Review request: 8024650: Don't adjust MaxMetaspaceSize up to MetaspaceSize

[Stefan Karlsson]

On 9/17/13 2:08 AM, Jon Masamitsu wrote:
> To follow up on Thomas' point about
>
>   90   if (MetaspaceSize < 256*K) {
>   91     vm_exit_during_initialization("Too small initial Metaspace 
> size");
>   92   }
>
> There really isn't a value of MetaspaceSize that is too small.  It is 
> just
> a high-water-mark.  If it was zero, it should still work (just would get
> a GC  sooner rather than later).  I haven't tried it (hit the
> assertion) but it should work.  I'll file a bug if you don't want to 
> include
> it with this.

I've already pushed the fix for 8024650, so we would need a new bug 
report for that.

Related to this is the alignment of MetaspaceSize and MaxMetaspaeSize. 
It's done with min_alignment() and max_alignment(). These are the wrong 
alignments, since we don't have the same restrictions on the metaspaces 
as Java Heap has. We should be aligning to the memory commit granularity.

StefanK

>
> Jon
>
>
> On 9/13/2013 4:55 AM, Stefan Karlsson wrote:
>> http://cr.openjdk.java.net/~stefank/8024650/webrev.01/
>>
>> Changed in this revision:
>> - Removed incorrect FLAG_IS_DEFAULT check
>> - Added MetaspaceSize <= MaxMetaspaceSize assert
>> - Added boundary test cases where MetaspaceSize and MaxMetaspaceSize 
>> is set to 0.
>> - Added rudimentary tests for the new firstMatch function.
>>
>> Rejected proposals:
>> - Use MXBeans to get flag values.
>> - Use WBApi to get conservative max heap alignment.
>>
>> thanks,
>> StefanK
>>
>>
>> On 09/12/2013 05:18 PM, Stefan Karlsson wrote:
>>> http://cr.openjdk.java.net/~stefank/8024650/webrev.00/
>>>
>>> - Limit MetaspaceSize to MaxMetaspaceSize
>>> - Make sure we don't align down to 0.
>>> - jtreg test
>>>
>>> Note that this patch also adds/changes some functionality in the 
>>> OutputAnalyzer in test/testlibrary.
>>>
>>> thanks,
>>> StefanK
>>
>