1 /*
   2  * Copyright (c) 2003, 2021, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, 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 "compiler/compiler_globals.hpp"
  29 #include "gc/shared/barrierSet.hpp"
  30 #include "gc/shared/barrierSetAssembler.hpp"
  31 #include "interp_masm_aarch64.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "logging/log.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/markWord.hpp"
  37 #include "oops/method.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "prims/jvmtiExport.hpp"
  40 #include "prims/jvmtiThreadState.hpp"
  41 #include "runtime/basicLock.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   uint64_t 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, JavaThread::polling_word_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 // Apply stack watermark barrier.
 525 // Unlock the receiver if this is a synchronized method.
 526 // Unlock any Java monitors from syncronized blocks.
 527 // Remove the activation from the stack.
 528 //
 529 // If there are locked Java monitors
 530 //    If throw_monitor_exception
 531 //       throws IllegalMonitorStateException
 532 //    Else if install_monitor_exception
 533 //       installs IllegalMonitorStateException
 534 //    Else
 535 //       no error processing
 536 void InterpreterMacroAssembler::remove_activation(
 537         TosState state,
 538         bool throw_monitor_exception,
 539         bool install_monitor_exception,
 540         bool notify_jvmdi) {
 541   // Note: Registers r3 xmm0 may be in use for the
 542   // result check if synchronized method
 543   Label unlocked, unlock, no_unlock;
 544 
 545   // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
 546   // that would normally not be safe to use. Such bad returns into unsafe territory of
 547   // the stack, will call InterpreterRuntime::at_unwind.
 548   Label slow_path;
 549   Label fast_path;
 550   safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
 551   br(Assembler::AL, fast_path);
 552   bind(slow_path);
 553   push(state);
 554   set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
 555   super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
 556   reset_last_Java_frame(true);
 557   pop(state);
 558   bind(fast_path);
 559 
 560   // get the value of _do_not_unlock_if_synchronized into r3
 561   const Address do_not_unlock_if_synchronized(rthread,
 562     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 563   ldrb(r3, do_not_unlock_if_synchronized);
 564   strb(zr, do_not_unlock_if_synchronized); // reset the flag
 565 
 566  // get method access flags
 567   ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
 568   ldr(r2, Address(r1, Method::access_flags_offset()));
 569   tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
 570 
 571   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 572   // is set.
 573   cbnz(r3, no_unlock);
 574 
 575   // unlock monitor
 576   push(state); // save result
 577 
 578   // BasicObjectLock will be first in list, since this is a
 579   // synchronized method. However, need to check that the object has
 580   // not been unlocked by an explicit monitorexit bytecode.
 581   const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
 582                         wordSize - (int) sizeof(BasicObjectLock));
 583   // We use c_rarg1 so that if we go slow path it will be the correct
 584   // register for unlock_object to pass to VM directly
 585   lea(c_rarg1, monitor); // address of first monitor
 586 
 587   ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 588   cbnz(r0, unlock);
 589 
 590   pop(state);
 591   if (throw_monitor_exception) {
 592     // Entry already unlocked, need to throw exception
 593     call_VM(noreg, CAST_FROM_FN_PTR(address,
 594                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 595     should_not_reach_here();
 596   } else {
 597     // Monitor already unlocked during a stack unroll. If requested,
 598     // install an illegal_monitor_state_exception.  Continue with
 599     // stack unrolling.
 600     if (install_monitor_exception) {
 601       call_VM(noreg, CAST_FROM_FN_PTR(address,
 602                      InterpreterRuntime::new_illegal_monitor_state_exception));
 603     }
 604     b(unlocked);
 605   }
 606 
 607   bind(unlock);
 608   unlock_object(c_rarg1);
 609   pop(state);
 610 
 611   // Check that for block-structured locking (i.e., that all locked
 612   // objects has been unlocked)
 613   bind(unlocked);
 614 
 615   // r0: Might contain return value
 616 
 617   // Check that all monitors are unlocked
 618   {
 619     Label loop, exception, entry, restart;
 620     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 621     const Address monitor_block_top(
 622         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 623     const Address monitor_block_bot(
 624         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
 625 
 626     bind(restart);
 627     // We use c_rarg1 so that if we go slow path it will be the correct
 628     // register for unlock_object to pass to VM directly
 629     ldr(c_rarg1, monitor_block_top); // points to current entry, starting
 630                                      // with top-most entry
 631     lea(r19, monitor_block_bot);  // points to word before bottom of
 632                                   // monitor block
 633     b(entry);
 634 
 635     // Entry already locked, need to throw exception
 636     bind(exception);
 637 
 638     if (throw_monitor_exception) {
 639       // Throw exception
 640       MacroAssembler::call_VM(noreg,
 641                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 642                                    throw_illegal_monitor_state_exception));
 643       should_not_reach_here();
 644     } else {
 645       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 646       // Unlock does not block, so don't have to worry about the frame.
 647       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 648 
 649       push(state);
 650       unlock_object(c_rarg1);
 651       pop(state);
 652 
 653       if (install_monitor_exception) {
 654         call_VM(noreg, CAST_FROM_FN_PTR(address,
 655                                         InterpreterRuntime::
 656                                         new_illegal_monitor_state_exception));
 657       }
 658 
 659       b(restart);
 660     }
 661 
 662     bind(loop);
 663     // check if current entry is used
 664     ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 665     cbnz(rscratch1, exception);
 666 
 667     add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
 668     bind(entry);
 669     cmp(c_rarg1, r19); // check if bottom reached
 670     br(Assembler::NE, loop); // if not at bottom then check this entry
 671   }
 672 
 673   bind(no_unlock);
 674 
 675   // jvmti support
 676   if (notify_jvmdi) {
 677     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 678   } else {
 679     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 680   }
 681 
 682   // remove activation
 683   // get sender esp
 684   ldr(rscratch2,
 685       Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
 686   if (StackReservedPages > 0) {
 687     // testing if reserved zone needs to be re-enabled
 688     Label no_reserved_zone_enabling;
 689 
 690     // look for an overflow into the stack reserved zone, i.e.
 691     // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
 692     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 693     cmp(rscratch2, rscratch1);
 694     br(Assembler::LS, no_reserved_zone_enabling);
 695 
 696     call_VM_leaf(
 697       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
 698     call_VM(noreg, CAST_FROM_FN_PTR(address,
 699                    InterpreterRuntime::throw_delayed_StackOverflowError));
 700     should_not_reach_here();
 701 
 702     bind(no_reserved_zone_enabling);
 703   }
 704 
 705   // restore sender esp
 706   mov(esp, rscratch2);
 707   // remove frame anchor
 708   leave();
 709   // If we're returning to interpreted code we will shortly be
 710   // adjusting SP to allow some space for ESP.  If we're returning to
 711   // compiled code the saved sender SP was saved in sender_sp, so this
 712   // restores it.
 713   andr(sp, esp, -16);
 714 }
 715 
 716 // Lock object
 717 //
 718 // Args:
 719 //      c_rarg1: BasicObjectLock to be used for locking
 720 //
 721 // Kills:
 722 //      r0
 723 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 724 //      rscratch1, rscratch2 (scratch regs)
 725 void InterpreterMacroAssembler::lock_object(Register lock_reg)
 726 {
 727   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 728   if (UseHeavyMonitors) {
 729     call_VM(noreg,
 730             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 731             lock_reg);
 732   } else {
 733     Label done;
 734 
 735     const Register swap_reg = r0;
 736     const Register tmp = c_rarg2;
 737     const Register obj_reg = c_rarg3; // Will contain the oop
 738 
 739     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 740     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 741     const int mark_offset = lock_offset +
 742                             BasicLock::displaced_header_offset_in_bytes();
 743 
 744     Label slow_case;
 745 
 746     // Load object pointer into obj_reg %c_rarg3
 747     ldr(obj_reg, Address(lock_reg, obj_offset));
 748 
 749     if (DiagnoseSyncOnValueBasedClasses != 0) {
 750       load_klass(tmp, obj_reg);
 751       ldrw(tmp, Address(tmp, Klass::access_flags_offset()));
 752       tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
 753       br(Assembler::NE, slow_case);
 754     }
 755 
 756     // Load (object->mark() | 1) into swap_reg
 757     ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 758     orr(swap_reg, rscratch1, 1);
 759 
 760     // Save (object->mark() | 1) into BasicLock's displaced header
 761     str(swap_reg, Address(lock_reg, mark_offset));
 762 
 763     assert(lock_offset == 0,
 764            "displached header must be first word in BasicObjectLock");
 765 
 766     Label fail;
 767     cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 768 
 769     // Fast check for recursive lock.
 770     //
 771     // Can apply the optimization only if this is a stack lock
 772     // allocated in this thread. For efficiency, we can focus on
 773     // recently allocated stack locks (instead of reading the stack
 774     // base and checking whether 'mark' points inside the current
 775     // thread stack):
 776     //  1) (mark & 7) == 0, and
 777     //  2) sp <= mark < mark + os::pagesize()
 778     //
 779     // Warning: sp + os::pagesize can overflow the stack base. We must
 780     // neither apply the optimization for an inflated lock allocated
 781     // just above the thread stack (this is why condition 1 matters)
 782     // nor apply the optimization if the stack lock is inside the stack
 783     // of another thread. The latter is avoided even in case of overflow
 784     // because we have guard pages at the end of all stacks. Hence, if
 785     // we go over the stack base and hit the stack of another thread,
 786     // this should not be in a writeable area that could contain a
 787     // stack lock allocated by that thread. As a consequence, a stack
 788     // lock less than page size away from sp is guaranteed to be
 789     // owned by the current thread.
 790     //
 791     // These 3 tests can be done by evaluating the following
 792     // expression: ((mark - sp) & (7 - os::vm_page_size())),
 793     // assuming both stack pointer and pagesize have their
 794     // least significant 3 bits clear.
 795     // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
 796     // NOTE2: aarch64 does not like to subtract sp from rn so take a
 797     // copy
 798     mov(rscratch1, sp);
 799     sub(swap_reg, swap_reg, rscratch1);
 800     ands(swap_reg, swap_reg, (uint64_t)(7 - os::vm_page_size()));
 801 
 802     // Save the test result, for recursive case, the result is zero
 803     str(swap_reg, Address(lock_reg, mark_offset));
 804     br(Assembler::EQ, done);
 805 
 806     bind(slow_case);
 807 
 808     // Call the runtime routine for slow case
 809     call_VM(noreg,
 810             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 811             lock_reg);
 812 
 813     bind(done);
 814   }
 815 }
 816 
 817 
 818 // Unlocks an object. Used in monitorexit bytecode and
 819 // remove_activation.  Throws an IllegalMonitorException if object is
 820 // not locked by current thread.
 821 //
 822 // Args:
 823 //      c_rarg1: BasicObjectLock for lock
 824 //
 825 // Kills:
 826 //      r0
 827 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 828 //      rscratch1, rscratch2 (scratch regs)
 829 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
 830 {
 831   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 832 
 833   if (UseHeavyMonitors) {
 834     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 835   } else {
 836     Label done;
 837 
 838     const Register swap_reg   = r0;
 839     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 840     const Register obj_reg    = c_rarg3;  // Will contain the oop
 841 
 842     save_bcp(); // Save in case of exception
 843 
 844     // Convert from BasicObjectLock structure to object and BasicLock
 845     // structure Store the BasicLock address into %r0
 846     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 847 
 848     // Load oop into obj_reg(%c_rarg3)
 849     ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 850 
 851     // Free entry
 852     str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 853 
 854     // Load the old header from BasicLock structure
 855     ldr(header_reg, Address(swap_reg,
 856                             BasicLock::displaced_header_offset_in_bytes()));
 857 
 858     // Test for recursion
 859     cbz(header_reg, done);
 860 
 861     // Atomic swap back the old header
 862     cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 863 
 864     // Call the runtime routine for slow case.
 865     str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
 866     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 867 
 868     bind(done);
 869 
 870     restore_bcp();
 871   }
 872 }
 873 
 874 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 875                                                          Label& zero_continue) {
 876   assert(ProfileInterpreter, "must be profiling interpreter");
 877   ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 878   cbz(mdp, zero_continue);
 879 }
 880 
 881 // Set the method data pointer for the current bcp.
 882 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 883   assert(ProfileInterpreter, "must be profiling interpreter");
 884   Label set_mdp;
 885   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 886 
 887   // Test MDO to avoid the call if it is NULL.
 888   ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
 889   cbz(r0, set_mdp);
 890   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
 891   // r0: mdi
 892   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 893   ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
 894   lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
 895   add(r0, r1, r0);
 896   str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 897   bind(set_mdp);
 898   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 899 }
 900 
 901 void InterpreterMacroAssembler::verify_method_data_pointer() {
 902   assert(ProfileInterpreter, "must be profiling interpreter");
 903 #ifdef ASSERT
 904   Label verify_continue;
 905   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 906   stp(r2, r3, Address(pre(sp, -2 * wordSize)));
 907   test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
 908   get_method(r1);
 909 
 910   // If the mdp is valid, it will point to a DataLayout header which is
 911   // consistent with the bcp.  The converse is highly probable also.
 912   ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
 913   ldr(rscratch1, Address(r1, Method::const_offset()));
 914   add(r2, r2, rscratch1, Assembler::LSL);
 915   lea(r2, Address(r2, ConstMethod::codes_offset()));
 916   cmp(r2, rbcp);
 917   br(Assembler::EQ, verify_continue);
 918   // r1: method
 919   // rbcp: bcp // rbcp == 22
 920   // r3: mdp
 921   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 922                r1, rbcp, r3);
 923   bind(verify_continue);
 924   ldp(r2, r3, Address(post(sp, 2 * wordSize)));
 925   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 926 #endif // ASSERT
 927 }
 928 
 929 
 930 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 931                                                 int constant,
 932                                                 Register value) {
 933   assert(ProfileInterpreter, "must be profiling interpreter");
 934   Address data(mdp_in, constant);
 935   str(value, data);
 936 }
 937 
 938 
 939 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 940                                                       int constant,
 941                                                       bool decrement) {
 942   increment_mdp_data_at(mdp_in, noreg, constant, decrement);
 943 }
 944 
 945 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 946                                                       Register reg,
 947                                                       int constant,
 948                                                       bool decrement) {
 949   assert(ProfileInterpreter, "must be profiling interpreter");
 950   // %%% this does 64bit counters at best it is wasting space
 951   // at worst it is a rare bug when counters overflow
 952 
 953   assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
 954 
 955   Address addr1(mdp_in, constant);
 956   Address addr2(rscratch2, reg, Address::lsl(0));
 957   Address &addr = addr1;
 958   if (reg != noreg) {
 959     lea(rscratch2, addr1);
 960     addr = addr2;
 961   }
 962 
 963   if (decrement) {
 964     // Decrement the register.  Set condition codes.
 965     // Intel does this
 966     // addptr(data, (int32_t) -DataLayout::counter_increment);
 967     // If the decrement causes the counter to overflow, stay negative
 968     // Label L;
 969     // jcc(Assembler::negative, L);
 970     // addptr(data, (int32_t) DataLayout::counter_increment);
 971     // so we do this
 972     ldr(rscratch1, addr);
 973     subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
 974     Label L;
 975     br(Assembler::LO, L);       // skip store if counter underflow
 976     str(rscratch1, addr);
 977     bind(L);
 978   } else {
 979     assert(DataLayout::counter_increment == 1,
 980            "flow-free idiom only works with 1");
 981     // Intel does this
 982     // Increment the register.  Set carry flag.
 983     // addptr(data, DataLayout::counter_increment);
 984     // If the increment causes the counter to overflow, pull back by 1.
 985     // sbbptr(data, (int32_t)0);
 986     // so we do this
 987     ldr(rscratch1, addr);
 988     adds(rscratch1, rscratch1, DataLayout::counter_increment);
 989     Label L;
 990     br(Assembler::CS, L);       // skip store if counter overflow
 991     str(rscratch1, addr);
 992     bind(L);
 993   }
 994 }
 995 
 996 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 997                                                 int flag_byte_constant) {
 998   assert(ProfileInterpreter, "must be profiling interpreter");
 999   int flags_offset = in_bytes(DataLayout::flags_offset());
1000   // Set the flag
1001   ldrb(rscratch1, Address(mdp_in, flags_offset));
1002   orr(rscratch1, rscratch1, flag_byte_constant);
1003   strb(rscratch1, Address(mdp_in, flags_offset));
1004 }
1005 
1006 
1007 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1008                                                  int offset,
1009                                                  Register value,
1010                                                  Register test_value_out,
1011                                                  Label& not_equal_continue) {
1012   assert(ProfileInterpreter, "must be profiling interpreter");
1013   if (test_value_out == noreg) {
1014     ldr(rscratch1, Address(mdp_in, offset));
1015     cmp(value, rscratch1);
1016   } else {
1017     // Put the test value into a register, so caller can use it:
1018     ldr(test_value_out, Address(mdp_in, offset));
1019     cmp(value, test_value_out);
1020   }
1021   br(Assembler::NE, not_equal_continue);
1022 }
1023 
1024 
1025 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1026                                                      int offset_of_disp) {
1027   assert(ProfileInterpreter, "must be profiling interpreter");
1028   ldr(rscratch1, Address(mdp_in, offset_of_disp));
1029   add(mdp_in, mdp_in, rscratch1, LSL);
1030   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1031 }
1032 
1033 
1034 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1035                                                      Register reg,
1036                                                      int offset_of_disp) {
1037   assert(ProfileInterpreter, "must be profiling interpreter");
1038   lea(rscratch1, Address(mdp_in, offset_of_disp));
1039   ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1040   add(mdp_in, mdp_in, rscratch1, LSL);
1041   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1042 }
1043 
1044 
1045 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1046                                                        int constant) {
1047   assert(ProfileInterpreter, "must be profiling interpreter");
1048   add(mdp_in, mdp_in, (unsigned)constant);
1049   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1050 }
1051 
1052 
1053 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1054   assert(ProfileInterpreter, "must be profiling interpreter");
1055   // save/restore across call_VM
1056   stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1057   call_VM(noreg,
1058           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1059           return_bci);
1060   ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1061 }
1062 
1063 
1064 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1065                                                      Register bumped_count) {
1066   if (ProfileInterpreter) {
1067     Label profile_continue;
1068 
1069     // If no method data exists, go to profile_continue.
1070     // Otherwise, assign to mdp
1071     test_method_data_pointer(mdp, profile_continue);
1072 
1073     // We are taking a branch.  Increment the taken count.
1074     // We inline increment_mdp_data_at to return bumped_count in a register
1075     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1076     Address data(mdp, in_bytes(JumpData::taken_offset()));
1077     ldr(bumped_count, data);
1078     assert(DataLayout::counter_increment == 1,
1079             "flow-free idiom only works with 1");
1080     // Intel does this to catch overflow
1081     // addptr(bumped_count, DataLayout::counter_increment);
1082     // sbbptr(bumped_count, 0);
1083     // so we do this
1084     adds(bumped_count, bumped_count, DataLayout::counter_increment);
1085     Label L;
1086     br(Assembler::CS, L);       // skip store if counter overflow
1087     str(bumped_count, data);
1088     bind(L);
1089     // The method data pointer needs to be updated to reflect the new target.
1090     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1091     bind(profile_continue);
1092   }
1093 }
1094 
1095 
1096 void InterpreterMacroAssembler::profile_not_taken_branch(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 taking a branch.  Increment the not taken count.
1104     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1105 
1106     // The method data pointer needs to be updated to correspond to
1107     // the next bytecode
1108     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1109     bind(profile_continue);
1110   }
1111 }
1112 
1113 
1114 void InterpreterMacroAssembler::profile_call(Register mdp) {
1115   if (ProfileInterpreter) {
1116     Label profile_continue;
1117 
1118     // If no method data exists, go to profile_continue.
1119     test_method_data_pointer(mdp, profile_continue);
1120 
1121     // We are making a call.  Increment the count.
1122     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1123 
1124     // The method data pointer needs to be updated to reflect the new target.
1125     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1126     bind(profile_continue);
1127   }
1128 }
1129 
1130 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1131   if (ProfileInterpreter) {
1132     Label profile_continue;
1133 
1134     // If no method data exists, go to profile_continue.
1135     test_method_data_pointer(mdp, profile_continue);
1136 
1137     // We are making a call.  Increment the count.
1138     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1139 
1140     // The method data pointer needs to be updated to reflect the new target.
1141     update_mdp_by_constant(mdp,
1142                            in_bytes(VirtualCallData::
1143                                     virtual_call_data_size()));
1144     bind(profile_continue);
1145   }
1146 }
1147 
1148 
1149 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1150                                                      Register mdp,
1151                                                      Register reg2,
1152                                                      bool receiver_can_be_null) {
1153   if (ProfileInterpreter) {
1154     Label profile_continue;
1155 
1156     // If no method data exists, go to profile_continue.
1157     test_method_data_pointer(mdp, profile_continue);
1158 
1159     Label skip_receiver_profile;
1160     if (receiver_can_be_null) {
1161       Label not_null;
1162       // We are making a call.  Increment the count for null receiver.
1163       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1164       b(skip_receiver_profile);
1165       bind(not_null);
1166     }
1167 
1168     // Record the receiver type.
1169     record_klass_in_profile(receiver, mdp, reg2, true);
1170     bind(skip_receiver_profile);
1171 
1172     // The method data pointer needs to be updated to reflect the new target.
1173     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1174     bind(profile_continue);
1175   }
1176 }
1177 
1178 // This routine creates a state machine for updating the multi-row
1179 // type profile at a virtual call site (or other type-sensitive bytecode).
1180 // The machine visits each row (of receiver/count) until the receiver type
1181 // is found, or until it runs out of rows.  At the same time, it remembers
1182 // the location of the first empty row.  (An empty row records null for its
1183 // receiver, and can be allocated for a newly-observed receiver type.)
1184 // Because there are two degrees of freedom in the state, a simple linear
1185 // search will not work; it must be a decision tree.  Hence this helper
1186 // function is recursive, to generate the required tree structured code.
1187 // It's the interpreter, so we are trading off code space for speed.
1188 // See below for example code.
1189 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1190                                         Register receiver, Register mdp,
1191                                         Register reg2, int start_row,
1192                                         Label& done, bool is_virtual_call) {
1193   if (TypeProfileWidth == 0) {
1194     if (is_virtual_call) {
1195       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1196     }
1197 #if INCLUDE_JVMCI
1198     else if (EnableJVMCI) {
1199       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1200     }
1201 #endif // INCLUDE_JVMCI
1202   } else {
1203     int non_profiled_offset = -1;
1204     if (is_virtual_call) {
1205       non_profiled_offset = in_bytes(CounterData::count_offset());
1206     }
1207 #if INCLUDE_JVMCI
1208     else if (EnableJVMCI) {
1209       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1210     }
1211 #endif // INCLUDE_JVMCI
1212 
1213     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1214         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1215   }
1216 }
1217 
1218 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1219                                         Register reg2, int start_row, Label& done, int total_rows,
1220                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1221                                         int non_profiled_offset) {
1222   int last_row = total_rows - 1;
1223   assert(start_row <= last_row, "must be work left to do");
1224   // Test this row for both the item and for null.
1225   // Take any of three different outcomes:
1226   //   1. found item => increment count and goto done
1227   //   2. found null => keep looking for case 1, maybe allocate this cell
1228   //   3. found something else => keep looking for cases 1 and 2
1229   // Case 3 is handled by a recursive call.
1230   for (int row = start_row; row <= last_row; row++) {
1231     Label next_test;
1232     bool test_for_null_also = (row == start_row);
1233 
1234     // See if the item is item[n].
1235     int item_offset = in_bytes(item_offset_fn(row));
1236     test_mdp_data_at(mdp, item_offset, item,
1237                      (test_for_null_also ? reg2 : noreg),
1238                      next_test);
1239     // (Reg2 now contains the item from the CallData.)
1240 
1241     // The item is item[n].  Increment count[n].
1242     int count_offset = in_bytes(item_count_offset_fn(row));
1243     increment_mdp_data_at(mdp, count_offset);
1244     b(done);
1245     bind(next_test);
1246 
1247     if (test_for_null_also) {
1248       Label found_null;
1249       // Failed the equality check on item[n]...  Test for null.
1250       if (start_row == last_row) {
1251         // The only thing left to do is handle the null case.
1252         if (non_profiled_offset >= 0) {
1253           cbz(reg2, found_null);
1254           // Item did not match any saved item and there is no empty row for it.
1255           // Increment total counter to indicate polymorphic case.
1256           increment_mdp_data_at(mdp, non_profiled_offset);
1257           b(done);
1258           bind(found_null);
1259         } else {
1260           cbnz(reg2, done);
1261         }
1262         break;
1263       }
1264       // Since null is rare, make it be the branch-taken case.
1265       cbz(reg2, found_null);
1266 
1267       // Put all the "Case 3" tests here.
1268       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1269         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1270 
1271       // Found a null.  Keep searching for a matching item,
1272       // but remember that this is an empty (unused) slot.
1273       bind(found_null);
1274     }
1275   }
1276 
1277   // In the fall-through case, we found no matching item, but we
1278   // observed the item[start_row] is NULL.
1279 
1280   // Fill in the item field and increment the count.
1281   int item_offset = in_bytes(item_offset_fn(start_row));
1282   set_mdp_data_at(mdp, item_offset, item);
1283   int count_offset = in_bytes(item_count_offset_fn(start_row));
1284   mov(reg2, DataLayout::counter_increment);
1285   set_mdp_data_at(mdp, count_offset, reg2);
1286   if (start_row > 0) {
1287     b(done);
1288   }
1289 }
1290 
1291 // Example state machine code for three profile rows:
1292 //   // main copy of decision tree, rooted at row[1]
1293 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1294 //   if (row[0].rec != NULL) {
1295 //     // inner copy of decision tree, rooted at row[1]
1296 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1297 //     if (row[1].rec != NULL) {
1298 //       // degenerate decision tree, rooted at row[2]
1299 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1300 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1301 //       row[2].init(rec); goto done;
1302 //     } else {
1303 //       // remember row[1] is empty
1304 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1305 //       row[1].init(rec); goto done;
1306 //     }
1307 //   } else {
1308 //     // remember row[0] is empty
1309 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1310 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1311 //     row[0].init(rec); goto done;
1312 //   }
1313 //   done:
1314 
1315 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1316                                                         Register mdp, Register reg2,
1317                                                         bool is_virtual_call) {
1318   assert(ProfileInterpreter, "must be profiling");
1319   Label done;
1320 
1321   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1322 
1323   bind (done);
1324 }
1325 
1326 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1327                                             Register mdp) {
1328   if (ProfileInterpreter) {
1329     Label profile_continue;
1330     uint row;
1331 
1332     // If no method data exists, go to profile_continue.
1333     test_method_data_pointer(mdp, profile_continue);
1334 
1335     // Update the total ret count.
1336     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1337 
1338     for (row = 0; row < RetData::row_limit(); row++) {
1339       Label next_test;
1340 
1341       // See if return_bci is equal to bci[n]:
1342       test_mdp_data_at(mdp,
1343                        in_bytes(RetData::bci_offset(row)),
1344                        return_bci, noreg,
1345                        next_test);
1346 
1347       // return_bci is equal to bci[n].  Increment the count.
1348       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1349 
1350       // The method data pointer needs to be updated to reflect the new target.
1351       update_mdp_by_offset(mdp,
1352                            in_bytes(RetData::bci_displacement_offset(row)));
1353       b(profile_continue);
1354       bind(next_test);
1355     }
1356 
1357     update_mdp_for_ret(return_bci);
1358 
1359     bind(profile_continue);
1360   }
1361 }
1362 
1363 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1364   if (ProfileInterpreter) {
1365     Label profile_continue;
1366 
1367     // If no method data exists, go to profile_continue.
1368     test_method_data_pointer(mdp, profile_continue);
1369 
1370     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1371 
1372     // The method data pointer needs to be updated.
1373     int mdp_delta = in_bytes(BitData::bit_data_size());
1374     if (TypeProfileCasts) {
1375       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1376     }
1377     update_mdp_by_constant(mdp, mdp_delta);
1378 
1379     bind(profile_continue);
1380   }
1381 }
1382 
1383 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1384   if (ProfileInterpreter && TypeProfileCasts) {
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     int count_offset = in_bytes(CounterData::count_offset());
1391     // Back up the address, since we have already bumped the mdp.
1392     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1393 
1394     // *Decrement* the counter.  We expect to see zero or small negatives.
1395     increment_mdp_data_at(mdp, count_offset, true);
1396 
1397     bind (profile_continue);
1398   }
1399 }
1400 
1401 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1402   if (ProfileInterpreter) {
1403     Label profile_continue;
1404 
1405     // If no method data exists, go to profile_continue.
1406     test_method_data_pointer(mdp, profile_continue);
1407 
1408     // The method data pointer needs to be updated.
1409     int mdp_delta = in_bytes(BitData::bit_data_size());
1410     if (TypeProfileCasts) {
1411       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1412 
1413       // Record the object type.
1414       record_klass_in_profile(klass, mdp, reg2, false);
1415     }
1416     update_mdp_by_constant(mdp, mdp_delta);
1417 
1418     bind(profile_continue);
1419   }
1420 }
1421 
1422 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1423   if (ProfileInterpreter) {
1424     Label profile_continue;
1425 
1426     // If no method data exists, go to profile_continue.
1427     test_method_data_pointer(mdp, profile_continue);
1428 
1429     // Update the default case count
1430     increment_mdp_data_at(mdp,
1431                           in_bytes(MultiBranchData::default_count_offset()));
1432 
1433     // The method data pointer needs to be updated.
1434     update_mdp_by_offset(mdp,
1435                          in_bytes(MultiBranchData::
1436                                   default_displacement_offset()));
1437 
1438     bind(profile_continue);
1439   }
1440 }
1441 
1442 void InterpreterMacroAssembler::profile_switch_case(Register index,
1443                                                     Register mdp,
1444                                                     Register reg2) {
1445   if (ProfileInterpreter) {
1446     Label profile_continue;
1447 
1448     // If no method data exists, go to profile_continue.
1449     test_method_data_pointer(mdp, profile_continue);
1450 
1451     // Build the base (index * per_case_size_in_bytes()) +
1452     // case_array_offset_in_bytes()
1453     movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1454     movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1455     Assembler::maddw(index, index, reg2, rscratch1);
1456 
1457     // Update the case count
1458     increment_mdp_data_at(mdp,
1459                           index,
1460                           in_bytes(MultiBranchData::relative_count_offset()));
1461 
1462     // The method data pointer needs to be updated.
1463     update_mdp_by_offset(mdp,
1464                          index,
1465                          in_bytes(MultiBranchData::
1466                                   relative_displacement_offset()));
1467 
1468     bind(profile_continue);
1469   }
1470 }
1471 
1472 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1473   if (state == atos) {
1474     MacroAssembler::verify_oop(reg);
1475   }
1476 }
1477 
1478 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1479 
1480 
1481 void InterpreterMacroAssembler::notify_method_entry() {
1482   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1483   // track stack depth.  If it is possible to enter interp_only_mode we add
1484   // the code to check if the event should be sent.
1485   if (JvmtiExport::can_post_interpreter_events()) {
1486     Label L;
1487     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1488     cbzw(r3, L);
1489     call_VM(noreg, CAST_FROM_FN_PTR(address,
1490                                     InterpreterRuntime::post_method_entry));
1491     bind(L);
1492   }
1493 
1494   {
1495     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1496     get_method(c_rarg1);
1497     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1498                  rthread, c_rarg1);
1499   }
1500 
1501   // RedefineClasses() tracing support for obsolete method entry
1502   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1503     get_method(c_rarg1);
1504     call_VM_leaf(
1505       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1506       rthread, c_rarg1);
1507   }
1508 
1509  }
1510 
1511 
1512 void InterpreterMacroAssembler::notify_method_exit(
1513     TosState state, NotifyMethodExitMode mode) {
1514   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1515   // track stack depth.  If it is possible to enter interp_only_mode we add
1516   // the code to check if the event should be sent.
1517   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1518     Label L;
1519     // Note: frame::interpreter_frame_result has a dependency on how the
1520     // method result is saved across the call to post_method_exit. If this
1521     // is changed then the interpreter_frame_result implementation will
1522     // need to be updated too.
1523 
1524     // template interpreter will leave the result on the top of the stack.
1525     push(state);
1526     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1527     cbz(r3, L);
1528     call_VM(noreg,
1529             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1530     bind(L);
1531     pop(state);
1532   }
1533 
1534   {
1535     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1536     push(state);
1537     get_method(c_rarg1);
1538     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1539                  rthread, c_rarg1);
1540     pop(state);
1541   }
1542 }
1543 
1544 
1545 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1546 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1547                                                         int increment, Address mask,
1548                                                         Register scratch, Register scratch2,
1549                                                         bool preloaded, Condition cond,
1550                                                         Label* where) {
1551   if (!preloaded) {
1552     ldrw(scratch, counter_addr);
1553   }
1554   add(scratch, scratch, increment);
1555   strw(scratch, counter_addr);
1556   ldrw(scratch2, mask);
1557   ands(scratch, scratch, scratch2);
1558   br(cond, *where);
1559 }
1560 
1561 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1562                                                   int number_of_arguments) {
1563   // interpreter specific
1564   //
1565   // Note: No need to save/restore rbcp & rlocals pointer since these
1566   //       are callee saved registers and no blocking/ GC can happen
1567   //       in leaf calls.
1568 #ifdef ASSERT
1569   {
1570     Label L;
1571     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1572     cbz(rscratch1, L);
1573     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1574          " last_sp != NULL");
1575     bind(L);
1576   }
1577 #endif /* ASSERT */
1578   // super call
1579   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1580 }
1581 
1582 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1583                                              Register java_thread,
1584                                              Register last_java_sp,
1585                                              address  entry_point,
1586                                              int      number_of_arguments,
1587                                              bool     check_exceptions) {
1588   // interpreter specific
1589   //
1590   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1591   //       really make a difference for these runtime calls, since they are
1592   //       slow anyway. Btw., bcp must be saved/restored since it may change
1593   //       due to GC.
1594   // assert(java_thread == noreg , "not expecting a precomputed java thread");
1595   save_bcp();
1596 #ifdef ASSERT
1597   {
1598     Label L;
1599     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1600     cbz(rscratch1, L);
1601     stop("InterpreterMacroAssembler::call_VM_base:"
1602          " last_sp != NULL");
1603     bind(L);
1604   }
1605 #endif /* ASSERT */
1606   // super call
1607   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1608                                entry_point, number_of_arguments,
1609                      check_exceptions);
1610 // interpreter specific
1611   restore_bcp();
1612   restore_locals();
1613 }
1614 
1615 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1616   assert_different_registers(obj, rscratch1);
1617   Label update, next, none;
1618 
1619   verify_oop(obj);
1620 
1621   cbnz(obj, update);
1622   orptr(mdo_addr, TypeEntries::null_seen);
1623   b(next);
1624 
1625   bind(update);
1626   load_klass(obj, obj);
1627 
1628   ldr(rscratch1, mdo_addr);
1629   eor(obj, obj, rscratch1);
1630   tst(obj, TypeEntries::type_klass_mask);
1631   br(Assembler::EQ, next); // klass seen before, nothing to
1632                            // do. The unknown bit may have been
1633                            // set already but no need to check.
1634 
1635   tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1636   // already unknown. Nothing to do anymore.
1637 
1638   ldr(rscratch1, mdo_addr);
1639   cbz(rscratch1, none);
1640   cmp(rscratch1, (u1)TypeEntries::null_seen);
1641   br(Assembler::EQ, none);
1642   // There is a chance that the checks above (re-reading profiling
1643   // data from memory) fail if another thread has just set the
1644   // profiling to this obj's klass
1645   ldr(rscratch1, mdo_addr);
1646   eor(obj, obj, rscratch1);
1647   tst(obj, TypeEntries::type_klass_mask);
1648   br(Assembler::EQ, next);
1649 
1650   // different than before. Cannot keep accurate profile.
1651   orptr(mdo_addr, TypeEntries::type_unknown);
1652   b(next);
1653 
1654   bind(none);
1655   // first time here. Set profile type.
1656   str(obj, mdo_addr);
1657 
1658   bind(next);
1659 }
1660 
1661 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1662   if (!ProfileInterpreter) {
1663     return;
1664   }
1665 
1666   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1667     Label profile_continue;
1668 
1669     test_method_data_pointer(mdp, profile_continue);
1670 
1671     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1672 
1673     ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1674     cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1675     br(Assembler::NE, profile_continue);
1676 
1677     if (MethodData::profile_arguments()) {
1678       Label done;
1679       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1680 
1681       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1682         if (i > 0 || MethodData::profile_return()) {
1683           // If return value type is profiled we may have no argument to profile
1684           ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1685           sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1686           cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1687           add(rscratch1, mdp, off_to_args);
1688           br(Assembler::LT, done);
1689         }
1690         ldr(tmp, Address(callee, Method::const_offset()));
1691         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1692         // stack offset o (zero based) from the start of the argument
1693         // list, for n arguments translates into offset n - o - 1 from
1694         // the end of the argument list
1695         ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1696         sub(tmp, tmp, rscratch1);
1697         sub(tmp, tmp, 1);
1698         Address arg_addr = argument_address(tmp);
1699         ldr(tmp, arg_addr);
1700 
1701         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1702         profile_obj_type(tmp, mdo_arg_addr);
1703 
1704         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1705         off_to_args += to_add;
1706       }
1707 
1708       if (MethodData::profile_return()) {
1709         ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1710         sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1711       }
1712 
1713       add(rscratch1, mdp, off_to_args);
1714       bind(done);
1715       mov(mdp, rscratch1);
1716 
1717       if (MethodData::profile_return()) {
1718         // We're right after the type profile for the last
1719         // argument. tmp is the number of cells left in the
1720         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1721         // if there's a return to profile.
1722         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1723         add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1724       }
1725       str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1726     } else {
1727       assert(MethodData::profile_return(), "either profile call args or call ret");
1728       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1729     }
1730 
1731     // mdp points right after the end of the
1732     // CallTypeData/VirtualCallTypeData, right after the cells for the
1733     // return value type if there's one
1734 
1735     bind(profile_continue);
1736   }
1737 }
1738 
1739 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1740   assert_different_registers(mdp, ret, tmp, rbcp);
1741   if (ProfileInterpreter && MethodData::profile_return()) {
1742     Label profile_continue, done;
1743 
1744     test_method_data_pointer(mdp, profile_continue);
1745 
1746     if (MethodData::profile_return_jsr292_only()) {
1747       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1748 
1749       // If we don't profile all invoke bytecodes we must make sure
1750       // it's a bytecode we indeed profile. We can't go back to the
1751       // begining of the ProfileData we intend to update to check its
1752       // type because we're right after it and we don't known its
1753       // length
1754       Label do_profile;
1755       ldrb(rscratch1, Address(rbcp, 0));
1756       cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1757       br(Assembler::EQ, do_profile);
1758       cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1759       br(Assembler::EQ, do_profile);
1760       get_method(tmp);
1761       ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1762       subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1763       br(Assembler::NE, profile_continue);
1764 
1765       bind(do_profile);
1766     }
1767 
1768     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1769     mov(tmp, ret);
1770     profile_obj_type(tmp, mdo_ret_addr);
1771 
1772     bind(profile_continue);
1773   }
1774 }
1775 
1776 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1777   assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1778   if (ProfileInterpreter && MethodData::profile_parameters()) {
1779     Label profile_continue, done;
1780 
1781     test_method_data_pointer(mdp, profile_continue);
1782 
1783     // Load the offset of the area within the MDO used for
1784     // parameters. If it's negative we're not profiling any parameters
1785     ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1786     tbnz(tmp1, 31, profile_continue);  // i.e. sign bit set
1787 
1788     // Compute a pointer to the area for parameters from the offset
1789     // and move the pointer to the slot for the last
1790     // parameters. Collect profiling from last parameter down.
1791     // mdo start + parameters offset + array length - 1
1792     add(mdp, mdp, tmp1);
1793     ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1794     sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1795 
1796     Label loop;
1797     bind(loop);
1798 
1799     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1800     int type_base = in_bytes(ParametersTypeData::type_offset(0));
1801     int per_arg_scale = exact_log2(DataLayout::cell_size);
1802     add(rscratch1, mdp, off_base);
1803     add(rscratch2, mdp, type_base);
1804 
1805     Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1806     Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1807 
1808     // load offset on the stack from the slot for this parameter
1809     ldr(tmp2, arg_off);
1810     neg(tmp2, tmp2);
1811     // read the parameter from the local area
1812     ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1813 
1814     // profile the parameter
1815     profile_obj_type(tmp2, arg_type);
1816 
1817     // go to next parameter
1818     subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1819     br(Assembler::GE, loop);
1820 
1821     bind(profile_continue);
1822   }
1823 }