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   dec_held_monitor_count(rthread);
 610   pop(state);
 611 
 612   // Check that for block-structured locking (i.e., that all locked
 613   // objects has been unlocked)
 614   bind(unlocked);
 615 
 616   // r0: Might contain return value
 617 
 618   // Check that all monitors are unlocked
 619   {
 620     Label loop, exception, entry, restart;
 621     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 622     const Address monitor_block_top(
 623         rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
 624     const Address monitor_block_bot(
 625         rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
 626 
 627     bind(restart);
 628     // We use c_rarg1 so that if we go slow path it will be the correct
 629     // register for unlock_object to pass to VM directly
 630     ldr(c_rarg1, monitor_block_top); // points to current entry, starting
 631                                      // with top-most entry
 632     lea(r19, monitor_block_bot);  // points to word before bottom of
 633                                   // monitor block
 634     b(entry);
 635 
 636     // Entry already locked, need to throw exception
 637     bind(exception);
 638 
 639     if (throw_monitor_exception) {
 640       // Throw exception
 641       MacroAssembler::call_VM(noreg,
 642                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 643                                    throw_illegal_monitor_state_exception));
 644       should_not_reach_here();
 645     } else {
 646       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 647       // Unlock does not block, so don't have to worry about the frame.
 648       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 649 
 650       push(state);
 651       unlock_object(c_rarg1);
 652       dec_held_monitor_count(rthread);
 653       pop(state);
 654 
 655       if (install_monitor_exception) {
 656         call_VM(noreg, CAST_FROM_FN_PTR(address,
 657                                         InterpreterRuntime::
 658                                         new_illegal_monitor_state_exception));
 659       }
 660 
 661       b(restart);
 662     }
 663 
 664     bind(loop);
 665     // check if current entry is used
 666     ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 667     cbnz(rscratch1, exception);
 668 
 669     add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
 670     bind(entry);
 671     cmp(c_rarg1, r19); // check if bottom reached
 672     br(Assembler::NE, loop); // if not at bottom then check this entry
 673   }
 674 
 675   bind(no_unlock);
 676 
 677   // jvmti support
 678   if (notify_jvmdi) {
 679     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 680   } else {
 681     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 682   }
 683 
 684   // remove activation
 685   // get sender esp
 686   ldr(rscratch2,
 687       Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
 688   if (StackReservedPages > 0) {
 689     // testing if reserved zone needs to be re-enabled
 690     Label no_reserved_zone_enabling;
 691 
 692     // look for an overflow into the stack reserved zone, i.e.
 693     // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
 694     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 695     cmp(rscratch2, rscratch1);
 696     br(Assembler::LS, no_reserved_zone_enabling);
 697 
 698     call_VM_leaf(
 699       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
 700     call_VM(noreg, CAST_FROM_FN_PTR(address,
 701                    InterpreterRuntime::throw_delayed_StackOverflowError));
 702     should_not_reach_here();
 703 
 704     bind(no_reserved_zone_enabling);
 705   }
 706 
 707   // restore sender esp
 708   mov(esp, rscratch2);
 709   // remove frame anchor
 710   leave();
 711   // If we're returning to interpreted code we will shortly be
 712   // adjusting SP to allow some space for ESP.  If we're returning to
 713   // compiled code the saved sender SP was saved in sender_sp, so this
 714   // restores it.
 715   andr(sp, esp, -16);
 716 }
 717 
 718 // Lock object
 719 //
 720 // Args:
 721 //      c_rarg1: BasicObjectLock to be used for locking
 722 //
 723 // Kills:
 724 //      r0
 725 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 726 //      rscratch1, rscratch2 (scratch regs)
 727 void InterpreterMacroAssembler::lock_object(Register lock_reg)
 728 {
 729   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 730   if (UseHeavyMonitors) {
 731     call_VM(noreg,
 732             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 733             lock_reg);
 734   } else {
 735     Label done;
 736 
 737     const Register swap_reg = r0;
 738     const Register tmp = c_rarg2;
 739     const Register obj_reg = c_rarg3; // Will contain the oop
 740 
 741     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 742     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 743     const int mark_offset = lock_offset +
 744                             BasicLock::displaced_header_offset_in_bytes();
 745 
 746     Label slow_case;
 747 
 748     // Load object pointer into obj_reg %c_rarg3
 749     ldr(obj_reg, Address(lock_reg, obj_offset));
 750 
 751     if (DiagnoseSyncOnValueBasedClasses != 0) {
 752       load_klass(tmp, obj_reg);
 753       ldrw(tmp, Address(tmp, Klass::access_flags_offset()));
 754       tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
 755       br(Assembler::NE, slow_case);
 756     }
 757 
 758     // Load (object->mark() | 1) into swap_reg
 759     ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 760     orr(swap_reg, rscratch1, 1);
 761 
 762     // Save (object->mark() | 1) into BasicLock's displaced header
 763     str(swap_reg, Address(lock_reg, mark_offset));
 764 
 765     assert(lock_offset == 0,
 766            "displached header must be first word in BasicObjectLock");
 767 
 768     Label fail;
 769     cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 770 
 771     // Fast check for recursive lock.
 772     //
 773     // Can apply the optimization only if this is a stack lock
 774     // allocated in this thread. For efficiency, we can focus on
 775     // recently allocated stack locks (instead of reading the stack
 776     // base and checking whether 'mark' points inside the current
 777     // thread stack):
 778     //  1) (mark & 7) == 0, and
 779     //  2) sp <= mark < mark + os::pagesize()
 780     //
 781     // Warning: sp + os::pagesize can overflow the stack base. We must
 782     // neither apply the optimization for an inflated lock allocated
 783     // just above the thread stack (this is why condition 1 matters)
 784     // nor apply the optimization if the stack lock is inside the stack
 785     // of another thread. The latter is avoided even in case of overflow
 786     // because we have guard pages at the end of all stacks. Hence, if
 787     // we go over the stack base and hit the stack of another thread,
 788     // this should not be in a writeable area that could contain a
 789     // stack lock allocated by that thread. As a consequence, a stack
 790     // lock less than page size away from sp is guaranteed to be
 791     // owned by the current thread.
 792     //
 793     // These 3 tests can be done by evaluating the following
 794     // expression: ((mark - sp) & (7 - os::vm_page_size())),
 795     // assuming both stack pointer and pagesize have their
 796     // least significant 3 bits clear.
 797     // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
 798     // NOTE2: aarch64 does not like to subtract sp from rn so take a
 799     // copy
 800     mov(rscratch1, sp);
 801     sub(swap_reg, swap_reg, rscratch1);
 802     ands(swap_reg, swap_reg, (uint64_t)(7 - os::vm_page_size()));
 803 
 804     // Save the test result, for recursive case, the result is zero
 805     str(swap_reg, Address(lock_reg, mark_offset));
 806     br(Assembler::EQ, done);
 807 
 808     bind(slow_case);
 809 
 810     // Call the runtime routine for slow case
 811     call_VM(noreg,
 812             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 813             lock_reg);
 814 
 815     bind(done);
 816   }
 817 }
 818 
 819 
 820 // Unlocks an object. Used in monitorexit bytecode and
 821 // remove_activation.  Throws an IllegalMonitorException if object is
 822 // not locked by current thread.
 823 //
 824 // Args:
 825 //      c_rarg1: BasicObjectLock for lock
 826 //
 827 // Kills:
 828 //      r0
 829 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 830 //      rscratch1, rscratch2 (scratch regs)
 831 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
 832 {
 833   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 834 
 835   if (UseHeavyMonitors) {
 836     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 837   } else {
 838     Label done;
 839 
 840     const Register swap_reg   = r0;
 841     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 842     const Register obj_reg    = c_rarg3;  // Will contain the oop
 843 
 844     save_bcp(); // Save in case of exception
 845 
 846     // Convert from BasicObjectLock structure to object and BasicLock
 847     // structure Store the BasicLock address into %r0
 848     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 849 
 850     // Load oop into obj_reg(%c_rarg3)
 851     ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 852 
 853     // Free entry
 854     str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 855 
 856     // Load the old header from BasicLock structure
 857     ldr(header_reg, Address(swap_reg,
 858                             BasicLock::displaced_header_offset_in_bytes()));
 859 
 860     // Test for recursion
 861     cbz(header_reg, done);
 862 
 863     // Atomic swap back the old header
 864     cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 865 
 866     // Call the runtime routine for slow case.
 867     str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
 868     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 869 
 870     bind(done);
 871 
 872     restore_bcp();
 873   }
 874 }
 875 
 876 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 877                                                          Label& zero_continue) {
 878   assert(ProfileInterpreter, "must be profiling interpreter");
 879   ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 880   cbz(mdp, zero_continue);
 881 }
 882 
 883 // Set the method data pointer for the current bcp.
 884 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 885   assert(ProfileInterpreter, "must be profiling interpreter");
 886   Label set_mdp;
 887   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 888 
 889   // Test MDO to avoid the call if it is NULL.
 890   ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
 891   cbz(r0, set_mdp);
 892   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
 893   // r0: mdi
 894   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 895   ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
 896   lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
 897   add(r0, r1, r0);
 898   str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 899   bind(set_mdp);
 900   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 901 }
 902 
 903 void InterpreterMacroAssembler::verify_method_data_pointer() {
 904   assert(ProfileInterpreter, "must be profiling interpreter");
 905 #ifdef ASSERT
 906   Label verify_continue;
 907   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 908   stp(r2, r3, Address(pre(sp, -2 * wordSize)));
 909   test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
 910   get_method(r1);
 911 
 912   // If the mdp is valid, it will point to a DataLayout header which is
 913   // consistent with the bcp.  The converse is highly probable also.
 914   ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
 915   ldr(rscratch1, Address(r1, Method::const_offset()));
 916   add(r2, r2, rscratch1, Assembler::LSL);
 917   lea(r2, Address(r2, ConstMethod::codes_offset()));
 918   cmp(r2, rbcp);
 919   br(Assembler::EQ, verify_continue);
 920   // r1: method
 921   // rbcp: bcp // rbcp == 22
 922   // r3: mdp
 923   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 924                r1, rbcp, r3);
 925   bind(verify_continue);
 926   ldp(r2, r3, Address(post(sp, 2 * wordSize)));
 927   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 928 #endif // ASSERT
 929 }
 930 
 931 
 932 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 933                                                 int constant,
 934                                                 Register value) {
 935   assert(ProfileInterpreter, "must be profiling interpreter");
 936   Address data(mdp_in, constant);
 937   str(value, data);
 938 }
 939 
 940 
 941 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 942                                                       int constant,
 943                                                       bool decrement) {
 944   increment_mdp_data_at(mdp_in, noreg, constant, decrement);
 945 }
 946 
 947 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
 948                                                       Register reg,
 949                                                       int constant,
 950                                                       bool decrement) {
 951   assert(ProfileInterpreter, "must be profiling interpreter");
 952   // %%% this does 64bit counters at best it is wasting space
 953   // at worst it is a rare bug when counters overflow
 954 
 955   assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
 956 
 957   Address addr1(mdp_in, constant);
 958   Address addr2(rscratch2, reg, Address::lsl(0));
 959   Address &addr = addr1;
 960   if (reg != noreg) {
 961     lea(rscratch2, addr1);
 962     addr = addr2;
 963   }
 964 
 965   if (decrement) {
 966     // Decrement the register.  Set condition codes.
 967     // Intel does this
 968     // addptr(data, (int32_t) -DataLayout::counter_increment);
 969     // If the decrement causes the counter to overflow, stay negative
 970     // Label L;
 971     // jcc(Assembler::negative, L);
 972     // addptr(data, (int32_t) DataLayout::counter_increment);
 973     // so we do this
 974     ldr(rscratch1, addr);
 975     subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
 976     Label L;
 977     br(Assembler::LO, L);       // skip store if counter underflow
 978     str(rscratch1, addr);
 979     bind(L);
 980   } else {
 981     assert(DataLayout::counter_increment == 1,
 982            "flow-free idiom only works with 1");
 983     // Intel does this
 984     // Increment the register.  Set carry flag.
 985     // addptr(data, DataLayout::counter_increment);
 986     // If the increment causes the counter to overflow, pull back by 1.
 987     // sbbptr(data, (int32_t)0);
 988     // so we do this
 989     ldr(rscratch1, addr);
 990     adds(rscratch1, rscratch1, DataLayout::counter_increment);
 991     Label L;
 992     br(Assembler::CS, L);       // skip store if counter overflow
 993     str(rscratch1, addr);
 994     bind(L);
 995   }
 996 }
 997 
 998 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
 999                                                 int flag_byte_constant) {
1000   assert(ProfileInterpreter, "must be profiling interpreter");
1001   int flags_offset = in_bytes(DataLayout::flags_offset());
1002   // Set the flag
1003   ldrb(rscratch1, Address(mdp_in, flags_offset));
1004   orr(rscratch1, rscratch1, flag_byte_constant);
1005   strb(rscratch1, Address(mdp_in, flags_offset));
1006 }
1007 
1008 
1009 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1010                                                  int offset,
1011                                                  Register value,
1012                                                  Register test_value_out,
1013                                                  Label& not_equal_continue) {
1014   assert(ProfileInterpreter, "must be profiling interpreter");
1015   if (test_value_out == noreg) {
1016     ldr(rscratch1, Address(mdp_in, offset));
1017     cmp(value, rscratch1);
1018   } else {
1019     // Put the test value into a register, so caller can use it:
1020     ldr(test_value_out, Address(mdp_in, offset));
1021     cmp(value, test_value_out);
1022   }
1023   br(Assembler::NE, not_equal_continue);
1024 }
1025 
1026 
1027 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1028                                                      int offset_of_disp) {
1029   assert(ProfileInterpreter, "must be profiling interpreter");
1030   ldr(rscratch1, Address(mdp_in, offset_of_disp));
1031   add(mdp_in, mdp_in, rscratch1, LSL);
1032   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1033 }
1034 
1035 
1036 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1037                                                      Register reg,
1038                                                      int offset_of_disp) {
1039   assert(ProfileInterpreter, "must be profiling interpreter");
1040   lea(rscratch1, Address(mdp_in, offset_of_disp));
1041   ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1042   add(mdp_in, mdp_in, rscratch1, LSL);
1043   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1044 }
1045 
1046 
1047 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1048                                                        int constant) {
1049   assert(ProfileInterpreter, "must be profiling interpreter");
1050   add(mdp_in, mdp_in, (unsigned)constant);
1051   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1052 }
1053 
1054 
1055 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1056   assert(ProfileInterpreter, "must be profiling interpreter");
1057   // save/restore across call_VM
1058   stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1059   call_VM(noreg,
1060           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1061           return_bci);
1062   ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1063 }
1064 
1065 
1066 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1067                                                      Register bumped_count) {
1068   if (ProfileInterpreter) {
1069     Label profile_continue;
1070 
1071     // If no method data exists, go to profile_continue.
1072     // Otherwise, assign to mdp
1073     test_method_data_pointer(mdp, profile_continue);
1074 
1075     // We are taking a branch.  Increment the taken count.
1076     // We inline increment_mdp_data_at to return bumped_count in a register
1077     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1078     Address data(mdp, in_bytes(JumpData::taken_offset()));
1079     ldr(bumped_count, data);
1080     assert(DataLayout::counter_increment == 1,
1081             "flow-free idiom only works with 1");
1082     // Intel does this to catch overflow
1083     // addptr(bumped_count, DataLayout::counter_increment);
1084     // sbbptr(bumped_count, 0);
1085     // so we do this
1086     adds(bumped_count, bumped_count, DataLayout::counter_increment);
1087     Label L;
1088     br(Assembler::CS, L);       // skip store if counter overflow
1089     str(bumped_count, data);
1090     bind(L);
1091     // The method data pointer needs to be updated to reflect the new target.
1092     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1093     bind(profile_continue);
1094   }
1095 }
1096 
1097 
1098 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1099   if (ProfileInterpreter) {
1100     Label profile_continue;
1101 
1102     // If no method data exists, go to profile_continue.
1103     test_method_data_pointer(mdp, profile_continue);
1104 
1105     // We are taking a branch.  Increment the not taken count.
1106     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1107 
1108     // The method data pointer needs to be updated to correspond to
1109     // the next bytecode
1110     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1111     bind(profile_continue);
1112   }
1113 }
1114 
1115 
1116 void InterpreterMacroAssembler::profile_call(Register mdp) {
1117   if (ProfileInterpreter) {
1118     Label profile_continue;
1119 
1120     // If no method data exists, go to profile_continue.
1121     test_method_data_pointer(mdp, profile_continue);
1122 
1123     // We are making a call.  Increment the count.
1124     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1125 
1126     // The method data pointer needs to be updated to reflect the new target.
1127     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1128     bind(profile_continue);
1129   }
1130 }
1131 
1132 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1133   if (ProfileInterpreter) {
1134     Label profile_continue;
1135 
1136     // If no method data exists, go to profile_continue.
1137     test_method_data_pointer(mdp, profile_continue);
1138 
1139     // We are making a call.  Increment the count.
1140     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1141 
1142     // The method data pointer needs to be updated to reflect the new target.
1143     update_mdp_by_constant(mdp,
1144                            in_bytes(VirtualCallData::
1145                                     virtual_call_data_size()));
1146     bind(profile_continue);
1147   }
1148 }
1149 
1150 
1151 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1152                                                      Register mdp,
1153                                                      Register reg2,
1154                                                      bool receiver_can_be_null) {
1155   if (ProfileInterpreter) {
1156     Label profile_continue;
1157 
1158     // If no method data exists, go to profile_continue.
1159     test_method_data_pointer(mdp, profile_continue);
1160 
1161     Label skip_receiver_profile;
1162     if (receiver_can_be_null) {
1163       Label not_null;
1164       // We are making a call.  Increment the count for null receiver.
1165       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1166       b(skip_receiver_profile);
1167       bind(not_null);
1168     }
1169 
1170     // Record the receiver type.
1171     record_klass_in_profile(receiver, mdp, reg2, true);
1172     bind(skip_receiver_profile);
1173 
1174     // The method data pointer needs to be updated to reflect the new target.
1175     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1176     bind(profile_continue);
1177   }
1178 }
1179 
1180 // This routine creates a state machine for updating the multi-row
1181 // type profile at a virtual call site (or other type-sensitive bytecode).
1182 // The machine visits each row (of receiver/count) until the receiver type
1183 // is found, or until it runs out of rows.  At the same time, it remembers
1184 // the location of the first empty row.  (An empty row records null for its
1185 // receiver, and can be allocated for a newly-observed receiver type.)
1186 // Because there are two degrees of freedom in the state, a simple linear
1187 // search will not work; it must be a decision tree.  Hence this helper
1188 // function is recursive, to generate the required tree structured code.
1189 // It's the interpreter, so we are trading off code space for speed.
1190 // See below for example code.
1191 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1192                                         Register receiver, Register mdp,
1193                                         Register reg2, int start_row,
1194                                         Label& done, bool is_virtual_call) {
1195   if (TypeProfileWidth == 0) {
1196     if (is_virtual_call) {
1197       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1198     }
1199 #if INCLUDE_JVMCI
1200     else if (EnableJVMCI) {
1201       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1202     }
1203 #endif // INCLUDE_JVMCI
1204   } else {
1205     int non_profiled_offset = -1;
1206     if (is_virtual_call) {
1207       non_profiled_offset = in_bytes(CounterData::count_offset());
1208     }
1209 #if INCLUDE_JVMCI
1210     else if (EnableJVMCI) {
1211       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1212     }
1213 #endif // INCLUDE_JVMCI
1214 
1215     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1216         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1217   }
1218 }
1219 
1220 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1221                                         Register reg2, int start_row, Label& done, int total_rows,
1222                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1223                                         int non_profiled_offset) {
1224   int last_row = total_rows - 1;
1225   assert(start_row <= last_row, "must be work left to do");
1226   // Test this row for both the item and for null.
1227   // Take any of three different outcomes:
1228   //   1. found item => increment count and goto done
1229   //   2. found null => keep looking for case 1, maybe allocate this cell
1230   //   3. found something else => keep looking for cases 1 and 2
1231   // Case 3 is handled by a recursive call.
1232   for (int row = start_row; row <= last_row; row++) {
1233     Label next_test;
1234     bool test_for_null_also = (row == start_row);
1235 
1236     // See if the item is item[n].
1237     int item_offset = in_bytes(item_offset_fn(row));
1238     test_mdp_data_at(mdp, item_offset, item,
1239                      (test_for_null_also ? reg2 : noreg),
1240                      next_test);
1241     // (Reg2 now contains the item from the CallData.)
1242 
1243     // The item is item[n].  Increment count[n].
1244     int count_offset = in_bytes(item_count_offset_fn(row));
1245     increment_mdp_data_at(mdp, count_offset);
1246     b(done);
1247     bind(next_test);
1248 
1249     if (test_for_null_also) {
1250       Label found_null;
1251       // Failed the equality check on item[n]...  Test for null.
1252       if (start_row == last_row) {
1253         // The only thing left to do is handle the null case.
1254         if (non_profiled_offset >= 0) {
1255           cbz(reg2, found_null);
1256           // Item did not match any saved item and there is no empty row for it.
1257           // Increment total counter to indicate polymorphic case.
1258           increment_mdp_data_at(mdp, non_profiled_offset);
1259           b(done);
1260           bind(found_null);
1261         } else {
1262           cbnz(reg2, done);
1263         }
1264         break;
1265       }
1266       // Since null is rare, make it be the branch-taken case.
1267       cbz(reg2, found_null);
1268 
1269       // Put all the "Case 3" tests here.
1270       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1271         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1272 
1273       // Found a null.  Keep searching for a matching item,
1274       // but remember that this is an empty (unused) slot.
1275       bind(found_null);
1276     }
1277   }
1278 
1279   // In the fall-through case, we found no matching item, but we
1280   // observed the item[start_row] is NULL.
1281 
1282   // Fill in the item field and increment the count.
1283   int item_offset = in_bytes(item_offset_fn(start_row));
1284   set_mdp_data_at(mdp, item_offset, item);
1285   int count_offset = in_bytes(item_count_offset_fn(start_row));
1286   mov(reg2, DataLayout::counter_increment);
1287   set_mdp_data_at(mdp, count_offset, reg2);
1288   if (start_row > 0) {
1289     b(done);
1290   }
1291 }
1292 
1293 // Example state machine code for three profile rows:
1294 //   // main copy of decision tree, rooted at row[1]
1295 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1296 //   if (row[0].rec != NULL) {
1297 //     // inner copy of decision tree, rooted at row[1]
1298 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1299 //     if (row[1].rec != NULL) {
1300 //       // degenerate decision tree, rooted at row[2]
1301 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1302 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1303 //       row[2].init(rec); goto done;
1304 //     } else {
1305 //       // remember row[1] is empty
1306 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1307 //       row[1].init(rec); goto done;
1308 //     }
1309 //   } else {
1310 //     // remember row[0] is empty
1311 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1312 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1313 //     row[0].init(rec); goto done;
1314 //   }
1315 //   done:
1316 
1317 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1318                                                         Register mdp, Register reg2,
1319                                                         bool is_virtual_call) {
1320   assert(ProfileInterpreter, "must be profiling");
1321   Label done;
1322 
1323   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1324 
1325   bind (done);
1326 }
1327 
1328 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1329                                             Register mdp) {
1330   if (ProfileInterpreter) {
1331     Label profile_continue;
1332     uint row;
1333 
1334     // If no method data exists, go to profile_continue.
1335     test_method_data_pointer(mdp, profile_continue);
1336 
1337     // Update the total ret count.
1338     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1339 
1340     for (row = 0; row < RetData::row_limit(); row++) {
1341       Label next_test;
1342 
1343       // See if return_bci is equal to bci[n]:
1344       test_mdp_data_at(mdp,
1345                        in_bytes(RetData::bci_offset(row)),
1346                        return_bci, noreg,
1347                        next_test);
1348 
1349       // return_bci is equal to bci[n].  Increment the count.
1350       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1351 
1352       // The method data pointer needs to be updated to reflect the new target.
1353       update_mdp_by_offset(mdp,
1354                            in_bytes(RetData::bci_displacement_offset(row)));
1355       b(profile_continue);
1356       bind(next_test);
1357     }
1358 
1359     update_mdp_for_ret(return_bci);
1360 
1361     bind(profile_continue);
1362   }
1363 }
1364 
1365 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1366   if (ProfileInterpreter) {
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     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1373 
1374     // The method data pointer needs to be updated.
1375     int mdp_delta = in_bytes(BitData::bit_data_size());
1376     if (TypeProfileCasts) {
1377       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1378     }
1379     update_mdp_by_constant(mdp, mdp_delta);
1380 
1381     bind(profile_continue);
1382   }
1383 }
1384 
1385 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1386   if (ProfileInterpreter && TypeProfileCasts) {
1387     Label profile_continue;
1388 
1389     // If no method data exists, go to profile_continue.
1390     test_method_data_pointer(mdp, profile_continue);
1391 
1392     int count_offset = in_bytes(CounterData::count_offset());
1393     // Back up the address, since we have already bumped the mdp.
1394     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1395 
1396     // *Decrement* the counter.  We expect to see zero or small negatives.
1397     increment_mdp_data_at(mdp, count_offset, true);
1398 
1399     bind (profile_continue);
1400   }
1401 }
1402 
1403 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1404   if (ProfileInterpreter) {
1405     Label profile_continue;
1406 
1407     // If no method data exists, go to profile_continue.
1408     test_method_data_pointer(mdp, profile_continue);
1409 
1410     // The method data pointer needs to be updated.
1411     int mdp_delta = in_bytes(BitData::bit_data_size());
1412     if (TypeProfileCasts) {
1413       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1414 
1415       // Record the object type.
1416       record_klass_in_profile(klass, mdp, reg2, false);
1417     }
1418     update_mdp_by_constant(mdp, mdp_delta);
1419 
1420     bind(profile_continue);
1421   }
1422 }
1423 
1424 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1425   if (ProfileInterpreter) {
1426     Label profile_continue;
1427 
1428     // If no method data exists, go to profile_continue.
1429     test_method_data_pointer(mdp, profile_continue);
1430 
1431     // Update the default case count
1432     increment_mdp_data_at(mdp,
1433                           in_bytes(MultiBranchData::default_count_offset()));
1434 
1435     // The method data pointer needs to be updated.
1436     update_mdp_by_offset(mdp,
1437                          in_bytes(MultiBranchData::
1438                                   default_displacement_offset()));
1439 
1440     bind(profile_continue);
1441   }
1442 }
1443 
1444 void InterpreterMacroAssembler::profile_switch_case(Register index,
1445                                                     Register mdp,
1446                                                     Register reg2) {
1447   if (ProfileInterpreter) {
1448     Label profile_continue;
1449 
1450     // If no method data exists, go to profile_continue.
1451     test_method_data_pointer(mdp, profile_continue);
1452 
1453     // Build the base (index * per_case_size_in_bytes()) +
1454     // case_array_offset_in_bytes()
1455     movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1456     movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1457     Assembler::maddw(index, index, reg2, rscratch1);
1458 
1459     // Update the case count
1460     increment_mdp_data_at(mdp,
1461                           index,
1462                           in_bytes(MultiBranchData::relative_count_offset()));
1463 
1464     // The method data pointer needs to be updated.
1465     update_mdp_by_offset(mdp,
1466                          index,
1467                          in_bytes(MultiBranchData::
1468                                   relative_displacement_offset()));
1469 
1470     bind(profile_continue);
1471   }
1472 }
1473 
1474 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1475   if (state == atos) {
1476     MacroAssembler::verify_oop(reg);
1477   }
1478 }
1479 
1480 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1481 
1482 
1483 void InterpreterMacroAssembler::notify_method_entry() {
1484   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1485   // track stack depth.  If it is possible to enter interp_only_mode we add
1486   // the code to check if the event should be sent.
1487   if (JvmtiExport::can_post_interpreter_events()) {
1488     Label L;
1489     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1490     cbzw(r3, L);
1491     call_VM(noreg, CAST_FROM_FN_PTR(address,
1492                                     InterpreterRuntime::post_method_entry));
1493     bind(L);
1494   }
1495 
1496   {
1497     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1498     get_method(c_rarg1);
1499     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1500                  rthread, c_rarg1);
1501   }
1502 
1503   // RedefineClasses() tracing support for obsolete method entry
1504   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1505     get_method(c_rarg1);
1506     call_VM_leaf(
1507       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1508       rthread, c_rarg1);
1509   }
1510 
1511  }
1512 
1513 
1514 void InterpreterMacroAssembler::notify_method_exit(
1515     TosState state, NotifyMethodExitMode mode) {
1516   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1517   // track stack depth.  If it is possible to enter interp_only_mode we add
1518   // the code to check if the event should be sent.
1519   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1520     Label L;
1521     // Note: frame::interpreter_frame_result has a dependency on how the
1522     // method result is saved across the call to post_method_exit. If this
1523     // is changed then the interpreter_frame_result implementation will
1524     // need to be updated too.
1525 
1526     // template interpreter will leave the result on the top of the stack.
1527     push(state);
1528     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1529     cbz(r3, L);
1530     call_VM(noreg,
1531             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1532     bind(L);
1533     pop(state);
1534   }
1535 
1536   {
1537     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1538     push(state);
1539     get_method(c_rarg1);
1540     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1541                  rthread, c_rarg1);
1542     pop(state);
1543   }
1544 }
1545 
1546 
1547 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1548 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1549                                                         int increment, Address mask,
1550                                                         Register scratch, Register scratch2,
1551                                                         bool preloaded, Condition cond,
1552                                                         Label* where) {
1553   if (!preloaded) {
1554     ldrw(scratch, counter_addr);
1555   }
1556   add(scratch, scratch, increment);
1557   strw(scratch, counter_addr);
1558   ldrw(scratch2, mask);
1559   ands(scratch, scratch, scratch2);
1560   br(cond, *where);
1561 }
1562 
1563 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1564                                                   int number_of_arguments) {
1565   // interpreter specific
1566   //
1567   // Note: No need to save/restore rbcp & rlocals pointer since these
1568   //       are callee saved registers and no blocking/ GC can happen
1569   //       in leaf calls.
1570 #ifdef ASSERT
1571   {
1572     Label L;
1573     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1574     cbz(rscratch1, L);
1575     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1576          " last_sp != NULL");
1577     bind(L);
1578   }
1579 #endif /* ASSERT */
1580   // super call
1581   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1582 }
1583 
1584 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1585                                              Register java_thread,
1586                                              Register last_java_sp,
1587                                              address  entry_point,
1588                                              int      number_of_arguments,
1589                                              bool     check_exceptions) {
1590   // interpreter specific
1591   //
1592   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1593   //       really make a difference for these runtime calls, since they are
1594   //       slow anyway. Btw., bcp must be saved/restored since it may change
1595   //       due to GC.
1596   // assert(java_thread == noreg , "not expecting a precomputed java thread");
1597   save_bcp();
1598 #ifdef ASSERT
1599   {
1600     Label L;
1601     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1602     cbz(rscratch1, L);
1603     stop("InterpreterMacroAssembler::call_VM_base:"
1604          " last_sp != NULL");
1605     bind(L);
1606   }
1607 #endif /* ASSERT */
1608   // super call
1609   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1610                                entry_point, number_of_arguments,
1611                      check_exceptions);
1612 // interpreter specific
1613   restore_bcp();
1614   restore_locals();
1615 }
1616 
1617 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1618   assert_different_registers(obj, rscratch1);
1619   Label update, next, none;
1620 
1621   verify_oop(obj);
1622 
1623   cbnz(obj, update);
1624   orptr(mdo_addr, TypeEntries::null_seen);
1625   b(next);
1626 
1627   bind(update);
1628   load_klass(obj, obj);
1629 
1630   ldr(rscratch1, mdo_addr);
1631   eor(obj, obj, rscratch1);
1632   tst(obj, TypeEntries::type_klass_mask);
1633   br(Assembler::EQ, next); // klass seen before, nothing to
1634                            // do. The unknown bit may have been
1635                            // set already but no need to check.
1636 
1637   tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1638   // already unknown. Nothing to do anymore.
1639 
1640   ldr(rscratch1, mdo_addr);
1641   cbz(rscratch1, none);
1642   cmp(rscratch1, (u1)TypeEntries::null_seen);
1643   br(Assembler::EQ, none);
1644   // There is a chance that the checks above (re-reading profiling
1645   // data from memory) fail if another thread has just set the
1646   // profiling to this obj's klass
1647   ldr(rscratch1, mdo_addr);
1648   eor(obj, obj, rscratch1);
1649   tst(obj, TypeEntries::type_klass_mask);
1650   br(Assembler::EQ, next);
1651 
1652   // different than before. Cannot keep accurate profile.
1653   orptr(mdo_addr, TypeEntries::type_unknown);
1654   b(next);
1655 
1656   bind(none);
1657   // first time here. Set profile type.
1658   str(obj, mdo_addr);
1659 
1660   bind(next);
1661 }
1662 
1663 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1664   if (!ProfileInterpreter) {
1665     return;
1666   }
1667 
1668   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1669     Label profile_continue;
1670 
1671     test_method_data_pointer(mdp, profile_continue);
1672 
1673     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1674 
1675     ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1676     cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1677     br(Assembler::NE, profile_continue);
1678 
1679     if (MethodData::profile_arguments()) {
1680       Label done;
1681       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1682 
1683       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1684         if (i > 0 || MethodData::profile_return()) {
1685           // If return value type is profiled we may have no argument to profile
1686           ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1687           sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1688           cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1689           add(rscratch1, mdp, off_to_args);
1690           br(Assembler::LT, done);
1691         }
1692         ldr(tmp, Address(callee, Method::const_offset()));
1693         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1694         // stack offset o (zero based) from the start of the argument
1695         // list, for n arguments translates into offset n - o - 1 from
1696         // the end of the argument list
1697         ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1698         sub(tmp, tmp, rscratch1);
1699         sub(tmp, tmp, 1);
1700         Address arg_addr = argument_address(tmp);
1701         ldr(tmp, arg_addr);
1702 
1703         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1704         profile_obj_type(tmp, mdo_arg_addr);
1705 
1706         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1707         off_to_args += to_add;
1708       }
1709 
1710       if (MethodData::profile_return()) {
1711         ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1712         sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1713       }
1714 
1715       add(rscratch1, mdp, off_to_args);
1716       bind(done);
1717       mov(mdp, rscratch1);
1718 
1719       if (MethodData::profile_return()) {
1720         // We're right after the type profile for the last
1721         // argument. tmp is the number of cells left in the
1722         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1723         // if there's a return to profile.
1724         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1725         add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1726       }
1727       str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1728     } else {
1729       assert(MethodData::profile_return(), "either profile call args or call ret");
1730       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1731     }
1732 
1733     // mdp points right after the end of the
1734     // CallTypeData/VirtualCallTypeData, right after the cells for the
1735     // return value type if there's one
1736 
1737     bind(profile_continue);
1738   }
1739 }
1740 
1741 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1742   assert_different_registers(mdp, ret, tmp, rbcp);
1743   if (ProfileInterpreter && MethodData::profile_return()) {
1744     Label profile_continue, done;
1745 
1746     test_method_data_pointer(mdp, profile_continue);
1747 
1748     if (MethodData::profile_return_jsr292_only()) {
1749       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1750 
1751       // If we don't profile all invoke bytecodes we must make sure
1752       // it's a bytecode we indeed profile. We can't go back to the
1753       // begining of the ProfileData we intend to update to check its
1754       // type because we're right after it and we don't known its
1755       // length
1756       Label do_profile;
1757       ldrb(rscratch1, Address(rbcp, 0));
1758       cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1759       br(Assembler::EQ, do_profile);
1760       cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1761       br(Assembler::EQ, do_profile);
1762       get_method(tmp);
1763       ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1764       subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1765       br(Assembler::NE, profile_continue);
1766 
1767       bind(do_profile);
1768     }
1769 
1770     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1771     mov(tmp, ret);
1772     profile_obj_type(tmp, mdo_ret_addr);
1773 
1774     bind(profile_continue);
1775   }
1776 }
1777 
1778 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1779   assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1780   if (ProfileInterpreter && MethodData::profile_parameters()) {
1781     Label profile_continue, done;
1782 
1783     test_method_data_pointer(mdp, profile_continue);
1784 
1785     // Load the offset of the area within the MDO used for
1786     // parameters. If it's negative we're not profiling any parameters
1787     ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1788     tbnz(tmp1, 31, profile_continue);  // i.e. sign bit set
1789 
1790     // Compute a pointer to the area for parameters from the offset
1791     // and move the pointer to the slot for the last
1792     // parameters. Collect profiling from last parameter down.
1793     // mdo start + parameters offset + array length - 1
1794     add(mdp, mdp, tmp1);
1795     ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1796     sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1797 
1798     Label loop;
1799     bind(loop);
1800 
1801     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1802     int type_base = in_bytes(ParametersTypeData::type_offset(0));
1803     int per_arg_scale = exact_log2(DataLayout::cell_size);
1804     add(rscratch1, mdp, off_base);
1805     add(rscratch2, mdp, type_base);
1806 
1807     Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1808     Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1809 
1810     // load offset on the stack from the slot for this parameter
1811     ldr(tmp2, arg_off);
1812     neg(tmp2, tmp2);
1813     // read the parameter from the local area
1814     ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1815 
1816     // profile the parameter
1817     profile_obj_type(tmp2, arg_type);
1818 
1819     // go to next parameter
1820     subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1821     br(Assembler::GE, loop);
1822 
1823     bind(profile_continue);
1824   }
1825 }
--- EOF ---