RFR [sh/8u]: Do not enable UseCountedLoopSafepoints in Shenandoah by default

[Roman Kennke]

> On 10/23/19 12:28 PM, Roman Kennke wrote:
>> Enabling -XX:+UseCountedLoopSafepoints in 8u seems risky because it's
>> known to trigger bugs. In particular, it shows up with the upcoming LRB
>> backport and the TestMaybeNullUnsafeAccess.java testcase.
>>
>> Let's not enable it by default, and print a warning when the user
>> enables it. The downside of it is regressed TTSP for code-paths with
>> long counted loops.
>>
>> http://cr.openjdk.java.net/~rkennke/clsp/webrev.00/
> 
> Sounds fine.
> 
> *) Current checking code assumes the default is "false", maybe it should be more straight-forward
> instead:
> 
>   if (UseCountedLoopSafepoints) {
>     ...
>   }
> 
> *) Let's not put the URLs in the logs? Not sure that would be the only bug, and it is dubious to
> list just one. Put the reference to example bug in the code comment, if you want.

Ok. Like this then?

http://cr.openjdk.java.net/~rkennke/clsp/webrev.01/

BTW, testing: hotspot_gc_shenandoah (also, with the upcoming LRB
backport, which showed the bug).

Roman