[9] RFR (S): 8177963: Parallel GC fails fast when per-thread task log overflows

sangheon sangheon.kim at oracle.com
Wed Apr 5 17:47:23 UTC 2017


Hi Thomas,

On 04/05/2017 02:17 AM, Thomas Schatzl wrote:
> Hi Sangheon,
>
> On Tue, 2017-04-04 at 12:40 -0700, sangheon wrote:
>> Hi Thomas,
>>
>> Thank you for fixing this!
>>
>> http://cr.openjdk.java.net/~tschatzl/8177963/webrev/src/share/vm/gc/p
>> arallel/gcTaskThread.cpp.frames.html
>> 62   uint time = MIN2(_time_stamp_index, (uint)GCTaskTimeStampEntries
>> - 1);
>> Variable name seems not appropriate. 'index' instead of 'time'?
> Kind of fixed after incorporating Kim's suggestions.
OK.

>
>>     68   _time_stamp_index++;
>> We would face overflow here, so increase only if it is less than
>> GCTaskTimeStampEntries?
>> (Because you want to re-use the last buffer when buffers overflow, if
>> not line 58 also should be '>=')
>> Or simply rotate it, _time_stamp_index%=GCTaskTimeStampEntries
> I do not really care either way, we just should not crash. In this
> version I implemented Kim's suggestion to ignore all further time
> stamps.
OK.

>
>> http://cr.openjdk.java.net/~tschatzl/8177963/webrev/src/share/vm/gc/p
>> arallel/gcTaskThread.hpp.frames.html
>> You already updated all other files except this.
> Fixed.
>
> New webrevs at:
> http://cr.openjdk.java.net/~tschatzl/8177963/webrev.0_to_1 (diff)
> http://cr.openjdk.java.net/~tschatzl/8177963/webrev.1 (full)
webrev.1 seems good to me.

Thanks,
Sangheon


>
> Thanks,
>    Thomas
>




More information about the hotspot-gc-dev mailing list