RFR(T) 8219244: NMT: Move safepoint related memory allocations from mtInternal to mtSafepoint

[zgu at redhat.com]

On Tue, 2019-02-19 at 16:09 -0500, coleen.phillimore at oracle.com wrote:
> 
> On 2/19/19 4:07 PM, zgu at redhat.com wrote:
> > Hi Coleen,
> > 
> > On Tue, 2019-02-19 at 15:13 -0500, coleen.phillimore at oracle.com
> > wrote:
> > > This is such a small allocation, I think it makes more sense to
> > > add
> > > it to mtThread.
> > 
> > I get it if there is not mtSafepoint category, it is not worth to
> > have
> > one for a small allocation, but we already have it. Why is it
> > better to
> > add to mtThread?
> 
> Doesn't the current mtSafepoint category only report on mmap memory
> for 
> the polling page?  This would add a very tiny allocation for the 
> ThreadSafepointState (maybe too small to report?).

Okay, this makes sense. Probably won't show up if application has only
a few JavaThreads.

>  These allocations 
> really belong to the thread.
I changed bug synopsis to "NMT: Change ThreadSafepointState's
allocation type from mtInternal to mtThread"

Updated Webrev: http://cr.openjdk.java.net/~zgu/JDK-8219244/webrev.01/

Thanks,

-Zhengyu

> 
> Coleen
> 
> > 
> > Thanks,
> > 
> > -Zhengyu
> > 
> > 
> > Coleen
> > > On 2/18/19 3:12 PM, zgu at redhat.com wrote:
> > > > There is only one instance I can find.
> > > > 
> > > > Bug: https://bugs.openjdk.java.net/browse/JDK-8219244
> > > > Webrev: http://cr.openjdk.java.net/~zgu/JDK-8219244/webrev.00/
> > > > 
> > > > Test:
> > > > 
> > > >     hotspot_nmt on Linux x64 (fastdebug and release)
> > > >     JDK-submit tests.
> > > > 
> > > > Thanks,
> > > > 
> > > > -Zhengyu
> 
>