6960970: Debugger very slow during stepping

Co-authored-by: Patricio Chilano Mateo <pchilanomate@openjdk.org>
Reviewed-by: lmesnik, pchilanomate

Author: Serguei Spitsyn

src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp

@@ -1174,7 +1174,7 @@ void InterpreterMacroAssembler::notify_method_exit(
   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
   // track stack depth.  If it is possible to enter interp_only_mode we add
   // the code to check if the event should be sent.
-  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
+  if (mode == NotifyJVMTI && (JvmtiExport::can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     Label L;
     // Note: frame::interpreter_frame_result has a dependency on how the
     // method result is saved across the call to post_method_exit. If this
@@ -1183,8 +1183,15 @@ void InterpreterMacroAssembler::notify_method_exit(
 
     // template interpreter will leave the result on the top of the stack.
     push(state);
-    ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
-    cbz(r3, L);
+
+    ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
+    cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
+
+    ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::frame_pop_cnt_offset()));
+    ldrw(rscratch2, Address(rthread, JavaThread::interp_only_mode_offset()));
+    orrw(rscratch1, rscratch1, rscratch2);
+    cbzw(rscratch1, L);
+
     call_VM(noreg,
             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
     bind(L);

src/hotspot/cpu/arm/interp_masm_arm.cpp

@@ -1576,14 +1576,21 @@ void InterpreterMacroAssembler::notify_method_exit(
   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
   // track stack depth.  If it is possible to enter interp_only_mode we add
   // the code to check if the event should be sent.
-  if (mode == NotifyJVMTI && can_post_interpreter_events()) {
+  if (mode == NotifyJVMTI && (can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     Label L;
+    const Register thread_state = R2_tmp;
+
     // Note: frame::interpreter_frame_result has a dependency on how the
     // method result is saved across the call to post_method_exit. If this
     // is changed then the interpreter_frame_result implementation will
     // need to be updated too.
 
+    ldr(thread_state, Address(Rthread, JavaThread::jvmti_thread_state_offset()));
+    cbz(thread_state, L); // if (thread->jvmti_thread_state() == nullptr) exit;
+
+    ldr_s32(thread_state, Address(thread_state, JvmtiThreadState::frame_pop_cnt_offset()));
     ldr_s32(Rtemp, Address(Rthread, JavaThread::interp_only_mode_offset()));
+    orr(Rtemp, Rtemp, thread_state);
     cbz(Rtemp, L);
 
     if (native) {

src/hotspot/cpu/ppc/interp_masm_ppc_64.cpp

@@ -2345,18 +2345,12 @@ void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosSta
   // entry/exit events are sent for that thread to track stack
   // depth. If it is possible to enter interp_only_mode we add
   // the code to check if the event should be sent.
-  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
-    Label jvmti_post_done;
-
-    lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread);
-    cmpwi(CR0, R0, 0);
-    beq(CR0, jvmti_post_done);
+  if (mode == NotifyJVMTI && (JvmtiExport::can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     if (!is_native_method) { push(state); } // Expose tos to GC.
     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), check_exceptions);
     if (!is_native_method) { pop(state); }
 
     align(32, 12);
-    bind(jvmti_post_done);
   }
 
   // Dtrace support not implemented.

src/hotspot/cpu/riscv/interp_masm_riscv.cpp

@@ -1220,20 +1220,16 @@ void InterpreterMacroAssembler::notify_method_exit(
   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
   // track stack depth.  If it is possible to enter interp_only_mode we add
   // the code to check if the event should be sent.
-  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
-    Label L;
+  if (mode == NotifyJVMTI && (JvmtiExport::can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     // Note: frame::interpreter_frame_result has a dependency on how the
     // method result is saved across the call to post_method_exit. If this
     // is changed then the interpreter_frame_result implementation will
     // need to be updated too.
 
     // template interpreter will leave the result on the top of the stack.
     push(state);
-    lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
-    beqz(x13, L);
     call_VM(noreg,
             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
-    bind(L);
     pop(state);
   }
 

src/hotspot/cpu/s390/interp_masm_s390.cpp

@@ -2002,14 +2002,10 @@ void InterpreterMacroAssembler::notify_method_exit(bool native_method,
   // entry/exit events are sent for that thread to track stack
   // depth. If it is possible to enter interp_only_mode we add
   // the code to check if the event should be sent.
-  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
-    Label jvmti_post_done;
-    MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
-    z_bre(jvmti_post_done);
+  if (mode == NotifyJVMTI && (JvmtiExport::can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     if (!native_method) push(state); // see frame::interpreter_frame_result()
     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
     if (!native_method) pop(state);
-    bind(jvmti_post_done);
   }
 }
 

src/hotspot/cpu/x86/interp_masm_x86.cpp

@@ -1603,7 +1603,7 @@ void InterpreterMacroAssembler::notify_method_exit(
   // the code to check if the event should be sent.
   Register rthread = r15_thread;
   Register rarg = c_rarg1;
-  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
+  if (mode == NotifyJVMTI && (JvmtiExport::can_post_interpreter_events() || JvmtiExport::can_post_frame_pop())) {
     Label L;
     // Note: frame::interpreter_frame_result has a dependency on how the
     // method result is saved across the call to post_method_exit. If this
@@ -1612,9 +1612,18 @@ void InterpreterMacroAssembler::notify_method_exit(
 
     // template interpreter will leave the result on the top of the stack.
     push(state);
-    movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
-    testl(rdx, rdx);
+
+    movptr(rdx, Address(rthread, JavaThread::jvmti_thread_state_offset()));
+    testptr(rdx, rdx);
+    jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
+
+    movl(rdx, Address(rdx, JvmtiThreadState::frame_pop_cnt_offset()));
+    movl(rcx, Address(rthread, JavaThread::interp_only_mode_offset()));
+
+    orl(rdx, rcx);
+    testl(rdx,rdx);
     jcc(Assembler::zero, L);
+
     call_VM(noreg,
             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
     bind(L);

src/hotspot/share/interpreter/interpreterRuntime.cpp

@@ -1057,6 +1057,10 @@ JRT_ENTRY(nmethod*,
 
   LastFrameAccessor last_frame(current);
   assert(last_frame.is_interpreted_frame(), "must come from interpreter");
+
+  if (JvmtiExport::can_post_frame_pop() && JvmtiExport::has_frame_pop_for_top_frame(current)) {
+    return nullptr; // no OSR if there is a FramePop event request for top frame
+  }
   methodHandle method(current, last_frame.method());
   const int branch_bci = branch_bcp != nullptr ? method->bci_from(branch_bcp) : InvocationEntryBci;
   const int bci = branch_bcp != nullptr ? method->bci_from(last_frame.bcp()) : InvocationEntryBci;

src/hotspot/share/interpreter/zero/bytecodeInterpreter.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2002, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 2026, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -3145,10 +3145,14 @@ void BytecodeInterpreter::run(interpreterState istate) {
     // Whenever JVMTI puts a thread in interp_only_mode, method
     // entry/exit events are sent for that thread to track stack depth.
 
-    if (JVMTI_ENABLED && !suppress_exit_event && THREAD->is_interp_only_mode()) {
-      // Prevent any HandleMarkCleaner from freeing our live handles
-      HandleMark __hm(THREAD);
-      CALL_VM_NOCHECK(InterpreterRuntime::post_method_exit(THREAD));
+    if (JVMTI_ENABLED && !suppress_exit_event) {
+      JvmtiThreadState* state = THREAD->jvmti_thread_state();
+      int frame_pop_cnt = state == nullptr ? 0 : state->frame_pop_cnt();
+      if (THREAD->is_interp_only_mode() || frame_pop_cnt) {
+        // Prevent any HandleMarkCleaner from freeing our live handles
+        HandleMark __hm(THREAD);
+        CALL_VM_NOCHECK(InterpreterRuntime::post_method_exit(THREAD));
+      }
     }
 
     //

src/hotspot/share/oops/stackChunkOop.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -154,6 +154,7 @@ class stackChunkOopDesc : public instanceOopDesc {
   inline void set_has_bitmap(bool value);
 
   inline bool has_thaw_slowpath_condition() const;
+  inline void force_slow_path();
 
   inline bool requires_barriers();
 

src/hotspot/share/oops/stackChunkOop.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -192,6 +192,15 @@ inline void stackChunkOopDesc::set_has_bitmap(bool value)          { set_flag(FL
 
 inline bool stackChunkOopDesc::has_thaw_slowpath_condition() const { return flags() != 0; }
 
+inline void stackChunkOopDesc::force_slow_path() {
+#if INCLUDE_ZGC || INCLUDE_SHENANDOAHGC
+  if (UseZGC || UseShenandoahGC) {
+    relativize_derived_pointers_concurrently();
+  }
+#endif
+  set_flag(FLAG_HAS_INTERPRETED_FRAMES, true);
+}
+
 inline bool stackChunkOopDesc::requires_barriers() {
   return Universe::heap()->requires_barriers(this);
 }