-cpu vs. on-runqueue-waiting-to-run states then you 
> should use the already existing task_curr(p) call.

Thanks. task_curr is what we needed.
Would exporting task_curr be ok or should we continue to wrap in a 
separate function ?


--Shailabh