On Thursday, December 12, 2013 10:50:15 AM Tony Camuso wrote:
> (Slightly rephrased from a bug report submitted by Stuart Hayes)
>
> The purpose of the acpi_pad driver is to implement the "processor power
> aggregator" device as described in the ACPI 4.0 spec section 8.5. It
> takes requests from the BIOS (via ACPI) to put a specified number of
> CPUs into idle, in order to save power, until further notice.
>
> It does this by creating high-priority threads that try to keep the CPUs
> in a high C-state (using the monitor/mwait CPU instructions). The
> mwait() call is in a loop that checks periodically if the thread should
> end and a few other things.
>
> It was discovered through testing that the power_saving threads were
> causing the system to consume more power than the system was consuming
> before the threads were created. A counter in the main loop of
> power_saving_thread() revealed that it was spinning. The mwait()
> instruction was not keeping the CPU in a high C state very much if at
> all.
>
> Here is a simplification of the loop in function power_saving_thread() in
> drivers/acpi/acpi_pad.c
>
> while (!kthread_should_stop()) {
> :
> try_to_freeze()
> :
> while (!need_resched()) {
> :
> if (!need_resched())
> __mwait(power_saving_mwait_eax, 1);
> :
> if (jiffies > expire_time) {
> do_sleep = 1;
> break;
> }
> }
> }
>
> If need_resched() returns true, then mwait() is not called. It was
> returning true because of things like timer interrupts, as in the
> following sequence.
>
> hrtimer_interrupt->__run_hrtimer->tick_sched_timer-> update_process_times->
> rcu_check_callbacks->rcu_pending->__rcu_pending->set_need_resched
>
> Kernels 3.5.0-rc2+ do not exhibit this problem, because a patch to
> try_to_freeze() in include/linux/freezer.h introduces a call to
> might_sleep(), which ultimately calls schedule() to clear the reschedule
> flag and allows the the loop to execute the call to mwait().
>
> However, the changes to try_to_freeze are unrelated to acpi_pad, and it
> does not seem like a good idea to rely on an unrelated patch in a
> function that could later be changed and reintroduce this bug.
>
> Therefore, it seems better to make an explicit call to schedule() in the
> outer loop when the need_resched flag is set.
>
> Reported-by: Stuart Hayes <stuart_hayes@xxxxxxxx>
> Tested-by: Stuart Hayes <stuart_hayes@xxxxxxxx>
> Signed-off-by: Tony Camuso <tcamuso@xxxxxxxxxx>
Peter Zijlstra has an acpi_pad patch that I think will conflict with this, so
please resubmit after the 3.14 merge window if still necessary.
Thanks!
> ---
> drivers/acpi/acpi_pad.c | 9 ++++++++-
> 1 files changed, 8 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/acpi/acpi_pad.c b/drivers/acpi/acpi_pad.c
> index fc6008f..2c67e59 100644
> --- a/drivers/acpi/acpi_pad.c
> +++ b/drivers/acpi/acpi_pad.c
> @@ -219,8 +219,15 @@ static int power_saving_thread(void *data)
> * borrow CPU time from this CPU and cause RT task use > 95%
> * CPU time. To make 'avoid starvation' work, takes a nap here.
> */
> - if (do_sleep)
> + if (unlikely(do_sleep))
> schedule_timeout_killable(HZ * idle_pct / 100);
> +
> + /* If an external event has set the need_resched flag, then
> + * we need to deal with it, or this loop will continue to
> + * spin without calling __mwait().
> + */
> + if (unlikely(need_resched()))
> + schedule();
> }
>
> exit_round_robin(tsk_index);
>
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
