8286476: x86_32: Fix crashes with non-preview mode after JDK-8284161 (Virtual Threads)

Reviewed-by: coleenp, stuefe

src/hotspot/cpu/x86/stubGenerator_x86_32.cpp

@@ -3866,6 +3866,104 @@ class StubGenerator: public StubCodeGenerator {
     StubRoutines::x86::_fpu_subnormal_bias2[2]= 0x7bff;
   }
 
+  RuntimeStub* generate_cont_doYield() {
+    if (!Continuations::enabled()) return nullptr;
+    Unimplemented();
+    return nullptr;
+  }
+
+  address generate_cont_thaw(bool return_barrier, bool exception) {
+    if (!Continuations::enabled()) return nullptr;
+    Unimplemented();
+    return nullptr;
+  }
+
+  address generate_cont_thaw() {
+    if (!Continuations::enabled()) return nullptr;
+    Unimplemented();
+    return nullptr;
+  }
+
+  address generate_cont_returnBarrier() {
+    if (!Continuations::enabled()) return nullptr;
+    Unimplemented();
+    return nullptr;
+  }
+
+  address generate_cont_returnBarrier_exception() {
+    if (!Continuations::enabled()) return nullptr;
+    Unimplemented();
+    return nullptr;
+  }
+
+#if INCLUDE_JFR
+
+  static void jfr_prologue(address the_pc, MacroAssembler* masm) {
+    Register java_thread = rdi;
+    __ get_thread(java_thread);
+    __ set_last_Java_frame(java_thread, rsp, rbp, the_pc);
+    __ movptr(Address(rsp, 0), java_thread);
+  }
+
+  // The handle is dereferenced through a load barrier.
+  static void jfr_epilogue(MacroAssembler* masm) {
+    Register java_thread = rdi;
+    __ get_thread(java_thread);
+    __ reset_last_Java_frame(java_thread, true);
+    Label null_jobject;
+    __ testptr(rax, rax);
+    __ jcc(Assembler::zero, null_jobject);
+    DecoratorSet decorators = ACCESS_READ | IN_NATIVE;
+    BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+    bs->load_at(masm, decorators, T_OBJECT, rax, Address(rax, 0), noreg, java_thread);
+    __ bind(null_jobject);
+  }
+
+  // For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
+  // It returns a jobject handle to the event writer.
+  // The handle is dereferenced and the return value is the event writer oop.
+  static RuntimeStub* generate_jfr_write_checkpoint() {
+    enum layout {
+      FPUState_off         = 0,
+      rbp_off              = FPUStateSizeInWords,
+      rdi_off,
+      rsi_off,
+      rcx_off,
+      rbx_off,
+      saved_argument_off,
+      saved_argument_off2, // 2nd half of double
+      framesize
+    };
+
+    int insts_size = 512;
+    int locs_size = 64;
+    CodeBuffer code("jfr_write_checkpoint", insts_size, locs_size);
+    OopMapSet* oop_maps = new OopMapSet();
+    MacroAssembler* masm = new MacroAssembler(&code);
+    MacroAssembler* _masm = masm;
+
+    address start = __ pc();
+    __ enter();
+    int frame_complete = __ pc() - start;
+    address the_pc = __ pc();
+    jfr_prologue(the_pc, _masm);
+    __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
+    jfr_epilogue(_masm);
+    __ leave();
+    __ ret(0);
+
+    OopMap* map = new OopMap(framesize, 1); // rbp
+    oop_maps->add_gc_map(the_pc - start, map);
+
+    RuntimeStub* stub = // codeBlob framesize is in words (not VMRegImpl::slot_size)
+      RuntimeStub::new_runtime_stub("jfr_write_checkpoint", &code, frame_complete,
+                                    (framesize >> (LogBytesPerWord - LogBytesPerInt)),
+                                    oop_maps, false);
+    return stub;
+  }
+
+#endif // INCLUDE_JFR
+
   //---------------------------------------------------------------------------
   // Initialization
 
@@ -3953,6 +4051,19 @@ class StubGenerator: public StubCodeGenerator {
     }
   }
 
+  void generate_phase1() {
+    // Continuation stubs:
+    StubRoutines::_cont_thaw          = generate_cont_thaw();
+    StubRoutines::_cont_returnBarrier = generate_cont_returnBarrier();
+    StubRoutines::_cont_returnBarrierExc = generate_cont_returnBarrier_exception();
+    StubRoutines::_cont_doYield_stub = generate_cont_doYield();
+    StubRoutines::_cont_doYield      = StubRoutines::_cont_doYield_stub == nullptr ? nullptr
+                                                                                   : StubRoutines::_cont_doYield_stub->entry_point();
+
+    JFR_ONLY(StubRoutines::_jfr_write_checkpoint_stub = generate_jfr_write_checkpoint();)
+    JFR_ONLY(StubRoutines::_jfr_write_checkpoint = StubRoutines::_jfr_write_checkpoint_stub->entry_point();)
+  }
+
   void generate_all() {
     // Generates all stubs and initializes the entry points
 
@@ -4044,11 +4155,13 @@ class StubGenerator: public StubCodeGenerator {
 
 
  public:
-  StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
-    if (all) {
-      generate_all();
-    } else {
+  StubGenerator(CodeBuffer* code, int phase) : StubCodeGenerator(code) {
+    if (phase == 0) {
       generate_initial();
+    } else if (phase == 1) {
+      generate_phase1(); // stubs that must be available for the interpreter
+    } else {
+      generate_all();
     }
   }
 }; // end class declaration

src/hotspot/share/runtime/signature.cpp

@@ -116,7 +116,7 @@ ReferenceArgumentCount::ReferenceArgumentCount(Symbol* signature)
   do_parameters_on(this);  // non-virtual template execution
 }
 
-#ifdef ASSERT
+#if !defined(_LP64) || defined(ZERO) || defined(ASSERT)
 static int compute_num_stack_arg_slots(Symbol* signature, int sizeargs, bool is_static) {
   ResourceMark rm;
   BasicType* sig_bt = NEW_RESOURCE_ARRAY(BasicType, sizeargs);
@@ -138,7 +138,7 @@ static int compute_num_stack_arg_slots(Symbol* signature, int sizeargs, bool is_
 
   return SharedRuntime::java_calling_convention(sig_bt, regs, sizeargs);
 }
-#endif // ASSERT
+#endif
 
 void Fingerprinter::compute_fingerprint_and_return_type(bool static_flag) {
   // See if we fingerprinted this method already
@@ -177,13 +177,15 @@ void Fingerprinter::compute_fingerprint_and_return_type(bool static_flag) {
     _param_size += 1;  // this is the convention for Method::compute_size_of_parameters
   }
 
+#if defined(_LP64) && !defined(ZERO)
   _stack_arg_slots = align_up(_stack_arg_slots, 2);
-
 #ifdef ASSERT
   int dbg_stack_arg_slots = compute_num_stack_arg_slots(_signature, _param_size, static_flag);
-#if defined(_LP64) && !defined(ZERO)
   assert(_stack_arg_slots == dbg_stack_arg_slots, "fingerprinter: %d full: %d", _stack_arg_slots, dbg_stack_arg_slots);
 #endif
+#else
+  // Fallback: computed _stack_arg_slots is unreliable, compute directly.
+  _stack_arg_slots = compute_num_stack_arg_slots(_signature, _param_size, static_flag);
 #endif
 
   // Detect overflow.  (We counted _param_size correctly.)
@@ -220,7 +222,10 @@ void Fingerprinter::initialize_calling_convention(bool static_flag) {
 
 void Fingerprinter::do_type_calling_convention(BasicType type) {
   // We compute the number of slots for stack-passed arguments in compiled calls.
-  // The value computed for 32-bit ports and for zero is bogus, and will need to be fixed.
+  // TODO: SharedRuntime::java_calling_convention is the shared code that knows all details
+  // about the platform-specific calling conventions. This method tries to compute the stack
+  // args number... poorly, at least for 32-bit ports and for zero. Current code has the fallback
+  // that recomputes the stack args number from SharedRuntime::java_calling_convention.
 #if defined(_LP64) && !defined(ZERO)
   switch (type) {
   case T_VOID: