8254231: Implementation of Foreign Linker API (Incubator)

Reviewed-by: coleenp, ihse, dholmes, vlivanov

Author: mcimadamore

src/hotspot/cpu/aarch64/aarch64.ad

@@ -1770,6 +1770,11 @@ int MachCallRuntimeNode::ret_addr_offset() {
   }
 }
 
+int MachCallNativeNode::ret_addr_offset() {
+  ShouldNotReachHere();
+  return -1;
+}
+
 // Indicate if the safepoint node needs the polling page as an input
 
 // the shared code plants the oop data at the start of the generated

src/hotspot/cpu/aarch64/foreign_globals_aarch64.cpp

@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, Arm Limited. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+#include "precompiled.hpp"
+#include "runtime/jniHandles.hpp"
+#include "runtime/jniHandles.inline.hpp"
+#include "oops/typeArrayOop.inline.hpp"
+#include "prims/foreign_globals.hpp"
+#include "prims/foreign_globals.inline.hpp"
+
+bool ABIDescriptor::is_volatile_reg(Register reg) const {
+  return _integer_argument_registers.contains(reg)
+    || _integer_additional_volatile_registers.contains(reg);
+}
+
+bool ABIDescriptor::is_volatile_reg(FloatRegister reg) const {
+    return _vector_argument_registers.contains(reg)
+        || _vector_additional_volatile_registers.contains(reg);
+}
+
+#define INTEGER_TYPE 0
+#define VECTOR_TYPE 1
+
+const ABIDescriptor ForeignGlobals::parse_abi_descriptor_impl(jobject jabi) const {
+  oop abi_oop = JNIHandles::resolve_non_null(jabi);
+  ABIDescriptor abi;
+
+  objArrayOop inputStorage = cast<objArrayOop>(abi_oop->obj_field(ABI.inputStorage_offset));
+  loadArray(inputStorage, INTEGER_TYPE, abi._integer_argument_registers, as_Register);
+  loadArray(inputStorage, VECTOR_TYPE, abi._vector_argument_registers, as_FloatRegister);
+
+  objArrayOop outputStorage = cast<objArrayOop>(abi_oop->obj_field(ABI.outputStorage_offset));
+  loadArray(outputStorage, INTEGER_TYPE, abi._integer_return_registers, as_Register);
+  loadArray(outputStorage, VECTOR_TYPE, abi._vector_return_registers, as_FloatRegister);
+
+  objArrayOop volatileStorage = cast<objArrayOop>(abi_oop->obj_field(ABI.volatileStorage_offset));
+  loadArray(volatileStorage, INTEGER_TYPE, abi._integer_additional_volatile_registers, as_Register);
+  loadArray(volatileStorage, VECTOR_TYPE, abi._vector_additional_volatile_registers, as_FloatRegister);
+
+  abi._stack_alignment_bytes = abi_oop->int_field(ABI.stackAlignment_offset);
+  abi._shadow_space_bytes = abi_oop->int_field(ABI.shadowSpace_offset);
+
+  return abi;
+}
+
+const BufferLayout ForeignGlobals::parse_buffer_layout_impl(jobject jlayout) const {
+  oop layout_oop = JNIHandles::resolve_non_null(jlayout);
+  BufferLayout layout;
+
+  layout.stack_args_bytes = layout_oop->long_field(BL.stack_args_bytes_offset);
+  layout.stack_args = layout_oop->long_field(BL.stack_args_offset);
+  layout.arguments_next_pc = layout_oop->long_field(BL.arguments_next_pc_offset);
+
+  typeArrayOop input_offsets = cast<typeArrayOop>(layout_oop->obj_field(BL.input_type_offsets_offset));
+  layout.arguments_integer = (size_t) input_offsets->long_at(INTEGER_TYPE);
+  layout.arguments_vector = (size_t) input_offsets->long_at(VECTOR_TYPE);
+
+  typeArrayOop output_offsets = cast<typeArrayOop>(layout_oop->obj_field(BL.output_type_offsets_offset));
+  layout.returns_integer = (size_t) output_offsets->long_at(INTEGER_TYPE);
+  layout.returns_vector = (size_t) output_offsets->long_at(VECTOR_TYPE);
+
+  layout.buffer_size = layout_oop->long_field(BL.size_offset);
+
+  return layout;
+}

src/hotspot/cpu/aarch64/foreign_globals_aarch64.hpp

@@ -0,0 +1,60 @@
+/*
+ * Copyright (c) 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, Arm Limited. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+#ifndef CPU_AARCH64_VM_FOREIGN_GLOBALS_AARCH64_HPP
+#define CPU_AARCH64_VM_FOREIGN_GLOBALS_AARCH64_HPP
+
+#include "asm/macroAssembler.hpp"
+#include "utilities/growableArray.hpp"
+
+constexpr size_t float_reg_size = 16; // bytes
+
+struct ABIDescriptor {
+  GrowableArray<Register> _integer_argument_registers;
+  GrowableArray<Register> _integer_return_registers;
+  GrowableArray<FloatRegister> _vector_argument_registers;
+  GrowableArray<FloatRegister> _vector_return_registers;
+
+  GrowableArray<Register> _integer_additional_volatile_registers;
+  GrowableArray<FloatRegister> _vector_additional_volatile_registers;
+
+  int32_t _stack_alignment_bytes;
+  int32_t _shadow_space_bytes;
+
+  bool is_volatile_reg(Register reg) const;
+  bool is_volatile_reg(FloatRegister reg) const;
+};
+
+struct BufferLayout {
+  size_t stack_args_bytes;
+  size_t stack_args;
+  size_t arguments_vector;
+  size_t arguments_integer;
+  size_t arguments_next_pc;
+  size_t returns_vector;
+  size_t returns_integer;
+  size_t buffer_size;
+};
+
+#endif // CPU_AARCH64_VM_FOREIGN_GLOBALS_AARCH64_HPP

src/hotspot/cpu/aarch64/methodHandles_aarch64.cpp

@@ -178,6 +178,13 @@ address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler*
     return NULL;
   }
 
+  // No need in interpreter entry for linkToNative for now.
+  // Interpreter calls compiled entry through i2c.
+  if (iid == vmIntrinsics::_linkToNative) {
+    __ hlt(0);
+    return NULL;
+  }
+
   // r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
   // rmethod: Method*
   // r3: argument locator (parameter slot count, added to rsp)
@@ -272,7 +279,10 @@ void MethodHandles::generate_method_handle_dispatch(MacroAssembler* _masm,
   assert_different_registers(temp1, temp2, temp3, receiver_reg);
   assert_different_registers(temp1, temp2, temp3, member_reg);
 
-  if (iid == vmIntrinsics::_invokeBasic) {
+  if (iid == vmIntrinsics::_invokeBasic || iid == vmIntrinsics::_linkToNative) {
+    if (iid == vmIntrinsics::_linkToNative) {
+      assert(for_compiler_entry, "only compiler entry is supported");
+    }
     // indirect through MH.form.vmentry.vmtarget
     jump_to_lambda_form(_masm, receiver_reg, rmethod, temp1, for_compiler_entry);
 

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -1195,7 +1195,7 @@ static void gen_special_dispatch(MacroAssembler* masm,
     member_arg_pos = method->size_of_parameters() - 1;  // trailing MemberName argument
     member_reg = r19;  // known to be free at this point
     has_receiver = MethodHandles::ref_kind_has_receiver(ref_kind);
-  } else if (iid == vmIntrinsics::_invokeBasic) {
+  } else if (iid == vmIntrinsics::_invokeBasic || iid == vmIntrinsics::_linkToNative) {
     has_receiver = true;
   } else {
     fatal("unexpected intrinsic id %d", iid);
@@ -3069,3 +3069,10 @@ void OptoRuntime::generate_exception_blob() {
   _exception_blob =  ExceptionBlob::create(&buffer, oop_maps, SimpleRuntimeFrame::framesize >> 1);
 }
 #endif // COMPILER2
+
+BufferBlob* SharedRuntime::make_native_invoker(address call_target,
+                                           int shadow_space_bytes,
+                                           const GrowableArray<VMReg>& input_registers,
+                                           const GrowableArray<VMReg>& output_registers) {
+  return NULL;
+}

src/hotspot/cpu/aarch64/universalNativeInvoker_aarch64.cpp

@@ -0,0 +1,127 @@
+/*
+ * Copyright (c) 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, Arm Limited. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+#include "precompiled.hpp"
+#include "asm/macroAssembler.hpp"
+#include "code/codeBlob.hpp"
+#include "memory/resourceArea.hpp"
+#include "prims/universalNativeInvoker.hpp"
+
+#define __ _masm->
+
+void ProgrammableInvoker::Generator::generate() {
+  __ enter();
+
+  // Name registers used in the stub code. These are all caller-save so
+  // may be clobbered by the call to the native function. Avoid using
+  // rscratch1 here as it's r8 which is the indirect result register in
+  // the standard ABI.
+  Register Rctx = r10, Rstack_size = r11;
+  Register Rwords = r12, Rtmp = r13;
+  Register Rsrc_ptr = r14, Rdst_ptr = r15;
+
+  assert_different_registers(Rctx, Rstack_size, rscratch1, rscratch2);
+
+  // TODO: if the callee is not using the standard C ABI then we need to
+  //       preserve more registers here.
+
+  __ block_comment("init_and_alloc_stack");
+
+  __ mov(Rctx, c_rarg0);
+  __ str(Rctx, Address(__ pre(sp, -2 * wordSize)));
+
+  assert(_abi->_stack_alignment_bytes % 16 == 0, "stack must be 16 byte aligned");
+
+  __ block_comment("allocate_stack");
+  __ ldr(Rstack_size, Address(Rctx, (int) _layout->stack_args_bytes));
+  __ add(rscratch2, Rstack_size, _abi->_stack_alignment_bytes - 1);
+  __ andr(rscratch2, rscratch2, -_abi->_stack_alignment_bytes);
+  __ sub(sp, sp, rscratch2);
+
+  __ block_comment("load_arguments");
+
+  __ ldr(Rsrc_ptr, Address(Rctx, (int) _layout->stack_args));
+  __ lsr(Rwords, Rstack_size, LogBytesPerWord);
+  __ mov(Rdst_ptr, sp);
+
+  Label Ldone, Lnext;
+  __ bind(Lnext);
+  __ cbz(Rwords, Ldone);
+  __ ldr(Rtmp, __ post(Rsrc_ptr, wordSize));
+  __ str(Rtmp, __ post(Rdst_ptr, wordSize));
+  __ sub(Rwords, Rwords, 1);
+  __ b(Lnext);
+  __ bind(Ldone);
+
+  for (int i = 0; i < _abi->_vector_argument_registers.length(); i++) {
+    ssize_t offs = _layout->arguments_vector + i * float_reg_size;
+    __ ldrq(_abi->_vector_argument_registers.at(i), Address(Rctx, offs));
+  }
+
+  for (int i = 0; i < _abi->_integer_argument_registers.length(); i++) {
+    ssize_t offs = _layout->arguments_integer + i * sizeof(uintptr_t);
+    __ ldr(_abi->_integer_argument_registers.at(i), Address(Rctx, offs));
+  }
+
+  assert(_abi->_shadow_space_bytes == 0, "shadow space not supported on AArch64");
+
+  // call target function
+  __ block_comment("call target function");
+  __ ldr(rscratch2, Address(Rctx, (int) _layout->arguments_next_pc));
+  __ blr(rscratch2);
+
+  __ ldr(Rctx, Address(rfp, -2 * wordSize));   // Might have clobbered Rctx
+
+  __ block_comment("store_registers");
+
+  for (int i = 0; i < _abi->_integer_return_registers.length(); i++) {
+    ssize_t offs = _layout->returns_integer + i * sizeof(uintptr_t);
+    __ str(_abi->_integer_return_registers.at(i), Address(Rctx, offs));
+  }
+
+  for (int i = 0; i < _abi->_vector_return_registers.length(); i++) {
+    ssize_t offs = _layout->returns_vector + i * float_reg_size;
+    __ strq(_abi->_vector_return_registers.at(i), Address(Rctx, offs));
+  }
+
+  __ leave();
+  __ ret(lr);
+
+  __ flush();
+}
+
+address ProgrammableInvoker::generate_adapter(jobject jabi, jobject jlayout) {
+  ResourceMark rm;
+  const ABIDescriptor abi = ForeignGlobals::parse_abi_descriptor(jabi);
+  const BufferLayout layout = ForeignGlobals::parse_buffer_layout(jlayout);
+
+  BufferBlob* _invoke_native_blob = BufferBlob::create("invoke_native_blob", native_invoker_size);
+
+  CodeBuffer code2(_invoke_native_blob);
+  ProgrammableInvoker::Generator g2(&code2, &abi, &layout);
+  g2.generate();
+  code2.log_section_sizes("InvokeNativeBlob");
+
+  return _invoke_native_blob->code_begin();
+}