RFR: 8258032: Reconsider LEAF entry restrictions

[Coleen Phillimore]

On Tue, 5 Jan 2021 11:46:30 GMT, Ron Pressler <rpressler at openjdk.org> wrote:

>> This change allows JRT_LEAF functions to create Handles because it doesn't need them but some code inside the VM needs Handles.  There's a NoSafepointVerifier in JRT_LEAF functions.
>> The JNI_LEAF and JVM_LEAF functions have NoHandleMark (so you can't create handles) but the transition ThreadInVMfromNative will reenable the HandleMarks, so that all this code doesn't have to do it itself.
>> Tested with tier1-6.
>
> src/hotspot/share/code/compiledMethod.cpp line 707:
> 
>> 705:     Thread* thread = Thread::current();
>> 706:     ResetNoHandleMark rnm; // Might be called from LEAF/QUICK ENTRY
>> 707:     HandleMark hm(thread);
> 
> Why was the `HandleMark` removed here? Is it not needed?

There are no Handles in this block.  There are a lot of rogue HandleMarks around because of PermGen.  They don't do anything harmful, but they don't hold any Handles either.

> src/hotspot/share/runtime/interfaceSupport.inline.hpp line 353:
> 
>> 351: #define VM_LEAF_BASE(result_type, header)                            \
>> 352:   TRACE_CALL(result_type, header)                                    \
>> 353:   debug_only(NoHandleMark __hm;)                                     \
> 
> Shouldn't `NoSafepointVerifier` be added here and then it can be removed from `JRT_LEAF`?

I don't want a NoSafepointVerifier in JNI_LEAF, and the equivalent fromNative JVM_LEAF, which "calls" this.  I only want it in JRT_LEAF.  JNI_LEAF isn't really a leaf, it often calls JNI functions that transition to the VM.

> src/hotspot/share/runtime/interfaceSupport.inline.hpp line 484:
> 
>> 482:   result_type JNICALL header {                                       \
>> 483:     VM_Exit::block_if_vm_exited();                                   \
>> 484:     debug_only(NoHandleMark __hm;)                                   \
> 
> Why does `JVM_LEAF` need `NoHandleMark`?

JVM_LEAF comes from _thread_in_native so there shouldn't be oops or Handles.  It's really equivalent to JNI_LEAF.  I tried to make sense of this in the CR report, mostly so I'd remember what I was thinking.

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

PR: https://git.openjdk.java.net/jdk/pull/1846