RFR: 8351487: [ubsan] jvmti.h runtime error: load of value which is not a valid value

[Chris Plummer]

On Mon, 7 Jul 2025 12:21:06 GMT, Matthias Baesken <mbaesken at openjdk.org> wrote:

> When running with ubsan - enabled binaries, the following 2 issues are observed in jvmti.h (on macOS aarch64).
> 
> 
> jtreg test vmTestbase/nsk/jvmti/SetVerboseFlag/setvrbflag002/TestDescription.java 
> 
> /priv/jenkins/client-home/workspace/openjdk-jdk-weekly-macos_aarch64-opt/build/support/modules_include/java.base/jvmti.h:2645:44: runtime error: load of value 4294967295, which is not a valid value for type 'jvmtiVerboseFlag'
>     #0 0x104453540 in _jvmtiEnv::SetVerboseFlag(jvmtiVerboseFlag, unsigned char) jvmti.h:2645
>     #1 0x10444ed5c in agentProc(_jvmtiEnv*, JNIEnv_*, void*) setvrbflag002.cpp:49
>     #2 0x10444f1c8 in agentThreadWrapper(_jvmtiEnv*, JNIEnv_*, void*) agent_tools.cpp:151
>     #3 0x10914895c in JvmtiAgentThread::call_start_function() jvmtiImpl.cpp:89
>     #4 0x108ebd7f0 in JavaThread::thread_main_inner() javaThread.cpp:776
>     #5 0x1096f6094 in Thread::call_run() thread.cpp:231
>     #6 0x10941c37c in thread_native_entry(Thread*) os_bsd.cpp:601
>     #7 0x1936fef90 in _pthread_start+0x84 (libsystem_pthread.dylib:arm64e+0x6f90)
>     #8 0x1936f9d30 in thread_start+0x4 (libsystem_pthread.dylib:arm64e+0x1d30)
> 
> vmTestbase/nsk/jvmti/GetErrorName/geterrname002/TestDescription.java
> 
> /priv/jenkins/client-home/workspace/openjdk-jdk-weekly-macos_aarch64-opt/build/support/modules_include/java.base/jvmti.h:2640:42: runtime error: load of value 4294967295, which is not a valid value for type 'jvmtiError'
>     #0 0x1002b3504 in _jvmtiEnv::GetErrorName(jvmtiError, char**) jvmti.h:2640
>     #1 0x1002aec9c in agentProc(_jvmtiEnv*, JNIEnv_*, void*) geterrname002.cpp:50
>     #2 0x1002af198 in agentThreadWrapper(_jvmtiEnv*, JNIEnv_*, void*) agent_tools.cpp:151
>     #3 0x104f4895c in JvmtiAgentThread::call_start_function() jvmtiImpl.cpp:89
>     #4 0x104cbd7f0 in JavaThread::thread_main_inner() javaThread.cpp:776
>     #5 0x1054f6094 in Thread::call_run() thread.cpp:231
>     #6 0x10521c37c in thread_native_entry(Thread*) os_bsd.cpp:601
>     #7 0x1936fef90 in _pthread_start+0x84 (libsystem_pthread.dylib:arm64e+0x6f90)
>     #8 0x1936f9d30 in thread_start+0x4 (libsystem_pthread.dylib:arm64e+0x1d30)
> 
> 
> So the problem reported is SetVerboseFlag method (which is part of the VM) calls SetVerboseFlag function with that bad argument ;  but the test code seems to intentionally use 'bad' arguments . But this does not work well with ubsan ;  so better exclude those tests from ubsan checking .
> Please note that jvmti.h is shipped so adding some macros/attributes to methods for avoiding ubsan  is probab...

This is somewhat of an odd situation. SetVerboseFlag() says the argument type is an enum, but it treats it like an int and does error checking to make sure it one of the enum values. If ubsan dug a little deeper to see how the argument is handled, it would see that and not complain.

test/hotspot/jtreg/vmTestbase/nsk/jvmti/GetErrorName/geterrname002/TestDescription.java line 38:

> 36:  * COMMENTS
> 37:  *
> 38:  * @requires !vm.ubsan

Needs an @comment

test/hotspot/jtreg/vmTestbase/nsk/jvmti/SetVerboseFlag/setvrbflag002/TestDescription.java line 37:

> 35:  * COMMENTS
> 36:  *
> 37:  * @requires !vm.ubsan

Needs an @comment

-------------

Changes requested by cjplummer (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/26159#pullrequestreview-2995047256
PR Review Comment: https://git.openjdk.org/jdk/pull/26159#discussion_r2190865703
PR Review Comment: https://git.openjdk.org/jdk/pull/26159#discussion_r2190866138