RFR(XS): JDK-8214272: Don't use memset to initialize arrays of MemoryUsage in memoryManager.cpp

Dmitry Chuyko dmitry.chuyko at bell-sw.com
Tue Nov 27 14:30:35 UTC 2018


Doesn't "= MemoryUsage()" actually imply "delete onstack MemoryUsage" 
after copying? Probably safe here but it may be not safe in general.

-Dmitry

On 11/27/18 5:14 PM, Roman Kennke wrote:
> I think Aleksey's version is clearer. I used a similar approach in a fix
> for the same issue:
>
> http://hg.openjdk.java.net/jdk/jdk/rev/710e5a66a64e
>
> Roman
>
>
>> Placement new sets values placed in arrays. The arrays as a whole are
>> managed as before. I.e.:
>>
>> GCStatInfo::~GCStatInfo() {
>>    FREE_C_HEAP_ARRAY(MemoryUsage*, _before_gc_usage_array);
>>    FREE_C_HEAP_ARRAY(MemoryUsage*, _after_gc_usage_array);
>> }
>>
>> -Dmitry
>>
>> On 11/27/18 5:00 PM, Aleksey Shipilev wrote:
>>> On 11/27/18 2:51 PM, Dmitry Chuyko wrote:
>>>> rfe: https://bugs.openjdk.java.net/browse/JDK-8214272
>>>> webrev: http://cr.openjdk.java.net/~dchuyko/8214272/webrev.00/
>>> Ummmm. Aren't you leaking memory by calling "new" without "delete"?
>>>
>>> Consider this instead:
>>>
>>> diff -r a554db76b2e9 src/hotspot/share/services/memoryManager.cpp
>>> --- a/src/hotspot/share/services/memoryManager.cpp      Fri Nov 23
>>> 11:22:31 2018 +0100
>>> +++ b/src/hotspot/share/services/memoryManager.cpp      Tue Nov 27
>>> 14:56:09 2018 +0100
>>> @@ -169,7 +169,8 @@
>>>      _start_time = 0L;
>>>      _end_time = 0L;
>>> -  size_t len = _usage_array_size * sizeof(MemoryUsage);
>>> -  memset(_before_gc_usage_array, 0, len);
>>> -  memset(_after_gc_usage_array, 0, len);
>>> +  for (int i = 0; i < _usage_array_size; i++) {
>>> +    _before_gc_usage_array[i] = MemoryUsage();
>>> +    _after_gc_usage_array[i]  = MemoryUsage();
>>> +  }
>>>    }
>>>
>>> See, for example:
>>>    https://bugs.openjdk.java.net/browse/JDK-8213745
>>>
>>> -Aleksey
>>>


More information about the hotspot-gc-dev mailing list