8247746: [lworld] TestNullableArray::test78 fails with ZGC due to incorrect result

[TobiHartmann]

A missing control dependency between an oop load from an array and the corresponding "is-not-flat" check led to re-ordering of the load to before the check. Although the result is only used in the "not-flat" case, ZGC load barriers are still executed on that "oop" that is actually a non-oop field of the flattened inline type array element. In rare cases, the following barrier code screws up the contents of that field:

#0 0x00007fa990daf32e in Atomic::PlatformCmpxchg<8ul>::operator() (this=, exchange_value=, compare_value=,
dest=0x100000f17350) at /oracle/valhalla_int/open/src/hotspot/os_cpu/linux_x86/atomic_linux_x86.hpp:126
#1 Atomic::CmpxchgImpl<unsigned long, unsigned long, unsigned long, void>::operator() (this=, order=memory_order_conservative, exchange_value=21990232555242,
compare_value=, dest=0x100000f17350) at /oracle/valhalla_int/open/src/hotspot/share/runtime/atomic.hpp:745
#2 Atomic::cmpxchg<unsigned long, unsigned long, unsigned long> (order=memory_order_conservative, exchange_value=21990232555242, compare_value=, dest=0x100000f17350)
at /oracle/valhalla_int/open/src/hotspot/share/runtime/atomic.hpp:721
#3 ZBarrier::self_heal<&ZBarrier::is_good_or_null_fast_path> (heal_addr=, addr=, p=0x100000f17350)
at /oracle/valhalla_int/open/src/hotspot/share/gc/z/zBarrier.inline.hpp:120
#4 ZBarrier::self_heal<&ZBarrier::is_good_or_null_fast_path> (heal_addr=, addr=, p=0x100000f17350)
at /oracle/valhalla_int/open/src/hotspot/share/gc/z/zBarrier.inline.hpp:106
#5 ZBarrier::barrier<&ZBarrier::is_good_or_null_fast_path, &ZBarrier::load_barrier_on_oop_slow_path> (o=..., p=)
at /oracle/valhalla_int/open/src/hotspot/share/gc/z/zBarrier.inline.hpp:152
#6 ZBarrier::load_barrier_on_oop_field_preloaded (o=..., p=0x100000f17350) at /oracle/valhalla_int/open/src/hotspot/share/gc/z/zBarrier.inline.hpp:236
#7 ZBarrierSetRuntime::load_barrier_on_oop_field_preloaded (o=, p=0x100000f17350) at /oracle/valhalla_int/open/src/hotspot/share/gc/z/zBarrierSetRuntime.cpp:30
#8 0x00007fa9851a8774 in ?? ()
#9 0x0000000000000000 in ?? ()

---

Progress

• Change must not contain extraneous whitespace

Issue

• JDK-8247746: [lworld] TestNullableArray::test78 fails with ZGC due to incorrect result

Download

$ git fetch https://git.openjdk.java.net/valhalla pull/87/head:pull/87
$ git checkout pull/87

[bridgekeeper]

👋 Welcome back thartmann! A progress list of the required criteria for merging this PR into lworld will be added to the body of your pull request.

[openjdk]

@TobiHartmann This change now passes all automated pre-integration checks, type /integrate in a new comment to proceed. After integration, the commit message will be:

8247746: [lworld] TestNullableArray::test78 fails with ZGC due to incorrect result

• If you would like to add a summary, use the /summary command.
• To credit additional contributors, use the /contributor command.
• To add additional solved issues, use the /issue command.

There are currently no new commits on the lworld branch since the last update of the source branch of this PR. If another commit should be pushed before you perform the /integrate command, your PR will be automatically rebased. If you would like to avoid potential automatic rebasing, specify the current head hash when integrating, like this: /integrate 8072069c3685c772cbe65aa682a765d2c6802c7c.

➡️ To integrate this PR with the above commit message to the lworld branch, type /integrate in a new comment.

[mlbridge]

Webrevs

• 00: Full (3acd55d)

[mlbridge]

Mailing list message from Roland Westrelin on valhalla-dev:

Webrev: https://webrevs.openjdk.java.net/valhalla/87/webrev.00

Looks good to me.

Roland.

[mlbridge]

Mailing list message from Tobias Hartmann on valhalla-dev:

Thanks Roland!

Best regards,
Tobias

On 17.06.20 12:18, Roland Westrelin wrote:

Webrev: https://webrevs.openjdk.java.net/valhalla/87/webrev.00

Looks good to me.

Roland.

[TobiHartmann]

/integrate

[openjdk]

@TobiHartmann
Pushed as commit dc585c2.