RFR (XS) 8229000: Fix lock and reenable assert in Monitor::check_safepoint_state

[coleen.phillimore at oracle.com]

On 8/2/19 9:43 AM, David Holmes wrote:
> Hi Coleen,
>
> On 2/08/2019 10:37 pm, coleen.phillimore at oracle.com wrote:
>> There was only one place in CMS that violated the condition in the 
>> new assert, fixed with this patch.  See linked bugs for details.
>>
>> open webrev at 
>> http://cr.openjdk.java.net/~coleenp/2019/8229000.01/webrev
>> bug link https://bugs.openjdk.java.net/browse/JDK-8229000
>
> The fix seems okay in the sense that we have a NonJavaThread and so 
> should be using lock_without_safepoint_check(). But I'm a bit puzzled 
> as the code you fixed is not what was failing in JDK-8223190 when the 
> assert was removed. ??

The stack doesn't show the inlined function, do_yield_work from 
should_return, but it is the same lock.

Thanks,
Coleen

>
> Thanks,
> David
>
>> Ran mach5 hs-tier1-3 (all failed tests were in these tiers) on 
>> linux-x64-debug, and tier1 on all Oracle platforms.
>>
>> Thanks,
>> Coleen