RFR 8u: Cherry-pick 8140584: nmethod::oops_do_marking_epilogue always runs verification code

[Aleksey Shipilev]

On 05/23/2017 03:41 PM, Andrew Haley wrote:
> On 23/05/17 13:18, Aleksey Shipilev wrote:
> 
>> Shenandoah-8u runs into the serious performance problem with code
>> cache scan -- we are wasting several milliseconds per pause -- and
>> it seems to be fixed in 8u152:
>> https://bugs.openjdk.java.net/browse/JDK-8140584
>>
>> I propose we cherry-pick it to our shenandoah-8u repo, while we wait for it to
>> arrive from upstream.
>>
>> Andrew, any objections?
> 
> We have to think about the criteria we use to decide such things.  It
> seems to me that this is a very low-risk patch, but on the other hand
> the gains are slight. 

The gains are slight for STW GC, where 5ms is a tiny drop in a large pause. For
Shenandoah, it is more than a half of the pause :)

> On the gripping hand, if we allow in many small
> such patches we have a significant divergence from upstream.  Each
> patch on its own is no big deal, but they all add up.
> 
> So, I'm on the horns of a dilemma.  (Ouch!)

Yeah. In this particular case, we are not diverging from upstream, for some
notion of "upstream". Since we are planning to pull 8b152 once released anyway,
what we do here is fast-forwarding to the future today. I guess simplicity of
the patch + already being in upstream is enough to qualify for cherry-pick?

-Aleksey