ACPI / PAD: call schedule() when need_resched() is true
authorTony Camuso <tcamuso@redhat.com>
Fri, 25 Apr 2014 18:19:29 +0000 (14:19 -0400)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Mon, 28 Apr 2014 10:26:00 +0000 (12:26 +0200)
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-and-tested-by: Stuart Hayes <stuart_hayes@dell.com>
Signed-off-by: Tony Camuso <tcamuso@redhat.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/acpi/acpi_pad.c

index 37d73024b82e4f9a7df2820e331bd873e0805947..e20708f2b8e56a89924adf0c5fa8517ccadb4d5d 100644 (file)
@@ -215,8 +215,15 @@ 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);