8303210: [linux, Windows] Make UseSystemMemoryBarrier available as pr…

…oduct flag

Reviewed-by: dholmes, rehn

src/hotspot/cpu/arm/sharedRuntime_arm.cpp

@@ -1211,7 +1211,9 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
   __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
 
   // make sure the store is observed before reading the SafepointSynchronize state and further mem refs
-  __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
+  if (!UseSystemMemoryBarrier) {
+    __ membar(MacroAssembler::Membar_mask_bits(MacroAssembler::StoreLoad | MacroAssembler::StoreStore), Rtemp);
+  }
 
   __ safepoint_poll(R2, call_safepoint_runtime);
   __ ldr_u32(R3, Address(Rthread, JavaThread::suspend_flags_offset()));

src/hotspot/cpu/arm/templateInterpreterGenerator_arm.cpp

@@ -1007,8 +1007,10 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
   __ mov(Rtemp, _thread_in_native_trans);
   __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
 
-    // Force this write out before the read below
-  __ membar(MacroAssembler::StoreLoad, Rtemp);
+  // Force this write out before the read below
+  if (!UseSystemMemoryBarrier) {
+    __ membar(MacroAssembler::StoreLoad, Rtemp);
+  }
 
   // Protect the return value in the interleaved code: save it to callee-save registers.
   __ mov(Rsaved_result_lo, R0);

src/hotspot/cpu/riscv/downcallLinker_riscv.cpp

@@ -265,7 +265,9 @@ void DowncallStubGenerator::generate() {
   __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
 
   // Force this write out before the read below
-  __ membar(MacroAssembler::AnyAny);
+  if (!UseSystemMemoryBarrier) {
+    __ membar(MacroAssembler::AnyAny);
+  }
 
   Label L_after_safepoint_poll;
   Label L_safepoint_poll_slow_path;

src/hotspot/cpu/riscv/sharedRuntime_riscv.cpp

@@ -44,6 +44,7 @@
 #include "prims/methodHandles.hpp"
 #include "runtime/continuation.hpp"
 #include "runtime/continuationEntry.inline.hpp"
+#include "runtime/globals.hpp"
 #include "runtime/jniHandles.hpp"
 #include "runtime/safepointMechanism.hpp"
 #include "runtime/sharedRuntime.hpp"
@@ -1746,7 +1747,9 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
   __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
 
   // Force this write out before the read below
-  __ membar(MacroAssembler::AnyAny);
+  if (!UseSystemMemoryBarrier) {
+    __ membar(MacroAssembler::AnyAny);
+  }
 
   // check for safepoint operation in progress and/or pending suspend requests
   {

src/hotspot/cpu/riscv/templateInterpreterGenerator_riscv.cpp

@@ -45,6 +45,7 @@
 #include "runtime/arguments.hpp"
 #include "runtime/deoptimization.hpp"
 #include "runtime/frame.inline.hpp"
+#include "runtime/globals.hpp"
 #include "runtime/jniHandles.hpp"
 #include "runtime/sharedRuntime.hpp"
 #include "runtime/stubRoutines.hpp"
@@ -1159,7 +1160,9 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
   __ sw(t0, Address(xthread, JavaThread::thread_state_offset()));
 
   // Force this write out before the read below
-  __ membar(MacroAssembler::AnyAny);
+  if (!UseSystemMemoryBarrier) {
+    __ membar(MacroAssembler::AnyAny);
+  }
 
   // check for safepoint operation in progress and/or pending suspend requests
   {

src/hotspot/cpu/s390/sharedRuntime_s390.cpp

@@ -1838,7 +1838,9 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
     save_native_result(masm, ret_type, workspace_slot_offset); // Make Z_R2 available as work reg.
 
     // Force this write out before the read below.
-    __ z_fence();
+    if (!UseSystemMemoryBarrier) {
+      __ z_fence();
+    }
 
     __ safepoint_poll(sync, Z_R1);
 

src/hotspot/cpu/s390/templateInterpreterGenerator_s390.cpp

@@ -1525,7 +1525,9 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
   // synchronization is progress, and escapes.
 
   __ set_thread_state(_thread_in_native_trans);
-  __ z_fence();
+  if (!UseSystemMemoryBarrier) {
+    __ z_fence();
+  }
 
   // Now before we return to java we must look for a current safepoint
   // (a new safepoint can not start since we entered native_trans).

src/hotspot/cpu/x86/sharedRuntime_x86_32.cpp

@@ -1772,9 +1772,11 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
 
   // Force this write out before the read below
-  __ membar(Assembler::Membar_mask_bits(
-            Assembler::LoadLoad | Assembler::LoadStore |
-            Assembler::StoreLoad | Assembler::StoreStore));
+  if (!UseSystemMemoryBarrier) {
+    __ membar(Assembler::Membar_mask_bits(
+              Assembler::LoadLoad | Assembler::LoadStore |
+              Assembler::StoreLoad | Assembler::StoreStore));
+  }
 
   if (AlwaysRestoreFPU) {
     // Make sure the control word is correct.

src/hotspot/os/linux/systemMemoryBarrier_linux.cpp

@@ -72,21 +72,21 @@ static int membarrier(int cmd, unsigned int flags, int cpu_id) {
 bool LinuxSystemMemoryBarrier::initialize() {
   int ret = membarrier(MEMBARRIER_CMD_QUERY, 0, 0);
   if (ret < 0) {
-    log_error(os)("MEMBARRIER_CMD_QUERY unsupported");
+    log_info(os)("MEMBARRIER_CMD_QUERY unsupported");
     return false;
   }
   if (!(ret & MEMBARRIER_CMD_PRIVATE_EXPEDITED) ||
       !(ret & MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED)) {
-    log_error(os)("MEMBARRIER PRIVATE_EXPEDITED unsupported");
+    log_info(os)("MEMBARRIER PRIVATE_EXPEDITED unsupported");
     return false;
   }
   ret = membarrier(MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED, 0, 0);
   guarantee_with_errno(ret == 0, "MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED failed");
+  log_info(os)("Using MEMBARRIER PRIVATE_EXPEDITED");
   return true;
 }
 
 void LinuxSystemMemoryBarrier::emit() {
   int s = membarrier(MEMBARRIER_CMD_PRIVATE_EXPEDITED, 0, 0);
   guarantee_with_errno(s >= 0, "MEMBARRIER_CMD_PRIVATE_EXPEDITED failed");
 }
-

src/hotspot/share/runtime/arguments.cpp

@@ -64,6 +64,7 @@
 #include "utilities/parseInteger.hpp"
 #include "utilities/powerOfTwo.hpp"
 #include "utilities/stringUtils.hpp"
+#include "utilities/systemMemoryBarrier.hpp"
 #if INCLUDE_JFR
 #include "jfr/jfr.hpp"
 #endif
@@ -2153,6 +2154,8 @@ jint Arguments::parse_vm_init_args(const JavaVMInitArgs *vm_options_args,
 
   os::init_container_support();
 
+  SystemMemoryBarrier::initialize();
+
   // Do final processing now that all arguments have been parsed
   result = finalize_vm_init_args(patch_mod_javabase);
   if (result != JNI_OK) {

src/hotspot/share/runtime/globals.hpp

@@ -1293,8 +1293,8 @@ const int ObjectAlignmentInBytes = 8;
           "Delay in milliseconds for option SafepointTimeout")              \
           range(0, max_intx LP64_ONLY(/MICROUNITS))                         \
                                                                             \
-  product(bool, UseSystemMemoryBarrier, false, EXPERIMENTAL,                \
-          "Try to enable system memory barrier")                            \
+  product(bool, UseSystemMemoryBarrier, false,                              \
+          "Try to enable system memory barrier if supported by OS")         \
                                                                             \
   product(intx, NmethodSweepActivity, 4,                                    \
           "Removes cold nmethods from code cache if > 0. Higher values "    \

src/hotspot/share/runtime/threads.cpp

@@ -95,7 +95,6 @@
 #include "utilities/dtrace.hpp"
 #include "utilities/events.hpp"
 #include "utilities/macros.hpp"
-#include "utilities/systemMemoryBarrier.hpp"
 #include "utilities/vmError.hpp"
 #if INCLUDE_JVMCI
 #include "jvmci/jvmci.hpp"
@@ -552,14 +551,6 @@ jint Threads::create_vm(JavaVMInitArgs* args, bool* canTryAgain) {
   // crash Linux VM, see notes in os_linux.cpp.
   main_thread->stack_overflow_state()->create_stack_guard_pages();
 
-  if (UseSystemMemoryBarrier) {
-    if (!SystemMemoryBarrier::initialize()) {
-      vm_shutdown_during_initialization("Failed to initialize the requested system memory barrier synchronization.");
-      return JNI_EINVAL;
-    }
-    log_debug(os)("Using experimental system memory barrier synchronization");
-  }
-
   // Initialize Java-Level synchronization subsystem
   ObjectMonitor::Initialize();
   ObjectSynchronizer::initialize();

src/hotspot/share/utilities/systemMemoryBarrier.hpp

@@ -26,6 +26,8 @@
 #define SHARE_UTILITIES_SYSTEMMEMORYBARRIER_HPP
 
 #include "logging/log.hpp"
+#include "runtime/globals.hpp"
+#include "runtime/globals_extension.hpp"
 #include "utilities/globalDefinitions.hpp"
 
 #if defined(LINUX)
@@ -38,7 +40,7 @@ typedef WindowsSystemMemoryBarrier SystemMemoryBarrierDefault;
 class NoSystemMemoryBarrier {
  public:
   static bool initialize() {
-    log_error(os)("SystemMemoryBarrier not supported on this platform");
+    log_info(os)("SystemMemoryBarrier not supported on this platform");
     return false;
   }
   static void emit() {
@@ -51,8 +53,17 @@ typedef NoSystemMemoryBarrier SystemMemoryBarrierDefault;
 template <typename SystemMemoryBarrierImpl>
 class SystemMemoryBarrierType : public AllStatic {
  public:
-  static bool initialize() { return SystemMemoryBarrierImpl::initialize(); }
-  static void emit()       { SystemMemoryBarrierImpl::emit(); }
+  static void initialize() {
+    if (UseSystemMemoryBarrier) {
+      if (!SystemMemoryBarrierImpl::initialize()) {
+        if (!FLAG_IS_DEFAULT(UseSystemMemoryBarrier)) {
+          warning("UseSystemMemoryBarrier specified, but not supported on this OS version. Use -Xlog:os=info for details.");
+        }
+        FLAG_SET_ERGO(UseSystemMemoryBarrier, false);
+      }
+    }
+  }
+  static void emit() { SystemMemoryBarrierImpl::emit(); }
 };
 
 typedef SystemMemoryBarrierType<SystemMemoryBarrierDefault> SystemMemoryBarrier;