8257882: Implement linkToNative intrinsic on AArch64

Reviewed-by: jvernee, mcimadamore, aph

src/hotspot/cpu/aarch64/aarch64.ad

@@ -1772,8 +1772,13 @@ int MachCallRuntimeNode::ret_addr_offset() {
 }
 
 int MachCallNativeNode::ret_addr_offset() {
-  ShouldNotReachHere();
-  return -1;
+  // This is implemented using aarch64_enc_java_to_runtime as above.
+  CodeBlob *cb = CodeCache::find_blob(_entry_point);
+  if (cb) {
+    return 1 * NativeInstruction::instruction_size;
+  } else {
+    return 6 * NativeInstruction::instruction_size;
+  }
 }
 
 // Indicate if the safepoint node needs the polling page as an input
@@ -16044,6 +16049,21 @@ instruct CallLeafNoFPDirect(method meth)
   ins_pipe(pipe_class_call);
 %}
 
+instruct CallNativeDirect(method meth)
+%{
+  match(CallNative);
+
+  effect(USE meth);
+
+  ins_cost(CALL_COST);
+
+  format %{ "CALL, native $meth" %}
+
+  ins_encode( aarch64_enc_java_to_runtime(meth) );
+
+  ins_pipe(pipe_class_call);
+%}
+
 // Tail Call; Jump from runtime stub to Java code.
 // Also known as an 'interprocedural jump'.
 // Target of jump will eventually return to caller.

src/hotspot/cpu/aarch64/frame_aarch64.cpp

@@ -354,6 +354,11 @@ frame frame::sender_for_entry_frame(RegisterMap* map) const {
   assert(map->include_argument_oops(), "should be set by clear");
   vmassert(jfa->last_Java_pc() != NULL, "not walkable");
   frame fr(jfa->last_Java_sp(), jfa->last_Java_fp(), jfa->last_Java_pc());
+
+  if (jfa->saved_fp_address()) {
+    update_map_with_saved_link(map, jfa->saved_fp_address());
+  }
+
   return fr;
 }
 

src/hotspot/cpu/aarch64/gc/z/zBarrierSetAssembler_aarch64.cpp

@@ -314,7 +314,7 @@ class ZSaveLiveRegisters {
 private:
   MacroAssembler* const _masm;
   RegSet                _gp_regs;
-  RegSet                _fp_regs;
+  FloatRegSet           _fp_regs;
 
 public:
   void initialize(ZLoadBarrierStubC2* stub) {
@@ -327,7 +327,7 @@ public:
         if (vm_reg->is_Register()) {
           _gp_regs += RegSet::of(vm_reg->as_Register());
         } else if (vm_reg->is_FloatRegister()) {
-          _fp_regs += RegSet::of((Register)vm_reg->as_FloatRegister());
+          _fp_regs += FloatRegSet::of(vm_reg->as_FloatRegister());
         } else {
           fatal("Unknown register type");
         }

src/hotspot/cpu/aarch64/javaFrameAnchor_aarch64.hpp

@@ -31,6 +31,9 @@ private:
   // FP value associated with _last_Java_sp:
   intptr_t* volatile        _last_Java_fp;           // pointer is volatile not what it points to
 
+  // (Optional) location of saved FP register, which GCs want to inspect
+  intptr_t** volatile _saved_fp_address;
+
 public:
   // Each arch must define reset, save, restore
   // These are used by objects that only care about:
@@ -44,6 +47,7 @@ public:
     OrderAccess::release();
     _last_Java_fp = NULL;
     _last_Java_pc = NULL;
+    _saved_fp_address = NULL;
   }
 
   void copy(JavaFrameAnchor* src) {
@@ -62,6 +66,8 @@ public:
     _last_Java_pc = src->_last_Java_pc;
     // Must be last so profiler will always see valid frame if has_last_frame() is true
     _last_Java_sp = src->_last_Java_sp;
+
+    _saved_fp_address = src->_saved_fp_address;
   }
 
   bool walkable(void)                            { return _last_Java_sp != NULL && _last_Java_pc != NULL; }
@@ -72,9 +78,12 @@ public:
 
   address last_Java_pc(void)                     { return _last_Java_pc; }
 
+  intptr_t** saved_fp_address(void) const        { return _saved_fp_address; }
+
 private:
 
   static ByteSize last_Java_fp_offset()          { return byte_offset_of(JavaFrameAnchor, _last_Java_fp); }
+  static ByteSize saved_fp_address_offset()      { return byte_offset_of(JavaFrameAnchor, _saved_fp_address); }
 
 public:
 

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -318,6 +318,8 @@ void MacroAssembler::reset_last_Java_frame(bool clear_fp) {
 
   // Always clear the pc because it could have been set by make_walkable()
   str(zr, Address(rthread, JavaThread::last_Java_pc_offset()));
+
+  str(zr, Address(rthread, JavaThread::saved_fp_address_offset()));
 }
 
 // Calls to C land
@@ -5293,8 +5295,9 @@ void MacroAssembler::cache_wbsync(bool is_pre) {
 }
 
 void MacroAssembler::verify_sve_vector_length() {
+  // Make sure that native code does not change SVE vector length.
+  if (!UseSVE) return;
   Label verify_ok;
-  assert(UseSVE > 0, "should only be used for SVE");
   movw(rscratch1, zr);
   sve_inc(rscratch1, B);
   subsw(zr, rscratch1, VM_Version::get_initial_sve_vector_length());

src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp

@@ -474,8 +474,8 @@ public:
   void push(RegSet regs, Register stack) { if (regs.bits()) push(regs.bits(), stack); }
   void pop(RegSet regs, Register stack) { if (regs.bits()) pop(regs.bits(), stack); }
 
-  void push_fp(RegSet regs, Register stack) { if (regs.bits()) push_fp(regs.bits(), stack); }
-  void pop_fp(RegSet regs, Register stack) { if (regs.bits()) pop_fp(regs.bits(), stack); }
+  void push_fp(FloatRegSet regs, Register stack) { if (regs.bits()) push_fp(regs.bits(), stack); }
+  void pop_fp(FloatRegSet regs, Register stack) { if (regs.bits()) pop_fp(regs.bits(), stack); }
 
   static RegSet call_clobbered_registers();
 

src/hotspot/cpu/aarch64/register_aarch64.hpp

@@ -63,10 +63,6 @@ class RegisterImpl: public AbstractRegisterImpl {
   bool  has_byte_register() const                { return 0 <= (intptr_t)this && (intptr_t)this < number_of_byte_registers; }
   const char* name() const;
   int   encoding_nocheck() const                 { return (intptr_t)this; }
-
-  // Return the bit which represents this register.  This is intended
-  // to be ORed into a bitmask: for usage see class RegSet below.
-  uint64_t bit(bool should_set = true) const { return should_set ? 1 << encoding() : 0; }
 };
 
 // The integer registers of the aarch64 architecture
@@ -304,91 +300,93 @@ class ConcreteRegisterImpl : public AbstractRegisterImpl {
   static const int max_pr;
 };
 
-class RegSetIterator;
+template <class RegImpl = Register> class RegSetIterator;
 
 // A set of registers
-class RegSet {
+template <class RegImpl>
+class AbstractRegSet {
   uint32_t _bitset;
 
-  RegSet(uint32_t bitset) : _bitset(bitset) { }
+  AbstractRegSet(uint32_t bitset) : _bitset(bitset) { }
 
 public:
 
-  RegSet() : _bitset(0) { }
+  AbstractRegSet() : _bitset(0) { }
 
-  RegSet(Register r1) : _bitset(r1->bit()) { }
+  AbstractRegSet(RegImpl r1) : _bitset(1 << r1->encoding()) { }
 
-  RegSet operator+(const RegSet aSet) const {
-    RegSet result(_bitset | aSet._bitset);
+  AbstractRegSet operator+(const AbstractRegSet aSet) const {
+    AbstractRegSet result(_bitset | aSet._bitset);
     return result;
   }
 
-  RegSet operator-(const RegSet aSet) const {
-    RegSet result(_bitset & ~aSet._bitset);
+  AbstractRegSet operator-(const AbstractRegSet aSet) const {
+    AbstractRegSet result(_bitset & ~aSet._bitset);
     return result;
   }
 
-  RegSet &operator+=(const RegSet aSet) {
+  AbstractRegSet &operator+=(const AbstractRegSet aSet) {
     *this = *this + aSet;
     return *this;
   }
 
-  RegSet &operator-=(const RegSet aSet) {
+  AbstractRegSet &operator-=(const AbstractRegSet aSet) {
     *this = *this - aSet;
     return *this;
   }
 
-  static RegSet of(Register r1) {
-    return RegSet(r1);
+  static AbstractRegSet of(RegImpl r1) {
+    return AbstractRegSet(r1);
   }
 
-  static RegSet of(Register r1, Register r2) {
+  static AbstractRegSet of(RegImpl r1, RegImpl r2) {
     return of(r1) + r2;
   }
 
-  static RegSet of(Register r1, Register r2, Register r3) {
+  static AbstractRegSet of(RegImpl r1, RegImpl r2, RegImpl r3) {
     return of(r1, r2) + r3;
   }
 
-  static RegSet of(Register r1, Register r2, Register r3, Register r4) {
+  static AbstractRegSet of(RegImpl r1, RegImpl r2, RegImpl r3, RegImpl r4) {
     return of(r1, r2, r3) + r4;
   }
 
-  static RegSet range(Register start, Register end) {
+  static AbstractRegSet range(RegImpl start, RegImpl end) {
     uint32_t bits = ~0;
     bits <<= start->encoding();
     bits <<= 31 - end->encoding();
     bits >>= 31 - end->encoding();
 
-    return RegSet(bits);
+    return AbstractRegSet(bits);
   }
 
   uint32_t bits() const { return _bitset; }
 
 private:
 
-  Register first() {
-    uint32_t first = _bitset & -_bitset;
-    return first ? as_Register(exact_log2(first)) : noreg;
-  }
+  RegImpl first();
 
 public:
 
-  friend class RegSetIterator;
+  friend class RegSetIterator<RegImpl>;
 
-  RegSetIterator begin();
+  RegSetIterator<RegImpl> begin();
 };
 
+typedef AbstractRegSet<Register> RegSet;
+typedef AbstractRegSet<FloatRegister> FloatRegSet;
+
+template <class RegImpl>
 class RegSetIterator {
-  RegSet _regs;
+  AbstractRegSet<RegImpl> _regs;
 
 public:
-  RegSetIterator(RegSet x): _regs(x) {}
+  RegSetIterator(AbstractRegSet<RegImpl> x): _regs(x) {}
   RegSetIterator(const RegSetIterator& mit) : _regs(mit._regs) {}
 
   RegSetIterator& operator++() {
-    Register r = _regs.first();
-    if (r != noreg)
+    RegImpl r = _regs.first();
+    if (r->is_valid())
       _regs -= r;
     return *this;
   }
@@ -400,13 +398,26 @@ public:
     return ! (rhs == *this);
   }
 
-  Register operator*() {
+  RegImpl operator*() {
     return _regs.first();
   }
 };
 
-inline RegSetIterator RegSet::begin() {
-  return RegSetIterator(*this);
+template <class RegImpl>
+inline RegSetIterator<RegImpl> AbstractRegSet<RegImpl>::begin() {
+  return RegSetIterator<RegImpl>(*this);
+}
+
+template <>
+inline Register AbstractRegSet<Register>::first() {
+  uint32_t first = _bitset & -_bitset;
+  return first ? as_Register(exact_log2(first)) : noreg;
+}
+
+template <>
+inline FloatRegister AbstractRegSet<FloatRegister>::first() {
+  uint32_t first = _bitset & -_bitset;
+  return first ? as_FloatRegister(exact_log2(first)) : fnoreg;
 }
 
 #endif // CPU_AARCH64_REGISTER_AARCH64_HPP

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -1855,10 +1855,7 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
   // Force this write out before the read below
   __ dmb(Assembler::ISH);
 
-  if (UseSVE > 0) {
-    // Make sure that jni code does not change SVE vector length.
-    __ verify_sve_vector_length();
-  }
+  __ verify_sve_vector_length();
 
   // Check for safepoint operation in progress and/or pending suspend requests.
   {
@@ -3071,9 +3068,168 @@ void OptoRuntime::generate_exception_blob() {
 }
 #endif // COMPILER2
 
+// ---------------------------------------------------------------
+
+class NativeInvokerGenerator : public StubCodeGenerator {
+  address _call_target;
+  int _shadow_space_bytes;
+
+  const GrowableArray<VMReg>& _input_registers;
+  const GrowableArray<VMReg>& _output_registers;
+public:
+  NativeInvokerGenerator(CodeBuffer* buffer,
+                         address call_target,
+                         int shadow_space_bytes,
+                         const GrowableArray<VMReg>& input_registers,
+                         const GrowableArray<VMReg>& output_registers)
+   : StubCodeGenerator(buffer, PrintMethodHandleStubs),
+     _call_target(call_target),
+     _shadow_space_bytes(shadow_space_bytes),
+     _input_registers(input_registers),
+     _output_registers(output_registers) {}
+  void generate();
+
+private:
+#ifdef ASSERT
+  bool target_uses_register(VMReg reg) {
+    return _input_registers.contains(reg) || _output_registers.contains(reg);
+  }
+#endif
+};
+
+static const int native_invoker_code_size = 1024;
+
 BufferBlob* SharedRuntime::make_native_invoker(address call_target,
-                                           int shadow_space_bytes,
-                                           const GrowableArray<VMReg>& input_registers,
-                                           const GrowableArray<VMReg>& output_registers) {
-  return NULL;
+                                               int shadow_space_bytes,
+                                               const GrowableArray<VMReg>& input_registers,
+                                               const GrowableArray<VMReg>& output_registers) {
+  BufferBlob* _invoke_native_blob =
+    BufferBlob::create("nep_invoker_blob", native_invoker_code_size);
+  if (_invoke_native_blob == NULL)
+    return NULL; // allocation failure
+
+  CodeBuffer code(_invoke_native_blob);
+  NativeInvokerGenerator g(&code, call_target, shadow_space_bytes, input_registers, output_registers);
+  g.generate();
+  code.log_section_sizes("nep_invoker_blob");
+
+  return _invoke_native_blob;
+}
+
+void NativeInvokerGenerator::generate() {
+  assert(!(target_uses_register(rscratch1->as_VMReg())
+           || target_uses_register(rscratch2->as_VMReg())
+           || target_uses_register(rthread->as_VMReg())),
+         "Register conflict");
+
+  MacroAssembler* masm = _masm;
+
+  __ set_last_Java_frame(sp, noreg, lr, rscratch1);
+
+  __ enter();
+
+  // Store a pointer to the previous R29 (RFP) saved on the stack as it
+  // may contain an oop if PreserveFramePointer is off. This value is
+  // retrieved later by frame::sender_for_entry_frame() when the stack
+  // is walked.
+  __ mov(rscratch1, sp);
+  __ str(rscratch1, Address(rthread, JavaThread::saved_fp_address_offset()));
+
+  // State transition
+  __ mov(rscratch1, _thread_in_native);
+  __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
+  __ stlrw(rscratch1, rscratch2);
+
+  assert(_shadow_space_bytes == 0, "not expecting shadow space on AArch64");
+
+  rt_call(masm, _call_target);
+
+  __ mov(rscratch1, _thread_in_native_trans);
+  __ strw(rscratch1, Address(rthread, JavaThread::thread_state_offset()));
+
+  // Force this write out before the read below
+  __ membar(Assembler::LoadLoad | Assembler::LoadStore |
+            Assembler::StoreLoad | Assembler::StoreStore);
+
+  __ verify_sve_vector_length();
+
+  Label L_after_safepoint_poll;
+  Label L_safepoint_poll_slow_path;
+
+  __ safepoint_poll(L_safepoint_poll_slow_path, true /* at_return */, true /* acquire */, false /* in_nmethod */);
+
+  __ ldrw(rscratch1, Address(rthread, JavaThread::suspend_flags_offset()));
+  __ cbnzw(rscratch1, L_safepoint_poll_slow_path);
+
+  __ bind(L_after_safepoint_poll);
+
+  // change thread state
+  __ mov(rscratch1, _thread_in_Java);
+  __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
+  __ stlrw(rscratch1, rscratch2);
+
+  __ block_comment("reguard stack check");
+  Label L_reguard;
+  Label L_after_reguard;
+  __ ldrb(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
+  __ cmpw(rscratch1, StackOverflow::stack_guard_yellow_reserved_disabled);
+  __ br(Assembler::EQ, L_reguard);
+  __ bind(L_after_reguard);
+
+  __ reset_last_Java_frame(true);
+
+  __ leave(); // required for proper stackwalking of RuntimeStub frame
+  __ ret(lr);
+
+  //////////////////////////////////////////////////////////////////////////////
+
+  __ block_comment("{ L_safepoint_poll_slow_path");
+  __ bind(L_safepoint_poll_slow_path);
+
+  // Need to save the native result registers around any runtime calls.
+  RegSet spills;
+  FloatRegSet fp_spills;
+  for (int i = 0; i < _output_registers.length(); i++) {
+    VMReg output = _output_registers.at(i);
+    if (output->is_Register()) {
+      spills += RegSet::of(output->as_Register());
+    } else if (output->is_FloatRegister()) {
+      fp_spills += FloatRegSet::of(output->as_FloatRegister());
+    }
+  }
+
+  __ push(spills, sp);
+  __ push_fp(fp_spills, sp);
+
+  __ mov(c_rarg0, rthread);
+  assert(frame::arg_reg_save_area_bytes == 0, "not expecting frame reg save area");
+  __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
+  __ blr(rscratch1);
+
+  __ pop_fp(fp_spills, sp);
+  __ pop(spills, sp);
+
+  __ b(L_after_safepoint_poll);
+  __ block_comment("} L_safepoint_poll_slow_path");
+
+  //////////////////////////////////////////////////////////////////////////////
+
+  __ block_comment("{ L_reguard");
+  __ bind(L_reguard);
+
+  __ push(spills, sp);
+  __ push_fp(fp_spills, sp);
+
+  rt_call(masm, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
+
+  __ pop_fp(fp_spills, sp);
+  __ pop(spills, sp);
+
+  __ b(L_after_reguard);
+
+  __ block_comment("} L_reguard");
+
+  //////////////////////////////////////////////////////////////////////////////
+
+  __ flush();
 }

src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp

@@ -1319,7 +1319,7 @@ class StubGenerator: public StubCodeGenerator {
       = MacroAssembler::call_clobbered_registers() - rscratch1;
     __ mov(rscratch1, (uint64_t)0xdeadbeef);
     __ orr(rscratch1, rscratch1, rscratch1, Assembler::LSL, 32);
-    for (RegSetIterator it = clobbered.begin(); *it != noreg; ++it) {
+    for (RegSetIterator<> it = clobbered.begin(); *it != noreg; ++it) {
       __ mov(*it, rscratch1);
     }
 #endif
@@ -5696,7 +5696,7 @@ class StubGenerator: public StubCodeGenerator {
 
       // Register allocation
 
-      RegSetIterator regs = (RegSet::range(r0, r26) - r18_tls).begin();
+      RegSetIterator<> regs = (RegSet::range(r0, r26) - r18_tls).begin();
       Pa_base = *regs;       // Argument registers
       if (squaring)
         Pb_base = Pa_base;

src/hotspot/cpu/aarch64/templateInterpreterGenerator_aarch64.cpp

@@ -1373,10 +1373,7 @@ address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
   __ push(dtos);
   __ push(ltos);
 
-  if (UseSVE > 0) {
-    // Make sure that jni code does not change SVE vector length.
-    __ verify_sve_vector_length();
-  }
+  __ verify_sve_vector_length();
 
   // change thread state
   __ mov(rscratch1, _thread_in_native_trans);

src/hotspot/os_cpu/linux_aarch64/thread_linux_aarch64.hpp

@@ -39,6 +39,10 @@
     return byte_offset_of(JavaThread, _anchor) + JavaFrameAnchor::last_Java_fp_offset();
   }
 
+  static ByteSize saved_fp_address_offset() {
+    return byte_offset_of(JavaThread, _anchor) + JavaFrameAnchor::saved_fp_address_offset();
+  }
+
   bool pd_get_top_frame_for_signal_handler(frame* fr_addr, void* ucontext,
     bool isInJava);
 

src/hotspot/os_cpu/windows_aarch64/thread_windows_aarch64.hpp

@@ -38,6 +38,10 @@
     return byte_offset_of(JavaThread, _anchor) + JavaFrameAnchor::last_Java_fp_offset();
   }
 
+  static ByteSize saved_fp_address_offset() {
+    return byte_offset_of(JavaThread, _anchor) + JavaFrameAnchor::saved_fp_address_offset();
+  }
+
   bool pd_get_top_frame_for_signal_handler(frame* fr_addr, void* ucontext,
     bool isInJava);
 

test/jdk/java/foreign/TestIntrinsics.java

@@ -23,7 +23,7 @@
 
 /*
  * @test
- * @requires os.arch=="amd64" | os.arch=="x86_64"
+ * @requires os.arch=="amd64" | os.arch=="x86_64" | os.arch=="aarch64"
  * @run testng/othervm
  *   -Djdk.internal.foreign.ProgrammableInvoker.USE_SPEC=true
  *   -Djdk.internal.foreign.ProgrammableInvoker.USE_INTRINSICS=true