Introduce JVMPortableRestoreMode

[tajila]

Previously, the CRIURestoreNonPortableMode was used to determine if a checkpoint could be taken after restore. If CRIURestoreNonPortableMode was enabled (default mode) only a single checkpoint is allowed and the JVM doesn't need to generate portable code upon restore. This capability also allows the JVM to use standard security providers since there is no risk for that state to be serialized.

There is a need to separate out the portability capabilites and the behavioural changes that CRIURestoreNonPortableMode offers. There are cases where one wants portability as well as standard security capabilites for debugging purposes.

This PR separates out the portability capabilites and the behavioural aspects by providing an option that forces portability upon restore regardless of what CRIURestoreNonPortableMode is set to.

In the future we can add another option that forces portability in non-CRIU mode, it its desired.

[tajila]

@dsouzai is there a way to turn on JIT logging to confirm it is generating portable JIT code?

[dsouzai]

I don't see any logging. What you could do is something like this:

diff --git a/runtime/compiler/control/HookedByTheJit.cpp b/runtime/compiler/control/HookedByTheJit.cpp
index 22dfb172c..79bef5591 100644
--- a/runtime/compiler/control/HookedByTheJit.cpp
+++ b/runtime/compiler/control/HookedByTheJit.cpp
@@ -1862,6 +1862,12 @@ static void jitHookPrepareRestore(J9HookInterface * * hookInterface, UDATA event

       /* Reinitialize the (technically unused) targetProcesssorInfo on x86 to prevent asserts */
       TR::Compiler->target.cpu.initializeTargetProcessorInfo(true);
+
+      printf("Will no longer generate portable code\n");
+      }
+   else
+      {
+      printf("Will continue to generate portable code\n");
       }

    TR::CompilationInfo * compInfo = TR::CompilationInfo::get(jitConfig);
diff --git a/runtime/compiler/control/rossa.cpp b/runtime/compiler/control/rossa.cpp
index 835fbde10..8a104f1e3 100644
--- a/runtime/compiler/control/rossa.cpp
+++ b/runtime/compiler/control/rossa.cpp
@@ -2005,6 +2005,8 @@ aboutToBootstrap(J9JavaVM * javaVM, J9JITConfig * jitConfig)
       if (!J9_ARE_ANY_BITS_SET(javaVM->extendedRuntimeFlags2, J9_EXTENDED_RUNTIME2_ENABLE_PORTABLE_SHARED_CACHE))
          TR::Compiler->relocatableTarget.cpu = TR::CPU::detectRelocatable(TR::Compiler->omrPortLib);
       jitConfig->targetProcessor = TR::Compiler->target.cpu.getProcessorDescription();
+
+      printf("Will generate portable code");
       }

[dsouzai]

The definition of isJVMInPortableRestoreMode is

BOOLEAN
isJVMInPortableRestoreMode(J9VMThread *currentThread)
{
	return !isNonPortableRestoreMode(currentThread) || J9_ARE_ALL_BITS_SET(currentThread->javaVM->checkpointState.flags, J9VM_CRIU_IS_PORTABLE_JVM_RESTORE_MODE);
}

which means that it'll return true if J9VM_CRIU_IS_NON_PORTABLE_RESTORE_MODE is not set. However, this query doesn't check whether CRIU Support is enabled. So, in all the places where isCheckpointAllowed is replaced by isJVMInPortableRestoreMode, aren't we going to execute that code even when CRIU is disabled?

[tajila]

So, in all the places where isCheckpointAllowed is replaced by isJVMInPortableRestoreMode, aren't we going to execute that code even when CRIU is disabled?

Ya, Ive corrected that.

[dsouzai]

Overall LGTM; just requesting a couple of changes.

[dsouzai]

Oh! I just recalled, you also need to change the MINOR_NUMBER in CommunicationStream.hpp since a new front end query is being added.

[tajila]

Ready for another look

[dsouzai]

Minor changes requested.

[tajila]

Ready for another look

[dsouzai]

jenkins test sanity all jdk17,jdk21

[tajila]

All test passed in the windows run, just an infra issue: https://openj9-jenkins.osuosl.org/job/Test_openjdk17_j9_sanity.functional_x86-64_windows_Personal/315/tapResults/.

testServerComesUpAfterClient, testServerGoesDownAnotherComesUp, testServerUnreachableForAWhile failed on PPC (17 and 21), known issue #14594