8221554: aarch64 cross-modifying code

Reviewed-by: rehn, aph

Author: a74nh

src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp

@@ -2922,7 +2922,6 @@ void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* arg
   if (info != NULL) {
     add_call_info_here(info);
   }
-  __ maybe_isb();
 }
 
 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {

src/hotspot/cpu/aarch64/c1_Runtime1_aarch64.cpp

@@ -80,7 +80,6 @@ int StubAssembler::call_RT(Register oop_result1, Register metadata_result, addre
   pop(r0, sp);
 #endif
   reset_last_Java_frame(true);
-  maybe_isb();
 
   // check for pending exceptions
   { Label L;

src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp

@@ -1617,7 +1617,7 @@ void InterpreterMacroAssembler::call_VM_base(Register oop_result,
     Label L;
     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
     cbz(rscratch1, L);
-    stop("InterpreterMacroAssembler::call_VM_leaf_base:"
+    stop("InterpreterMacroAssembler::call_VM_base:"
          " last_sp != NULL");
     bind(L);
   }

src/hotspot/cpu/aarch64/jniFastGetField_aarch64.cpp

@@ -159,7 +159,6 @@ address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
     __ enter();
     __ lea(rscratch1, ExternalAddress(slow_case_addr));
     __ blr(rscratch1);
-    __ maybe_isb();
     __ leave();
     __ ret(lr);
   }

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -1381,7 +1381,6 @@ void MacroAssembler::call_VM_leaf_base(address entry_point,
     bind(*retaddr);
 
   ldp(rscratch1, rmethod, Address(post(sp, 2 * wordSize)));
-  maybe_isb();
 }
 
 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) {
@@ -4387,10 +4386,15 @@ void MacroAssembler::get_polling_page(Register dest, relocInfo::relocType rtype)
 // Read the polling page.  The address of the polling page must
 // already be in r.
 address MacroAssembler::read_polling_page(Register r, relocInfo::relocType rtype) {
-  InstructionMark im(this);
-  code_section()->relocate(inst_mark(), rtype);
-  ldrw(zr, Address(r, 0));
-  return inst_mark();
+  address mark;
+  {
+    InstructionMark im(this);
+    code_section()->relocate(inst_mark(), rtype);
+    ldrw(zr, Address(r, 0));
+    mark = inst_mark();
+  }
+  verify_cross_modify_fence_not_required();
+  return mark;
 }
 
 void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_offset) {
@@ -4455,6 +4459,7 @@ void MacroAssembler::build_frame(int framesize) {
       sub(sp, sp, rscratch1);
     }
   }
+  verify_cross_modify_fence_not_required();
 }
 
 void MacroAssembler::remove_frame(int framesize) {
@@ -5315,3 +5320,29 @@ void MacroAssembler::verify_ptrue() {
   stop("Error: the preserved predicate register (p7) elements are not all true");
   bind(verify_ok);
 }
+
+void MacroAssembler::safepoint_isb() {
+  isb();
+#ifndef PRODUCT
+  if (VerifyCrossModifyFence) {
+    // Clear the thread state.
+    strb(zr, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
+  }
+#endif
+}
+
+#ifndef PRODUCT
+void MacroAssembler::verify_cross_modify_fence_not_required() {
+  if (VerifyCrossModifyFence) {
+    // Check if thread needs a cross modify fence.
+    ldrb(rscratch1, Address(rthread, in_bytes(JavaThread::requires_cross_modify_fence_offset())));
+    Label fence_not_required;
+    cbz(rscratch1, fence_not_required);
+    // If it does then fail.
+    lea(rscratch1, CAST_FROM_FN_PTR(address, JavaThread::verify_cross_modify_fence_failure));
+    mov(c_rarg0, rthread);
+    blr(rscratch1);
+    bind(fence_not_required);
+  }
+}
+#endif

src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp

@@ -1315,8 +1315,9 @@ class MacroAssembler: public Assembler {
                        Register zlen, Register tmp1, Register tmp2, Register tmp3,
                        Register tmp4, Register tmp5, Register tmp6, Register tmp7);
   void mul_add(Register out, Register in, Register offs, Register len, Register k);
-  // ISB may be needed because of a safepoint
-  void maybe_isb() { isb(); }
+
+  // Place an ISB after code may have been modified due to a safepoint.
+  void safepoint_isb();
 
 private:
   // Return the effective address r + (r1 << ext) + offset.
@@ -1392,6 +1393,11 @@ class MacroAssembler: public Assembler {
   }
   void cache_wb(Address line);
   void cache_wbsync(bool is_pre);
+
+private:
+  // Check the current thread doesn't need a cross modify fence.
+  void verify_cross_modify_fence_not_required() PRODUCT_RETURN;
+
 };
 
 #ifdef ASSERT

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -373,7 +373,10 @@ static void patch_callers_callsite(MacroAssembler *masm) {
   __ mov(c_rarg1, lr);
   __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::fixup_callers_callsite)));
   __ blr(rscratch1);
-  __ maybe_isb();
+
+  // Explicit isb required because fixup_callers_callsite may change the code
+  // stream.
+  __ safepoint_isb();
 
   __ pop_CPU_state();
   // restore sp
@@ -1150,7 +1153,6 @@ static void rt_call(MacroAssembler* masm, address dest) {
   } else {
     __ lea(rscratch1, RuntimeAddress(dest));
     __ blr(rscratch1);
-    __ maybe_isb();
   }
 }
 
@@ -1857,7 +1859,7 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
     __ verify_sve_vector_length();
   }
 
-  // check for safepoint operation in progress and/or pending suspend requests
+  // Check for safepoint operation in progress and/or pending suspend requests.
   {
     // We need an acquire here to ensure that any subsequent load of the
     // global SafepointSynchronize::_state flag is ordered after this load
@@ -2081,7 +2083,7 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
 #endif
     __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
     __ blr(rscratch1);
-    __ maybe_isb();
+
     // Restore any method result value
     restore_native_result(masm, ret_type, stack_slots);
 
@@ -2787,7 +2789,6 @@ SafepointBlob* SharedRuntime::generate_handler_blob(address call_ptr, int poll_t
 
   __ reset_last_Java_frame(false);
 
-  __ maybe_isb();
   __ membar(Assembler::LoadLoad | Assembler::LoadStore);
 
   if (UseSVE > 0 && save_vectors) {
@@ -2894,8 +2895,6 @@ RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const cha
 
   oop_maps->add_gc_map( __ offset() - start, map);
 
-  __ maybe_isb();
-
   // r0 contains the address we are going to jump to assuming no exception got installed
 
   // clear last_Java_sp
@@ -3017,7 +3016,8 @@ void OptoRuntime::generate_exception_blob() {
   __ mov(c_rarg0, rthread);
   __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, OptoRuntime::handle_exception_C)));
   __ blr(rscratch1);
-  __ maybe_isb();
+  // handle_exception_C is a special VM call which does not require an explicit
+  // instruction sync afterwards.
 
   // Set an oopmap for the call site.  This oopmap will only be used if we
   // are unwinding the stack.  Hence, all locations will be dead.

src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp

@@ -5629,7 +5629,6 @@ class StubGenerator: public StubCodeGenerator {
     oop_maps->add_gc_map(the_pc - start, map);
 
     __ reset_last_Java_frame(true);
-    __ maybe_isb();
 
     if (UseSVE > 0) {
       // Reinitialize the ptrue predicate register, in case the external runtime

src/hotspot/cpu/aarch64/templateInterpreterGenerator_aarch64.cpp

@@ -1357,7 +1357,6 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
   // Call the native method.
   __ blr(r10);
   __ bind(native_return);
-  __ maybe_isb();
   __ get_method(rmethod);
   // result potentially in r0 or v0
 
@@ -1410,7 +1409,6 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
     __ mov(c_rarg0, rthread);
     __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
     __ blr(rscratch2);
-    __ maybe_isb();
     __ get_method(rmethod);
     __ reinit_heapbase();
     __ bind(Continue);

src/hotspot/os_cpu/aix_ppc/orderAccess_aix_ppc.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2019 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -73,7 +73,7 @@ inline void OrderAccess::storeload()  { inlasm_sync();   }
 inline void OrderAccess::acquire()    { inlasm_lwsync(); }
 inline void OrderAccess::release()    { inlasm_lwsync(); }
 inline void OrderAccess::fence()      { inlasm_sync();   }
-inline void OrderAccess::cross_modify_fence()
+inline void OrderAccess::cross_modify_fence_impl()
                                       { inlasm_isync();  }
 
 #undef inlasm_sync