[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-hotspot-runtime-dev
Subject:    Re: RFR: 8068451: Kitchensink fails with assert(_size >= sz) failed: Negative size
From:       Coleen Phillimore <coleen.phillimore () oracle ! com>
Date:       2015-02-10 18:33:26
Message-ID: 54DA4EF6.3030401 () oracle ! com
[Download RAW message or body]


Thanks Harold!
Coleen

On 2/10/15, 1:00 PM, harold seigel wrote:
> Hi Coleen,
>
> Your changes look good.
>
> Thanks, Harold
>
> On 2/9/2015 12:38 PM, Coleen Phillimore wrote:
>> Summary: Need ThreadCritical lock around chunk pool cleaning and 
>> copying for snapshot
>>
>> This change restores the TC lock around the code that frees chunks 
>> from the chunk pool.  NMT1 needed ThreadCritical lock to track 
>> malloc'ed memory, but NMT2 doesn't.  This lock is also used to 
>> protect the code that copies the running snapshot into a saved 
>> snapshot.  After the latter snapshot is created, NMT reporting 
>> adjusts the malloc'ed memory for chunks because this memory is 
>> already reported as the various arenas, so we don't want to 
>> double-count it.
>>
>> Tested Kitchensink with NMT for three days with code to sleep during 
>> the snapshot copy.  This code reproduced the assertion 18x in about 8 
>> hrs.  (Took out the sleep).  Also, fixed some whitespace and typos in 
>> the execution path of the bug.  Also ran NMT tests.  Can't add a 
>> test, it's noreg-hard.
>>
>> open webrev at http://cr.openjdk.java.net/~coleenp/8068451/
>> bug link https://bugs.openjdk.java.net/browse/JDK-8068451
>>
>> thanks,
>> Coleen
>>
>>
>>
>

[prev in list] [next in list] [prev in thread] [next in thread]