# HG changeset patch
# User dpochepk
# Date 1561731022 -10800
# Node ID f9a2f93a0c87b894219522b6de5637e6b2018a68
# Parent  33bb8c970770fd2cd2e907cc37f4751dc35552c4
8223173: Implement fast class initialization checks on AARCH64
Reviewed-by: aph

diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/aarch64.ad
--- a/src/hotspot/cpu/aarch64/aarch64.ad	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/aarch64.ad	Fri Jun 28 17:10:22 2019 +0300
@@ -1761,6 +1761,17 @@
   // branch if we need to invalidate the method later
   __ nop();
 
+  if (C->clinit_barrier_on_entry()) {
+    assert(!C->method()->holder()->is_not_initialized(), "initialization should have been started");
+
+    Label L_skip_barrier;
+
+    __ mov_metadata(rscratch2, C->method()->holder()->constant_encoding());
+    __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
+    __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+    __ bind(L_skip_barrier);
+  }
+  
   int bangsize = C->bang_size_in_bytes();
   if (C->need_stack_bang(bangsize) && UseStackBanging)
     __ generate_stack_overflow_check(bangsize);
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -317,7 +317,15 @@
 }
 
 void LIR_Assembler::clinit_barrier(ciMethod* method) {
-  ShouldNotReachHere(); // not implemented
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  assert(!method->holder()->is_not_initialized(), "initialization should have been started");
+
+  Label L_skip_barrier;
+
+  __ mov_metadata(rscratch2, method->holder()->constant_encoding());
+  __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier /*L_fast_path*/);
+  __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+  __ bind(L_skip_barrier);
 }
 
 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -331,11 +331,6 @@
 
 
 void C1_MacroAssembler::build_frame(int framesize, int bang_size_in_bytes) {
-  // If we have to make this method not-entrant we'll overwrite its
-  // first instruction with a jump.  For this action to be legal we
-  // must ensure that this first instruction is a B, BL, NOP, BKPT,
-  // SVC, HVC, or SMC.  Make it a NOP.
-  nop();
   assert(bang_size_in_bytes >= framesize, "stack bang size incorrect");
   // Make sure there is enough stack space for this method's activation.
   // Note that we do this before doing an enter().
@@ -355,6 +350,11 @@
 
 
 void C1_MacroAssembler::verified_entry() {
+  // If we have to make this method not-entrant we'll overwrite its
+  // first instruction with a jump.  For this action to be legal we
+  // must ensure that this first instruction is a B, BL, NOP, BKPT,
+  // SVC, HVC, or SMC.  Make it a NOP.
+  nop();
 }
 
 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -288,6 +288,18 @@
   ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
 }
 
+void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
+                                                              Register method,
+                                                              Register cache) {
+  const int method_offset = in_bytes(
+    ConstantPoolCache::base_offset() +
+      ((byte_no == TemplateTable::f2_byte)
+       ? ConstantPoolCacheEntry::f2_offset()
+       : ConstantPoolCacheEntry::f1_offset()));
+
+  ldr(method, Address(cache, method_offset)); // get f1 Method*
+}
+
 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 // subtype of super_klass.
 //
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp	Fri Jun 28 17:10:22 2019 +0300
@@ -124,6 +124,8 @@
   // load cpool->resolved_klass_at(index);
   void load_resolved_klass_at_offset(Register cpool, Register index, Register klass, Register temp);
 
+  void load_resolved_method_at_index(int byte_no, Register method, Register cache);
+
   void pop_ptr(Register r = r0);
   void pop_i(Register r = r0);
   void pop_l(Register r = r0);
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -1307,6 +1307,35 @@
   bind(L_fallthrough);
 }
 
+void MacroAssembler::clinit_barrier(Register klass, Register scratch, Label* L_fast_path, Label* L_slow_path) {
+  assert(L_fast_path != NULL || L_slow_path != NULL, "at least one is required");
+  assert_different_registers(klass, rthread, scratch);
+
+  Label L_fallthrough, L_tmp;
+  if (L_fast_path == NULL) {
+    L_fast_path = &L_fallthrough;
+  } else if (L_slow_path == NULL) {
+    L_slow_path = &L_fallthrough;
+  }
+  // Fast path check: class is fully initialized
+  ldrb(scratch, Address(klass, InstanceKlass::init_state_offset()));
+  subs(zr, scratch, InstanceKlass::fully_initialized);
+  br(Assembler::EQ, *L_fast_path);
+
+  // Fast path check: current thread is initializer thread
+  ldr(scratch, Address(klass, InstanceKlass::init_thread_offset()));
+  cmp(rthread, scratch);
+
+  if (L_slow_path == &L_fallthrough) {
+    br(Assembler::EQ, *L_fast_path);
+    bind(*L_slow_path);
+  } else if (L_fast_path == &L_fallthrough) {
+    br(Assembler::NE, *L_slow_path);
+    bind(*L_fast_path);
+  } else {
+    Unimplemented();
+  }
+}
 
 void MacroAssembler::verify_oop(Register reg, const char* s) {
   if (!VerifyOops) return;
@@ -3683,6 +3712,12 @@
   bs->obj_equals(this, obj1, obj2);
 }
 
+void MacroAssembler::load_method_holder(Register holder, Register method) {
+  ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
+  ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
+  ldr(holder, Address(holder, ConstantPool::pool_holder_offset_in_bytes())); // InstanceKlass*
+}
+
 void MacroAssembler::load_klass(Register dst, Register src) {
   if (UseCompressedClassPointers) {
     ldrw(dst, Address(src, oopDesc::klass_offset_in_bytes()));
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	Fri Jun 28 17:10:22 2019 +0300
@@ -788,6 +788,8 @@
   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
   void c2bool(Register x);
 
+  void load_method_holder(Register holder, Register method);
+
   // oop manipulations
   void load_klass(Register dst, Register src);
   void store_klass(Register dst, Register src);
@@ -926,6 +928,11 @@
                            Register temp_reg,
                            Label& L_success);
 
+  void clinit_barrier(Register klass,
+                      Register thread,
+                      Label* L_fast_path = NULL,
+                      Label* L_slow_path = NULL);
+
   Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
 
 
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -799,6 +799,22 @@
   }
 #endif
 
+  // Class initialization barrier for static methods
+  if (VM_Version::supports_fast_class_init_checks()) {
+    Label L_skip_barrier;
+
+    { // Bypass the barrier for non-static methods
+      __ ldrw(rscratch1, Address(rmethod, Method::access_flags_offset()));
+      __ andsw(zr, rscratch1, JVM_ACC_STATIC);
+      __ br(Assembler::EQ, L_skip_barrier); // non-static
+    }
+
+    __ load_method_holder(rscratch2, rmethod);
+    __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
+    __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+    __ bind(L_skip_barrier);
+  }
+
   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 
   __ flush();
@@ -1580,6 +1596,15 @@
   // SVC, HVC, or SMC.  Make it a NOP.
   __ nop();
 
+  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+    Label L_skip_barrier;
+    __ mov_metadata(rscratch2, method->method_holder()); // InstanceKlass*
+    __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
+    __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+
+    __ bind(L_skip_barrier);
+  }
+
   // Generate stack overflow check
   if (UseStackBanging) {
     __ bang_stack_with_offset(JavaThread::stack_shadow_zone_size());
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp	Fri Jun 28 17:10:22 2019 +0300
@@ -2323,7 +2323,7 @@
   const Register temp = r19;
   assert_different_registers(Rcache, index, temp);
 
-  Label resolved;
+  Label resolved, clinit_barrier_slow;
 
   Bytecodes::Code code = bytecode();
   switch (code) {
@@ -2338,6 +2338,8 @@
   __ br(Assembler::EQ, resolved);
 
   // resolve first time through
+  // Class initialization barrier slow path lands here as well.
+  __ bind(clinit_barrier_slow);
   address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
   __ mov(temp, (int) code);
   __ call_VM(noreg, entry, temp);
@@ -2347,6 +2349,13 @@
   // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
   // so all clients ofthis method must be modified accordingly
   __ bind(resolved);
+
+  // Class initialization barrier for static methods
+  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+    __ load_resolved_method_at_index(byte_no, temp, Rcache);
+    __ load_method_holder(temp, temp);
+    __ clinit_barrier(temp, rscratch1, NULL, &clinit_barrier_slow);
+  }
 }
 
 // The Rcache and index registers must be set before call
@@ -3418,9 +3427,8 @@
   __ profile_virtual_call(r3, r13, r19);
 
   // Get declaring interface class from method, and itable index
-  __ ldr(r0, Address(rmethod, Method::const_offset()));
-  __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
-  __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
+
+  __ load_method_holder(r0, rmethod);
   __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
   __ subw(rmethod, rmethod, Method::itable_index_max);
   __ negw(rmethod, rmethod);
diff -r 33bb8c970770 -r f9a2f93a0c87 src/hotspot/cpu/aarch64/vm_version_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp	Fri Jun 28 11:26:07 2019 +0100
+++ b/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp	Fri Jun 28 17:10:22 2019 +0300
@@ -124,6 +124,7 @@
   static int dcache_line_size() {
     return (1 << ((_psr_info.ctr_el0 >> 16) & 0x0f)) * 4;
   }
+  static bool supports_fast_class_init_checks() { return true; }
 };
 
 #endif // CPU_AARCH64_VM_VERSION_AARCH64_HPP