RFR: 8221102: Allow GC threads to participate in threads claiming protocol

[Kim Barrett]

> On Mar 26, 2019, at 6:31 AM, Roman Kennke <rkennke at redhat.com> wrote:
>> OK.  So does that count as a "review"?  And I still need a second.
> 
> No, not yet.
> 
> I've tested it against my failing stress test (together with your njt_lock patch) and it seems to fix it.
> 
> I'm also testing it against Shenandoah test suite, and so far looks good.

OK, thanks for doing the additional testing.

> I don't quite understand the resetting logic though. It first increments the global counter. If it turns 0, it resets all thread tokens to 0, using the claim_oops_do/cmpxchg protocol, but that is now expecting 0 too? Why not simply assign 0 instead? Maybe that code just isn't obvious enough…?

There isn't a setter for the per-thread token, only the claim
function. But note that we're using the non-parallel variant (the
first argument is false), so it effectively is just a simple
assignment to the new claim value (when not already that value), with
no cmpxchg involved.

> Also, the resetting logic warrants a test, it is so rare, it will otherwise not get executed, but if it fails, it would be catastrophic.

I was hoping the resetting code was simple enough to not require a
test.  I will see if I can come up with a sensible gtest.