[Fwd: Re: Review request: 6581734 CMS Old Gen's collection usage is zero after GC which is incorrect]

Kevin Walls kevin.walls at oracle.com
Fri Jul 30 13:30:35 PDT 2010


Great, thanks - respun as 
http://cr.openjdk.java.net/~kevinw/6581734/webrev.02/



Mandy Chung wrote:
> Hi Kevin,
>
> Thanks for fixing this.  Looks good with some minor comments:
>
> memoryManager.hpp line 181-182:
>   get_last_gc_stat() returns gc_index but the comment says it returns 
> the size of the array pool.
>
> management.cpp line 1908:
>   if (!mgr->get_last_gc_stat(stat) == 0) {
>
> You forgot to take out "!" when you changed the get_last_gc_stat() to 
> return size_t instead of bool.
>
> test/gc/6581734/Test6581734.java
>   @run main/othervm -server -Xmx512m -verbose:gc ...
>
> Why do you need "-server"?  We also want client VM to be tested, right?
>
> You can drop -verbose:gc and use MemoryMXBean.setVerbose(true) to turn 
> on gc tracing if you like.
>
> The test source isn't formatted properly, e.g. line 49-56, 69-75, 
> 79-124, 127, 131.  Can you fix the formatting - using 4 spaces as 
> indentation (not tab or 2 spaces)?
>
> Thanks
> Mandy
>
> Kevin Walls wrote:
>>
>> Hi Mandy,
>>
>> Yes, thanks for the discussions we've had so far. 8-)
>>
>> Here's an updated JDK7 version:
>>
>> http://cr.openjdk.java.net/~kevinw/6581734/webrev.01/
>>
>> Thanks
>> Kevin
>>
>> Mandy Chung wrote:
>>> Kevin,
>>>
>>> As I know, you're revising the fix for the 6 update release.   I'll 
>>> review your revised webrev when it's ready.
>>>
>>> Thanks
>>> Mandy
>>>
>>> Kevin Walls wrote:
>>>> Hi -
>>>>
>>>> This is a review request for the issue where cms doesn't update the 
>>>> statistics used by the MemoryPoolMXBean system: only concurrent 
>>>> mode failures update the stats.  We just aren't prepared for 
>>>> concurrent collections, so it takes a fair amount of customisation 
>>>> to add that flexibility...
>>>>
>>>> http://cr.openjdk.java.net/~kevinw/6581734/webrev/
>>>>
>>>> Thanks
>>>> Kevin
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>
>



More information about the hotspot-gc-dev mailing list