1 /*
   2  * Copyright (c) 2008, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.inline.hpp"
  27 #include "gc/shared/barrierSet.hpp"
  28 #include "gc/shared/cardTable.hpp"
  29 #include "gc/shared/cardTableBarrierSet.inline.hpp"
  30 #include "gc/shared/collectedHeap.hpp"
  31 #include "interp_masm_arm.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "jvm.h"
  35 #include "logging/log.hpp"
  36 #include "oops/arrayOop.hpp"
  37 #include "oops/markWord.hpp"
  38 #include "oops/method.hpp"
  39 #include "oops/methodData.hpp"
  40 #include "oops/resolvedFieldEntry.hpp"
  41 #include "oops/resolvedIndyEntry.hpp"
  42 #include "oops/resolvedMethodEntry.hpp"
  43 #include "prims/jvmtiExport.hpp"
  44 #include "prims/jvmtiThreadState.hpp"
  45 #include "runtime/basicLock.hpp"
  46 #include "runtime/frame.inline.hpp"
  47 #include "runtime/safepointMechanism.hpp"
  48 #include "runtime/sharedRuntime.hpp"
  49 #include "utilities/powerOfTwo.hpp"
  50 
  51 //--------------------------------------------------------------------
  52 // Implementation of InterpreterMacroAssembler
  53 
  54 
  55 
  56 
  57 InterpreterMacroAssembler::InterpreterMacroAssembler(CodeBuffer* code) : MacroAssembler(code) {
  58 }
  59 
  60 void InterpreterMacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) {
  61 #ifdef ASSERT
  62   // Ensure that last_sp is not filled.
  63   { Label L;
  64     ldr(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
  65     cbz(Rtemp, L);
  66     stop("InterpreterMacroAssembler::call_VM_helper: last_sp != nullptr");
  67     bind(L);
  68   }
  69 #endif // ASSERT
  70 
  71   // Rbcp must be saved/restored since it may change due to GC.
  72   save_bcp();
  73 
  74 
  75   // super call
  76   MacroAssembler::call_VM_helper(oop_result, entry_point, number_of_arguments, check_exceptions);
  77 
  78 
  79   // Restore interpreter specific registers.
  80   restore_bcp();
  81   restore_method();
  82 }
  83 
  84 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  85   assert(entry, "Entry must have been generated by now");
  86   b(entry);
  87 }
  88 
  89 void InterpreterMacroAssembler::check_and_handle_popframe() {
  90   if (can_pop_frame()) {
  91     Label L;
  92     const Register popframe_cond = R2_tmp;
  93 
  94     // Initiate popframe handling only if it is not already being processed.  If the flag
  95     // has the popframe_processing bit set, it means that this code is called *during* popframe
  96     // handling - we don't want to reenter.
  97 
  98     ldr_s32(popframe_cond, Address(Rthread, JavaThread::popframe_condition_offset()));
  99     tbz(popframe_cond, exact_log2(JavaThread::popframe_pending_bit), L);
 100     tbnz(popframe_cond, exact_log2(JavaThread::popframe_processing_bit), L);
 101 
 102     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 103     // address of the same-named entrypoint in the generated interpreter code.
 104     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 105 
 106     // Call indirectly to avoid generation ordering problem.
 107     jump(R0);
 108 
 109     bind(L);
 110   }
 111 }
 112 
 113 
 114 // Blows R2, Rtemp. Sets TOS cached value.
 115 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 116   const Register thread_state = R2_tmp;
 117 
 118   ldr(thread_state, Address(Rthread, JavaThread::jvmti_thread_state_offset()));
 119 
 120   const Address tos_addr(thread_state, JvmtiThreadState::earlyret_tos_offset());
 121   const Address oop_addr(thread_state, JvmtiThreadState::earlyret_oop_offset());
 122   const Address val_addr(thread_state, JvmtiThreadState::earlyret_value_offset());
 123   const Address val_addr_hi(thread_state, JvmtiThreadState::earlyret_value_offset()
 124                              + in_ByteSize(wordSize));
 125 
 126   Register zero = zero_register(Rtemp);
 127 
 128   switch (state) {
 129     case atos: ldr(R0_tos, oop_addr);
 130                str(zero, oop_addr);
 131                interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
 132                break;
 133 
 134     case ltos: ldr(R1_tos_hi, val_addr_hi);        // fall through
 135     case btos:                                     // fall through
 136     case ztos:                                     // fall through
 137     case ctos:                                     // fall through
 138     case stos:                                     // fall through
 139     case itos: ldr_s32(R0_tos, val_addr);          break;
 140 #ifdef __SOFTFP__
 141     case dtos: ldr(R1_tos_hi, val_addr_hi);        // fall through
 142     case ftos: ldr(R0_tos, val_addr);              break;
 143 #else
 144     case ftos: ldr_float (S0_tos, val_addr);       break;
 145     case dtos: ldr_double(D0_tos, val_addr);       break;
 146 #endif // __SOFTFP__
 147     case vtos: /* nothing to do */                 break;
 148     default  : ShouldNotReachHere();
 149   }
 150   // Clean up tos value in the thread object
 151   str(zero, val_addr);
 152   str(zero, val_addr_hi);
 153 
 154   mov(Rtemp, (int) ilgl);
 155   str_32(Rtemp, tos_addr);
 156 }
 157 
 158 
 159 // Blows R2, Rtemp.
 160 void InterpreterMacroAssembler::check_and_handle_earlyret() {
 161   if (can_force_early_return()) {
 162     Label L;
 163     const Register thread_state = R2_tmp;
 164 
 165     ldr(thread_state, Address(Rthread, JavaThread::jvmti_thread_state_offset()));
 166     cbz(thread_state, L); // if (thread->jvmti_thread_state() == nullptr) exit;
 167 
 168     // Initiate earlyret handling only if it is not already being processed.
 169     // If the flag has the earlyret_processing bit set, it means that this code
 170     // is called *during* earlyret handling - we don't want to reenter.
 171 
 172     ldr_s32(Rtemp, Address(thread_state, JvmtiThreadState::earlyret_state_offset()));
 173     cmp(Rtemp, JvmtiThreadState::earlyret_pending);
 174     b(L, ne);
 175 
 176     // Call Interpreter::remove_activation_early_entry() to get the address of the
 177     // same-named entrypoint in the generated interpreter code.
 178 
 179     ldr_s32(R0, Address(thread_state, JvmtiThreadState::earlyret_tos_offset()));
 180     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), R0);
 181 
 182     jump(R0);
 183 
 184     bind(L);
 185   }
 186 }
 187 
 188 
 189 // Sets reg. Blows Rtemp.
 190 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 191   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 192   assert(reg != Rtemp, "should be different registers");
 193 
 194   ldrb(Rtemp, Address(Rbcp, bcp_offset));
 195   ldrb(reg, Address(Rbcp, bcp_offset+1));
 196   orr(reg, reg, AsmOperand(Rtemp, lsl, BitsPerByte));
 197 }
 198 
 199 void InterpreterMacroAssembler::get_index_at_bcp(Register index, int bcp_offset, Register tmp_reg, size_t index_size) {
 200   assert_different_registers(index, tmp_reg);
 201   if (index_size == sizeof(u2)) {
 202     // load bytes of index separately to avoid unaligned access
 203     ldrb(index, Address(Rbcp, bcp_offset+1));
 204     ldrb(tmp_reg, Address(Rbcp, bcp_offset));
 205     orr(index, tmp_reg, AsmOperand(index, lsl, BitsPerByte));
 206   } else if (index_size == sizeof(u4)) {
 207     ldrb(index, Address(Rbcp, bcp_offset+3));
 208     ldrb(tmp_reg, Address(Rbcp, bcp_offset+2));
 209     orr(index, tmp_reg, AsmOperand(index, lsl, BitsPerByte));
 210     ldrb(tmp_reg, Address(Rbcp, bcp_offset+1));
 211     orr(index, tmp_reg, AsmOperand(index, lsl, BitsPerByte));
 212     ldrb(tmp_reg, Address(Rbcp, bcp_offset));
 213     orr(index, tmp_reg, AsmOperand(index, lsl, BitsPerByte));
 214   } else if (index_size == sizeof(u1)) {
 215     ldrb(index, Address(Rbcp, bcp_offset));
 216   } else {
 217     ShouldNotReachHere();
 218   }
 219 }
 220 
 221 // Load object from cpool->resolved_references(index)
 222 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 223                                            Register result, Register index) {
 224   assert_different_registers(result, index);
 225   get_constant_pool(result);
 226 
 227   Register cache = result;
 228   // load pointer for resolved_references[] objArray
 229   ldr(cache, Address(result, ConstantPool::cache_offset()));
 230   ldr(cache, Address(result, ConstantPoolCache::resolved_references_offset()));
 231   resolve_oop_handle(cache);
 232   // Add in the index
 233   // convert from field index to resolved_references() index and from
 234   // word index to byte offset. Since this is a java object, it can be compressed
 235   logical_shift_left(index, index, LogBytesPerHeapOop);
 236   add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
 237   load_heap_oop(result, Address(cache, index));
 238 }
 239 
 240 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
 241                                            Register Rcpool, Register Rindex, Register Rklass) {
 242   add(Rtemp, Rcpool, AsmOperand(Rindex, lsl, LogBytesPerWord));
 243   ldrh(Rtemp, Address(Rtemp, sizeof(ConstantPool))); // Rtemp = resolved_klass_index
 244   ldr(Rklass, Address(Rcpool,  ConstantPool::resolved_klasses_offset())); // Rklass = cpool->_resolved_klasses
 245   add(Rklass, Rklass, AsmOperand(Rtemp, lsl, LogBytesPerWord));
 246   ldr(Rklass, Address(Rklass, Array<Klass*>::base_offset_in_bytes()));
 247 }
 248 
 249 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
 250   // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
 251   assert_different_registers(cache, index, Rtemp);
 252 
 253   get_index_at_bcp(index, 1, Rtemp, sizeof(u4));
 254 
 255   // load constant pool cache pointer
 256   ldr(cache, Address(FP, frame::interpreter_frame_cache_offset * wordSize));
 257 
 258   // Get address of invokedynamic array
 259   ldr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
 260 
 261   // Scale the index to be the entry index * sizeof(ResolvedInvokeDynamicInfo)
 262   // On ARM32 sizeof(ResolvedIndyEntry) is 12, use mul instead of lsl
 263   mov(Rtemp, sizeof(ResolvedIndyEntry));
 264   mul(index, index, Rtemp);
 265 
 266   add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
 267   add(cache, cache, index);
 268 }
 269 
 270 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
 271   // Get index out of bytecode pointer
 272   assert_different_registers(cache, index);
 273 
 274   get_index_at_bcp(index, bcp_offset, cache /*as tmp*/, sizeof(u2));
 275 
 276   // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
 277   // sizeof(ResolvedFieldEntry) is 16 on Arm, so using shift
 278   if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
 279     // load constant pool cache pointer
 280     ldr(cache, Address(FP, frame::interpreter_frame_cache_offset * wordSize));
 281     // Get address of field entries array
 282     ldr(cache, Address(cache, in_bytes(ConstantPoolCache::field_entries_offset())));
 283 
 284     add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
 285     add(cache, cache, AsmOperand(index, lsl, log2i_exact(sizeof(ResolvedFieldEntry))));
 286   }
 287   else {
 288     mov(cache, sizeof(ResolvedFieldEntry));
 289     mul(index, index, cache);
 290     // load constant pool cache pointer
 291     ldr(cache, Address(FP, frame::interpreter_frame_cache_offset * wordSize));
 292 
 293     // Get address of field entries array
 294     ldr(cache, Address(cache, in_bytes(ConstantPoolCache::field_entries_offset())));
 295     add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
 296     add(cache, cache, index);
 297   }
 298 }
 299 
 300 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
 301   assert_different_registers(cache, index);
 302 
 303   // Get index out of bytecode pointer
 304   get_index_at_bcp(index, bcp_offset, cache /* as tmp */, sizeof(u2));
 305 
 306   // sizeof(ResolvedMethodEntry) is not a power of 2 on Arm, so can't use shift
 307   mov(cache, sizeof(ResolvedMethodEntry));
 308   mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
 309 
 310   // load constant pool cache pointer
 311   ldr(cache, Address(FP, frame::interpreter_frame_cache_offset * wordSize));
 312   // Get address of method entries array
 313   ldr(cache, Address(cache, in_bytes(ConstantPoolCache::method_entries_offset())));
 314   add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
 315   add(cache, cache, index);
 316 }
 317 
 318 // Generate a subtype check: branch to not_subtype if sub_klass is
 319 // not a subtype of super_klass.
 320 // Profiling code for the subtype check failure (profile_typecheck_failed)
 321 // should be explicitly generated by the caller in the not_subtype case.
 322 // Blows Rtemp, tmp1, tmp2.
 323 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 324                                                   Register Rsuper_klass,
 325                                                   Label &not_subtype,
 326                                                   Register tmp1,
 327                                                   Register tmp2) {
 328 
 329   assert_different_registers(Rsub_klass, Rsuper_klass, tmp1, tmp2, Rtemp);
 330   Label ok_is_subtype, loop, update_cache;
 331 
 332   const Register super_check_offset = tmp1;
 333   const Register cached_super = tmp2;
 334 
 335   // Profile the not-null value's klass.
 336   profile_typecheck(tmp1, Rsub_klass);
 337 
 338   // Load the super-klass's check offset into
 339   ldr_u32(super_check_offset, Address(Rsuper_klass, Klass::super_check_offset_offset()));
 340 
 341   // Check for self
 342   cmp(Rsub_klass, Rsuper_klass);
 343 
 344   // Load from the sub-klass's super-class display list, or a 1-word cache of
 345   // the secondary superclass list, or a failing value with a sentinel offset
 346   // if the super-klass is an interface or exceptionally deep in the Java
 347   // hierarchy and we have to scan the secondary superclass list the hard way.
 348   // See if we get an immediate positive hit
 349   ldr(cached_super, Address(Rsub_klass, super_check_offset));
 350 
 351   cond_cmp(Rsuper_klass, cached_super, ne);
 352   b(ok_is_subtype, eq);
 353 
 354   // Check for immediate negative hit
 355   cmp(super_check_offset, in_bytes(Klass::secondary_super_cache_offset()));
 356   b(not_subtype, ne);
 357 
 358   // Now do a linear scan of the secondary super-klass chain.
 359   const Register supers_arr = tmp1;
 360   const Register supers_cnt = tmp2;
 361   const Register cur_super  = Rtemp;
 362 
 363   // Load objArrayOop of secondary supers.
 364   ldr(supers_arr, Address(Rsub_klass, Klass::secondary_supers_offset()));
 365 
 366   ldr_u32(supers_cnt, Address(supers_arr, Array<Klass*>::length_offset_in_bytes())); // Load the array length
 367   cmp(supers_cnt, 0);
 368 
 369   // Skip to the start of array elements and prefetch the first super-klass.
 370   ldr(cur_super, Address(supers_arr, Array<Klass*>::base_offset_in_bytes(), pre_indexed), ne);
 371   b(not_subtype, eq);
 372 
 373   bind(loop);
 374 
 375 
 376   cmp(cur_super, Rsuper_klass);
 377   b(update_cache, eq);
 378 
 379   subs(supers_cnt, supers_cnt, 1);
 380 
 381   ldr(cur_super, Address(supers_arr, wordSize, pre_indexed), ne);
 382 
 383   b(loop, ne);
 384 
 385   b(not_subtype);
 386 
 387   bind(update_cache);
 388   // Must be equal but missed in cache.  Update cache.
 389   str(Rsuper_klass, Address(Rsub_klass, Klass::secondary_super_cache_offset()));
 390 
 391   bind(ok_is_subtype);
 392 }
 393 
 394 
 395 //////////////////////////////////////////////////////////////////////////////////
 396 
 397 
 398 // Java Expression Stack
 399 
 400 void InterpreterMacroAssembler::pop_ptr(Register r) {
 401   assert(r != Rstack_top, "unpredictable instruction");
 402   ldr(r, Address(Rstack_top, wordSize, post_indexed));
 403 }
 404 
 405 void InterpreterMacroAssembler::pop_i(Register r) {
 406   assert(r != Rstack_top, "unpredictable instruction");
 407   ldr_s32(r, Address(Rstack_top, wordSize, post_indexed));
 408   zap_high_non_significant_bits(r);
 409 }
 410 
 411 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
 412   assert_different_registers(lo, hi);
 413   assert(lo < hi, "lo must be < hi");
 414   pop(RegisterSet(lo) | RegisterSet(hi));
 415 }
 416 
 417 void InterpreterMacroAssembler::pop_f(FloatRegister fd) {
 418   fpops(fd);
 419 }
 420 
 421 void InterpreterMacroAssembler::pop_d(FloatRegister fd) {
 422   fpopd(fd);
 423 }
 424 
 425 
 426 // Transition vtos -> state. Blows R0, R1. Sets TOS cached value.
 427 void InterpreterMacroAssembler::pop(TosState state) {
 428   switch (state) {
 429     case atos: pop_ptr(R0_tos);                              break;
 430     case btos:                                               // fall through
 431     case ztos:                                               // fall through
 432     case ctos:                                               // fall through
 433     case stos:                                               // fall through
 434     case itos: pop_i(R0_tos);                                break;
 435     case ltos: pop_l(R0_tos_lo, R1_tos_hi);                  break;
 436 #ifdef __SOFTFP__
 437     case ftos: pop_i(R0_tos);                                break;
 438     case dtos: pop_l(R0_tos_lo, R1_tos_hi);                  break;
 439 #else
 440     case ftos: pop_f(S0_tos);                                break;
 441     case dtos: pop_d(D0_tos);                                break;
 442 #endif // __SOFTFP__
 443     case vtos: /* nothing to do */                           break;
 444     default  : ShouldNotReachHere();
 445   }
 446   interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
 447 }
 448 
 449 void InterpreterMacroAssembler::push_ptr(Register r) {
 450   assert(r != Rstack_top, "unpredictable instruction");
 451   str(r, Address(Rstack_top, -wordSize, pre_indexed));
 452   check_stack_top_on_expansion();
 453 }
 454 
 455 void InterpreterMacroAssembler::push_i(Register r) {
 456   assert(r != Rstack_top, "unpredictable instruction");
 457   str_32(r, Address(Rstack_top, -wordSize, pre_indexed));
 458   check_stack_top_on_expansion();
 459 }
 460 
 461 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
 462   assert_different_registers(lo, hi);
 463   assert(lo < hi, "lo must be < hi");
 464   push(RegisterSet(lo) | RegisterSet(hi));
 465 }
 466 
 467 void InterpreterMacroAssembler::push_f() {
 468   fpushs(S0_tos);
 469 }
 470 
 471 void InterpreterMacroAssembler::push_d() {
 472   fpushd(D0_tos);
 473 }
 474 
 475 // Transition state -> vtos. Blows Rtemp.
 476 void InterpreterMacroAssembler::push(TosState state) {
 477   interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
 478   switch (state) {
 479     case atos: push_ptr(R0_tos);                              break;
 480     case btos:                                                // fall through
 481     case ztos:                                                // fall through
 482     case ctos:                                                // fall through
 483     case stos:                                                // fall through
 484     case itos: push_i(R0_tos);                                break;
 485     case ltos: push_l(R0_tos_lo, R1_tos_hi);                  break;
 486 #ifdef __SOFTFP__
 487     case ftos: push_i(R0_tos);                                break;
 488     case dtos: push_l(R0_tos_lo, R1_tos_hi);                  break;
 489 #else
 490     case ftos: push_f();                                      break;
 491     case dtos: push_d();                                      break;
 492 #endif // __SOFTFP__
 493     case vtos: /* nothing to do */                            break;
 494     default  : ShouldNotReachHere();
 495   }
 496 }
 497 
 498 
 499 
 500 // Converts return value in R0/R1 (interpreter calling conventions) to TOS cached value.
 501 void InterpreterMacroAssembler::convert_retval_to_tos(TosState state) {
 502 #if (!defined __SOFTFP__ && !defined __ABI_HARD__)
 503   // According to interpreter calling conventions, result is returned in R0/R1,
 504   // but templates expect ftos in S0, and dtos in D0.
 505   if (state == ftos) {
 506     fmsr(S0_tos, R0);
 507   } else if (state == dtos) {
 508     fmdrr(D0_tos, R0, R1);
 509   }
 510 #endif // !__SOFTFP__ && !__ABI_HARD__
 511 }
 512 
 513 // Converts TOS cached value to return value in R0/R1 (according to interpreter calling conventions).
 514 void InterpreterMacroAssembler::convert_tos_to_retval(TosState state) {
 515 #if (!defined __SOFTFP__ && !defined __ABI_HARD__)
 516   // According to interpreter calling conventions, result is returned in R0/R1,
 517   // so ftos (S0) and dtos (D0) are moved to R0/R1.
 518   if (state == ftos) {
 519     fmrs(R0, S0_tos);
 520   } else if (state == dtos) {
 521     fmrrd(R0, R1, D0_tos);
 522   }
 523 #endif // !__SOFTFP__ && !__ABI_HARD__
 524 }
 525 
 526 
 527 
 528 // Helpers for swap and dup
 529 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 530   ldr(val, Address(Rstack_top, Interpreter::expr_offset_in_bytes(n)));
 531 }
 532 
 533 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 534   str(val, Address(Rstack_top, Interpreter::expr_offset_in_bytes(n)));
 535 }
 536 
 537 
 538 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 539 
 540   // set sender sp
 541   mov(Rsender_sp, SP);
 542 
 543   // record last_sp
 544   str(Rsender_sp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
 545 }
 546 
 547 // Jump to from_interpreted entry of a call unless single stepping is possible
 548 // in this thread in which case we must call the i2i entry
 549 void InterpreterMacroAssembler::jump_from_interpreted(Register method) {
 550   assert_different_registers(method, Rtemp);
 551 
 552   prepare_to_jump_from_interpreted();
 553 
 554   if (can_post_interpreter_events()) {
 555     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 556     // compiled code in threads for which the event is enabled.  Check here for
 557     // interp_only_mode if these events CAN be enabled.
 558 
 559     ldr_s32(Rtemp, Address(Rthread, JavaThread::interp_only_mode_offset()));
 560     cmp(Rtemp, 0);
 561     ldr(PC, Address(method, Method::interpreter_entry_offset()), ne);
 562   }
 563 
 564   indirect_jump(Address(method, Method::from_interpreted_offset()), Rtemp);
 565 }
 566 
 567 
 568 void InterpreterMacroAssembler::restore_dispatch() {
 569   mov_slow(RdispatchTable, (address)Interpreter::dispatch_table(vtos));
 570 }
 571 
 572 
 573 // The following two routines provide a hook so that an implementation
 574 // can schedule the dispatch in two parts.
 575 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 576   // Nothing ARM-specific to be done here.
 577 }
 578 
 579 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 580   dispatch_next(state, step);
 581 }
 582 
 583 void InterpreterMacroAssembler::dispatch_base(TosState state,
 584                                               DispatchTableMode table_mode,
 585                                               bool verifyoop, bool generate_poll) {
 586   if (VerifyActivationFrameSize) {
 587     Label L;
 588     sub(Rtemp, FP, SP);
 589     int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
 590     cmp(Rtemp, min_frame_size);
 591     b(L, ge);
 592     stop("broken stack frame");
 593     bind(L);
 594   }
 595 
 596   if (verifyoop) {
 597     interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
 598   }
 599 
 600   Label safepoint;
 601   address* const safepoint_table = Interpreter::safept_table(state);
 602   address* const table           = Interpreter::dispatch_table(state);
 603   bool needs_thread_local_poll = generate_poll && table != safepoint_table;
 604 
 605   if (needs_thread_local_poll) {
 606     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 607     ldr(Rtemp, Address(Rthread, JavaThread::polling_word_offset()));
 608     tbnz(Rtemp, exact_log2(SafepointMechanism::poll_bit()), safepoint);
 609   }
 610 
 611   if((state == itos) || (state == btos) || (state == ztos) || (state == ctos) || (state == stos)) {
 612     zap_high_non_significant_bits(R0_tos);
 613   }
 614 
 615 #ifdef ASSERT
 616   Label L;
 617   mov_slow(Rtemp, (address)Interpreter::dispatch_table(vtos));
 618   cmp(Rtemp, RdispatchTable);
 619   b(L, eq);
 620   stop("invalid RdispatchTable");
 621   bind(L);
 622 #endif
 623 
 624   if (table_mode == DispatchDefault) {
 625     if (state == vtos) {
 626       indirect_jump(Address::indexed_ptr(RdispatchTable, R3_bytecode), Rtemp);
 627     } else {
 628       // on 32-bit ARM this method is faster than the one above.
 629       sub(Rtemp, RdispatchTable, (Interpreter::distance_from_dispatch_table(vtos) -
 630                            Interpreter::distance_from_dispatch_table(state)) * wordSize);
 631       indirect_jump(Address::indexed_ptr(Rtemp, R3_bytecode), Rtemp);
 632     }
 633   } else {
 634     assert(table_mode == DispatchNormal, "invalid dispatch table mode");
 635     address table = (address) Interpreter::normal_table(state);
 636     mov_slow(Rtemp, table);
 637     indirect_jump(Address::indexed_ptr(Rtemp, R3_bytecode), Rtemp);
 638   }
 639 
 640   if (needs_thread_local_poll) {
 641     bind(safepoint);
 642     lea(Rtemp, ExternalAddress((address)safepoint_table));
 643     indirect_jump(Address::indexed_ptr(Rtemp, R3_bytecode), Rtemp);
 644   }
 645 
 646   nop(); // to avoid filling CPU pipeline with invalid instructions
 647   nop();
 648 }
 649 
 650 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 651   dispatch_base(state, DispatchDefault, true, generate_poll);
 652 }
 653 
 654 
 655 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 656   dispatch_base(state, DispatchNormal);
 657 }
 658 
 659 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 660   dispatch_base(state, DispatchNormal, false);
 661 }
 662 
 663 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 664   // load next bytecode and advance Rbcp
 665   ldrb(R3_bytecode, Address(Rbcp, step, pre_indexed));
 666   dispatch_base(state, DispatchDefault, true, generate_poll);
 667 }
 668 
 669 void InterpreterMacroAssembler::narrow(Register result) {
 670   // mask integer result to narrower return type.
 671   const Register Rtmp = R2;
 672 
 673   // get method type
 674   ldr(Rtmp, Address(Rmethod, Method::const_offset()));
 675   ldrb(Rtmp, Address(Rtmp, ConstMethod::result_type_offset()));
 676 
 677   Label notBool, notByte, notChar, done;
 678   cmp(Rtmp, T_INT);
 679   b(done, eq);
 680 
 681   cmp(Rtmp, T_BOOLEAN);
 682   b(notBool, ne);
 683   and_32(result, result, 1);
 684   b(done);
 685 
 686   bind(notBool);
 687   cmp(Rtmp, T_BYTE);
 688   b(notByte, ne);
 689   sign_extend(result, result, 8);
 690   b(done);
 691 
 692   bind(notByte);
 693   cmp(Rtmp, T_CHAR);
 694   b(notChar, ne);
 695   zero_extend(result, result, 16);
 696   b(done);
 697 
 698   bind(notChar);
 699   // cmp(Rtmp, T_SHORT);
 700   // b(done, ne);
 701   sign_extend(result, result, 16);
 702 
 703   // Nothing to do
 704   bind(done);
 705 }
 706 
 707 // remove activation
 708 //
 709 // Unlock the receiver if this is a synchronized method.
 710 // Unlock any Java monitors from synchronized blocks.
 711 // Remove the activation from the stack.
 712 //
 713 // If there are locked Java monitors
 714 //    If throw_monitor_exception
 715 //       throws IllegalMonitorStateException
 716 //    Else if install_monitor_exception
 717 //       installs IllegalMonitorStateException
 718 //    Else
 719 //       no error processing
 720 void InterpreterMacroAssembler::remove_activation(TosState state, Register ret_addr,
 721                                                   bool throw_monitor_exception,
 722                                                   bool install_monitor_exception,
 723                                                   bool notify_jvmdi) {
 724   Label unlock, unlocked, no_unlock;
 725 
 726   // Note: Registers R0, R1, S0 and D0 (TOS cached value) may be in use for the result.
 727 
 728   const Address do_not_unlock_if_synchronized(Rthread,
 729                          JavaThread::do_not_unlock_if_synchronized_offset());
 730 
 731   const Register Rflag = R2;
 732   const Register Raccess_flags = R3;
 733 
 734   restore_method();
 735 
 736   ldrb(Rflag, do_not_unlock_if_synchronized);
 737 
 738   // get method access flags
 739   ldr_u32(Raccess_flags, Address(Rmethod, Method::access_flags_offset()));
 740 
 741   strb(zero_register(Rtemp), do_not_unlock_if_synchronized); // reset the flag
 742 
 743   // check if method is synchronized
 744 
 745   tbz(Raccess_flags, JVM_ACC_SYNCHRONIZED_BIT, unlocked);
 746 
 747   // Don't unlock anything if the _do_not_unlock_if_synchronized flag is set.
 748   cbnz(Rflag, no_unlock);
 749 
 750   // unlock monitor
 751   push(state);                                   // save result
 752 
 753   // BasicObjectLock will be first in list, since this is a synchronized method. However, need
 754   // to check that the object has not been unlocked by an explicit monitorexit bytecode.
 755 
 756   const Register Rmonitor = R0;                  // fixed in unlock_object()
 757   const Register Robj = R2;
 758 
 759   // address of first monitor
 760   sub(Rmonitor, FP, - frame::interpreter_frame_monitor_block_bottom_offset * wordSize + (int)sizeof(BasicObjectLock));
 761 
 762   ldr(Robj, Address(Rmonitor, BasicObjectLock::obj_offset()));
 763   cbnz(Robj, unlock);
 764 
 765   pop(state);
 766 
 767   if (throw_monitor_exception) {
 768     // Entry already unlocked, need to throw exception
 769     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
 770     should_not_reach_here();
 771   } else {
 772     // Monitor already unlocked during a stack unroll.
 773     // If requested, install an illegal_monitor_state_exception.
 774     // Continue with stack unrolling.
 775     if (install_monitor_exception) {
 776       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
 777     }
 778     b(unlocked);
 779   }
 780 
 781 
 782   // Exception case for the check that all monitors are unlocked.
 783   const Register Rcur = R2;
 784   Label restart_check_monitors_unlocked, exception_monitor_is_still_locked;
 785 
 786   bind(exception_monitor_is_still_locked);
 787   // Monitor entry is still locked, need to throw exception.
 788   // Rcur: monitor entry.
 789 
 790   if (throw_monitor_exception) {
 791     // Throw exception
 792     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
 793     should_not_reach_here();
 794   } else {
 795     // Stack unrolling. Unlock object and install illegal_monitor_exception
 796     // Unlock does not block, so don't have to worry about the frame
 797 
 798     push(state);
 799     mov(Rmonitor, Rcur);
 800     unlock_object(Rmonitor);
 801 
 802     if (install_monitor_exception) {
 803       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
 804     }
 805 
 806     pop(state);
 807     b(restart_check_monitors_unlocked);
 808   }
 809 
 810   bind(unlock);
 811   unlock_object(Rmonitor);
 812   pop(state);
 813 
 814   // Check that for block-structured locking (i.e., that all locked objects has been unlocked)
 815   bind(unlocked);
 816 
 817   // Check that all monitors are unlocked
 818   {
 819     Label loop;
 820 
 821     const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 822     const Register Rbottom = R3;
 823     const Register Rcur_obj = Rtemp;
 824 
 825     bind(restart_check_monitors_unlocked);
 826 
 827     ldr(Rcur, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
 828                                  // points to current entry, starting with top-most entry
 829     sub(Rbottom, FP, -frame::interpreter_frame_monitor_block_bottom_offset * wordSize);
 830                                  // points to word before bottom of monitor block
 831 
 832     cmp(Rcur, Rbottom);          // check if there are no monitors
 833     ldr(Rcur_obj, Address(Rcur, BasicObjectLock::obj_offset()), ne);
 834                                  // prefetch monitor's object
 835     b(no_unlock, eq);
 836 
 837     bind(loop);
 838     // check if current entry is used
 839     cbnz(Rcur_obj, exception_monitor_is_still_locked);
 840 
 841     add(Rcur, Rcur, entry_size);      // otherwise advance to next entry
 842     cmp(Rcur, Rbottom);               // check if bottom reached
 843     ldr(Rcur_obj, Address(Rcur, BasicObjectLock::obj_offset()), ne);
 844                                       // prefetch monitor's object
 845     b(loop, ne);                      // if not at bottom then check this entry
 846   }
 847 
 848   bind(no_unlock);
 849 
 850   // jvmti support
 851   if (notify_jvmdi) {
 852     notify_method_exit(state, NotifyJVMTI);     // preserve TOSCA
 853   } else {
 854     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
 855   }
 856 
 857   // remove activation
 858   mov(Rtemp, FP);
 859   ldmia(FP, RegisterSet(FP) | RegisterSet(LR));
 860   ldr(SP, Address(Rtemp, frame::interpreter_frame_sender_sp_offset * wordSize));
 861 
 862   if (ret_addr != LR) {
 863     mov(ret_addr, LR);
 864   }
 865 }
 866 
 867 
 868 // At certain points in the method invocation the monitor of
 869 // synchronized methods hasn't been entered yet.
 870 // To correctly handle exceptions at these points, we set the thread local
 871 // variable _do_not_unlock_if_synchronized to true. The remove_activation will
 872 // check this flag.
 873 void InterpreterMacroAssembler::set_do_not_unlock_if_synchronized(bool flag, Register tmp) {
 874   const Address do_not_unlock_if_synchronized(Rthread,
 875                          JavaThread::do_not_unlock_if_synchronized_offset());
 876   if (flag) {
 877     mov(tmp, 1);
 878     strb(tmp, do_not_unlock_if_synchronized);
 879   } else {
 880     strb(zero_register(tmp), do_not_unlock_if_synchronized);
 881   }
 882 }
 883 
 884 // Lock object
 885 //
 886 // Argument: R1 : Points to BasicObjectLock to be used for locking.
 887 // Must be initialized with object to lock.
 888 // Blows volatile registers R0-R3, Rtemp, LR. Calls VM.
 889 void InterpreterMacroAssembler::lock_object(Register Rlock) {
 890   assert(Rlock == R1, "the second argument");
 891 
 892   if (LockingMode == LM_MONITOR) {
 893     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), Rlock);
 894   } else {
 895     Label done;
 896 
 897     const Register Robj = R2;
 898     const Register Rmark = R3;
 899     assert_different_registers(Robj, Rmark, Rlock, R0, Rtemp);
 900 
 901     const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
 902     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
 903     const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
 904 
 905     Label already_locked, slow_case;
 906 
 907     // Load object pointer
 908     ldr(Robj, Address(Rlock, obj_offset));
 909 
 910     if (DiagnoseSyncOnValueBasedClasses != 0) {
 911       load_klass(R0, Robj);
 912       ldr_u32(R0, Address(R0, Klass::access_flags_offset()));
 913       tst(R0, JVM_ACC_IS_VALUE_BASED_CLASS);
 914       b(slow_case, ne);
 915     }
 916 
 917     if (LockingMode == LM_LIGHTWEIGHT) {
 918       lightweight_lock(Robj, R0 /* t1 */, Rmark /* t2 */, Rtemp /* t3 */, 0 /* savemask */, slow_case);
 919       b(done);
 920     } else if (LockingMode == LM_LEGACY) {
 921       // On MP platforms the next load could return a 'stale' value if the memory location has been modified by another thread.
 922       // That would be acceptable as ether CAS or slow case path is taken in that case.
 923       // Exception to that is if the object is locked by the calling thread, then the recursive test will pass (guaranteed as
 924       // loads are satisfied from a store queue if performed on the same processor).
 925 
 926       assert(oopDesc::mark_offset_in_bytes() == 0, "must be");
 927       ldr(Rmark, Address(Robj, oopDesc::mark_offset_in_bytes()));
 928 
 929       // Test if object is already locked
 930       tst(Rmark, markWord::unlocked_value);
 931       b(already_locked, eq);
 932 
 933       // Save old object->mark() into BasicLock's displaced header
 934       str(Rmark, Address(Rlock, mark_offset));
 935 
 936       cas_for_lock_acquire(Rmark, Rlock, Robj, Rtemp, slow_case);
 937 
 938       b(done);
 939 
 940       // If we got here that means the object is locked by ether calling thread or another thread.
 941       bind(already_locked);
 942       // Handling of locked objects: recursive locks and slow case.
 943 
 944       // Fast check for recursive lock.
 945       //
 946       // Can apply the optimization only if this is a stack lock
 947       // allocated in this thread. For efficiency, we can focus on
 948       // recently allocated stack locks (instead of reading the stack
 949       // base and checking whether 'mark' points inside the current
 950       // thread stack):
 951       //  1) (mark & 3) == 0
 952       //  2) SP <= mark < SP + os::pagesize()
 953       //
 954       // Warning: SP + os::pagesize can overflow the stack base. We must
 955       // neither apply the optimization for an inflated lock allocated
 956       // just above the thread stack (this is why condition 1 matters)
 957       // nor apply the optimization if the stack lock is inside the stack
 958       // of another thread. The latter is avoided even in case of overflow
 959       // because we have guard pages at the end of all stacks. Hence, if
 960       // we go over the stack base and hit the stack of another thread,
 961       // this should not be in a writeable area that could contain a
 962       // stack lock allocated by that thread. As a consequence, a stack
 963       // lock less than page size away from SP is guaranteed to be
 964       // owned by the current thread.
 965       //
 966       // Note: assuming SP is aligned, we can check the low bits of
 967       // (mark-SP) instead of the low bits of mark. In that case,
 968       // assuming page size is a power of 2, we can merge the two
 969       // conditions into a single test:
 970       // => ((mark - SP) & (3 - os::pagesize())) == 0
 971 
 972       // (3 - os::pagesize()) cannot be encoded as an ARM immediate operand.
 973       // Check independently the low bits and the distance to SP.
 974       // -1- test low 2 bits
 975       movs(R0, AsmOperand(Rmark, lsl, 30));
 976       // -2- test (mark - SP) if the low two bits are 0
 977       sub(R0, Rmark, SP, eq);
 978       movs(R0, AsmOperand(R0, lsr, exact_log2(os::vm_page_size())), eq);
 979       // If still 'eq' then recursive locking OK: store 0 into lock record
 980       str(R0, Address(Rlock, mark_offset), eq);
 981 
 982       b(done, eq);
 983     }
 984 
 985     bind(slow_case);
 986 
 987     // Call the runtime routine for slow case
 988     if (LockingMode == LM_LIGHTWEIGHT) {
 989       // Pass oop, not lock, in fast lock case. call_VM wants R1 though.
 990       push(R1);
 991       mov(R1, Robj);
 992       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj), R1);
 993       pop(R1);
 994     } else {
 995       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), Rlock);
 996     }
 997     bind(done);
 998   }
 999 }
1000 
1001 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1002 //
1003 // Argument: R0: Points to BasicObjectLock structure for lock
1004 // Throw an IllegalMonitorException if object is not locked by current thread
1005 // Blows volatile registers R0-R3, Rtemp, LR. Calls VM.
1006 void InterpreterMacroAssembler::unlock_object(Register Rlock) {
1007   assert(Rlock == R0, "the first argument");
1008 
1009   if (LockingMode == LM_MONITOR) {
1010     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), Rlock);
1011   } else {
1012     Label done, slow_case;
1013 
1014     const Register Robj = R2;
1015     const Register Rmark = R3;
1016     assert_different_registers(Robj, Rmark, Rlock, Rtemp);
1017 
1018     const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1019     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1020     const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
1021 
1022     const Register Rzero = zero_register(Rtemp);
1023 
1024     // Load oop into Robj
1025     ldr(Robj, Address(Rlock, obj_offset));
1026 
1027     // Free entry
1028     str(Rzero, Address(Rlock, obj_offset));
1029 
1030     if (LockingMode == LM_LIGHTWEIGHT) {
1031 
1032       // Check for non-symmetric locking. This is allowed by the spec and the interpreter
1033       // must handle it.
1034       ldr(Rtemp, Address(Rthread, JavaThread::lock_stack_top_offset()));
1035       sub(Rtemp, Rtemp, oopSize);
1036       ldr(Rtemp, Address(Rthread, Rtemp));
1037       cmpoop(Rtemp, Robj);
1038       b(slow_case, ne);
1039 
1040       lightweight_unlock(Robj /* obj */, Rlock /* t1 */, Rmark /* t2 */, Rtemp /* t3 */,
1041                          1 /* savemask (save t1) */, slow_case);
1042 
1043       b(done);
1044 
1045     } else if (LockingMode == LM_LEGACY) {
1046 
1047       // Load the old header from BasicLock structure
1048       ldr(Rmark, Address(Rlock, mark_offset));
1049 
1050       // Test for recursion (zero mark in BasicLock)
1051       cbz(Rmark, done);
1052 
1053       bool allow_fallthrough_on_failure = true;
1054 
1055       cas_for_lock_release(Rlock, Rmark, Robj, Rtemp, slow_case, allow_fallthrough_on_failure);
1056 
1057       b(done, eq);
1058 
1059     }
1060     bind(slow_case);
1061 
1062     // Call the runtime routine for slow case.
1063     str(Robj, Address(Rlock, obj_offset)); // restore obj
1064     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), Rlock);
1065 
1066     bind(done);
1067   }
1068 }
1069 
1070 // Test ImethodDataPtr.  If it is null, continue at the specified label
1071 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1072   assert(ProfileInterpreter, "must be profiling interpreter");
1073   ldr(mdp, Address(FP, frame::interpreter_frame_mdp_offset * wordSize));
1074   cbz(mdp, zero_continue);
1075 }
1076 
1077 
1078 // Set the method data pointer for the current bcp.
1079 // Blows volatile registers R0-R3, Rtemp, LR.
1080 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1081   assert(ProfileInterpreter, "must be profiling interpreter");
1082   Label set_mdp;
1083 
1084   // Test MDO to avoid the call if it is null.
1085   ldr(Rtemp, Address(Rmethod, Method::method_data_offset()));
1086   cbz(Rtemp, set_mdp);
1087 
1088   mov(R0, Rmethod);
1089   mov(R1, Rbcp);
1090   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R0, R1);
1091   // R0/W0: mdi
1092 
1093   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1094   ldr(Rtemp, Address(Rmethod, Method::method_data_offset()));
1095   add(Rtemp, Rtemp, in_bytes(MethodData::data_offset()));
1096   add_ptr_scaled_int32(Rtemp, Rtemp, R0, 0);
1097 
1098   bind(set_mdp);
1099   str(Rtemp, Address(FP, frame::interpreter_frame_mdp_offset * wordSize));
1100 }
1101 
1102 
1103 void InterpreterMacroAssembler::verify_method_data_pointer() {
1104   assert(ProfileInterpreter, "must be profiling interpreter");
1105 #ifdef ASSERT
1106   Label verify_continue;
1107   save_caller_save_registers();
1108 
1109   const Register Rmdp = R2;
1110   test_method_data_pointer(Rmdp, verify_continue); // If mdp is zero, continue
1111 
1112   // If the mdp is valid, it will point to a DataLayout header which is
1113   // consistent with the bcp.  The converse is highly probable also.
1114 
1115   ldrh(R3, Address(Rmdp, DataLayout::bci_offset()));
1116   ldr(Rtemp, Address(Rmethod, Method::const_offset()));
1117   add(R3, R3, Rtemp);
1118   add(R3, R3, in_bytes(ConstMethod::codes_offset()));
1119   cmp(R3, Rbcp);
1120   b(verify_continue, eq);
1121 
1122   mov(R0, Rmethod);
1123   mov(R1, Rbcp);
1124   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), R0, R1, Rmdp);
1125 
1126   bind(verify_continue);
1127   restore_caller_save_registers();
1128 #endif // ASSERT
1129 }
1130 
1131 
1132 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int offset, Register value) {
1133   assert(ProfileInterpreter, "must be profiling interpreter");
1134   assert_different_registers(mdp_in, value);
1135   str(value, Address(mdp_in, offset));
1136 }
1137 
1138 
1139 // Increments mdp data. Sets bumped_count register to adjusted counter.
1140 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1141                                                       int offset,
1142                                                       Register bumped_count,
1143                                                       bool decrement) {
1144   assert(ProfileInterpreter, "must be profiling interpreter");
1145 
1146   // Counter address
1147   Address data(mdp_in, offset);
1148   assert_different_registers(mdp_in, bumped_count);
1149 
1150   increment_mdp_data_at(data, bumped_count, decrement);
1151 }
1152 
1153 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, int flag_byte_constant) {
1154   assert_different_registers(mdp_in, Rtemp);
1155   assert(ProfileInterpreter, "must be profiling interpreter");
1156   assert((0 < flag_byte_constant) && (flag_byte_constant < (1 << BitsPerByte)), "flag mask is out of range");
1157 
1158   // Set the flag
1159   ldrb(Rtemp, Address(mdp_in, in_bytes(DataLayout::flags_offset())));
1160   orr(Rtemp, Rtemp, (unsigned)flag_byte_constant);
1161   strb(Rtemp, Address(mdp_in, in_bytes(DataLayout::flags_offset())));
1162 }
1163 
1164 
1165 // Increments mdp data. Sets bumped_count register to adjusted counter.
1166 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1167                                                       Register bumped_count,
1168                                                       bool decrement) {
1169   assert(ProfileInterpreter, "must be profiling interpreter");
1170 
1171   ldr(bumped_count, data);
1172   if (decrement) {
1173     // Decrement the register. Set condition codes.
1174     subs(bumped_count, bumped_count, DataLayout::counter_increment);
1175     // Avoid overflow.
1176     add(bumped_count, bumped_count, DataLayout::counter_increment, pl);
1177   } else {
1178     // Increment the register. Set condition codes.
1179     adds(bumped_count, bumped_count, DataLayout::counter_increment);
1180     // Avoid overflow.
1181     sub(bumped_count, bumped_count, DataLayout::counter_increment, mi);
1182   }
1183   str(bumped_count, data);
1184 }
1185 
1186 
1187 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1188                                                  int offset,
1189                                                  Register value,
1190                                                  Register test_value_out,
1191                                                  Label& not_equal_continue) {
1192   assert(ProfileInterpreter, "must be profiling interpreter");
1193   assert_different_registers(mdp_in, test_value_out, value);
1194 
1195   ldr(test_value_out, Address(mdp_in, offset));
1196   cmp(test_value_out, value);
1197 
1198   b(not_equal_continue, ne);
1199 }
1200 
1201 
1202 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp, Register reg_temp) {
1203   assert(ProfileInterpreter, "must be profiling interpreter");
1204   assert_different_registers(mdp_in, reg_temp);
1205 
1206   ldr(reg_temp, Address(mdp_in, offset_of_disp));
1207   add(mdp_in, mdp_in, reg_temp);
1208   str(mdp_in, Address(FP, frame::interpreter_frame_mdp_offset * wordSize));
1209 }
1210 
1211 
1212 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, Register reg_offset, Register reg_tmp) {
1213   assert(ProfileInterpreter, "must be profiling interpreter");
1214   assert_different_registers(mdp_in, reg_offset, reg_tmp);
1215 
1216   ldr(reg_tmp, Address(mdp_in, reg_offset));
1217   add(mdp_in, mdp_in, reg_tmp);
1218   str(mdp_in, Address(FP, frame::interpreter_frame_mdp_offset * wordSize));
1219 }
1220 
1221 
1222 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1223   assert(ProfileInterpreter, "must be profiling interpreter");
1224   add(mdp_in, mdp_in, constant);
1225   str(mdp_in, Address(FP, frame::interpreter_frame_mdp_offset * wordSize));
1226 }
1227 
1228 
1229 // Blows volatile registers R0-R3, Rtemp, LR).
1230 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1231   assert(ProfileInterpreter, "must be profiling interpreter");
1232   assert_different_registers(return_bci, R0, R1, R2, R3, Rtemp);
1233 
1234   mov(R1, return_bci);
1235   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), R1);
1236 }
1237 
1238 
1239 // Sets mdp, bumped_count registers, blows Rtemp.
1240 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1241   assert_different_registers(mdp, bumped_count);
1242 
1243   if (ProfileInterpreter) {
1244     Label profile_continue;
1245 
1246     // If no method data exists, go to profile_continue.
1247     // Otherwise, assign to mdp
1248     test_method_data_pointer(mdp, profile_continue);
1249 
1250     // We are taking a branch. Increment the taken count.
1251     increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()), bumped_count);
1252 
1253     // The method data pointer needs to be updated to reflect the new target.
1254     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()), Rtemp);
1255 
1256     bind (profile_continue);
1257   }
1258 }
1259 
1260 
1261 // Sets mdp, blows Rtemp.
1262 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1263   assert_different_registers(mdp, Rtemp);
1264 
1265   if (ProfileInterpreter) {
1266     Label profile_continue;
1267 
1268     // If no method data exists, go to profile_continue.
1269     test_method_data_pointer(mdp, profile_continue);
1270 
1271     // We are taking a branch.  Increment the not taken count.
1272     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Rtemp);
1273 
1274     // The method data pointer needs to be updated to correspond to the next bytecode
1275     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1276 
1277     bind (profile_continue);
1278   }
1279 }
1280 
1281 
1282 // Sets mdp, blows Rtemp.
1283 void InterpreterMacroAssembler::profile_call(Register mdp) {
1284   assert_different_registers(mdp, Rtemp);
1285 
1286   if (ProfileInterpreter) {
1287     Label profile_continue;
1288 
1289     // If no method data exists, go to profile_continue.
1290     test_method_data_pointer(mdp, profile_continue);
1291 
1292     // We are making a call.  Increment the count.
1293     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()), Rtemp);
1294 
1295     // The method data pointer needs to be updated to reflect the new target.
1296     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1297 
1298     bind (profile_continue);
1299   }
1300 }
1301 
1302 
1303 // Sets mdp, blows Rtemp.
1304 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1305   if (ProfileInterpreter) {
1306     Label profile_continue;
1307 
1308     // If no method data exists, go to profile_continue.
1309     test_method_data_pointer(mdp, profile_continue);
1310 
1311     // We are making a call.  Increment the count.
1312     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()), Rtemp);
1313 
1314     // The method data pointer needs to be updated to reflect the new target.
1315     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1316 
1317     bind (profile_continue);
1318   }
1319 }
1320 
1321 
1322 // Sets mdp, blows Rtemp.
1323 void InterpreterMacroAssembler::profile_virtual_call(Register mdp, Register receiver, bool receiver_can_be_null) {
1324   assert_different_registers(mdp, receiver, Rtemp);
1325 
1326   if (ProfileInterpreter) {
1327     Label profile_continue;
1328 
1329     // If no method data exists, go to profile_continue.
1330     test_method_data_pointer(mdp, profile_continue);
1331 
1332     Label skip_receiver_profile;
1333     if (receiver_can_be_null) {
1334       Label not_null;
1335       cbnz(receiver, not_null);
1336       // We are making a call.  Increment the count for null receiver.
1337       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()), Rtemp);
1338       b(skip_receiver_profile);
1339       bind(not_null);
1340     }
1341 
1342     // Record the receiver type.
1343     record_klass_in_profile(receiver, mdp, Rtemp, true);
1344     bind(skip_receiver_profile);
1345 
1346     // The method data pointer needs to be updated to reflect the new target.
1347     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1348     bind(profile_continue);
1349   }
1350 }
1351 
1352 
1353 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1354                                         Register receiver, Register mdp,
1355                                         Register reg_tmp,
1356                                         int start_row, Label& done, bool is_virtual_call) {
1357   if (TypeProfileWidth == 0)
1358     return;
1359 
1360   assert_different_registers(receiver, mdp, reg_tmp);
1361 
1362   int last_row = VirtualCallData::row_limit() - 1;
1363   assert(start_row <= last_row, "must be work left to do");
1364   // Test this row for both the receiver and for null.
1365   // Take any of three different outcomes:
1366   //   1. found receiver => increment count and goto done
1367   //   2. found null => keep looking for case 1, maybe allocate this cell
1368   //   3. found something else => keep looking for cases 1 and 2
1369   // Case 3 is handled by a recursive call.
1370   for (int row = start_row; row <= last_row; row++) {
1371     Label next_test;
1372 
1373     // See if the receiver is receiver[n].
1374     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1375 
1376     test_mdp_data_at(mdp, recvr_offset, receiver, reg_tmp, next_test);
1377 
1378     // The receiver is receiver[n].  Increment count[n].
1379     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1380     increment_mdp_data_at(mdp, count_offset, reg_tmp);
1381     b(done);
1382 
1383     bind(next_test);
1384     // reg_tmp now contains the receiver from the CallData.
1385 
1386     if (row == start_row) {
1387       Label found_null;
1388       // Failed the equality check on receiver[n]...  Test for null.
1389       if (start_row == last_row) {
1390         // The only thing left to do is handle the null case.
1391         if (is_virtual_call) {
1392           cbz(reg_tmp, found_null);
1393           // Receiver did not match any saved receiver and there is no empty row for it.
1394           // Increment total counter to indicate polymorphic case.
1395           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()), reg_tmp);
1396           b(done);
1397           bind(found_null);
1398         } else {
1399           cbnz(reg_tmp, done);
1400         }
1401         break;
1402       }
1403       // Since null is rare, make it be the branch-taken case.
1404       cbz(reg_tmp, found_null);
1405 
1406       // Put all the "Case 3" tests here.
1407       record_klass_in_profile_helper(receiver, mdp, reg_tmp, start_row + 1, done, is_virtual_call);
1408 
1409       // Found a null.  Keep searching for a matching receiver,
1410       // but remember that this is an empty (unused) slot.
1411       bind(found_null);
1412     }
1413   }
1414 
1415   // In the fall-through case, we found no matching receiver, but we
1416   // observed the receiver[start_row] is null.
1417 
1418   // Fill in the receiver field and increment the count.
1419   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1420   set_mdp_data_at(mdp, recvr_offset, receiver);
1421   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1422   mov(reg_tmp, DataLayout::counter_increment);
1423   set_mdp_data_at(mdp, count_offset, reg_tmp);
1424   if (start_row > 0) {
1425     b(done);
1426   }
1427 }
1428 
1429 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1430                                                         Register mdp,
1431                                                         Register reg_tmp,
1432                                                         bool is_virtual_call) {
1433   assert(ProfileInterpreter, "must be profiling");
1434   assert_different_registers(receiver, mdp, reg_tmp);
1435 
1436   Label done;
1437 
1438   record_klass_in_profile_helper(receiver, mdp, reg_tmp, 0, done, is_virtual_call);
1439 
1440   bind (done);
1441 }
1442 
1443 // Sets mdp, blows volatile registers R0-R3, Rtemp, LR).
1444 void InterpreterMacroAssembler::profile_ret(Register mdp, Register return_bci) {
1445   assert_different_registers(mdp, return_bci, Rtemp, R0, R1, R2, R3);
1446 
1447   if (ProfileInterpreter) {
1448     Label profile_continue;
1449     uint row;
1450 
1451     // If no method data exists, go to profile_continue.
1452     test_method_data_pointer(mdp, profile_continue);
1453 
1454     // Update the total ret count.
1455     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()), Rtemp);
1456 
1457     for (row = 0; row < RetData::row_limit(); row++) {
1458       Label next_test;
1459 
1460       // See if return_bci is equal to bci[n]:
1461       test_mdp_data_at(mdp, in_bytes(RetData::bci_offset(row)), return_bci,
1462                        Rtemp, next_test);
1463 
1464       // return_bci is equal to bci[n].  Increment the count.
1465       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)), Rtemp);
1466 
1467       // The method data pointer needs to be updated to reflect the new target.
1468       update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)), Rtemp);
1469       b(profile_continue);
1470       bind(next_test);
1471     }
1472 
1473     update_mdp_for_ret(return_bci);
1474 
1475     bind(profile_continue);
1476   }
1477 }
1478 
1479 
1480 // Sets mdp.
1481 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1482   if (ProfileInterpreter) {
1483     Label profile_continue;
1484 
1485     // If no method data exists, go to profile_continue.
1486     test_method_data_pointer(mdp, profile_continue);
1487 
1488     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1489 
1490     // The method data pointer needs to be updated.
1491     int mdp_delta = in_bytes(BitData::bit_data_size());
1492     if (TypeProfileCasts) {
1493       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1494     }
1495     update_mdp_by_constant(mdp, mdp_delta);
1496 
1497     bind (profile_continue);
1498   }
1499 }
1500 
1501 
1502 // Sets mdp, blows Rtemp.
1503 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1504   assert_different_registers(mdp, Rtemp);
1505 
1506   if (ProfileInterpreter && TypeProfileCasts) {
1507     Label profile_continue;
1508 
1509     // If no method data exists, go to profile_continue.
1510     test_method_data_pointer(mdp, profile_continue);
1511 
1512     int count_offset = in_bytes(CounterData::count_offset());
1513     // Back up the address, since we have already bumped the mdp.
1514     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1515 
1516     // *Decrement* the counter.  We expect to see zero or small negatives.
1517     increment_mdp_data_at(mdp, count_offset, Rtemp, true);
1518 
1519     bind (profile_continue);
1520   }
1521 }
1522 
1523 
1524 // Sets mdp, blows Rtemp.
1525 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass)
1526 {
1527   assert_different_registers(mdp, klass, Rtemp);
1528 
1529   if (ProfileInterpreter) {
1530     Label profile_continue;
1531 
1532     // If no method data exists, go to profile_continue.
1533     test_method_data_pointer(mdp, profile_continue);
1534 
1535     // The method data pointer needs to be updated.
1536     int mdp_delta = in_bytes(BitData::bit_data_size());
1537     if (TypeProfileCasts) {
1538       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1539 
1540       // Record the object type.
1541       record_klass_in_profile(klass, mdp, Rtemp, false);
1542     }
1543     update_mdp_by_constant(mdp, mdp_delta);
1544 
1545     bind(profile_continue);
1546   }
1547 }
1548 
1549 
1550 // Sets mdp, blows Rtemp.
1551 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1552   assert_different_registers(mdp, Rtemp);
1553 
1554   if (ProfileInterpreter) {
1555     Label profile_continue;
1556 
1557     // If no method data exists, go to profile_continue.
1558     test_method_data_pointer(mdp, profile_continue);
1559 
1560     // Update the default case count
1561     increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()), Rtemp);
1562 
1563     // The method data pointer needs to be updated.
1564     update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()), Rtemp);
1565 
1566     bind(profile_continue);
1567   }
1568 }
1569 
1570 
1571 // Sets mdp. Blows reg_tmp1, reg_tmp2. Index could be the same as reg_tmp2.
1572 void InterpreterMacroAssembler::profile_switch_case(Register mdp, Register index, Register reg_tmp1, Register reg_tmp2) {
1573   assert_different_registers(mdp, reg_tmp1, reg_tmp2);
1574   assert_different_registers(mdp, reg_tmp1, index);
1575 
1576   if (ProfileInterpreter) {
1577     Label profile_continue;
1578 
1579     const int count_offset = in_bytes(MultiBranchData::case_array_offset()) +
1580                               in_bytes(MultiBranchData::relative_count_offset());
1581 
1582     const int displacement_offset = in_bytes(MultiBranchData::case_array_offset()) +
1583                               in_bytes(MultiBranchData::relative_displacement_offset());
1584 
1585     // If no method data exists, go to profile_continue.
1586     test_method_data_pointer(mdp, profile_continue);
1587 
1588     // Build the base (index * per_case_size_in_bytes())
1589     logical_shift_left(reg_tmp1, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1590 
1591     // Update the case count
1592     add(reg_tmp1, reg_tmp1, count_offset);
1593     increment_mdp_data_at(Address(mdp, reg_tmp1), reg_tmp2);
1594 
1595     // The method data pointer needs to be updated.
1596     add(reg_tmp1, reg_tmp1, displacement_offset - count_offset);
1597     update_mdp_by_offset(mdp, reg_tmp1, reg_tmp2);
1598 
1599     bind (profile_continue);
1600   }
1601 }
1602 
1603 
1604 void InterpreterMacroAssembler::byteswap_u32(Register r, Register rtmp1, Register rtmp2) {
1605   if (VM_Version::supports_rev()) {
1606     rev(r, r);
1607   } else {
1608     eor(rtmp1, r, AsmOperand(r, ror, 16));
1609     mvn(rtmp2, 0x0000ff00);
1610     andr(rtmp1, rtmp2, AsmOperand(rtmp1, lsr, 8));
1611     eor(r, rtmp1, AsmOperand(r, ror, 8));
1612   }
1613 }
1614 
1615 
1616 void InterpreterMacroAssembler::inc_global_counter(address address_of_counter, int offset, Register tmp1, Register tmp2, bool avoid_overflow) {
1617   const intx addr = (intx) (address_of_counter + offset);
1618 
1619   assert ((addr & 0x3) == 0, "address of counter should be aligned");
1620   const intx offset_mask = right_n_bits(12);
1621 
1622   const address base = (address) (addr & ~offset_mask);
1623   const int offs = (int) (addr & offset_mask);
1624 
1625   const Register addr_base = tmp1;
1626   const Register val = tmp2;
1627 
1628   mov_slow(addr_base, base);
1629   ldr_s32(val, Address(addr_base, offs));
1630 
1631   if (avoid_overflow) {
1632     adds_32(val, val, 1);
1633     str(val, Address(addr_base, offs), pl);
1634   } else {
1635     add_32(val, val, 1);
1636     str_32(val, Address(addr_base, offs));
1637   }
1638 }
1639 
1640 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char *file, int line) {
1641   if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop", file, line); }
1642 }
1643 
1644 // Inline assembly for:
1645 //
1646 // if (thread is in interp_only_mode) {
1647 //   InterpreterRuntime::post_method_entry();
1648 // }
1649 // if (DTraceMethodProbes) {
1650 //   SharedRuntime::dtrace_method_entry(method, receiver);
1651 // }
1652 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1653 //   SharedRuntime::rc_trace_method_entry(method, receiver);
1654 // }
1655 
1656 void InterpreterMacroAssembler::notify_method_entry() {
1657   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1658   // track stack depth.  If it is possible to enter interp_only_mode we add
1659   // the code to check if the event should be sent.
1660   if (can_post_interpreter_events()) {
1661     Label L;
1662 
1663     ldr_s32(Rtemp, Address(Rthread, JavaThread::interp_only_mode_offset()));
1664     cbz(Rtemp, L);
1665 
1666     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
1667 
1668     bind(L);
1669   }
1670 
1671   // Note: Disable DTrace runtime check for now to eliminate overhead on each method entry
1672   if (DTraceMethodProbes) {
1673     Label Lcontinue;
1674 
1675     ldrb_global(Rtemp, (address)&DTraceMethodProbes);
1676     cbz(Rtemp, Lcontinue);
1677 
1678     mov(R0, Rthread);
1679     mov(R1, Rmethod);
1680     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), R0, R1);
1681 
1682     bind(Lcontinue);
1683   }
1684   // RedefineClasses() tracing support for obsolete method entry
1685   if (log_is_enabled(Trace, redefine, class, obsolete)) {
1686     mov(R0, Rthread);
1687     mov(R1, Rmethod);
1688     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1689                  R0, R1);
1690   }
1691 }
1692 
1693 
1694 void InterpreterMacroAssembler::notify_method_exit(
1695                  TosState state, NotifyMethodExitMode mode,
1696                  bool native, Register result_lo, Register result_hi, FloatRegister result_fp) {
1697   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1698   // track stack depth.  If it is possible to enter interp_only_mode we add
1699   // the code to check if the event should be sent.
1700   if (mode == NotifyJVMTI && can_post_interpreter_events()) {
1701     Label L;
1702     // Note: frame::interpreter_frame_result has a dependency on how the
1703     // method result is saved across the call to post_method_exit. If this
1704     // is changed then the interpreter_frame_result implementation will
1705     // need to be updated too.
1706 
1707     ldr_s32(Rtemp, Address(Rthread, JavaThread::interp_only_mode_offset()));
1708     cbz(Rtemp, L);
1709 
1710     if (native) {
1711       // For c++ and template interpreter push both result registers on the
1712       // stack in native, we don't know the state.
1713       // See frame::interpreter_frame_result for code that gets the result values from here.
1714       assert(result_lo != noreg, "result registers should be defined");
1715 
1716       assert(result_hi != noreg, "result registers should be defined");
1717 
1718 #ifdef __ABI_HARD__
1719       assert(result_fp != fnoreg, "FP result register must be defined");
1720       sub(SP, SP, 2 * wordSize);
1721       fstd(result_fp, Address(SP));
1722 #endif // __ABI_HARD__
1723 
1724       push(RegisterSet(result_lo) | RegisterSet(result_hi));
1725 
1726       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1727 
1728       pop(RegisterSet(result_lo) | RegisterSet(result_hi));
1729 #ifdef __ABI_HARD__
1730       fldd(result_fp, Address(SP));
1731       add(SP, SP, 2 * wordSize);
1732 #endif // __ABI_HARD__
1733 
1734     } else {
1735       // For the template interpreter, the value on tos is the size of the
1736       // state. (c++ interpreter calls jvmti somewhere else).
1737       push(state);
1738       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1739       pop(state);
1740     }
1741 
1742     bind(L);
1743   }
1744 
1745   // Note: Disable DTrace runtime check for now to eliminate overhead on each method exit
1746   if (DTraceMethodProbes) {
1747     Label Lcontinue;
1748 
1749     ldrb_global(Rtemp, (address)&DTraceMethodProbes);
1750     cbz(Rtemp, Lcontinue);
1751 
1752     push(state);
1753 
1754     mov(R0, Rthread);
1755     mov(R1, Rmethod);
1756 
1757     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), R0, R1);
1758 
1759     pop(state);
1760 
1761     bind(Lcontinue);
1762   }
1763 }
1764 
1765 
1766 #ifndef PRODUCT
1767 
1768 void InterpreterMacroAssembler::trace_state(const char* msg) {
1769   int push_size = save_caller_save_registers();
1770 
1771   Label Lcontinue;
1772   InlinedString Lmsg0("%s: FP=" INTPTR_FORMAT ", SP=" INTPTR_FORMAT "\n");
1773   InlinedString Lmsg(msg);
1774   InlinedAddress Lprintf((address)printf);
1775 
1776   ldr_literal(R0, Lmsg0);
1777   ldr_literal(R1, Lmsg);
1778   mov(R2, FP);
1779   add(R3, SP, push_size);  // original SP (without saved registers)
1780   ldr_literal(Rtemp, Lprintf);
1781   call(Rtemp);
1782 
1783   b(Lcontinue);
1784 
1785   bind_literal(Lmsg0);
1786   bind_literal(Lmsg);
1787   bind_literal(Lprintf);
1788 
1789 
1790   bind(Lcontinue);
1791 
1792   restore_caller_save_registers();
1793 }
1794 
1795 #endif
1796 
1797 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1798 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1799                                                         int increment, Address mask_addr,
1800                                                         Register scratch, Register scratch2,
1801                                                         AsmCondition cond, Label* where) {
1802   // caution: scratch2 and base address of counter_addr can be the same
1803   assert_different_registers(scratch, scratch2);
1804   ldr_u32(scratch, counter_addr);
1805   add(scratch, scratch, increment);
1806   str_32(scratch, counter_addr);
1807 
1808   ldr(scratch2, mask_addr);
1809   andrs(scratch, scratch, scratch2);
1810   b(*where, cond);
1811 }
1812 
1813 void InterpreterMacroAssembler::get_method_counters(Register method,
1814                                                     Register Rcounters,
1815                                                     Label& skip,
1816                                                     bool saveRegs,
1817                                                     Register reg1,
1818                                                     Register reg2,
1819                                                     Register reg3) {
1820   const Address method_counters(method, Method::method_counters_offset());
1821   Label has_counters;
1822 
1823   ldr(Rcounters, method_counters);
1824   cbnz(Rcounters, has_counters);
1825 
1826   if (saveRegs) {
1827     // Save and restore in use caller-saved registers since they will be trashed by call_VM
1828     assert(reg1 != noreg, "must specify reg1");
1829     assert(reg2 != noreg, "must specify reg2");
1830     assert(reg3 == noreg, "must not specify reg3");
1831     push(RegisterSet(reg1) | RegisterSet(reg2));
1832   }
1833 
1834   mov(R1, method);
1835   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), R1);
1836 
1837   if (saveRegs) {
1838     pop(RegisterSet(reg1) | RegisterSet(reg2));
1839   }
1840 
1841   ldr(Rcounters, method_counters);
1842   cbz(Rcounters, skip); // No MethodCounters created, OutOfMemory
1843 
1844   bind(has_counters);
1845 }