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

Mandy Chung mandy.chung at oracle.com
Fri Jul 30 10:31:49 PDT 2010


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