RFR: 8071462: Remove G1ParGCAllocator::alloc_buffer_waste

Michail Chernov michail.chernov at oracle.com
Wed Apr 8 12:58:03 UTC 2015


Hi,

Please could someone review this update?

Changes were updated - fixed code formatting.
http://cr.openjdk.java.net/~akulyakh/8071462/webrev.01/

Thanks,
Michail

On 07.04.2015 17:26, Michail Chernov wrote:
> Hi,
>
> Can I have couple of reviews for next change?
>
> Bug: https://bugs.openjdk.java.net/browse/JDK-8071462
> Webrev: http://cr.openjdk.java.net/~eistepan/~mchernov/8071462/webrev.00/
>
> G1ParGCAllocator has methods add_to_alloc_buffer_waste(size_t waste) 
> and add_to_undo_waste(size_t waste) which are used wrong.
> The main idea of fix is to track waste and undo to waste in PLAB. We 
> cannot use PLABStats to store such values because PLABStats is shared 
> between all PLAB and cannot store separate values of waste in itself. 
> We can get waste and undo to waste values from ParGCAllocBuffer before 
> ParGCAllocBuffer::flush_and_reture() which is called from 
> G1ParScanThreadState destructor (destructor calls 
> G1DefaultParGCAllocator::retire_alloc_buffers which calls 
> ParGCAllocBuffer::flush_and_reture() ). Now destructor calls separate 
> method which is invoked before G1ParScanThreadState destruction, so we 
> can get waste and undo to waste to get termination stats.
>
> Also removed unused methods in G1ParScanThreadState: 
> add_to_alloc_buffer_waste and  add_to_undo_waste.
>
> Fix was checked locally with -XX:+UseG1GC -XX:+PrintTerminationStats 
> and was checked by JPRT.
>
> Thanks,
> Michail
>
>



More information about the hotspot-gc-dev mailing list