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/biasedLocking.hpp"
  43 #include "runtime/frame.inline.hpp"
  44 #include "runtime/safepointMechanism.hpp"
  45 #include "runtime/sharedRuntime.hpp"
  46 #include "runtime/thread.inline.hpp"
  47 #include "utilities/powerOfTwo.hpp"
  48 
  49 void InterpreterMacroAssembler::narrow(Register result) {
  50 
  51   // Get method->_constMethod->_result_type
  52   ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
  53   ldr(rscratch1, Address(rscratch1, Method::const_offset()));
  54   ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
  55 
  56   Label done, notBool, notByte, notChar;
  57 
  58   // common case first
  59   cmpw(rscratch1, T_INT);
  60   br(Assembler::EQ, done);
  61 
  62   // mask integer result to narrower return type.
  63   cmpw(rscratch1, T_BOOLEAN);
  64   br(Assembler::NE, notBool);
  65   andw(result, result, 0x1);
  66   b(done);
  67 
  68   bind(notBool);
  69   cmpw(rscratch1, T_BYTE);
  70   br(Assembler::NE, notByte);
  71   sbfx(result, result, 0, 8);
  72   b(done);
  73 
  74   bind(notByte);
  75   cmpw(rscratch1, T_CHAR);
  76   br(Assembler::NE, notChar);
  77   ubfx(result, result, 0, 16);  // truncate upper 16 bits
  78   b(done);
  79 
  80   bind(notChar);
  81   sbfx(result, result, 0, 16);     // sign-extend short
  82 
  83   // Nothing to do for T_INT
  84   bind(done);
  85 }
  86 
  87 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  88   assert(entry, "Entry must have been generated by now");
  89   b(entry);
  90 }
  91 
  92 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
  93   if (JvmtiExport::can_pop_frame()) {
  94     Label L;
  95     // Initiate popframe handling only if it is not already being
  96     // processed.  If the flag has the popframe_processing bit set, it
  97     // means that this code is called *during* popframe handling - we
  98     // don't want to reenter.
  99     // This method is only called just after the call into the vm in
 100     // call_VM_base, so the arg registers are available.
 101     ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
 102     tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
 103     tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
 104     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 105     // address of the same-named entrypoint in the generated interpreter code.
 106     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 107     br(r0);
 108     bind(L);
 109   }
 110 }
 111 
 112 
 113 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 114   ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 115   const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
 116   const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
 117   const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
 118   switch (state) {
 119     case atos: ldr(r0, oop_addr);
 120                str(zr, oop_addr);
 121                verify_oop(r0, state);               break;
 122     case ltos: ldr(r0, val_addr);                   break;
 123     case btos:                                   // fall through
 124     case ztos:                                   // fall through
 125     case ctos:                                   // fall through
 126     case stos:                                   // fall through
 127     case itos: ldrw(r0, val_addr);                  break;
 128     case ftos: ldrs(v0, val_addr);                  break;
 129     case dtos: ldrd(v0, val_addr);                  break;
 130     case vtos: /* nothing to do */                  break;
 131     default  : ShouldNotReachHere();
 132   }
 133   // Clean up tos value in the thread object
 134   movw(rscratch1, (int) ilgl);
 135   strw(rscratch1, tos_addr);
 136   strw(zr, val_addr);
 137 }
 138 
 139 
 140 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 141   if (JvmtiExport::can_force_early_return()) {
 142     Label L;
 143     ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 144     cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit;
 145 
 146     // Initiate earlyret handling only if it is not already being processed.
 147     // If the flag has the earlyret_processing bit set, it means that this code
 148     // is called *during* earlyret handling - we don't want to reenter.
 149     ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
 150     cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
 151     br(Assembler::NE, L);
 152 
 153     // Call Interpreter::remove_activation_early_entry() to get the address of the
 154     // same-named entrypoint in the generated interpreter code.
 155     ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 156     ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
 157     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
 158     br(r0);
 159     bind(L);
 160   }
 161 }
 162 
 163 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
 164   Register reg,
 165   int bcp_offset) {
 166   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 167   ldrh(reg, Address(rbcp, bcp_offset));
 168   rev16(reg, reg);
 169 }
 170 
 171 void InterpreterMacroAssembler::get_dispatch() {
 172   uint64_t offset;
 173   adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
 174   lea(rdispatch, Address(rdispatch, offset));
 175 }
 176 
 177 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 178                                                        int bcp_offset,
 179                                                        size_t index_size) {
 180   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 181   if (index_size == sizeof(u2)) {
 182     load_unsigned_short(index, Address(rbcp, bcp_offset));
 183   } else if (index_size == sizeof(u4)) {
 184     // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
 185     ldrw(index, Address(rbcp, bcp_offset));
 186     // Check if the secondary index definition is still ~x, otherwise
 187     // we have to change the following assembler code to calculate the
 188     // plain index.
 189     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 190     eonw(index, index, zr);  // convert to plain index
 191   } else if (index_size == sizeof(u1)) {
 192     load_unsigned_byte(index, Address(rbcp, bcp_offset));
 193   } else {
 194     ShouldNotReachHere();
 195   }
 196 }
 197 
 198 // Return
 199 // Rindex: index into constant pool
 200 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
 201 //
 202 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
 203 // the true address of the cache entry.
 204 //
 205 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 206                                                            Register index,
 207                                                            int bcp_offset,
 208                                                            size_t index_size) {
 209   assert_different_registers(cache, index);
 210   assert_different_registers(cache, rcpool);
 211   get_cache_index_at_bcp(index, bcp_offset, index_size);
 212   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 213   // convert from field index to ConstantPoolCacheEntry
 214   // aarch64 already has the cache in rcpool so there is no need to
 215   // install it in cache. instead we pre-add the indexed offset to
 216   // rcpool and return it in cache. All clients of this method need to
 217   // be modified accordingly.
 218   add(cache, rcpool, index, Assembler::LSL, 5);
 219 }
 220 
 221 
 222 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 223                                                                         Register index,
 224                                                                         Register bytecode,
 225                                                                         int byte_no,
 226                                                                         int bcp_offset,
 227                                                                         size_t index_size) {
 228   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
 229   // We use a 32-bit load here since the layout of 64-bit words on
 230   // little-endian machines allow us that.
 231   // n.b. unlike x86 cache already includes the index offset
 232   lea(bytecode, Address(cache,
 233                          ConstantPoolCache::base_offset()
 234                          + ConstantPoolCacheEntry::indices_offset()));
 235   ldarw(bytecode, bytecode);
 236   const int shift_count = (1 + byte_no) * BitsPerByte;
 237   ubfx(bytecode, bytecode, shift_count, BitsPerByte);
 238 }
 239 
 240 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 241                                                                Register tmp,
 242                                                                int bcp_offset,
 243                                                                size_t index_size) {
 244   assert(cache != tmp, "must use different register");
 245   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 246   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 247   // convert from field index to ConstantPoolCacheEntry index
 248   // and from word offset to byte offset
 249   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
 250   ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
 251   // skip past the header
 252   add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
 253   add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord);  // construct pointer to cache entry
 254 }
 255 
 256 void InterpreterMacroAssembler::get_method_counters(Register method,
 257                                                     Register mcs, Label& skip) {
 258   Label has_counters;
 259   ldr(mcs, Address(method, Method::method_counters_offset()));
 260   cbnz(mcs, has_counters);
 261   call_VM(noreg, CAST_FROM_FN_PTR(address,
 262           InterpreterRuntime::build_method_counters), method);
 263   ldr(mcs, Address(method, Method::method_counters_offset()));
 264   cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
 265   bind(has_counters);
 266 }
 267 
 268 // Load object from cpool->resolved_references(index)
 269 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 270                                            Register result, Register index, Register tmp) {
 271   assert_different_registers(result, index);
 272 
 273   get_constant_pool(result);
 274   // load pointer for resolved_references[] objArray
 275   ldr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
 276   ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
 277   resolve_oop_handle(result, tmp);
 278   // Add in the index
 279   add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
 280   load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)));
 281 }
 282 
 283 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
 284                              Register cpool, Register index, Register klass, Register temp) {
 285   add(temp, cpool, index, LSL, LogBytesPerWord);
 286   ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
 287   ldr(klass, Address(cpool,  ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
 288   add(klass, klass, temp, LSL, LogBytesPerWord);
 289   ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
 290 }
 291 
 292 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
 293                                                               Register method,
 294                                                               Register cache) {
 295   const int method_offset = in_bytes(
 296     ConstantPoolCache::base_offset() +
 297       ((byte_no == TemplateTable::f2_byte)
 298        ? ConstantPoolCacheEntry::f2_offset()
 299        : ConstantPoolCacheEntry::f1_offset()));
 300 
 301   ldr(method, Address(cache, method_offset)); // get f1 Method*
 302 }
 303 
 304 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 305 // subtype of super_klass.
 306 //
 307 // Args:
 308 //      r0: superklass
 309 //      Rsub_klass: subklass
 310 //
 311 // Kills:
 312 //      r2, r5
 313 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 314                                                   Label& ok_is_subtype) {
 315   assert(Rsub_klass != r0, "r0 holds superklass");
 316   assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
 317   assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
 318 
 319   // Profile the not-null value's klass.
 320   profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
 321 
 322   // Do the check.
 323   check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
 324 
 325   // Profile the failure of the check.
 326   profile_typecheck_failed(r2); // blows r2
 327 }
 328 
 329 // Java Expression Stack
 330 
 331 void InterpreterMacroAssembler::pop_ptr(Register r) {
 332   ldr(r, post(esp, wordSize));
 333 }
 334 
 335 void InterpreterMacroAssembler::pop_i(Register r) {
 336   ldrw(r, post(esp, wordSize));
 337 }
 338 
 339 void InterpreterMacroAssembler::pop_l(Register r) {
 340   ldr(r, post(esp, 2 * Interpreter::stackElementSize));
 341 }
 342 
 343 void InterpreterMacroAssembler::push_ptr(Register r) {
 344   str(r, pre(esp, -wordSize));
 345  }
 346 
 347 void InterpreterMacroAssembler::push_i(Register r) {
 348   str(r, pre(esp, -wordSize));
 349 }
 350 
 351 void InterpreterMacroAssembler::push_l(Register r) {
 352   str(zr, pre(esp, -wordSize));
 353   str(r, pre(esp, - wordSize));
 354 }
 355 
 356 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
 357   ldrs(r, post(esp, wordSize));
 358 }
 359 
 360 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
 361   ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
 362 }
 363 
 364 void InterpreterMacroAssembler::push_f(FloatRegister r) {
 365   strs(r, pre(esp, -wordSize));
 366 }
 367 
 368 void InterpreterMacroAssembler::push_d(FloatRegister r) {
 369   strd(r, pre(esp, 2* -wordSize));
 370 }
 371 
 372 void InterpreterMacroAssembler::pop(TosState state) {
 373   switch (state) {
 374   case atos: pop_ptr();                 break;
 375   case btos:
 376   case ztos:
 377   case ctos:
 378   case stos:
 379   case itos: pop_i();                   break;
 380   case ltos: pop_l();                   break;
 381   case ftos: pop_f();                   break;
 382   case dtos: pop_d();                   break;
 383   case vtos: /* nothing to do */        break;
 384   default:   ShouldNotReachHere();
 385   }
 386   verify_oop(r0, state);
 387 }
 388 
 389 void InterpreterMacroAssembler::push(TosState state) {
 390   verify_oop(r0, state);
 391   switch (state) {
 392   case atos: push_ptr();                break;
 393   case btos:
 394   case ztos:
 395   case ctos:
 396   case stos:
 397   case itos: push_i();                  break;
 398   case ltos: push_l();                  break;
 399   case ftos: push_f();                  break;
 400   case dtos: push_d();                  break;
 401   case vtos: /* nothing to do */        break;
 402   default  : ShouldNotReachHere();
 403   }
 404 }
 405 
 406 // Helpers for swap and dup
 407 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 408   ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
 409 }
 410 
 411 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 412   str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
 413 }
 414 
 415 void InterpreterMacroAssembler::load_float(Address src) {
 416   ldrs(v0, src);
 417 }
 418 
 419 void InterpreterMacroAssembler::load_double(Address src) {
 420   ldrd(v0, src);
 421 }
 422 
 423 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 424   // set sender sp
 425   mov(r13, sp);
 426   // record last_sp
 427   str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 428 }
 429 
 430 // Jump to from_interpreted entry of a call unless single stepping is possible
 431 // in this thread in which case we must call the i2i entry
 432 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 433   prepare_to_jump_from_interpreted();
 434 
 435   if (JvmtiExport::can_post_interpreter_events()) {
 436     Label run_compiled_code;
 437     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 438     // compiled code in threads for which the event is enabled.  Check here for
 439     // interp_only_mode if these events CAN be enabled.
 440     ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
 441     cbzw(rscratch1, run_compiled_code);
 442     ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
 443     br(rscratch1);
 444     bind(run_compiled_code);
 445   }
 446 
 447   ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
 448   br(rscratch1);
 449 }
 450 
 451 // The following two routines provide a hook so that an implementation
 452 // can schedule the dispatch in two parts.  amd64 does not do this.
 453 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 454 }
 455 
 456 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 457     dispatch_next(state, step);
 458 }
 459 
 460 void InterpreterMacroAssembler::dispatch_base(TosState state,
 461                                               address* table,
 462                                               bool verifyoop,
 463                                               bool generate_poll) {
 464   if (VerifyActivationFrameSize) {
 465     Unimplemented();
 466   }
 467   if (verifyoop) {
 468     verify_oop(r0, state);
 469   }
 470 
 471   Label safepoint;
 472   address* const safepoint_table = Interpreter::safept_table(state);
 473   bool needs_thread_local_poll = generate_poll && table != safepoint_table;
 474 
 475   if (needs_thread_local_poll) {
 476     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 477     ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
 478     tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
 479   }
 480 
 481   if (table == Interpreter::dispatch_table(state)) {
 482     addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
 483     ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
 484   } else {
 485     mov(rscratch2, (address)table);
 486     ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
 487   }
 488   br(rscratch2);
 489 
 490   if (needs_thread_local_poll) {
 491     bind(safepoint);
 492     lea(rscratch2, ExternalAddress((address)safepoint_table));
 493     ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
 494     br(rscratch2);
 495   }
 496 }
 497 
 498 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 499   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 500 }
 501 
 502 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 503   dispatch_base(state, Interpreter::normal_table(state));
 504 }
 505 
 506 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 507   dispatch_base(state, Interpreter::normal_table(state), false);
 508 }
 509 
 510 
 511 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 512   // load next bytecode
 513   ldrb(rscratch1, Address(pre(rbcp, step)));
 514   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
 515 }
 516 
 517 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 518   // load current bytecode
 519   ldrb(rscratch1, Address(rbcp, 0));
 520   dispatch_base(state, table);
 521 }
 522 
 523 // remove activation
 524 //
 525 // Apply stack watermark barrier.
 526 // Unlock the receiver if this is a synchronized method.
 527 // Unlock any Java monitors from syncronized blocks.
 528 // Remove the activation from the stack.
 529 //
 530 // If there are locked Java monitors
 531 //    If throw_monitor_exception
 532 //       throws IllegalMonitorStateException
 533 //    Else if install_monitor_exception
 534 //       installs IllegalMonitorStateException
 535 //    Else
 536 //       no error processing
 537 void InterpreterMacroAssembler::remove_activation(
 538         TosState state,
 539         bool throw_monitor_exception,
 540         bool install_monitor_exception,
 541         bool notify_jvmdi) {
 542   // Note: Registers r3 xmm0 may be in use for the
 543   // result check if synchronized method
 544   Label unlocked, unlock, no_unlock;
 545 
 546   // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
 547   // that would normally not be safe to use. Such bad returns into unsafe territory of
 548   // the stack, will call InterpreterRuntime::at_unwind.
 549   Label slow_path;
 550   Label fast_path;
 551   safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
 552   br(Assembler::AL, fast_path);
 553   bind(slow_path);
 554   push(state);
 555   set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
 556   super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
 557   reset_last_Java_frame(true);
 558   pop(state);
 559   bind(fast_path);
 560 
 561   // get the value of _do_not_unlock_if_synchronized into r3
 562   const Address do_not_unlock_if_synchronized(rthread,
 563     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 564   ldrb(r3, do_not_unlock_if_synchronized);
 565   strb(zr, do_not_unlock_if_synchronized); // reset the flag
 566 
 567  // get method access flags
 568   ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
 569   ldr(r2, Address(r1, Method::access_flags_offset()));
 570   tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
 571 
 572   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 573   // is set.
 574   cbnz(r3, no_unlock);
 575 
 576   // unlock monitor
 577   push(state); // save result
 578 
 579   // BasicObjectLock will be first in list, since this is a
 580   // synchronized method. However, need to check that the object has
 581   // not been unlocked by an explicit monitorexit bytecode.
 582   const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
 583                         wordSize - (int) sizeof(BasicObjectLock));
 584   // We use c_rarg1 so that if we go slow path it will be the correct
 585   // register for unlock_object to pass to VM directly
 586   lea(c_rarg1, monitor); // address of first monitor
 587 
 588   ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 589   cbnz(r0, unlock);
 590 
 591   pop(state);
 592   if (throw_monitor_exception) {
 593     // Entry already unlocked, need to throw exception
 594     call_VM(noreg, CAST_FROM_FN_PTR(address,
 595                    InterpreterRuntime::throw_illegal_monitor_state_exception));
 596     should_not_reach_here();
 597   } else {
 598     // Monitor already unlocked during a stack unroll. If requested,
 599     // install an illegal_monitor_state_exception.  Continue with
 600     // stack unrolling.
 601     if (install_monitor_exception) {
 602       call_VM(noreg, CAST_FROM_FN_PTR(address,
 603                      InterpreterRuntime::new_illegal_monitor_state_exception));
 604     }
 605     b(unlocked);
 606   }
 607 
 608   bind(unlock);
 609   unlock_object(c_rarg1);
 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       pop(state);
 653 
 654       if (install_monitor_exception) {
 655         call_VM(noreg, CAST_FROM_FN_PTR(address,
 656                                         InterpreterRuntime::
 657                                         new_illegal_monitor_state_exception));
 658       }
 659 
 660       b(restart);
 661     }
 662 
 663     bind(loop);
 664     // check if current entry is used
 665     ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
 666     cbnz(rscratch1, exception);
 667 
 668     add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
 669     bind(entry);
 670     cmp(c_rarg1, r19); // check if bottom reached
 671     br(Assembler::NE, loop); // if not at bottom then check this entry
 672   }
 673 
 674   bind(no_unlock);
 675 
 676   // jvmti support
 677   if (notify_jvmdi) {
 678     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
 679   } else {
 680     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 681   }
 682 
 683   // remove activation
 684   // get sender esp
 685   ldr(rscratch2,
 686       Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
 687   if (StackReservedPages > 0) {
 688     // testing if reserved zone needs to be re-enabled
 689     Label no_reserved_zone_enabling;
 690 
 691     // look for an overflow into the stack reserved zone, i.e.
 692     // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
 693     ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
 694     cmp(rscratch2, rscratch1);
 695     br(Assembler::LS, no_reserved_zone_enabling);
 696 
 697     call_VM_leaf(
 698       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
 699     call_VM(noreg, CAST_FROM_FN_PTR(address,
 700                    InterpreterRuntime::throw_delayed_StackOverflowError));
 701     should_not_reach_here();
 702 
 703     bind(no_reserved_zone_enabling);
 704   }
 705 
 706   // restore sender esp
 707   mov(esp, rscratch2);
 708   // remove frame anchor
 709   leave();
 710   // If we're returning to interpreted code we will shortly be
 711   // adjusting SP to allow some space for ESP.  If we're returning to
 712   // compiled code the saved sender SP was saved in sender_sp, so this
 713   // restores it.
 714   andr(sp, esp, -16);
 715 }
 716 
 717 // Lock object
 718 //
 719 // Args:
 720 //      c_rarg1: BasicObjectLock to be used for locking
 721 //
 722 // Kills:
 723 //      r0
 724 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
 725 //      rscratch1, rscratch2 (scratch regs)
 726 void InterpreterMacroAssembler::lock_object(Register lock_reg)
 727 {
 728   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
 729   if (LockingMode == LM_MONITOR) {
 730     call_VM(noreg,
 731             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 732             lock_reg);
 733   } else {
 734     Label done;
 735 
 736     const Register swap_reg = r0;
 737     const Register tmp = c_rarg2;
 738     const Register obj_reg = c_rarg3; // Will contain the oop
 739 
 740     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
 741     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
 742     const int mark_offset = lock_offset +
 743                             BasicLock::displaced_header_offset_in_bytes();
 744 
 745     Label slow_case;
 746 
 747     // Load object pointer into obj_reg %c_rarg3
 748     ldr(obj_reg, Address(lock_reg, obj_offset));
 749 
 750     if (DiagnoseSyncOnValueBasedClasses != 0) {
 751       load_klass(tmp, obj_reg);
 752       ldrw(tmp, Address(tmp, Klass::access_flags_offset()));
 753       tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
 754       br(Assembler::NE, slow_case);
 755     }
 756 
 757     if (LockingMode == LM_LIGHTWEIGHT) {
 758       ldr(tmp, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 759       fast_lock(obj_reg, tmp, rscratch1, rscratch2, slow_case);
 760       b(done);
 761     } else {
 762       if (UseBiasedLocking) {
 763         biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
 764       }
 765 
 766       // Load (object->mark() | 1) into swap_reg
 767       ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 768       orr(swap_reg, rscratch1, 1);
 769 
 770       // Save (object->mark() | 1) into BasicLock's displaced header
 771       str(swap_reg, Address(lock_reg, mark_offset));
 772 
 773       assert(lock_offset == 0,
 774              "displached header must be first word in BasicObjectLock");
 775 
 776       Label fail;
 777       if (PrintBiasedLockingStatistics) {
 778         Label fast;
 779         cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail);
 780         bind(fast);
 781         atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
 782                     rscratch2, rscratch1, tmp);
 783         b(done);
 784         bind(fail);
 785       } else {
 786         cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 787       }
 788 
 789       // Fast check for recursive lock.
 790       //
 791       // Can apply the optimization only if this is a stack lock
 792       // allocated in this thread. For efficiency, we can focus on
 793       // recently allocated stack locks (instead of reading the stack
 794       // base and checking whether 'mark' points inside the current
 795       // thread stack):
 796       //  1) (mark & 7) == 0, and
 797       //  2) sp <= mark < mark + os::pagesize()
 798       //
 799       // Warning: sp + os::pagesize can overflow the stack base. We must
 800       // neither apply the optimization for an inflated lock allocated
 801       // just above the thread stack (this is why condition 1 matters)
 802       // nor apply the optimization if the stack lock is inside the stack
 803       // of another thread. The latter is avoided even in case of overflow
 804       // because we have guard pages at the end of all stacks. Hence, if
 805       // we go over the stack base and hit the stack of another thread,
 806       // this should not be in a writeable area that could contain a
 807       // stack lock allocated by that thread. As a consequence, a stack
 808       // lock less than page size away from sp is guaranteed to be
 809       // owned by the current thread.
 810       //
 811       // These 3 tests can be done by evaluating the following
 812       // expression: ((mark - sp) & (7 - os::vm_page_size())),
 813       // assuming both stack pointer and pagesize have their
 814       // least significant 3 bits clear.
 815       // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
 816       // NOTE2: aarch64 does not like to subtract sp from rn so take a
 817       // copy
 818       mov(rscratch1, sp);
 819       sub(swap_reg, swap_reg, rscratch1);
 820       ands(swap_reg, swap_reg, (uint64_t)(7 - os::vm_page_size()));
 821 
 822       // Save the test result, for recursive case, the result is zero
 823       str(swap_reg, Address(lock_reg, mark_offset));
 824 
 825       if (PrintBiasedLockingStatistics) {
 826         br(Assembler::NE, slow_case);
 827         atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
 828                     rscratch2, rscratch1, tmp);
 829       }
 830       br(Assembler::EQ, done);
 831     }
 832     bind(slow_case);
 833 
 834     // Call the runtime routine for slow case
 835     if (LockingMode == LM_LIGHTWEIGHT) {
 836       call_VM(noreg,
 837               CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
 838               obj_reg);
 839     } else {
 840       call_VM(noreg,
 841               CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 842               lock_reg);
 843     }
 844 
 845     bind(done);
 846   }
 847 }
 848 
 849 
 850 // Unlocks an object. Used in monitorexit bytecode and
 851 // remove_activation.  Throws an IllegalMonitorException if object is
 852 // not locked by current thread.
 853 //
 854 // Args:
 855 //      c_rarg1: BasicObjectLock for lock
 856 //
 857 // Kills:
 858 //      r0
 859 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
 860 //      rscratch1, rscratch2 (scratch regs)
 861 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
 862 {
 863   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
 864 
 865   if (LockingMode == LM_MONITOR) {
 866     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 867   } else {
 868     Label done;
 869 
 870     const Register swap_reg   = r0;
 871     const Register header_reg = c_rarg2;  // Will contain the old oopMark
 872     const Register obj_reg    = c_rarg3;  // Will contain the oop
 873 
 874     save_bcp(); // Save in case of exception
 875 
 876     if (LockingMode != LM_LIGHTWEIGHT) {
 877       // Convert from BasicObjectLock structure to object and BasicLock
 878       // structure Store the BasicLock address into %r0
 879       lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
 880     }
 881 
 882     // Load oop into obj_reg(%c_rarg3)
 883     ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 884 
 885     // Free entry
 886     str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
 887 
 888     if (LockingMode == LM_LIGHTWEIGHT) {
 889       Label slow_case;
 890 
 891       // Check for non-symmetric locking. This is allowed by the spec and the interpreter
 892       // must handle it.
 893       Register tmp = rscratch1;
 894       // First check for lock-stack underflow.
 895       ldrw(tmp, Address(rthread, JavaThread::lock_stack_top_offset()));
 896       cmpw(tmp, (unsigned)LockStack::start_offset());
 897       br(Assembler::LE, slow_case);
 898       // Then check if the top of the lock-stack matches the unlocked object.
 899       subw(tmp, tmp, oopSize);
 900       ldr(tmp, Address(rthread, tmp));
 901       cmpoop(tmp, obj_reg);
 902       br(Assembler::NE, slow_case);
 903 
 904       ldr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
 905       tbnz(header_reg, exact_log2(markWord::monitor_value), slow_case);
 906       fast_unlock(obj_reg, header_reg, swap_reg, rscratch1, slow_case);
 907       b(done);
 908       bind(slow_case);
 909     } else {
 910       if (UseBiasedLocking) {
 911         biased_locking_exit(obj_reg, header_reg, done);
 912       }
 913 
 914       // Load the old header from BasicLock structure
 915       ldr(header_reg, Address(swap_reg,
 916                               BasicLock::displaced_header_offset_in_bytes()));
 917 
 918       // Test for recursion
 919       cbz(header_reg, done);
 920 
 921       // Atomic swap back the old header
 922       cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
 923     }
 924     // Call the runtime routine for slow case.
 925     str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
 926     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
 927 
 928     bind(done);
 929 
 930     restore_bcp();
 931   }
 932 }
 933 
 934 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
 935                                                          Label& zero_continue) {
 936   assert(ProfileInterpreter, "must be profiling interpreter");
 937   ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 938   cbz(mdp, zero_continue);
 939 }
 940 
 941 // Set the method data pointer for the current bcp.
 942 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
 943   assert(ProfileInterpreter, "must be profiling interpreter");
 944   Label set_mdp;
 945   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 946 
 947   // Test MDO to avoid the call if it is NULL.
 948   ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
 949   cbz(r0, set_mdp);
 950   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
 951   // r0: mdi
 952   // mdo is guaranteed to be non-zero here, we checked for it before the call.
 953   ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
 954   lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
 955   add(r0, r1, r0);
 956   str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
 957   bind(set_mdp);
 958   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 959 }
 960 
 961 void InterpreterMacroAssembler::verify_method_data_pointer() {
 962   assert(ProfileInterpreter, "must be profiling interpreter");
 963 #ifdef ASSERT
 964   Label verify_continue;
 965   stp(r0, r1, Address(pre(sp, -2 * wordSize)));
 966   stp(r2, r3, Address(pre(sp, -2 * wordSize)));
 967   test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
 968   get_method(r1);
 969 
 970   // If the mdp is valid, it will point to a DataLayout header which is
 971   // consistent with the bcp.  The converse is highly probable also.
 972   ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
 973   ldr(rscratch1, Address(r1, Method::const_offset()));
 974   add(r2, r2, rscratch1, Assembler::LSL);
 975   lea(r2, Address(r2, ConstMethod::codes_offset()));
 976   cmp(r2, rbcp);
 977   br(Assembler::EQ, verify_continue);
 978   // r1: method
 979   // rbcp: bcp // rbcp == 22
 980   // r3: mdp
 981   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
 982                r1, rbcp, r3);
 983   bind(verify_continue);
 984   ldp(r2, r3, Address(post(sp, 2 * wordSize)));
 985   ldp(r0, r1, Address(post(sp, 2 * wordSize)));
 986 #endif // ASSERT
 987 }
 988 
 989 
 990 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
 991                                                 int constant,
 992                                                 Register value) {
 993   assert(ProfileInterpreter, "must be profiling interpreter");
 994   Address data(mdp_in, constant);
 995   str(value, data);
 996 }
 997 
 998 
 999 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1000                                                       int constant,
1001                                                       bool decrement) {
1002   increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1003 }
1004 
1005 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1006                                                       Register reg,
1007                                                       int constant,
1008                                                       bool decrement) {
1009   assert(ProfileInterpreter, "must be profiling interpreter");
1010   // %%% this does 64bit counters at best it is wasting space
1011   // at worst it is a rare bug when counters overflow
1012 
1013   assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1014 
1015   Address addr1(mdp_in, constant);
1016   Address addr2(rscratch2, reg, Address::lsl(0));
1017   Address &addr = addr1;
1018   if (reg != noreg) {
1019     lea(rscratch2, addr1);
1020     addr = addr2;
1021   }
1022 
1023   if (decrement) {
1024     // Decrement the register.  Set condition codes.
1025     // Intel does this
1026     // addptr(data, (int32_t) -DataLayout::counter_increment);
1027     // If the decrement causes the counter to overflow, stay negative
1028     // Label L;
1029     // jcc(Assembler::negative, L);
1030     // addptr(data, (int32_t) DataLayout::counter_increment);
1031     // so we do this
1032     ldr(rscratch1, addr);
1033     subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1034     Label L;
1035     br(Assembler::LO, L);       // skip store if counter underflow
1036     str(rscratch1, addr);
1037     bind(L);
1038   } else {
1039     assert(DataLayout::counter_increment == 1,
1040            "flow-free idiom only works with 1");
1041     // Intel does this
1042     // Increment the register.  Set carry flag.
1043     // addptr(data, DataLayout::counter_increment);
1044     // If the increment causes the counter to overflow, pull back by 1.
1045     // sbbptr(data, (int32_t)0);
1046     // so we do this
1047     ldr(rscratch1, addr);
1048     adds(rscratch1, rscratch1, DataLayout::counter_increment);
1049     Label L;
1050     br(Assembler::CS, L);       // skip store if counter overflow
1051     str(rscratch1, addr);
1052     bind(L);
1053   }
1054 }
1055 
1056 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1057                                                 int flag_byte_constant) {
1058   assert(ProfileInterpreter, "must be profiling interpreter");
1059   int flags_offset = in_bytes(DataLayout::flags_offset());
1060   // Set the flag
1061   ldrb(rscratch1, Address(mdp_in, flags_offset));
1062   orr(rscratch1, rscratch1, flag_byte_constant);
1063   strb(rscratch1, Address(mdp_in, flags_offset));
1064 }
1065 
1066 
1067 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1068                                                  int offset,
1069                                                  Register value,
1070                                                  Register test_value_out,
1071                                                  Label& not_equal_continue) {
1072   assert(ProfileInterpreter, "must be profiling interpreter");
1073   if (test_value_out == noreg) {
1074     ldr(rscratch1, Address(mdp_in, offset));
1075     cmp(value, rscratch1);
1076   } else {
1077     // Put the test value into a register, so caller can use it:
1078     ldr(test_value_out, Address(mdp_in, offset));
1079     cmp(value, test_value_out);
1080   }
1081   br(Assembler::NE, not_equal_continue);
1082 }
1083 
1084 
1085 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1086                                                      int offset_of_disp) {
1087   assert(ProfileInterpreter, "must be profiling interpreter");
1088   ldr(rscratch1, Address(mdp_in, offset_of_disp));
1089   add(mdp_in, mdp_in, rscratch1, LSL);
1090   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1091 }
1092 
1093 
1094 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1095                                                      Register reg,
1096                                                      int offset_of_disp) {
1097   assert(ProfileInterpreter, "must be profiling interpreter");
1098   lea(rscratch1, Address(mdp_in, offset_of_disp));
1099   ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1100   add(mdp_in, mdp_in, rscratch1, LSL);
1101   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1102 }
1103 
1104 
1105 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1106                                                        int constant) {
1107   assert(ProfileInterpreter, "must be profiling interpreter");
1108   add(mdp_in, mdp_in, (unsigned)constant);
1109   str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1110 }
1111 
1112 
1113 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1114   assert(ProfileInterpreter, "must be profiling interpreter");
1115   // save/restore across call_VM
1116   stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1117   call_VM(noreg,
1118           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1119           return_bci);
1120   ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1121 }
1122 
1123 
1124 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1125                                                      Register bumped_count) {
1126   if (ProfileInterpreter) {
1127     Label profile_continue;
1128 
1129     // If no method data exists, go to profile_continue.
1130     // Otherwise, assign to mdp
1131     test_method_data_pointer(mdp, profile_continue);
1132 
1133     // We are taking a branch.  Increment the taken count.
1134     // We inline increment_mdp_data_at to return bumped_count in a register
1135     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1136     Address data(mdp, in_bytes(JumpData::taken_offset()));
1137     ldr(bumped_count, data);
1138     assert(DataLayout::counter_increment == 1,
1139             "flow-free idiom only works with 1");
1140     // Intel does this to catch overflow
1141     // addptr(bumped_count, DataLayout::counter_increment);
1142     // sbbptr(bumped_count, 0);
1143     // so we do this
1144     adds(bumped_count, bumped_count, DataLayout::counter_increment);
1145     Label L;
1146     br(Assembler::CS, L);       // skip store if counter overflow
1147     str(bumped_count, data);
1148     bind(L);
1149     // The method data pointer needs to be updated to reflect the new target.
1150     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1151     bind(profile_continue);
1152   }
1153 }
1154 
1155 
1156 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1157   if (ProfileInterpreter) {
1158     Label profile_continue;
1159 
1160     // If no method data exists, go to profile_continue.
1161     test_method_data_pointer(mdp, profile_continue);
1162 
1163     // We are taking a branch.  Increment the not taken count.
1164     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1165 
1166     // The method data pointer needs to be updated to correspond to
1167     // the next bytecode
1168     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1169     bind(profile_continue);
1170   }
1171 }
1172 
1173 
1174 void InterpreterMacroAssembler::profile_call(Register mdp) {
1175   if (ProfileInterpreter) {
1176     Label profile_continue;
1177 
1178     // If no method data exists, go to profile_continue.
1179     test_method_data_pointer(mdp, profile_continue);
1180 
1181     // We are making a call.  Increment the count.
1182     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1183 
1184     // The method data pointer needs to be updated to reflect the new target.
1185     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1186     bind(profile_continue);
1187   }
1188 }
1189 
1190 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1191   if (ProfileInterpreter) {
1192     Label profile_continue;
1193 
1194     // If no method data exists, go to profile_continue.
1195     test_method_data_pointer(mdp, profile_continue);
1196 
1197     // We are making a call.  Increment the count.
1198     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1199 
1200     // The method data pointer needs to be updated to reflect the new target.
1201     update_mdp_by_constant(mdp,
1202                            in_bytes(VirtualCallData::
1203                                     virtual_call_data_size()));
1204     bind(profile_continue);
1205   }
1206 }
1207 
1208 
1209 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1210                                                      Register mdp,
1211                                                      Register reg2,
1212                                                      bool receiver_can_be_null) {
1213   if (ProfileInterpreter) {
1214     Label profile_continue;
1215 
1216     // If no method data exists, go to profile_continue.
1217     test_method_data_pointer(mdp, profile_continue);
1218 
1219     Label skip_receiver_profile;
1220     if (receiver_can_be_null) {
1221       Label not_null;
1222       // We are making a call.  Increment the count for null receiver.
1223       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1224       b(skip_receiver_profile);
1225       bind(not_null);
1226     }
1227 
1228     // Record the receiver type.
1229     record_klass_in_profile(receiver, mdp, reg2, true);
1230     bind(skip_receiver_profile);
1231 
1232     // The method data pointer needs to be updated to reflect the new target.
1233     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1234     bind(profile_continue);
1235   }
1236 }
1237 
1238 // This routine creates a state machine for updating the multi-row
1239 // type profile at a virtual call site (or other type-sensitive bytecode).
1240 // The machine visits each row (of receiver/count) until the receiver type
1241 // is found, or until it runs out of rows.  At the same time, it remembers
1242 // the location of the first empty row.  (An empty row records null for its
1243 // receiver, and can be allocated for a newly-observed receiver type.)
1244 // Because there are two degrees of freedom in the state, a simple linear
1245 // search will not work; it must be a decision tree.  Hence this helper
1246 // function is recursive, to generate the required tree structured code.
1247 // It's the interpreter, so we are trading off code space for speed.
1248 // See below for example code.
1249 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1250                                         Register receiver, Register mdp,
1251                                         Register reg2, int start_row,
1252                                         Label& done, bool is_virtual_call) {
1253   if (TypeProfileWidth == 0) {
1254     if (is_virtual_call) {
1255       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1256     }
1257 #if INCLUDE_JVMCI
1258     else if (EnableJVMCI) {
1259       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1260     }
1261 #endif // INCLUDE_JVMCI
1262   } else {
1263     int non_profiled_offset = -1;
1264     if (is_virtual_call) {
1265       non_profiled_offset = in_bytes(CounterData::count_offset());
1266     }
1267 #if INCLUDE_JVMCI
1268     else if (EnableJVMCI) {
1269       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1270     }
1271 #endif // INCLUDE_JVMCI
1272 
1273     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1274         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1275   }
1276 }
1277 
1278 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1279                                         Register reg2, int start_row, Label& done, int total_rows,
1280                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1281                                         int non_profiled_offset) {
1282   int last_row = total_rows - 1;
1283   assert(start_row <= last_row, "must be work left to do");
1284   // Test this row for both the item and for null.
1285   // Take any of three different outcomes:
1286   //   1. found item => increment count and goto done
1287   //   2. found null => keep looking for case 1, maybe allocate this cell
1288   //   3. found something else => keep looking for cases 1 and 2
1289   // Case 3 is handled by a recursive call.
1290   for (int row = start_row; row <= last_row; row++) {
1291     Label next_test;
1292     bool test_for_null_also = (row == start_row);
1293 
1294     // See if the item is item[n].
1295     int item_offset = in_bytes(item_offset_fn(row));
1296     test_mdp_data_at(mdp, item_offset, item,
1297                      (test_for_null_also ? reg2 : noreg),
1298                      next_test);
1299     // (Reg2 now contains the item from the CallData.)
1300 
1301     // The item is item[n].  Increment count[n].
1302     int count_offset = in_bytes(item_count_offset_fn(row));
1303     increment_mdp_data_at(mdp, count_offset);
1304     b(done);
1305     bind(next_test);
1306 
1307     if (test_for_null_also) {
1308       Label found_null;
1309       // Failed the equality check on item[n]...  Test for null.
1310       if (start_row == last_row) {
1311         // The only thing left to do is handle the null case.
1312         if (non_profiled_offset >= 0) {
1313           cbz(reg2, found_null);
1314           // Item did not match any saved item and there is no empty row for it.
1315           // Increment total counter to indicate polymorphic case.
1316           increment_mdp_data_at(mdp, non_profiled_offset);
1317           b(done);
1318           bind(found_null);
1319         } else {
1320           cbnz(reg2, done);
1321         }
1322         break;
1323       }
1324       // Since null is rare, make it be the branch-taken case.
1325       cbz(reg2, found_null);
1326 
1327       // Put all the "Case 3" tests here.
1328       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1329         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1330 
1331       // Found a null.  Keep searching for a matching item,
1332       // but remember that this is an empty (unused) slot.
1333       bind(found_null);
1334     }
1335   }
1336 
1337   // In the fall-through case, we found no matching item, but we
1338   // observed the item[start_row] is NULL.
1339 
1340   // Fill in the item field and increment the count.
1341   int item_offset = in_bytes(item_offset_fn(start_row));
1342   set_mdp_data_at(mdp, item_offset, item);
1343   int count_offset = in_bytes(item_count_offset_fn(start_row));
1344   mov(reg2, DataLayout::counter_increment);
1345   set_mdp_data_at(mdp, count_offset, reg2);
1346   if (start_row > 0) {
1347     b(done);
1348   }
1349 }
1350 
1351 // Example state machine code for three profile rows:
1352 //   // main copy of decision tree, rooted at row[1]
1353 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1354 //   if (row[0].rec != NULL) {
1355 //     // inner copy of decision tree, rooted at row[1]
1356 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1357 //     if (row[1].rec != NULL) {
1358 //       // degenerate decision tree, rooted at row[2]
1359 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1360 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1361 //       row[2].init(rec); goto done;
1362 //     } else {
1363 //       // remember row[1] is empty
1364 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1365 //       row[1].init(rec); goto done;
1366 //     }
1367 //   } else {
1368 //     // remember row[0] is empty
1369 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1370 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1371 //     row[0].init(rec); goto done;
1372 //   }
1373 //   done:
1374 
1375 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1376                                                         Register mdp, Register reg2,
1377                                                         bool is_virtual_call) {
1378   assert(ProfileInterpreter, "must be profiling");
1379   Label done;
1380 
1381   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1382 
1383   bind (done);
1384 }
1385 
1386 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1387                                             Register mdp) {
1388   if (ProfileInterpreter) {
1389     Label profile_continue;
1390     uint row;
1391 
1392     // If no method data exists, go to profile_continue.
1393     test_method_data_pointer(mdp, profile_continue);
1394 
1395     // Update the total ret count.
1396     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1397 
1398     for (row = 0; row < RetData::row_limit(); row++) {
1399       Label next_test;
1400 
1401       // See if return_bci is equal to bci[n]:
1402       test_mdp_data_at(mdp,
1403                        in_bytes(RetData::bci_offset(row)),
1404                        return_bci, noreg,
1405                        next_test);
1406 
1407       // return_bci is equal to bci[n].  Increment the count.
1408       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1409 
1410       // The method data pointer needs to be updated to reflect the new target.
1411       update_mdp_by_offset(mdp,
1412                            in_bytes(RetData::bci_displacement_offset(row)));
1413       b(profile_continue);
1414       bind(next_test);
1415     }
1416 
1417     update_mdp_for_ret(return_bci);
1418 
1419     bind(profile_continue);
1420   }
1421 }
1422 
1423 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1424   if (ProfileInterpreter) {
1425     Label profile_continue;
1426 
1427     // If no method data exists, go to profile_continue.
1428     test_method_data_pointer(mdp, profile_continue);
1429 
1430     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1431 
1432     // The method data pointer needs to be updated.
1433     int mdp_delta = in_bytes(BitData::bit_data_size());
1434     if (TypeProfileCasts) {
1435       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1436     }
1437     update_mdp_by_constant(mdp, mdp_delta);
1438 
1439     bind(profile_continue);
1440   }
1441 }
1442 
1443 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1444   if (ProfileInterpreter && TypeProfileCasts) {
1445     Label profile_continue;
1446 
1447     // If no method data exists, go to profile_continue.
1448     test_method_data_pointer(mdp, profile_continue);
1449 
1450     int count_offset = in_bytes(CounterData::count_offset());
1451     // Back up the address, since we have already bumped the mdp.
1452     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1453 
1454     // *Decrement* the counter.  We expect to see zero or small negatives.
1455     increment_mdp_data_at(mdp, count_offset, true);
1456 
1457     bind (profile_continue);
1458   }
1459 }
1460 
1461 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1462   if (ProfileInterpreter) {
1463     Label profile_continue;
1464 
1465     // If no method data exists, go to profile_continue.
1466     test_method_data_pointer(mdp, profile_continue);
1467 
1468     // The method data pointer needs to be updated.
1469     int mdp_delta = in_bytes(BitData::bit_data_size());
1470     if (TypeProfileCasts) {
1471       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1472 
1473       // Record the object type.
1474       record_klass_in_profile(klass, mdp, reg2, false);
1475     }
1476     update_mdp_by_constant(mdp, mdp_delta);
1477 
1478     bind(profile_continue);
1479   }
1480 }
1481 
1482 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1483   if (ProfileInterpreter) {
1484     Label profile_continue;
1485 
1486     // If no method data exists, go to profile_continue.
1487     test_method_data_pointer(mdp, profile_continue);
1488 
1489     // Update the default case count
1490     increment_mdp_data_at(mdp,
1491                           in_bytes(MultiBranchData::default_count_offset()));
1492 
1493     // The method data pointer needs to be updated.
1494     update_mdp_by_offset(mdp,
1495                          in_bytes(MultiBranchData::
1496                                   default_displacement_offset()));
1497 
1498     bind(profile_continue);
1499   }
1500 }
1501 
1502 void InterpreterMacroAssembler::profile_switch_case(Register index,
1503                                                     Register mdp,
1504                                                     Register reg2) {
1505   if (ProfileInterpreter) {
1506     Label profile_continue;
1507 
1508     // If no method data exists, go to profile_continue.
1509     test_method_data_pointer(mdp, profile_continue);
1510 
1511     // Build the base (index * per_case_size_in_bytes()) +
1512     // case_array_offset_in_bytes()
1513     movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1514     movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1515     Assembler::maddw(index, index, reg2, rscratch1);
1516 
1517     // Update the case count
1518     increment_mdp_data_at(mdp,
1519                           index,
1520                           in_bytes(MultiBranchData::relative_count_offset()));
1521 
1522     // The method data pointer needs to be updated.
1523     update_mdp_by_offset(mdp,
1524                          index,
1525                          in_bytes(MultiBranchData::
1526                                   relative_displacement_offset()));
1527 
1528     bind(profile_continue);
1529   }
1530 }
1531 
1532 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1533   if (state == atos) {
1534     MacroAssembler::verify_oop(reg);
1535   }
1536 }
1537 
1538 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1539 
1540 
1541 void InterpreterMacroAssembler::notify_method_entry() {
1542   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1543   // track stack depth.  If it is possible to enter interp_only_mode we add
1544   // the code to check if the event should be sent.
1545   if (JvmtiExport::can_post_interpreter_events()) {
1546     Label L;
1547     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1548     cbzw(r3, L);
1549     call_VM(noreg, CAST_FROM_FN_PTR(address,
1550                                     InterpreterRuntime::post_method_entry));
1551     bind(L);
1552   }
1553 
1554   {
1555     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1556     get_method(c_rarg1);
1557     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1558                  rthread, c_rarg1);
1559   }
1560 
1561   // RedefineClasses() tracing support for obsolete method entry
1562   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1563     get_method(c_rarg1);
1564     call_VM_leaf(
1565       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1566       rthread, c_rarg1);
1567   }
1568 
1569  }
1570 
1571 
1572 void InterpreterMacroAssembler::notify_method_exit(
1573     TosState state, NotifyMethodExitMode mode) {
1574   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1575   // track stack depth.  If it is possible to enter interp_only_mode we add
1576   // the code to check if the event should be sent.
1577   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1578     Label L;
1579     // Note: frame::interpreter_frame_result has a dependency on how the
1580     // method result is saved across the call to post_method_exit. If this
1581     // is changed then the interpreter_frame_result implementation will
1582     // need to be updated too.
1583 
1584     // template interpreter will leave the result on the top of the stack.
1585     push(state);
1586     ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1587     cbz(r3, L);
1588     call_VM(noreg,
1589             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1590     bind(L);
1591     pop(state);
1592   }
1593 
1594   {
1595     SkipIfEqual skip(this, &DTraceMethodProbes, false);
1596     push(state);
1597     get_method(c_rarg1);
1598     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1599                  rthread, c_rarg1);
1600     pop(state);
1601   }
1602 }
1603 
1604 
1605 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1606 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1607                                                         int increment, Address mask,
1608                                                         Register scratch, Register scratch2,
1609                                                         bool preloaded, Condition cond,
1610                                                         Label* where) {
1611   if (!preloaded) {
1612     ldrw(scratch, counter_addr);
1613   }
1614   add(scratch, scratch, increment);
1615   strw(scratch, counter_addr);
1616   ldrw(scratch2, mask);
1617   ands(scratch, scratch, scratch2);
1618   br(cond, *where);
1619 }
1620 
1621 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1622                                                   int number_of_arguments) {
1623   // interpreter specific
1624   //
1625   // Note: No need to save/restore rbcp & rlocals pointer since these
1626   //       are callee saved registers and no blocking/ GC can happen
1627   //       in leaf calls.
1628 #ifdef ASSERT
1629   {
1630     Label L;
1631     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1632     cbz(rscratch1, L);
1633     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1634          " last_sp != NULL");
1635     bind(L);
1636   }
1637 #endif /* ASSERT */
1638   // super call
1639   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1640 }
1641 
1642 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1643                                              Register java_thread,
1644                                              Register last_java_sp,
1645                                              address  entry_point,
1646                                              int      number_of_arguments,
1647                                              bool     check_exceptions) {
1648   // interpreter specific
1649   //
1650   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1651   //       really make a difference for these runtime calls, since they are
1652   //       slow anyway. Btw., bcp must be saved/restored since it may change
1653   //       due to GC.
1654   // assert(java_thread == noreg , "not expecting a precomputed java thread");
1655   save_bcp();
1656 #ifdef ASSERT
1657   {
1658     Label L;
1659     ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1660     cbz(rscratch1, L);
1661     stop("InterpreterMacroAssembler::call_VM_base:"
1662          " last_sp != NULL");
1663     bind(L);
1664   }
1665 #endif /* ASSERT */
1666   // super call
1667   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1668                                entry_point, number_of_arguments,
1669                      check_exceptions);
1670 // interpreter specific
1671   restore_bcp();
1672   restore_locals();
1673 }
1674 
1675 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1676   assert_different_registers(obj, rscratch1);
1677   Label update, next, none;
1678 
1679   verify_oop(obj);
1680 
1681   cbnz(obj, update);
1682   orptr(mdo_addr, TypeEntries::null_seen);
1683   b(next);
1684 
1685   bind(update);
1686   load_klass(obj, obj);
1687 
1688   ldr(rscratch1, mdo_addr);
1689   eor(obj, obj, rscratch1);
1690   tst(obj, TypeEntries::type_klass_mask);
1691   br(Assembler::EQ, next); // klass seen before, nothing to
1692                            // do. The unknown bit may have been
1693                            // set already but no need to check.
1694 
1695   tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1696   // already unknown. Nothing to do anymore.
1697 
1698   ldr(rscratch1, mdo_addr);
1699   cbz(rscratch1, none);
1700   cmp(rscratch1, (u1)TypeEntries::null_seen);
1701   br(Assembler::EQ, none);
1702   // There is a chance that the checks above (re-reading profiling
1703   // data from memory) fail if another thread has just set the
1704   // profiling to this obj's klass
1705   ldr(rscratch1, mdo_addr);
1706   eor(obj, obj, rscratch1);
1707   tst(obj, TypeEntries::type_klass_mask);
1708   br(Assembler::EQ, next);
1709 
1710   // different than before. Cannot keep accurate profile.
1711   orptr(mdo_addr, TypeEntries::type_unknown);
1712   b(next);
1713 
1714   bind(none);
1715   // first time here. Set profile type.
1716   str(obj, mdo_addr);
1717 
1718   bind(next);
1719 }
1720 
1721 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1722   if (!ProfileInterpreter) {
1723     return;
1724   }
1725 
1726   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1727     Label profile_continue;
1728 
1729     test_method_data_pointer(mdp, profile_continue);
1730 
1731     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1732 
1733     ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1734     cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1735     br(Assembler::NE, profile_continue);
1736 
1737     if (MethodData::profile_arguments()) {
1738       Label done;
1739       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1740 
1741       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1742         if (i > 0 || MethodData::profile_return()) {
1743           // If return value type is profiled we may have no argument to profile
1744           ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1745           sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1746           cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1747           add(rscratch1, mdp, off_to_args);
1748           br(Assembler::LT, done);
1749         }
1750         ldr(tmp, Address(callee, Method::const_offset()));
1751         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1752         // stack offset o (zero based) from the start of the argument
1753         // list, for n arguments translates into offset n - o - 1 from
1754         // the end of the argument list
1755         ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1756         sub(tmp, tmp, rscratch1);
1757         sub(tmp, tmp, 1);
1758         Address arg_addr = argument_address(tmp);
1759         ldr(tmp, arg_addr);
1760 
1761         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1762         profile_obj_type(tmp, mdo_arg_addr);
1763 
1764         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1765         off_to_args += to_add;
1766       }
1767 
1768       if (MethodData::profile_return()) {
1769         ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1770         sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1771       }
1772 
1773       add(rscratch1, mdp, off_to_args);
1774       bind(done);
1775       mov(mdp, rscratch1);
1776 
1777       if (MethodData::profile_return()) {
1778         // We're right after the type profile for the last
1779         // argument. tmp is the number of cells left in the
1780         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1781         // if there's a return to profile.
1782         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1783         add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1784       }
1785       str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1786     } else {
1787       assert(MethodData::profile_return(), "either profile call args or call ret");
1788       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1789     }
1790 
1791     // mdp points right after the end of the
1792     // CallTypeData/VirtualCallTypeData, right after the cells for the
1793     // return value type if there's one
1794 
1795     bind(profile_continue);
1796   }
1797 }
1798 
1799 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1800   assert_different_registers(mdp, ret, tmp, rbcp);
1801   if (ProfileInterpreter && MethodData::profile_return()) {
1802     Label profile_continue, done;
1803 
1804     test_method_data_pointer(mdp, profile_continue);
1805 
1806     if (MethodData::profile_return_jsr292_only()) {
1807       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1808 
1809       // If we don't profile all invoke bytecodes we must make sure
1810       // it's a bytecode we indeed profile. We can't go back to the
1811       // begining of the ProfileData we intend to update to check its
1812       // type because we're right after it and we don't known its
1813       // length
1814       Label do_profile;
1815       ldrb(rscratch1, Address(rbcp, 0));
1816       cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1817       br(Assembler::EQ, do_profile);
1818       cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1819       br(Assembler::EQ, do_profile);
1820       get_method(tmp);
1821       ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1822       subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1823       br(Assembler::NE, profile_continue);
1824 
1825       bind(do_profile);
1826     }
1827 
1828     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1829     mov(tmp, ret);
1830     profile_obj_type(tmp, mdo_ret_addr);
1831 
1832     bind(profile_continue);
1833   }
1834 }
1835 
1836 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1837   assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1838   if (ProfileInterpreter && MethodData::profile_parameters()) {
1839     Label profile_continue, done;
1840 
1841     test_method_data_pointer(mdp, profile_continue);
1842 
1843     // Load the offset of the area within the MDO used for
1844     // parameters. If it's negative we're not profiling any parameters
1845     ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1846     tbnz(tmp1, 31, profile_continue);  // i.e. sign bit set
1847 
1848     // Compute a pointer to the area for parameters from the offset
1849     // and move the pointer to the slot for the last
1850     // parameters. Collect profiling from last parameter down.
1851     // mdo start + parameters offset + array length - 1
1852     add(mdp, mdp, tmp1);
1853     ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1854     sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1855 
1856     Label loop;
1857     bind(loop);
1858 
1859     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1860     int type_base = in_bytes(ParametersTypeData::type_offset(0));
1861     int per_arg_scale = exact_log2(DataLayout::cell_size);
1862     add(rscratch1, mdp, off_base);
1863     add(rscratch2, mdp, type_base);
1864 
1865     Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1866     Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1867 
1868     // load offset on the stack from the slot for this parameter
1869     ldr(tmp2, arg_off);
1870     neg(tmp2, tmp2);
1871     // read the parameter from the local area
1872     ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1873 
1874     // profile the parameter
1875     profile_obj_type(tmp2, arg_type);
1876 
1877     // go to next parameter
1878     subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1879     br(Assembler::GE, loop);
1880 
1881     bind(profile_continue);
1882   }
1883 }