1 /*
   2  * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "gc/shared/barrierSet.hpp"
  29 #include "gc/shared/barrierSetAssembler.hpp"
  30 #include "interp_masm_aarch64.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "interpreter/interpreterRuntime.hpp"
  33 #include "logging/log.hpp"
  34 #include "oops/arrayOop.hpp"
  35 #include "oops/markWord.hpp"
  36 #include "oops/method.hpp"
  37 #include "oops/methodData.hpp"
  38 #include "prims/jvmtiExport.hpp"
  39 #include "prims/jvmtiThreadState.hpp"
  40 #include "runtime/basicLock.hpp"
  41 #include "runtime/biasedLocking.hpp"
  42 #include "runtime/frame.inline.hpp"
  43 #include "runtime/safepointMechanism.hpp"
  44 #include "runtime/sharedRuntime.hpp"
  45 #include "runtime/thread.inline.hpp"
  46 #include "utilities/powerOfTwo.hpp"
  47 
  48 void InterpreterMacroAssembler::narrow(Register result) {
  49 
  50   // Get method->_constMethod->_result_type
  51   ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
  52   ldr(rscratch1, Address(rscratch1, Method::const_offset()));
  53   ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
  54 
  55   Label done, notBool, notByte, notChar;
  56 
  57   // common case first
  58   cmpw(rscratch1, T_INT);
  59   br(Assembler::EQ, done);
  60 
  61   // mask integer result to narrower return type.
  62   cmpw(rscratch1, T_BOOLEAN);
  63   br(Assembler::NE, notBool);
  64   andw(result, result, 0x1);
  65   b(done);
  66 
  67   bind(notBool);
  68   cmpw(rscratch1, T_BYTE);
  69   br(Assembler::NE, notByte);
  70   sbfx(result, result, 0, 8);
  71   b(done);
  72 
  73   bind(notByte);
  74   cmpw(rscratch1, T_CHAR);
  75   br(Assembler::NE, notChar);
  76   ubfx(result, result, 0, 16);  // truncate upper 16 bits
  77   b(done);
  78 
  79   bind(notChar);
  80   sbfx(result, result, 0, 16);     // sign-extend short
  81 
  82   // Nothing to do for T_INT
  83   bind(done);
  84 }
  85 
  86 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  87   assert(entry, "Entry must have been generated by now");
  88   b(entry);
  89 }
  90 
  91 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
  92   if (JvmtiExport::can_pop_frame()) {
  93     Label L;
  94     // Initiate popframe handling only if it is not already being
  95     // processed.  If the flag has the popframe_processing bit set, it
  96     // means that this code is called *during* popframe handling - we
  97     // don't want to reenter.
  98     // This method is only called just after the call into the vm in
  99     // call_VM_base, so the arg registers are available.
 100     ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
 101     tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
 102     tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
 103     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 104     // address of the same-named entrypoint in the generated interpreter code.
 105     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 106     br(r0);
 107     bind(L);
 108   }
 109 }
 110 
 111 
 112 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 113   ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 114   const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
 115   const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
 116   const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
 117   switch (state) {
 118     case atos: ldr(r0, oop_addr);
 119                str(zr, oop_addr);
 120                verify_oop(r0, state);               break;
 121     case ltos: ldr(r0, val_addr);                   break;
 122     case btos:                                   // fall through
 123     case ztos:                                   // fall through
 124     case ctos:                                   // fall through
 125     case stos:                                   // fall through
 126     case itos: ldrw(r0, val_addr);                  break;
 127     case ftos: ldrs(v0, val_addr);                  break;
 128     case dtos: ldrd(v0, val_addr);                  break;
 129     case vtos: /* nothing to do */                  break;
 130     default  : ShouldNotReachHere();
 131   }
 132   // Clean up tos value in the thread object
 133   movw(rscratch1, (int) ilgl);
 134   strw(rscratch1, tos_addr);
 135   strw(zr, val_addr);
 136 }
 137 
 138 
 139 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 140   if (JvmtiExport::can_force_early_return()) {
 141     Label L;
 142     ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 143     cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit;
 144 
 145     // Initiate earlyret handling only if it is not already being processed.
 146     // If the flag has the earlyret_processing bit set, it means that this code
 147     // is called *during* earlyret handling - we don't want to reenter.
 148     ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
 149     cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
 150     br(Assembler::NE, L);
 151 
 152     // Call Interpreter::remove_activation_early_entry() to get the address of the
 153     // same-named entrypoint in the generated interpreter code.
 154     ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 155     ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
 156     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
 157     br(r0);
 158     bind(L);
 159   }
 160 }
 161 
 162 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
 163   Register reg,
 164   int bcp_offset) {
 165   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 166   ldrh(reg, Address(rbcp, bcp_offset));
 167   rev16(reg, reg);
 168 }
 169 
 170 void InterpreterMacroAssembler::get_dispatch() {
 171   unsigned long offset;
 172   adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
 173   lea(rdispatch, Address(rdispatch, offset));
 174 }
 175 
 176 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 177                                                        int bcp_offset,
 178                                                        size_t index_size) {
 179   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 180   if (index_size == sizeof(u2)) {
 181     load_unsigned_short(index, Address(rbcp, bcp_offset));
 182   } else if (index_size == sizeof(u4)) {
 183     // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
 184     ldrw(index, Address(rbcp, bcp_offset));
 185     // Check if the secondary index definition is still ~x, otherwise
 186     // we have to change the following assembler code to calculate the
 187     // plain index.
 188     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 189     eonw(index, index, zr);  // convert to plain index
 190   } else if (index_size == sizeof(u1)) {
 191     load_unsigned_byte(index, Address(rbcp, bcp_offset));
 192   } else {
 193     ShouldNotReachHere();
 194   }
 195 }
 196 
 197 // Return
 198 // Rindex: index into constant pool
 199 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
 200 //
 201 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
 202 // the true address of the cache entry.
 203 //
 204 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 205                                                            Register index,
 206                                                            int bcp_offset,
 207                                                            size_t index_size) {
 208   assert_different_registers(cache, index);
 209   assert_different_registers(cache, rcpool);
 210   get_cache_index_at_bcp(index, bcp_offset, index_size);
 211   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 212   // convert from field index to ConstantPoolCacheEntry
 213   // aarch64 already has the cache in rcpool so there is no need to
 214   // install it in cache. instead we pre-add the indexed offset to
 215   // rcpool and return it in cache. All clients of this method need to
 216   // be modified accordingly.
 217   add(cache, rcpool, index, Assembler::LSL, 5);
 218 }
 219 
 220 
 221 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 222                                                                         Register index,
 223                                                                         Register bytecode,
 224                                                                         int byte_no,
 225                                                                         int bcp_offset,
 226                                                                         size_t index_size) {
 227   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
 228   // We use a 32-bit load here since the layout of 64-bit words on
 229   // little-endian machines allow us that.
 230   // n.b. unlike x86 cache already includes the index offset
 231   lea(bytecode, Address(cache,
 232                          ConstantPoolCache::base_offset()
 233                          + ConstantPoolCacheEntry::indices_offset()));
 234   ldarw(bytecode, bytecode);
 235   const int shift_count = (1 + byte_no) * BitsPerByte;
 236   ubfx(bytecode, bytecode, shift_count, BitsPerByte);
 237 }
 238 
 239 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 240                                                                Register tmp,
 241                                                                int bcp_offset,
 242                                                                size_t index_size) {
 243   assert(cache != tmp, "must use different register");
 244   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 245   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 246   // convert from field index to ConstantPoolCacheEntry index
 247   // and from word offset to byte offset
 248   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
 249   ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
 250   // skip past the header
 251   add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
 252   add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord);  // construct pointer to cache entry
 253 }
 254 
 255 void InterpreterMacroAssembler::get_method_counters(Register method,
 256                                                     Register mcs, Label& skip) {
 257   Label has_counters;
 258   ldr(mcs, Address(method, Method::method_counters_offset()));
 259   cbnz(mcs, has_counters);
 260   call_VM(noreg, CAST_FROM_FN_PTR(address,
 261           InterpreterRuntime::build_method_counters), method);
 262   ldr(mcs, Address(method, Method::method_counters_offset()));
 263   cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
 264   bind(has_counters);
 265 }
 266 
 267 // Load object from cpool->resolved_references(index)
 268 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 269                                            Register result, Register index, Register tmp) {
 270   assert_different_registers(result, index);
 271 
 272   get_constant_pool(result);
 273   // load pointer for resolved_references[] objArray
 274   ldr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
 275   ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
 276   resolve_oop_handle(result, tmp);
 277   // Add in the index
 278   add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 279   load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)));
 280 }
 281 
 282 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
 283                              Register cpool, Register index, Register klass, Register temp) {
 284   add(temp, cpool, index, LSL, LogBytesPerWord);
 285   ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
 286   ldr(klass, Address(cpool,  ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
 287   add(klass, klass, temp, LSL, LogBytesPerWord);
 288   ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
 289 }
 290 
 291 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
 292                                                               Register method,
 293                                                               Register cache) {
 294   const int method_offset = in_bytes(
 295     ConstantPoolCache::base_offset() +
 296       ((byte_no == TemplateTable::f2_byte)
 297        ? ConstantPoolCacheEntry::f2_offset()
 298        : ConstantPoolCacheEntry::f1_offset()));
 299 
 300   ldr(method, Address(cache, method_offset)); // get f1 Method*
 301 }
 302 
 303 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 304 // subtype of super_klass.
 305 //
 306 // Args:
 307 //      r0: superklass
 308 //      Rsub_klass: subklass
 309 //
 310 // Kills:
 311 //      r2, r5
 312 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 313                                                   Label& ok_is_subtype) {
 314   assert(Rsub_klass != r0, "r0 holds superklass");
 315   assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
 316   assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
 317 
 318   // Profile the not-null value's klass.
 319   profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
 320 
 321   // Do the check.
 322   check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
 323 
 324   // Profile the failure of the check.
 325   profile_typecheck_failed(r2); // blows r2
 326 }
 327 
 328 // Java Expression Stack
 329 
 330 void InterpreterMacroAssembler::pop_ptr(Register r) {
 331   ldr(r, post(esp, wordSize));
 332 }
 333 
 334 void InterpreterMacroAssembler::pop_i(Register r) {
 335   ldrw(r, post(esp, wordSize));
 336 }
 337 
 338 void InterpreterMacroAssembler::pop_l(Register r) {
 339   ldr(r, post(esp, 2 * Interpreter::stackElementSize));
 340 }
 341 
 342 void InterpreterMacroAssembler::push_ptr(Register r) {
 343   str(r, pre(esp, -wordSize));
 344  }
 345 
 346 void InterpreterMacroAssembler::push_i(Register r) {
 347   str(r, pre(esp, -wordSize));
 348 }
 349 
 350 void InterpreterMacroAssembler::push_l(Register r) {
 351   str(zr, pre(esp, -wordSize));
 352   str(r, pre(esp, - wordSize));
 353 }
 354 
 355 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
 356   ldrs(r, post(esp, wordSize));
 357 }
 358 
 359 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
 360   ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
 361 }
 362 
 363 void InterpreterMacroAssembler::push_f(FloatRegister r) {
 364   strs(r, pre(esp, -wordSize));
 365 }
 366 
 367 void InterpreterMacroAssembler::push_d(FloatRegister r) {
 368   strd(r, pre(esp, 2* -wordSize));
 369 }
 370 
 371 void InterpreterMacroAssembler::pop(TosState state) {
 372   switch (state) {
 373   case atos: pop_ptr();                 break;
 374   case btos:
 375   case ztos:
 376   case ctos:
 377   case stos:
 378   case itos: pop_i();                   break;
 379   case ltos: pop_l();                   break;
 380   case ftos: pop_f();                   break;
 381   case dtos: pop_d();                   break;
 382   case vtos: /* nothing to do */        break;
 383   default:   ShouldNotReachHere();
 384   }
 385   verify_oop(r0, state);
 386 }
 387 
 388 void InterpreterMacroAssembler::push(TosState state) {
 389   verify_oop(r0, state);
 390   switch (state) {
 391   case atos: push_ptr();                break;
 392   case btos:
 393   case ztos:
 394   case ctos:
 395   case stos:
 396   case itos: push_i();                  break;
 397   case ltos: push_l();                  break;
 398   case ftos: push_f();                  break;
 399   case dtos: push_d();                  break;
 400   case vtos: /* nothing to do */        break;
 401   default  : ShouldNotReachHere();
 402   }
 403 }
 404 
 405 // Helpers for swap and dup
 406 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 407   ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
 408 }
 409 
 410 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 411   str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
 412 }
 413 
 414 void InterpreterMacroAssembler::load_float(Address src) {
 415   ldrs(v0, src);
 416 }
 417 
 418 void InterpreterMacroAssembler::load_double(Address src) {
 419   ldrd(v0, src);
 420 }
 421 
 422 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 423   // set sender sp
 424   mov(r13, sp);
 425   // record last_sp
 426   str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 427 }
 428 
 429 // Jump to from_interpreted entry of a call unless single stepping is possible
 430 // in this thread in which case we must call the i2i entry
 431 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 432   prepare_to_jump_from_interpreted();
 433 
 434   if (JvmtiExport::can_post_interpreter_events()) {
 435     Label run_compiled_code;
 436     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 437     // compiled code in threads for which the event is enabled.  Check here for
 438     // interp_only_mode if these events CAN be enabled.
 439     ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
 440     cbzw(rscratch1, run_compiled_code);
 441     ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
 442     br(rscratch1);
 443     bind(run_compiled_code);
 444   }
 445 
 446   ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
 447   br(rscratch1);
 448 }
 449 
 450 // The following two routines provide a hook so that an implementation
 451 // can schedule the dispatch in two parts.  amd64 does not do this.
 452 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 453 }
 454 
 455 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 456     dispatch_next(state, step);
 457 }
 458 
 459 void InterpreterMacroAssembler::dispatch_base(TosState state,
 460                                               address* table,
 461                                               bool verifyoop,
 462                                               bool generate_poll) {
 463   if (VerifyActivationFrameSize) {
 464     Unimplemented();
 465   }
 466   if (verifyoop) {
 467     verify_oop(r0, state);
 468   }
 469 
 470   Label safepoint;
 471   address* const safepoint_table = Interpreter::safept_table(state);
 472   bool needs_thread_local_poll = generate_poll && table != safepoint_table;
 473 
 474   if (needs_thread_local_poll) {
 475     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 476     ldr(rscratch2, Address(rthread, Thread::polling_page_offset()));
 477     tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
 478   }
 479 
 480   if (table == Interpreter::dispatch_table(state)) {
 481     addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
 482     ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
 483   } else {
 484     mov(rscratch2, (address)table);
 485     ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
 486   }
 487   br(rscratch2);
 488 
 489   if (needs_thread_local_poll) {
 490     bind(safepoint);
 491     lea(rscratch2, ExternalAddress((address)safepoint_table));
 492     ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
 493     br(rscratch2);
 494   }
 495 }
 496 
 497 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 498   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 499 }
 500 
 501 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 502   dispatch_base(state, Interpreter::normal_table(state));
 503 }
 504 
 505 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 506   dispatch_base(state, Interpreter::normal_table(state), false);
 507 }
 508 
 509 
 510 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 511   // load next bytecode
 512   ldrb(rscratch1, Address(pre(rbcp, step)));
 513   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
 514 }
 515 
 516 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 517   // load current bytecode
 518   ldrb(rscratch1, Address(rbcp, 0));
 519   dispatch_base(state, table);
 520 }
 521 
 522 // remove activation
 523 //
 524 // Unlock the receiver if this is a synchronized method.
 525 // Unlock any Java monitors from syncronized blocks.
 526 // Remove the activation from the stack.
 527 //
 528 // If there are locked Java monitors
 529 //    If throw_monitor_exception
 530 //       throws IllegalMonitorStateException
 531 //    Else if install_monitor_exception
 532 //       installs IllegalMonitorStateException
 533 //    Else
 534 //       no error processing
 535 void InterpreterMacroAssembler::remove_activation(
 536         TosState state,
 537         bool throw_monitor_exception,
 538         bool install_monitor_exception,
 539         bool notify_jvmdi) {
 540   // Note: Registers r3 xmm0 may be in use for the
 541   // result check if synchronized method
 542   Label unlocked, unlock, no_unlock;
 543 
 544   // get the value of _do_not_unlock_if_synchronized into r3
 545   const Address do_not_unlock_if_synchronized(rthread,
 546     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 547   ldrb(r3, do_not_unlock_if_synchronized);
 548   strb(zr, do_not_unlock_if_synchronized); // reset the flag
 549 
 550  // get method access flags
 551   ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
 552   ldr(r2, Address(r1, Method::access_flags_offset()));
 553   tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
 554 
 555   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 556   // is set.
 557   cbnz(r3, no_unlock);
 558 
 559   // unlock monitor
 560   push(state); // save result
 561 
 562   // BasicObjectLock will be first in list, since this is a
 563   // synchronized method. However, need to check that the object has
 564   // not been unlocked by an explicit monitorexit bytecode.
 565   const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
 566                         wordSize - (int) sizeof(BasicObjectLock));
 567   // We use c_rarg1 so that if we go slow path it will be the correct
 568   // register for unlock_object to pass to VM directly
 569   lea(c_rarg1, monitor); // address of first monitor
 570 
 571   ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 572   cbnz(r0, unlock);
 573 
 574   pop(state);
 575   if (throw_monitor_exception) {
 576     // Entry already unlocked, need to throw exception
 577     call_VM(noreg, CAST_FROM_FN_PTR(address,
 578                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 579     should_not_reach_here();
 580   } else {
 581     // Monitor already unlocked during a stack unroll. If requested,
 582     // install an illegal_monitor_state_exception.  Continue with
 583     // stack unrolling.
 584     if (install_monitor_exception) {
 585       call_VM(noreg, CAST_FROM_FN_PTR(address,
 586                      InterpreterRuntime::new_illegal_monitor_state_exception));
 587     }
 588     b(unlocked);
 589   }
 590 
 591   bind(unlock);
 592   unlock_object(c_rarg1);
 593   pop(state);
 594 
 595   // Check that for block-structured locking (i.e., that all locked
 596   // objects has been unlocked)
 597   bind(unlocked);
 598 
 599   // r0: Might contain return value
 600 
 601   // Check that all monitors are unlocked
 602   {
 603     Label loop, exception, entry, restart;
 604     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 605     const Address monitor_block_top(
 606         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 607     const Address monitor_block_bot(
 608         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
 609 
 610     bind(restart);
 611     // We use c_rarg1 so that if we go slow path it will be the correct
 612     // register for unlock_object to pass to VM directly
 613     ldr(c_rarg1, monitor_block_top); // points to current entry, starting
 614                                      // with top-most entry
 615     lea(r19, monitor_block_bot);  // points to word before bottom of
 616                                   // monitor block
 617     b(entry);
 618 
 619     // Entry already locked, need to throw exception
 620     bind(exception);
 621 
 622     if (throw_monitor_exception) {
 623       // Throw exception
 624       MacroAssembler::call_VM(noreg,
 625                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 626                                    throw_illegal_monitor_state_exception));
 627       should_not_reach_here();
 628     } else {
 629       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 630       // Unlock does not block, so don't have to worry about the frame.
 631       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 632 
 633       push(state);
 634       unlock_object(c_rarg1);
 635       pop(state);
 636 
 637       if (install_monitor_exception) {
 638         call_VM(noreg, CAST_FROM_FN_PTR(address,
 639                                         InterpreterRuntime::
 640                                         new_illegal_monitor_state_exception));
 641       }
 642 
 643       b(restart);
 644     }
 645 
 646     bind(loop);
 647     // check if current entry is used
 648     ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 649     cbnz(rscratch1, exception);
 650 
 651     add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
 652     bind(entry);
 653     cmp(c_rarg1, r19); // check if bottom reached
 654     br(Assembler::NE, loop); // if not at bottom then check this entry
 655   }
 656 
 657   bind(no_unlock);
 658 
 659   // jvmti support
 660   if (notify_jvmdi) {
 661     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 662   } else {
 663     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 664   }
 665 
 666   // remove activation
 667   // get sender esp
 668   ldr(esp,
 669       Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
 670   if (StackReservedPages > 0) {
 671     // testing if reserved zone needs to be re-enabled
 672     Label no_reserved_zone_enabling;
 673 
 674     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 675     cmp(esp, rscratch1);
 676     br(Assembler::LS, no_reserved_zone_enabling);
 677 
 678     call_VM_leaf(
 679       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
 680     call_VM(noreg, CAST_FROM_FN_PTR(address,
 681                    InterpreterRuntime::throw_delayed_StackOverflowError));
 682     should_not_reach_here();
 683 
 684     bind(no_reserved_zone_enabling);
 685   }
 686   // remove frame anchor
 687   leave();
 688   // If we're returning to interpreted code we will shortly be
 689   // adjusting SP to allow some space for ESP.  If we're returning to
 690   // compiled code the saved sender SP was saved in sender_sp, so this
 691   // restores it.
 692   andr(sp, esp, -16);
 693 }
 694 
 695 // Lock object
 696 //
 697 // Args:
 698 //      c_rarg1: BasicObjectLock to be used for locking
 699 //
 700 // Kills:
 701 //      r0
 702 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 703 //      rscratch1, rscratch2 (scratch regs)
 704 void InterpreterMacroAssembler::lock_object(Register lock_reg)
 705 {
 706   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 707   if (UseHeavyMonitors) {
 708     call_VM(noreg,
 709             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 710             lock_reg);
 711   } else {
 712     Label done;
 713 
 714     const Register swap_reg = r0;
 715     const Register tmp = c_rarg2;
 716     const Register obj_reg = c_rarg3; // Will contain the oop
 717 
 718     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 719     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 720     const int mark_offset = lock_offset +
 721                             BasicLock::displaced_header_offset_in_bytes();
 722 
 723     Label slow_case;
 724 
 725     // Load object pointer into obj_reg %c_rarg3
 726     ldr(obj_reg, Address(lock_reg, obj_offset));
 727 
 728     if (UseBiasedLocking) {
 729       biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
 730     }
 731 
 732     // Load (object->mark() | 1) into swap_reg
 733     ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 734     orr(swap_reg, rscratch1, 1);
 735 
 736     // Save (object->mark() | 1) into BasicLock's displaced header
 737     str(swap_reg, Address(lock_reg, mark_offset));
 738 
 739     assert(lock_offset == 0,
 740            "displached header must be first word in BasicObjectLock");
 741 
 742     Label fail;
 743     if (PrintBiasedLockingStatistics) {
 744       Label fast;
 745       cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail);
 746       bind(fast);
 747       atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
 748                   rscratch2, rscratch1, tmp);
 749       b(done);
 750       bind(fail);
 751     } else {
 752       cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 753     }
 754 
 755     // Test if the oopMark is an obvious stack pointer, i.e.,
 756     //  1) (mark & 7) == 0, and
 757     //  2) rsp <= mark < mark + os::pagesize()
 758     //
 759     // These 3 tests can be done by evaluating the following
 760     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
 761     // assuming both stack pointer and pagesize have their
 762     // least significant 3 bits clear.
 763     // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg
 764     // NOTE2: aarch64 does not like to subtract sp from rn so take a
 765     // copy
 766     mov(rscratch1, sp);
 767     sub(swap_reg, swap_reg, rscratch1);
 768     ands(swap_reg, swap_reg, (unsigned long)(7 - os::vm_page_size()));
 769 
 770     // Save the test result, for recursive case, the result is zero
 771     str(swap_reg, Address(lock_reg, mark_offset));
 772 
 773     if (PrintBiasedLockingStatistics) {
 774       br(Assembler::NE, slow_case);
 775       atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
 776                   rscratch2, rscratch1, tmp);
 777     }
 778     br(Assembler::EQ, done);
 779 
 780     bind(slow_case);
 781 
 782     // Call the runtime routine for slow case
 783     call_VM(noreg,
 784             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 785             lock_reg);
 786 
 787     bind(done);
 788   }
 789 }
 790 
 791 
 792 // Unlocks an object. Used in monitorexit bytecode and
 793 // remove_activation.  Throws an IllegalMonitorException if object is
 794 // not locked by current thread.
 795 //
 796 // Args:
 797 //      c_rarg1: BasicObjectLock for lock
 798 //
 799 // Kills:
 800 //      r0
 801 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 802 //      rscratch1, rscratch2 (scratch regs)
 803 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
 804 {
 805   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 806 
 807   if (UseHeavyMonitors) {
 808     call_VM(noreg,
 809             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 810             lock_reg);
 811   } else {
 812     Label done;
 813 
 814     const Register swap_reg   = r0;
 815     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 816     const Register obj_reg    = c_rarg3;  // Will contain the oop
 817 
 818     save_bcp(); // Save in case of exception
 819 
 820     // Convert from BasicObjectLock structure to object and BasicLock
 821     // structure Store the BasicLock address into %r0
 822     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 823 
 824     // Load oop into obj_reg(%c_rarg3)
 825     ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 826 
 827     // Free entry
 828     str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 829 
 830     if (UseBiasedLocking) {
 831       biased_locking_exit(obj_reg, header_reg, done);
 832     }
 833 
 834     // Load the old header from BasicLock structure
 835     ldr(header_reg, Address(swap_reg,
 836                             BasicLock::displaced_header_offset_in_bytes()));
 837 
 838     // Test for recursion
 839     cbz(header_reg, done);
 840 
 841     // Atomic swap back the old header
 842     cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 843 
 844     // Call the runtime routine for slow case.
 845     str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
 846     call_VM(noreg,
 847             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
 848             lock_reg);
 849 
 850     bind(done);
 851 
 852     restore_bcp();
 853   }
 854 }
 855 
 856 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 857                                                          Label& zero_continue) {
 858   assert(ProfileInterpreter, "must be profiling interpreter");
 859   ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 860   cbz(mdp, zero_continue);
 861 }
 862 
 863 // Set the method data pointer for the current bcp.
 864 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 865   assert(ProfileInterpreter, "must be profiling interpreter");
 866   Label set_mdp;
 867   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 868 
 869   // Test MDO to avoid the call if it is NULL.
 870   ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
 871   cbz(r0, set_mdp);
 872   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
 873   // r0: mdi
 874   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 875   ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
 876   lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
 877   add(r0, r1, r0);
 878   str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 879   bind(set_mdp);
 880   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 881 }
 882 
 883 void InterpreterMacroAssembler::verify_method_data_pointer() {
 884   assert(ProfileInterpreter, "must be profiling interpreter");
 885 #ifdef ASSERT
 886   Label verify_continue;
 887   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 888   stp(r2, r3, Address(pre(sp, -2 * wordSize)));
 889   test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
 890   get_method(r1);
 891 
 892   // If the mdp is valid, it will point to a DataLayout header which is
 893   // consistent with the bcp.  The converse is highly probable also.
 894   ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
 895   ldr(rscratch1, Address(r1, Method::const_offset()));
 896   add(r2, r2, rscratch1, Assembler::LSL);
 897   lea(r2, Address(r2, ConstMethod::codes_offset()));
 898   cmp(r2, rbcp);
 899   br(Assembler::EQ, verify_continue);
 900   // r1: method
 901   // rbcp: bcp // rbcp == 22
 902   // r3: mdp
 903   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 904                r1, rbcp, r3);
 905   bind(verify_continue);
 906   ldp(r2, r3, Address(post(sp, 2 * wordSize)));
 907   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 908 #endif // ASSERT
 909 }
 910 
 911 
 912 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 913                                                 int constant,
 914                                                 Register value) {
 915   assert(ProfileInterpreter, "must be profiling interpreter");
 916   Address data(mdp_in, constant);
 917   str(value, data);
 918 }
 919 
 920 
 921 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 922                                                       int constant,
 923                                                       bool decrement) {
 924   increment_mdp_data_at(mdp_in, noreg, constant, decrement);
 925 }
 926 
 927 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 928                                                       Register reg,
 929                                                       int constant,
 930                                                       bool decrement) {
 931   assert(ProfileInterpreter, "must be profiling interpreter");
 932   // %%% this does 64bit counters at best it is wasting space
 933   // at worst it is a rare bug when counters overflow
 934 
 935   assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
 936 
 937   Address addr1(mdp_in, constant);
 938   Address addr2(rscratch2, reg, Address::lsl(0));
 939   Address &addr = addr1;
 940   if (reg != noreg) {
 941     lea(rscratch2, addr1);
 942     addr = addr2;
 943   }
 944 
 945   if (decrement) {
 946     // Decrement the register.  Set condition codes.
 947     // Intel does this
 948     // addptr(data, (int32_t) -DataLayout::counter_increment);
 949     // If the decrement causes the counter to overflow, stay negative
 950     // Label L;
 951     // jcc(Assembler::negative, L);
 952     // addptr(data, (int32_t) DataLayout::counter_increment);
 953     // so we do this
 954     ldr(rscratch1, addr);
 955     subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
 956     Label L;
 957     br(Assembler::LO, L);       // skip store if counter underflow
 958     str(rscratch1, addr);
 959     bind(L);
 960   } else {
 961     assert(DataLayout::counter_increment == 1,
 962            "flow-free idiom only works with 1");
 963     // Intel does this
 964     // Increment the register.  Set carry flag.
 965     // addptr(data, DataLayout::counter_increment);
 966     // If the increment causes the counter to overflow, pull back by 1.
 967     // sbbptr(data, (int32_t)0);
 968     // so we do this
 969     ldr(rscratch1, addr);
 970     adds(rscratch1, rscratch1, DataLayout::counter_increment);
 971     Label L;
 972     br(Assembler::CS, L);       // skip store if counter overflow
 973     str(rscratch1, addr);
 974     bind(L);
 975   }
 976 }
 977 
 978 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 979                                                 int flag_byte_constant) {
 980   assert(ProfileInterpreter, "must be profiling interpreter");
 981   int flags_offset = in_bytes(DataLayout::flags_offset());
 982   // Set the flag
 983   ldrb(rscratch1, Address(mdp_in, flags_offset));
 984   orr(rscratch1, rscratch1, flag_byte_constant);
 985   strb(rscratch1, Address(mdp_in, flags_offset));
 986 }
 987 
 988 
 989 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
 990                                                  int offset,
 991                                                  Register value,
 992                                                  Register test_value_out,
 993                                                  Label& not_equal_continue) {
 994   assert(ProfileInterpreter, "must be profiling interpreter");
 995   if (test_value_out == noreg) {
 996     ldr(rscratch1, Address(mdp_in, offset));
 997     cmp(value, rscratch1);
 998   } else {
 999     // Put the test value into a register, so caller can use it:
1000     ldr(test_value_out, Address(mdp_in, offset));
1001     cmp(value, test_value_out);
1002   }
1003   br(Assembler::NE, not_equal_continue);
1004 }
1005 
1006 
1007 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1008                                                      int offset_of_disp) {
1009   assert(ProfileInterpreter, "must be profiling interpreter");
1010   ldr(rscratch1, Address(mdp_in, offset_of_disp));
1011   add(mdp_in, mdp_in, rscratch1, LSL);
1012   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1013 }
1014 
1015 
1016 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1017                                                      Register reg,
1018                                                      int offset_of_disp) {
1019   assert(ProfileInterpreter, "must be profiling interpreter");
1020   lea(rscratch1, Address(mdp_in, offset_of_disp));
1021   ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1022   add(mdp_in, mdp_in, rscratch1, LSL);
1023   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1024 }
1025 
1026 
1027 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1028                                                        int constant) {
1029   assert(ProfileInterpreter, "must be profiling interpreter");
1030   add(mdp_in, mdp_in, (unsigned)constant);
1031   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1032 }
1033 
1034 
1035 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1036   assert(ProfileInterpreter, "must be profiling interpreter");
1037   // save/restore across call_VM
1038   stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1039   call_VM(noreg,
1040           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1041           return_bci);
1042   ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1043 }
1044 
1045 
1046 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1047                                                      Register bumped_count) {
1048   if (ProfileInterpreter) {
1049     Label profile_continue;
1050 
1051     // If no method data exists, go to profile_continue.
1052     // Otherwise, assign to mdp
1053     test_method_data_pointer(mdp, profile_continue);
1054 
1055     // We are taking a branch.  Increment the taken count.
1056     // We inline increment_mdp_data_at to return bumped_count in a register
1057     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1058     Address data(mdp, in_bytes(JumpData::taken_offset()));
1059     ldr(bumped_count, data);
1060     assert(DataLayout::counter_increment == 1,
1061             "flow-free idiom only works with 1");
1062     // Intel does this to catch overflow
1063     // addptr(bumped_count, DataLayout::counter_increment);
1064     // sbbptr(bumped_count, 0);
1065     // so we do this
1066     adds(bumped_count, bumped_count, DataLayout::counter_increment);
1067     Label L;
1068     br(Assembler::CS, L);       // skip store if counter overflow
1069     str(bumped_count, data);
1070     bind(L);
1071     // The method data pointer needs to be updated to reflect the new target.
1072     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1073     bind(profile_continue);
1074   }
1075 }
1076 
1077 
1078 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1079   if (ProfileInterpreter) {
1080     Label profile_continue;
1081 
1082     // If no method data exists, go to profile_continue.
1083     test_method_data_pointer(mdp, profile_continue);
1084 
1085     // We are taking a branch.  Increment the not taken count.
1086     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1087 
1088     // The method data pointer needs to be updated to correspond to
1089     // the next bytecode
1090     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1091     bind(profile_continue);
1092   }
1093 }
1094 
1095 
1096 void InterpreterMacroAssembler::profile_call(Register mdp) {
1097   if (ProfileInterpreter) {
1098     Label profile_continue;
1099 
1100     // If no method data exists, go to profile_continue.
1101     test_method_data_pointer(mdp, profile_continue);
1102 
1103     // We are making a call.  Increment the count.
1104     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1105 
1106     // The method data pointer needs to be updated to reflect the new target.
1107     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1108     bind(profile_continue);
1109   }
1110 }
1111 
1112 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1113   if (ProfileInterpreter) {
1114     Label profile_continue;
1115 
1116     // If no method data exists, go to profile_continue.
1117     test_method_data_pointer(mdp, profile_continue);
1118 
1119     // We are making a call.  Increment the count.
1120     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1121 
1122     // The method data pointer needs to be updated to reflect the new target.
1123     update_mdp_by_constant(mdp,
1124                            in_bytes(VirtualCallData::
1125                                     virtual_call_data_size()));
1126     bind(profile_continue);
1127   }
1128 }
1129 
1130 
1131 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1132                                                      Register mdp,
1133                                                      Register reg2,
1134                                                      bool receiver_can_be_null) {
1135   if (ProfileInterpreter) {
1136     Label profile_continue;
1137 
1138     // If no method data exists, go to profile_continue.
1139     test_method_data_pointer(mdp, profile_continue);
1140 
1141     Label skip_receiver_profile;
1142     if (receiver_can_be_null) {
1143       Label not_null;
1144       // We are making a call.  Increment the count for null receiver.
1145       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1146       b(skip_receiver_profile);
1147       bind(not_null);
1148     }
1149 
1150     // Record the receiver type.
1151     record_klass_in_profile(receiver, mdp, reg2, true);
1152     bind(skip_receiver_profile);
1153 
1154     // The method data pointer needs to be updated to reflect the new target.
1155     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1156     bind(profile_continue);
1157   }
1158 }
1159 
1160 // This routine creates a state machine for updating the multi-row
1161 // type profile at a virtual call site (or other type-sensitive bytecode).
1162 // The machine visits each row (of receiver/count) until the receiver type
1163 // is found, or until it runs out of rows.  At the same time, it remembers
1164 // the location of the first empty row.  (An empty row records null for its
1165 // receiver, and can be allocated for a newly-observed receiver type.)
1166 // Because there are two degrees of freedom in the state, a simple linear
1167 // search will not work; it must be a decision tree.  Hence this helper
1168 // function is recursive, to generate the required tree structured code.
1169 // It's the interpreter, so we are trading off code space for speed.
1170 // See below for example code.
1171 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1172                                         Register receiver, Register mdp,
1173                                         Register reg2, int start_row,
1174                                         Label& done, bool is_virtual_call) {
1175   if (TypeProfileWidth == 0) {
1176     if (is_virtual_call) {
1177       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1178     }
1179 #if INCLUDE_JVMCI
1180     else if (EnableJVMCI) {
1181       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1182     }
1183 #endif // INCLUDE_JVMCI
1184   } else {
1185     int non_profiled_offset = -1;
1186     if (is_virtual_call) {
1187       non_profiled_offset = in_bytes(CounterData::count_offset());
1188     }
1189 #if INCLUDE_JVMCI
1190     else if (EnableJVMCI) {
1191       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1192     }
1193 #endif // INCLUDE_JVMCI
1194 
1195     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1196         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1197   }
1198 }
1199 
1200 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1201                                         Register reg2, int start_row, Label& done, int total_rows,
1202                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1203                                         int non_profiled_offset) {
1204   int last_row = total_rows - 1;
1205   assert(start_row <= last_row, "must be work left to do");
1206   // Test this row for both the item and for null.
1207   // Take any of three different outcomes:
1208   //   1. found item => increment count and goto done
1209   //   2. found null => keep looking for case 1, maybe allocate this cell
1210   //   3. found something else => keep looking for cases 1 and 2
1211   // Case 3 is handled by a recursive call.
1212   for (int row = start_row; row <= last_row; row++) {
1213     Label next_test;
1214     bool test_for_null_also = (row == start_row);
1215 
1216     // See if the item is item[n].
1217     int item_offset = in_bytes(item_offset_fn(row));
1218     test_mdp_data_at(mdp, item_offset, item,
1219                      (test_for_null_also ? reg2 : noreg),
1220                      next_test);
1221     // (Reg2 now contains the item from the CallData.)
1222 
1223     // The item is item[n].  Increment count[n].
1224     int count_offset = in_bytes(item_count_offset_fn(row));
1225     increment_mdp_data_at(mdp, count_offset);
1226     b(done);
1227     bind(next_test);
1228 
1229     if (test_for_null_also) {
1230       Label found_null;
1231       // Failed the equality check on item[n]...  Test for null.
1232       if (start_row == last_row) {
1233         // The only thing left to do is handle the null case.
1234         if (non_profiled_offset >= 0) {
1235           cbz(reg2, found_null);
1236           // Item did not match any saved item and there is no empty row for it.
1237           // Increment total counter to indicate polymorphic case.
1238           increment_mdp_data_at(mdp, non_profiled_offset);
1239           b(done);
1240           bind(found_null);
1241         } else {
1242           cbnz(reg2, done);
1243         }
1244         break;
1245       }
1246       // Since null is rare, make it be the branch-taken case.
1247       cbz(reg2, found_null);
1248 
1249       // Put all the "Case 3" tests here.
1250       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1251         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1252 
1253       // Found a null.  Keep searching for a matching item,
1254       // but remember that this is an empty (unused) slot.
1255       bind(found_null);
1256     }
1257   }
1258 
1259   // In the fall-through case, we found no matching item, but we
1260   // observed the item[start_row] is NULL.
1261 
1262   // Fill in the item field and increment the count.
1263   int item_offset = in_bytes(item_offset_fn(start_row));
1264   set_mdp_data_at(mdp, item_offset, item);
1265   int count_offset = in_bytes(item_count_offset_fn(start_row));
1266   mov(reg2, DataLayout::counter_increment);
1267   set_mdp_data_at(mdp, count_offset, reg2);
1268   if (start_row > 0) {
1269     b(done);
1270   }
1271 }
1272 
1273 // Example state machine code for three profile rows:
1274 //   // main copy of decision tree, rooted at row[1]
1275 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1276 //   if (row[0].rec != NULL) {
1277 //     // inner copy of decision tree, rooted at row[1]
1278 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1279 //     if (row[1].rec != NULL) {
1280 //       // degenerate decision tree, rooted at row[2]
1281 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1282 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1283 //       row[2].init(rec); goto done;
1284 //     } else {
1285 //       // remember row[1] is empty
1286 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1287 //       row[1].init(rec); goto done;
1288 //     }
1289 //   } else {
1290 //     // remember row[0] is empty
1291 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1292 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1293 //     row[0].init(rec); goto done;
1294 //   }
1295 //   done:
1296 
1297 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1298                                                         Register mdp, Register reg2,
1299                                                         bool is_virtual_call) {
1300   assert(ProfileInterpreter, "must be profiling");
1301   Label done;
1302 
1303   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1304 
1305   bind (done);
1306 }
1307 
1308 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1309                                             Register mdp) {
1310   if (ProfileInterpreter) {
1311     Label profile_continue;
1312     uint row;
1313 
1314     // If no method data exists, go to profile_continue.
1315     test_method_data_pointer(mdp, profile_continue);
1316 
1317     // Update the total ret count.
1318     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1319 
1320     for (row = 0; row < RetData::row_limit(); row++) {
1321       Label next_test;
1322 
1323       // See if return_bci is equal to bci[n]:
1324       test_mdp_data_at(mdp,
1325                        in_bytes(RetData::bci_offset(row)),
1326                        return_bci, noreg,
1327                        next_test);
1328 
1329       // return_bci is equal to bci[n].  Increment the count.
1330       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1331 
1332       // The method data pointer needs to be updated to reflect the new target.
1333       update_mdp_by_offset(mdp,
1334                            in_bytes(RetData::bci_displacement_offset(row)));
1335       b(profile_continue);
1336       bind(next_test);
1337     }
1338 
1339     update_mdp_for_ret(return_bci);
1340 
1341     bind(profile_continue);
1342   }
1343 }
1344 
1345 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1346   if (ProfileInterpreter) {
1347     Label profile_continue;
1348 
1349     // If no method data exists, go to profile_continue.
1350     test_method_data_pointer(mdp, profile_continue);
1351 
1352     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1353 
1354     // The method data pointer needs to be updated.
1355     int mdp_delta = in_bytes(BitData::bit_data_size());
1356     if (TypeProfileCasts) {
1357       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1358     }
1359     update_mdp_by_constant(mdp, mdp_delta);
1360 
1361     bind(profile_continue);
1362   }
1363 }
1364 
1365 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1366   if (ProfileInterpreter && TypeProfileCasts) {
1367     Label profile_continue;
1368 
1369     // If no method data exists, go to profile_continue.
1370     test_method_data_pointer(mdp, profile_continue);
1371 
1372     int count_offset = in_bytes(CounterData::count_offset());
1373     // Back up the address, since we have already bumped the mdp.
1374     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1375 
1376     // *Decrement* the counter.  We expect to see zero or small negatives.
1377     increment_mdp_data_at(mdp, count_offset, true);
1378 
1379     bind (profile_continue);
1380   }
1381 }
1382 
1383 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1384   if (ProfileInterpreter) {
1385     Label profile_continue;
1386 
1387     // If no method data exists, go to profile_continue.
1388     test_method_data_pointer(mdp, profile_continue);
1389 
1390     // The method data pointer needs to be updated.
1391     int mdp_delta = in_bytes(BitData::bit_data_size());
1392     if (TypeProfileCasts) {
1393       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1394 
1395       // Record the object type.
1396       record_klass_in_profile(klass, mdp, reg2, false);
1397     }
1398     update_mdp_by_constant(mdp, mdp_delta);
1399 
1400     bind(profile_continue);
1401   }
1402 }
1403 
1404 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1405   if (ProfileInterpreter) {
1406     Label profile_continue;
1407 
1408     // If no method data exists, go to profile_continue.
1409     test_method_data_pointer(mdp, profile_continue);
1410 
1411     // Update the default case count
1412     increment_mdp_data_at(mdp,
1413                           in_bytes(MultiBranchData::default_count_offset()));
1414 
1415     // The method data pointer needs to be updated.
1416     update_mdp_by_offset(mdp,
1417                          in_bytes(MultiBranchData::
1418                                   default_displacement_offset()));
1419 
1420     bind(profile_continue);
1421   }
1422 }
1423 
1424 void InterpreterMacroAssembler::profile_switch_case(Register index,
1425                                                     Register mdp,
1426                                                     Register reg2) {
1427   if (ProfileInterpreter) {
1428     Label profile_continue;
1429 
1430     // If no method data exists, go to profile_continue.
1431     test_method_data_pointer(mdp, profile_continue);
1432 
1433     // Build the base (index * per_case_size_in_bytes()) +
1434     // case_array_offset_in_bytes()
1435     movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1436     movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1437     Assembler::maddw(index, index, reg2, rscratch1);
1438 
1439     // Update the case count
1440     increment_mdp_data_at(mdp,
1441                           index,
1442                           in_bytes(MultiBranchData::relative_count_offset()));
1443 
1444     // The method data pointer needs to be updated.
1445     update_mdp_by_offset(mdp,
1446                          index,
1447                          in_bytes(MultiBranchData::
1448                                   relative_displacement_offset()));
1449 
1450     bind(profile_continue);
1451   }
1452 }
1453 
1454 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1455   if (state == atos) {
1456     MacroAssembler::verify_oop(reg);
1457   }
1458 }
1459 
1460 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1461 
1462 
1463 void InterpreterMacroAssembler::notify_method_entry() {
1464   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1465   // track stack depth.  If it is possible to enter interp_only_mode we add
1466   // the code to check if the event should be sent.
1467   if (JvmtiExport::can_post_interpreter_events()) {
1468     Label L;
1469     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1470     cbzw(r3, L);
1471     call_VM(noreg, CAST_FROM_FN_PTR(address,
1472                                     InterpreterRuntime::post_method_entry));
1473     bind(L);
1474   }
1475 
1476   {
1477     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1478     get_method(c_rarg1);
1479     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1480                  rthread, c_rarg1);
1481   }
1482 
1483   // RedefineClasses() tracing support for obsolete method entry
1484   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1485     get_method(c_rarg1);
1486     call_VM_leaf(
1487       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1488       rthread, c_rarg1);
1489   }
1490 
1491  }
1492 
1493 
1494 void InterpreterMacroAssembler::notify_method_exit(
1495     TosState state, NotifyMethodExitMode mode) {
1496   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1497   // track stack depth.  If it is possible to enter interp_only_mode we add
1498   // the code to check if the event should be sent.
1499   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1500     Label L;
1501     // Note: frame::interpreter_frame_result has a dependency on how the
1502     // method result is saved across the call to post_method_exit. If this
1503     // is changed then the interpreter_frame_result implementation will
1504     // need to be updated too.
1505 
1506     // template interpreter will leave the result on the top of the stack.
1507     push(state);
1508     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1509     cbz(r3, L);
1510     call_VM(noreg,
1511             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1512     bind(L);
1513     pop(state);
1514   }
1515 
1516   {
1517     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1518     push(state);
1519     get_method(c_rarg1);
1520     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1521                  rthread, c_rarg1);
1522     pop(state);
1523   }
1524 }
1525 
1526 
1527 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1528 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1529                                                         int increment, Address mask,
1530                                                         Register scratch, Register scratch2,
1531                                                         bool preloaded, Condition cond,
1532                                                         Label* where) {
1533   if (!preloaded) {
1534     ldrw(scratch, counter_addr);
1535   }
1536   add(scratch, scratch, increment);
1537   strw(scratch, counter_addr);
1538   ldrw(scratch2, mask);
1539   ands(scratch, scratch, scratch2);
1540   br(cond, *where);
1541 }
1542 
1543 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1544                                                   int number_of_arguments) {
1545   // interpreter specific
1546   //
1547   // Note: No need to save/restore rbcp & rlocals pointer since these
1548   //       are callee saved registers and no blocking/ GC can happen
1549   //       in leaf calls.
1550 #ifdef ASSERT
1551   {
1552     Label L;
1553     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1554     cbz(rscratch1, L);
1555     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1556          " last_sp != NULL");
1557     bind(L);
1558   }
1559 #endif /* ASSERT */
1560   // super call
1561   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1562 }
1563 
1564 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1565                                              Register java_thread,
1566                                              Register last_java_sp,
1567                                              address  entry_point,
1568                                              int      number_of_arguments,
1569                                              bool     check_exceptions) {
1570   // interpreter specific
1571   //
1572   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1573   //       really make a difference for these runtime calls, since they are
1574   //       slow anyway. Btw., bcp must be saved/restored since it may change
1575   //       due to GC.
1576   // assert(java_thread == noreg , "not expecting a precomputed java thread");
1577   save_bcp();
1578 #ifdef ASSERT
1579   {
1580     Label L;
1581     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1582     cbz(rscratch1, L);
1583     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1584          " last_sp != NULL");
1585     bind(L);
1586   }
1587 #endif /* ASSERT */
1588   // super call
1589   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1590                                entry_point, number_of_arguments,
1591                      check_exceptions);
1592 // interpreter specific
1593   restore_bcp();
1594   restore_locals();
1595 }
1596 
1597 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1598   assert_different_registers(obj, rscratch1);
1599   Label update, next, none;
1600 
1601   verify_oop(obj);
1602 
1603   cbnz(obj, update);
1604   orptr(mdo_addr, TypeEntries::null_seen);
1605   b(next);
1606 
1607   bind(update);
1608   load_klass(obj, obj);
1609 
1610   ldr(rscratch1, mdo_addr);
1611   eor(obj, obj, rscratch1);
1612   tst(obj, TypeEntries::type_klass_mask);
1613   br(Assembler::EQ, next); // klass seen before, nothing to
1614                            // do. The unknown bit may have been
1615                            // set already but no need to check.
1616 
1617   tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1618   // already unknown. Nothing to do anymore.
1619 
1620   ldr(rscratch1, mdo_addr);
1621   cbz(rscratch1, none);
1622   cmp(rscratch1, (u1)TypeEntries::null_seen);
1623   br(Assembler::EQ, none);
1624   // There is a chance that the checks above (re-reading profiling
1625   // data from memory) fail if another thread has just set the
1626   // profiling to this obj's klass
1627   ldr(rscratch1, mdo_addr);
1628   eor(obj, obj, rscratch1);
1629   tst(obj, TypeEntries::type_klass_mask);
1630   br(Assembler::EQ, next);
1631 
1632   // different than before. Cannot keep accurate profile.
1633   orptr(mdo_addr, TypeEntries::type_unknown);
1634   b(next);
1635 
1636   bind(none);
1637   // first time here. Set profile type.
1638   str(obj, mdo_addr);
1639 
1640   bind(next);
1641 }
1642 
1643 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1644   if (!ProfileInterpreter) {
1645     return;
1646   }
1647 
1648   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1649     Label profile_continue;
1650 
1651     test_method_data_pointer(mdp, profile_continue);
1652 
1653     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1654 
1655     ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1656     cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1657     br(Assembler::NE, profile_continue);
1658 
1659     if (MethodData::profile_arguments()) {
1660       Label done;
1661       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1662 
1663       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1664         if (i > 0 || MethodData::profile_return()) {
1665           // If return value type is profiled we may have no argument to profile
1666           ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1667           sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1668           cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1669           add(rscratch1, mdp, off_to_args);
1670           br(Assembler::LT, done);
1671         }
1672         ldr(tmp, Address(callee, Method::const_offset()));
1673         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1674         // stack offset o (zero based) from the start of the argument
1675         // list, for n arguments translates into offset n - o - 1 from
1676         // the end of the argument list
1677         ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1678         sub(tmp, tmp, rscratch1);
1679         sub(tmp, tmp, 1);
1680         Address arg_addr = argument_address(tmp);
1681         ldr(tmp, arg_addr);
1682 
1683         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1684         profile_obj_type(tmp, mdo_arg_addr);
1685 
1686         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1687         off_to_args += to_add;
1688       }
1689 
1690       if (MethodData::profile_return()) {
1691         ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1692         sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1693       }
1694 
1695       add(rscratch1, mdp, off_to_args);
1696       bind(done);
1697       mov(mdp, rscratch1);
1698 
1699       if (MethodData::profile_return()) {
1700         // We're right after the type profile for the last
1701         // argument. tmp is the number of cells left in the
1702         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1703         // if there's a return to profile.
1704         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1705         add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1706       }
1707       str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1708     } else {
1709       assert(MethodData::profile_return(), "either profile call args or call ret");
1710       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1711     }
1712 
1713     // mdp points right after the end of the
1714     // CallTypeData/VirtualCallTypeData, right after the cells for the
1715     // return value type if there's one
1716 
1717     bind(profile_continue);
1718   }
1719 }
1720 
1721 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1722   assert_different_registers(mdp, ret, tmp, rbcp);
1723   if (ProfileInterpreter && MethodData::profile_return()) {
1724     Label profile_continue, done;
1725 
1726     test_method_data_pointer(mdp, profile_continue);
1727 
1728     if (MethodData::profile_return_jsr292_only()) {
1729       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1730 
1731       // If we don't profile all invoke bytecodes we must make sure
1732       // it's a bytecode we indeed profile. We can't go back to the
1733       // begining of the ProfileData we intend to update to check its
1734       // type because we're right after it and we don't known its
1735       // length
1736       Label do_profile;
1737       ldrb(rscratch1, Address(rbcp, 0));
1738       cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1739       br(Assembler::EQ, do_profile);
1740       cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1741       br(Assembler::EQ, do_profile);
1742       get_method(tmp);
1743       ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1744       subs(zr, rscratch1, vmIntrinsics::_compiledLambdaForm);
1745       br(Assembler::NE, profile_continue);
1746 
1747       bind(do_profile);
1748     }
1749 
1750     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1751     mov(tmp, ret);
1752     profile_obj_type(tmp, mdo_ret_addr);
1753 
1754     bind(profile_continue);
1755   }
1756 }
1757 
1758 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1759   assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1760   if (ProfileInterpreter && MethodData::profile_parameters()) {
1761     Label profile_continue, done;
1762 
1763     test_method_data_pointer(mdp, profile_continue);
1764 
1765     // Load the offset of the area within the MDO used for
1766     // parameters. If it's negative we're not profiling any parameters
1767     ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1768     tbnz(tmp1, 31, profile_continue);  // i.e. sign bit set
1769 
1770     // Compute a pointer to the area for parameters from the offset
1771     // and move the pointer to the slot for the last
1772     // parameters. Collect profiling from last parameter down.
1773     // mdo start + parameters offset + array length - 1
1774     add(mdp, mdp, tmp1);
1775     ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1776     sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1777 
1778     Label loop;
1779     bind(loop);
1780 
1781     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1782     int type_base = in_bytes(ParametersTypeData::type_offset(0));
1783     int per_arg_scale = exact_log2(DataLayout::cell_size);
1784     add(rscratch1, mdp, off_base);
1785     add(rscratch2, mdp, type_base);
1786 
1787     Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1788     Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1789 
1790     // load offset on the stack from the slot for this parameter
1791     ldr(tmp2, arg_off);
1792     neg(tmp2, tmp2);
1793     // read the parameter from the local area
1794     ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1795 
1796     // profile the parameter
1797     profile_obj_type(tmp2, arg_type);
1798 
1799     // go to next parameter
1800     subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1801     br(Assembler::GE, loop);
1802 
1803     bind(profile_continue);
1804   }
1805 }