8302736: Major performance regression in Math.log on aarch64

Reviewed-by: thartmann, dlong, aph

src/hotspot/share/opto/runtime.cpp

@@ -1562,10 +1562,6 @@ address OptoRuntime::handle_exception_C(JavaThread* current) {
 // *THIS IS NOT RECOMMENDED PROGRAMMING STYLE*
 //
 address OptoRuntime::rethrow_C(oopDesc* exception, JavaThread* thread, address ret_pc) {
-
-  // Enable WXWrite: the function called directly by compiled code.
-  MACOS_AARCH64_ONLY(ThreadWXEnable wx(WXWrite, thread));
-
   // ret_pc will have been loaded from the stack, so for AArch64 will be signed.
   // This needs authenticating, but to do that here requires the fp of the previous frame.
   // A better way of doing it would be authenticate in the caller by adding a

src/hotspot/share/runtime/continuationFreezeThaw.cpp

@@ -248,6 +248,8 @@ static JRT_LEAF(intptr_t*, thaw(JavaThread* thread, int kind))
   // JRT_ENTRY instead?
   ResetNoHandleMark rnhm;
 
+  // we might modify the code cache via BarrierSetNMethod::nmethod_entry_barrier
+  MACOS_AARCH64_ONLY(ThreadWXEnable __wx(WXWrite, thread));
   return ConfigT::thaw(thread, (Continuation::thaw_kind)kind);
 JRT_END
 

src/hotspot/share/runtime/interfaceSupport.inline.hpp

@@ -260,8 +260,6 @@ class VMNativeEntryWrapper {
 
 #define VM_LEAF_BASE(result_type, header)                            \
   debug_only(NoHandleMark __hm;)                                     \
-  MACOS_AARCH64_ONLY(ThreadWXEnable __wx(WXWrite,                    \
-                                         JavaThread::current()));    \
   os::verify_stack_alignment();                                      \
   /* begin of body */
 

src/hotspot/share/runtime/sharedRuntime.cpp

@@ -479,6 +479,9 @@ address SharedRuntime::raw_exception_handler_for_return_address(JavaThread* curr
     return StubRoutines::cont_returnBarrierExc();
   }
 
+  // write lock needed because we might update the pc desc cache via PcDescCache::add_pc_desc
+  MACOS_AARCH64_ONLY(ThreadWXEnable wx(WXWrite, current));
+
   // The fastest case first
   CodeBlob* blob = CodeCache::find_blob(return_address);
   CompiledMethod* nm = (blob != nullptr) ? blob->as_compiled_method_or_null() : nullptr;
@@ -2058,6 +2061,9 @@ JRT_LEAF(void, SharedRuntime::fixup_callers_callsite(Method* method, address cal
     return;
   }
 
+  // write lock needed because we might update the pc desc cache via PcDescCache::add_pc_desc
+  MACOS_AARCH64_ONLY(ThreadWXEnable __wx(WXWrite, JavaThread::current()));
+
   CodeBlob* cb = CodeCache::find_blob(caller_pc);
   if (cb == nullptr || !cb->is_compiled() || callee->is_unloading()) {
     return;