RFR: Use os::naked_short_sleep instead of naked Thread events for sleeping

[Roman Kennke]

Am 07.03.2018 um 11:35 schrieb Aleksey Shipilev:
> On 03/07/2018 11:26 AM, Roman Kennke wrote:
>> In some cases I actually question why we are waiting at all, and don't
>> we want to SpinPause() instead? E.g:
> 
> SpinPause is for very short busy-waits. We cannot afford busy-waiting on most of those paths.
> 
>>>    // Wait for the actual stop(), can't leave run_service() earlier.
>>>    while (!should_terminate()) {
>>> -    Thread::current()->_ParkEvent->park(10);
>>> +    os::naked_short_sleep(10);
>>>    }
>>>  }
>>
>> And hardcoded 10ms? ^^
> 
> Fixed with ShenandoahControlLoopInterval.
> 

Good.

>>> @@ -444,7 +444,7 @@
>>>                   GCCause::to_string(GCCause::_shenandoah_allocation_failure_evac),
>>>                   Thread::current()->name());
>>>      while (heap->is_evacuation_in_progress()) { // wait.
>>> -      Thread::current()->_ParkEvent->park(1);
>>> +      os::naked_short_sleep(1);
>>>      }
>>>    }
>>>  }
>>
>> This loop doesn't exist anymore (in dev repo).
> 
> It does exist. It does not exist with your OOM-during-evac changes.
> 

Oops. Stupid me.

>>> diff -r 58279d273cc4 src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
>>> --- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp	Mon Mar 05 17:05:30 2018 +0100
>>> +++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp	Wed Mar 07 11:13:54 2018 +0100
>>> @@ -2849,7 +2849,7 @@
>>>  void ShenandoahHeap::try_inject_alloc_failure() {
>>>    if (ShenandoahAllocFailureALot && !cancelled_concgc() && ((os::random() % 1000) > 950)) {
>>>      _inject_alloc_failure.set();
>>> -    Thread::current()->_ParkEvent->park(1);
>>> +    os::naked_short_sleep(1);
>>>      if (cancelled_concgc()) {
>>>        log_info(gc)("Allocation failure was successfully injected");
>>>      }
>>
>> What is this waiting for?
> 
> It waits for concurrent thread to react on cancellation, mostly for logging purposes.

Uh. Well I guess it doesn't hurt.

Push it.

Roman