java.lang.IllegalAccessException thrown for jdk.internal.misc.Unsafe.theUnsafe in Valhalla vt_standard testing

[hzongaro]

Many test failures are seen in recent PR sanity.functional testing with {x|z|a|p}linuxvalst (i.e., vt_standard) for pull requests #20112 and #20332. In most, if not all, cases IllegalAccessExceptions are thrown. I was able to reproduce this in an internal personal build with the latest merged source.

12:45:24  openjdk version "24-internal" 2025-03-18
12:45:24  OpenJDK Runtime Environment (build 24-internal-adhoc.jenkins.BuildJDKnextx86-64linuxvtstandardPersonal)
12:45:24  Eclipse OpenJ9 VM (build master-d51176e7598, JRE 24 Linux amd64-64-Bit Compressed References 20241016_390 (JIT enabled, AOT enabled)
12:45:24  OpenJ9   - d51176e7598
12:45:24  OMR      - 0de486a739f
12:45:24  JCL      - 972edea3789 based on jdk-24+12)

13:00:49  Testing: Create and Restore Criu Checkpoint Image once - CheckpointDeadlock
13:00:49  Test start time: 2024/10/16 10:00:48 Pacific Standard Time
13:00:49  Running command: bash /home/jenkins/workspace/Test_openjdknext_j9_sanity.functional_x86-64_linux_vt_standard_Personal_testList_1/aqa-tests/TKG/../../jvmtest/functional/cmdLineTests/criu/criuScript.sh /home/jenkins/workspace/Test_openjdknext_j9_sanity.functional_x86-64_linux_vt_standard_Personal_testList_1/aqa-tests/TKG/../../jvmtest/functional/cmdLineTests/criu /home/jenkins/workspace/Test_openjdknext_j9_sanity.functional_x86-64_linux_vt_standard_Personal_testList_1/jdkbinary/j2sdk-image/bin/java " -Xjit -XX:+CRIURestoreNonPortableMode  --add-opens java.base/jdk.internal.misc=ALL-UNNAMED  --add-exports java.base/openj9.internal.criu=ALL-UNNAMED" org.openj9.criu.DeadlockTest CheckpointDeadlock 1 false false
13:00:49  Time spent starting: 2 milliseconds
13:00:51  Time spent executing: 2266 milliseconds
13:00:51  Test result: FAILED
13:00:51  Output from test:
13:00:51   [OUT] start running script
13:00:51   [OUT] export GLIBC_TUNABLES=glibc.cpu.hwcaps=-XSAVEC,-XSAVE,-AVX2,-ERMS,-AVX,-AVX_Fast_Unaligned_Load
13:00:51   [OUT] export LD_BIND_NOT=on
13:00:51   [OUT] /home/jenkins/workspace/Test_openjdknext_j9_sanity.functional_x86-64_linux_vt_standard_Personal_testList_1/jdkbinary/j2sdk-image/bin/java -XX:+EnableCRIUSupport  -Xjit -XX:+CRIURestoreNonPortableMode  --add-opens java.base/jdk.internal.misc=ALL-UNNAMED  --add-exports java.base/openj9.internal.criu=ALL-UNNAMED -cp /home/jenkins/workspace/Test_openjdknext_j9_sanity.functional_x86-64_linux_vt_standard_Personal_testList_1/aqa-tests/TKG/../../jvmtest/functional/cmdLineTests/criu/criu.jar org.openj9.criu.DeadlockTest CheckpointDeadlock 1
13:00:51   [OUT] Exception in thread "main" java.lang.InternalError: java.lang.IllegalAccessException: static final field has no write access: jdk.internal.misc.Unsafe.theUnsafe/jdk.internal.misc.Unsafe/putStatic, from class java.lang.Object (module java.base)
13:00:51   [OUT] 	at java.base/jdk.internal.reflect.MethodHandleAccessorFactory.newFieldAccessor(MethodHandleAccessorFactory.java:207)
13:00:51   [OUT] 	at java.base/jdk.internal.reflect.ReflectionFactory.newFieldAccessor(ReflectionFactory.java:144)
13:00:51   [OUT] 	at java.base/java.lang.reflect.Field.acquireOverrideFieldAccessor(Field.java:1212)
13:00:51   [OUT] 	at java.base/java.lang.reflect.Field.getOverrideFieldAccessor(Field.java:1181)
13:00:51   [OUT] 	at java.base/java.lang.reflect.Field.get(Field.java:454)
13:00:51   [OUT] 	at org.openj9.criu.DeadlockTest.<clinit>(DeadlockTest.java:356)
13:00:51   [OUT] Caused by: java.lang.IllegalAccessException: static final field has no write access: jdk.internal.misc.Unsafe.theUnsafe/jdk.internal.misc.Unsafe/putStatic, from class java.lang.Object (module java.base)
13:00:51   [OUT] 	at java.base/java.lang.invoke.MemberName.makeAccessException(MemberName.java:913)
13:00:51   [OUT] 	at java.base/java.lang.invoke.MethodHandles$Lookup.unreflectField(MethodHandles.java:3608)
13:00:51   [OUT] 	at java.base/java.lang.invoke.MethodHandles$Lookup.unreflectSetter(MethodHandles.java:3599)
13:00:51   [OUT] 	at java.base/java.lang.invoke.MethodHandleImpl$1.unreflectField(MethodHandleImpl.java:1621)
13:00:51   [OUT] 	at java.base/jdk.internal.reflect.MethodHandleAccessorFactory.newFieldAccessor(MethodHandleAccessorFactory.java:185)
13:00:51   [OUT] 	... 5 more
13:00:51   [OUT] initiate restore
13:00:51   [OUT] Error (criu/protobuf.c:72): Unexpected EOF on (empty-image)
13:00:51   [OUT] Removed test output files
13:00:51   [OUT] finished script

[hzongaro]

@theresa-m, @a7ehuo - FYI

[github-actions]

Issue Number: 20372
Status: Open
Recommended Components: comp:test, comp:vm, comp:gc
Recommended Assignees: jasonfengj9, tajila, babsingh

[hangshao0]

The failed line is:

openj9/test/functional/cmdLineTests/criu/src/org/openj9/criu/DeadlockTest.java

Line 356 in 64cb0db

unsafe = (Unsafe) f.get(null);

@JasonFengJ9 Do you know if this fails on regular JDK24 build ?

[JasonFengJ9]

Do you know if this fails on regular JDK24 build ?

Currently, there are no JDK24 nightly builds.
Last openjdknext_j9_sanity.functional_aarch64_mac passed on Oct. 3rd.
Launched a JDKnext_x86-64_linux.

[JasonFengJ9]

Those failed tests in openjdknext_j9_sanity.functional_x86-64_linux_vt_standard passed in a regular openjdkNext_j9_sanity.functional_x86-64_linux

cmdLineTester_criu_nonPortableRestore_0_PASSED
...
cmdLineTester_libpathTestRtf_0_PASSED
...

FYI @hangshao0

[hangshao0]

Have you seen this failure before ? @babsingh @fengxue-IS

The https://github.com/ibmruntimes/openj9-openjdk-jdk.valuetypes repo is a few weeks behind https://github.com/ibmruntimes/openj9-openjdk-jdk though.

[keithc-ca]

Two possible solutions come to mind:

1. update https://github.com/ibmruntimes/openj9-openjdk-jdk.valuetypes - perhaps this is already fixed upstream
2. change the way the Unsafe instance is acquired in tests - we're already using --add-opens to gain access to the package so we could just use Unsafe.getUnsafe() rather than reflection

[babsingh]

Have you seen this failure before ?

It seems new. I haven't seen it before.

[theresa-m]

I tried the failing code with old Valhalla nightly builds and it started failing after the https://github.com/ibmruntimes/openj9-openjdk-jdk.valuetypes update in early May.

[theresa-m]

I looked through the merge commits of these files from the May extensions update but did not see anything inconsistent with the https://github.com/ibmruntimes/openj9-openjdk-jdk repository. This list is a subset of files identified as having had conflicts in ibmruntimes/openj9-openjdk-jdk.valuetypes#10, ibmruntimes/openj9-openjdk-jdk.valuetypes#11, ibmruntimes/openj9-openjdk-jdk.valuetypes#12

src/java.base/share/classes/sun/invoke/util/VerifyAccess.java
src/java.base/share/classes/java/lang/reflect/Modifier.java
src/java.base/share/classes/java/lang/invoke/MethodHandles.java
src/java.base/share/classes/java/lang/invoke/MethodHandleProxies.java
src/java.base/share/classes/jdk/internal/misc/Unsafe.java
src/java.base/share/classes/sun/invoke/util/Wrapper.java
src/java.base/share/classes/java/lang/reflect/ProxyGenerator.java
src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java

[keithc-ca]

I'm inclined to believe that openjdk hasn't finished updating the Field implementation. The usage pattern (seen in the description) should not require generation of a setter, but an override field is never considered read-only. See ReflectionFactory.java.