Add var handle method type lookup table walking

[cjjdespres]

The content of the varHandleMethodTypeLookupTable is now walked in allSlotsInROMClassDo.

The relevant section of the ROM class writer code is here, I believe:

openj9/runtime/bcutil/ROMClassWriter.cpp

Lines 887 to 910 in 6a6d7af

	#if defined(J9VM_OPT_METHOD_HANDLE)
	void writeVarHandleMethodTypeLookupTable()
	{
	if (!_markAndCountOnly) {
	U_16 lookupTableCount = _constantPoolMap->getVarHandleMethodTypeCount();
	if (lookupTableCount > 0) {
	U_16 *lookupTable = _constantPoolMap->getVarHandleMethodTypeLookupTable();
	/* The lookup table is a U_16[], so we need to pad it in order to keep the J9ROMClass alignment. */
	U_16 lookupTablePaddedCount = _constantPoolMap->getVarHandleMethodTypePaddedCount();
	U_16 *lookupTablePadding = lookupTable + lookupTableCount;
	UDATA lookupTablePaddingLength = lookupTablePaddedCount - lookupTableCount;
	/* Write the data section of the array */
	_cursor->writeData((U_8 *)lookupTable, lookupTableCount * sizeof(U_16), Cursor::GENERIC);
	/* Write the padding section of the array */
	if (lookupTablePaddingLength > 0) {
	_cursor->writeData((U_8 *)lookupTablePadding, lookupTablePaddingLength * sizeof(U_16), Cursor::GENERIC);
	}
	}
	}
	}
	#endif /* defined(J9VM_OPT_METHOD_HANDLE) */

I think the individual slots are "cpIndex", but I'm unsure. I also made up the new section name "varHandleMethodTypeLookupTable".

I also updated some of the documentation surrounding JITServerHelpers::packROMClass.

Fixes: #18957

[cjjdespres]

Attn @mpirvu. I think @tajila suggested someone to review my previous ROM class walking PRs. Thanks.

[mpirvu]

@babsingh Could you please review this PR or delegate? Thanks

[mpirvu]

jenkins test sanity plinuxjit,xlinuxjit,zlinuxjit,alinux64jit jdk11

[mpirvu]

jenkins test sanity.openjdk,extended.openjdk xlinuxjit,plinuxjit,zlinuxjit jdk11

[mpirvu]

The assert reported in #18957 is no longer seen with this fix

[babsingh]

I think the individual slots are "cpIndex", but I'm unsure.

Your changes are correct. varHandleMethodTypeLookupTable stores CP indices for VarHandle methodrefs. Processing them similar to the adjacent allSlotsIn*MethodRefIndexesDo functions is the correct approach.

[babsingh]

Commit message should be strengthened by adding why these changes are needed and what is fixed. Currently, the commit message only has a title.

[cjjdespres]

Force-pushed to address comments.

[babsingh]

LGTM. Old PR builds should still be good since the recent changes don't impact functionality. A JIT server file is also modified; I will defer the merge to a JIT member.

Old PR builds:

• https://openj9-jenkins.osuosl.org/view/Pull%20Requests/job/PullRequest-OpenJ9/5096
• https://openj9-jenkins.osuosl.org/view/Pull%20Requests/job/PullRequest-OpenJ9/5097

[mpirvu]

A JIT server file is also modified; I will defer the merge to a JIT member.

The JITServer file only contains comments. This PR should be good to merge.

[keithc-ca]

@hangshao0 Should the shared class cache version have been updated with this change?

Should debugtools/DDR_VM/src/com/ibm/j9ddr/vm29/tools/ddrinteractive/RomClassWalker.java not have been updated similarly?

[hangshao0]

@hangshao0 Should the shared class cache version have been updated with this change?

Is the layout of the rom class changed ? If yes, the shared cache version should be updated.

[cjjdespres]

The ROM class layout is unchanged. Only the ROM class walker in romclasswalk.c was changed, to walk an already-existing section that the walker had been missing.

[keithc-ca]

Ok, it would seem to be missing in DDR as well. Is someone working to correct that?

[cjjdespres]

Not that I know of. I wasn't aware of RomClassWalker.java until you mentioned it. I'm not familiar with the Java side of ROM class walking.

[babsingh]

Ok, it would seem to be missing in DDR as well. Is someone working to correct that?

I will take a look into it. It is associated to OpenJ9 MethodHandles, which we are trying to replace with OpenJDK MethodHandles in JDK8/JDK11. Since we plan to disable OJ9 MHs, this will take low priority.

[keithc-ca]

Even after we stop using OpenJ9 method handles, the DDR part will be useful for examining core files produced by VMs that predate the adoption of openjdk method handles.