RFR(one-liner): 8231321: compiler/codecache/stress/UnexpectedDeoptimizationAllTest.java failed assertion

[Tobias Hartmann]

Hi Robbin,

please note that we also hit the following assert (see comments in bug report):
Crash: Internal Error ...nmethod.cpp...assert(!found) failed: osr nmethod should have been invalidated

Thanks,
Tobias

On 25.09.19 10:12, Robbin Ehn wrote:
> Hi all, please review.
> 
> We no longer allow state to ever go backwards.
> So trying to make a zombie not-entrant is fine.
> What happens is something like:
> - make not-entrant
> - safepoint
> - deopt all => start iteration over nmethods
> - picks up this not-entrant method, since it is alive
> - sweeper turns it into a zombie
> - we try making the nmethod not-entrant once more => assert
> 
> If we remove the assert the method would just returns false instead, which is expected.
> 
> Thanks, Robbin
> 
> Issue:
> https://bugs.openjdk.java.net/browse/JDK-8231321
> 
> Changeset:
> diff -r 448fe2bfd505 src/hotspot/share/code/nmethod.cpp
> --- a/src/hotspot/share/code/nmethod.cpp    Tue Sep 24 08:54:39 2019 +0200
> +++ b/src/hotspot/share/code/nmethod.cpp    Wed Sep 25 10:02:45 2019 +0200
> @@ -1288,17 +1288,16 @@
>    }
>  }
> 
>  /**
>   * Common functionality for both make_not_entrant and make_zombie
>   */
>  bool nmethod::make_not_entrant_or_zombie(int state) {
>    assert(state == zombie || state == not_entrant, "must be zombie or not_entrant");
> -  assert(!is_zombie(), "should not already be a zombie");
> 
>    if (Atomic::load(&_state) >= state) {
>      // Avoid taking the lock if already in required state.
>      // This is safe from races because the state is an end-state,
>      // which the nmethod cannot back out of once entered.
>      // No need for fencing either.
>      return false;
>    }