1 /*
   2  * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2024 SAP SE. 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 // Major contributions by AHa, AS, JL, ML.
  27 
  28 #include "precompiled.hpp"
  29 #include "asm/macroAssembler.inline.hpp"
  30 #include "gc/shared/barrierSet.hpp"
  31 #include "gc/shared/barrierSetAssembler.hpp"
  32 #include "interp_masm_s390.hpp"
  33 #include "interpreter/interpreter.hpp"
  34 #include "interpreter/interpreterRuntime.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/markWord.hpp"
  37 #include "oops/methodCounters.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "oops/resolvedFieldEntry.hpp"
  40 #include "oops/resolvedIndyEntry.hpp"
  41 #include "oops/resolvedMethodEntry.hpp"
  42 #include "prims/jvmtiExport.hpp"
  43 #include "prims/jvmtiThreadState.hpp"
  44 #include "runtime/basicLock.hpp"
  45 #include "runtime/frame.inline.hpp"
  46 #include "runtime/javaThread.hpp"
  47 #include "runtime/safepointMechanism.hpp"
  48 #include "runtime/sharedRuntime.hpp"
  49 #include "utilities/macros.hpp"
  50 #include "utilities/powerOfTwo.hpp"
  51 
  52 // Implementation of InterpreterMacroAssembler.
  53 // This file specializes the assembler with interpreter-specific macros.
  54 
  55 #ifdef PRODUCT
  56 #define BLOCK_COMMENT(str)
  57 #define BIND(label)        bind(label);
  58 #else
  59 #define BLOCK_COMMENT(str) block_comment(str)
  60 #define BIND(label)        bind(label); BLOCK_COMMENT(#label ":")
  61 #endif
  62 
  63 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
  64   assert(entry != nullptr, "Entry must have been generated by now");
  65   assert(Rscratch != Z_R0, "Can't use R0 for addressing");
  66   branch_optimized(Assembler::bcondAlways, entry);
  67 }
  68 
  69 void InterpreterMacroAssembler::empty_expression_stack(void) {
  70   get_monitors(Z_R1_scratch);
  71   add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch);
  72 }
  73 
  74 // Dispatch code executed in the prolog of a bytecode which does not do it's
  75 // own dispatch.
  76 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
  77   // On z/Architecture we are short on registers, therefore we do not preload the
  78   // dispatch address of the next bytecode.
  79 }
  80 
  81 // Dispatch code executed in the epilog of a bytecode which does not do it's
  82 // own dispatch.
  83 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
  84   dispatch_next(state, step);
  85 }
  86 
  87 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
  88   z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp);  // Load next bytecode.
  89   add2reg(Z_bcp, bcp_incr);                   // Advance bcp. Add2reg produces optimal code.
  90   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
  91 }
  92 
  93 // Common code to dispatch and dispatch_only.
  94 // Dispatch value in Lbyte_code and increment Lbcp.
  95 
  96 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool generate_poll) {
  97   verify_FPU(1, state);
  98 
  99 #ifdef ASSERT
 100   address reentry = nullptr;
 101   { Label OK;
 102     // Check if the frame pointer in Z_fp is correct.
 103     z_cg(Z_fp, 0, Z_SP);
 104     z_bre(OK);
 105     reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE);
 106     bind(OK);
 107   }
 108   { Label OK;
 109     // check if the locals pointer in Z_locals is correct
 110     z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp);
 111     z_bre(OK);
 112     reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE);
 113     bind(OK);
 114   }
 115 #endif
 116 
 117   // TODO: Maybe implement +VerifyActivationFrameSize here.
 118   verify_oop(Z_tos, state);
 119 
 120   // Dispatch table to use.
 121   load_absolute_address(Z_tmp_1, (address)table);  // Z_tmp_1 = table;
 122 
 123   if (generate_poll) {
 124     address *sfpt_tbl = Interpreter::safept_table(state);
 125     if (table != sfpt_tbl) {
 126       Label dispatch;
 127       const Address poll_byte_addr(Z_thread, in_bytes(JavaThread::polling_word_offset()) + 7 /* Big Endian */);
 128       // Armed page has poll_bit set, if poll bit is cleared just continue.
 129       z_tm(poll_byte_addr, SafepointMechanism::poll_bit());
 130       z_braz(dispatch);
 131       load_absolute_address(Z_tmp_1, (address)sfpt_tbl);  // Z_tmp_1 = table;
 132       bind(dispatch);
 133     }
 134   }
 135 
 136   // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg.
 137   // Z_bytecode must have been loaded zero-extended for this approach to be correct.
 138   z_sll(Z_bytecode, LogBytesPerWord, Z_R0);   // Multiply by wordSize.
 139   z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1);      // Get entry addr.
 140 
 141   z_br(Z_tmp_1);
 142 }
 143 
 144 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 145   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
 146 }
 147 
 148 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 149   dispatch_base(state, Interpreter::normal_table(state));
 150 }
 151 
 152 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) {
 153   // Load current bytecode.
 154   z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0));
 155   dispatch_base(state, table);
 156 }
 157 
 158 // The following call_VM*_base() methods overload and mask the respective
 159 // declarations/definitions in class MacroAssembler. They are meant as a "detour"
 160 // to perform additional, template interpreter specific tasks before actually
 161 // calling their MacroAssembler counterparts.
 162 
 163 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) {
 164   bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
 165   // interpreter specific
 166   // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
 167   // saved registers and no blocking/ GC can happen in leaf calls.
 168 
 169   // super call
 170   MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
 171 }
 172 
 173 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) {
 174   // interpreter specific
 175   // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
 176   // saved registers and no blocking/ GC can happen in leaf calls.
 177 
 178   // super call
 179   MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
 180 }
 181 
 182 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
 183                                              address entry_point, bool check_exceptions) {
 184   bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
 185   // interpreter specific
 186 
 187   save_bcp();
 188   save_esp();
 189   // super call
 190   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 191                                entry_point, allow_relocation, check_exceptions);
 192   restore_bcp();
 193 }
 194 
 195 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
 196                                              address entry_point, bool allow_relocation,
 197                                              bool check_exceptions) {
 198   // interpreter specific
 199 
 200   save_bcp();
 201   save_esp();
 202   // super call
 203   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 204                                entry_point, allow_relocation, check_exceptions);
 205   restore_bcp();
 206 }
 207 
 208 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
 209   if (JvmtiExport::can_pop_frame()) {
 210     BLOCK_COMMENT("check_and_handle_popframe {");
 211     Label L;
 212     // Initiate popframe handling only if it is not already being
 213     // processed. If the flag has the popframe_processing bit set, it
 214     // means that this code is called *during* popframe handling - we
 215     // don't want to reenter.
 216     // TODO: Check if all four state combinations could be visible.
 217     // If (processing and !pending) is an invisible/impossible state,
 218     // there is optimization potential by testing both bits at once.
 219     // Then, All_Zeroes and All_Ones means skip, Mixed means doit.
 220     testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
 221             exact_log2(JavaThread::popframe_pending_bit));
 222     z_bfalse(L);
 223     testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
 224             exact_log2(JavaThread::popframe_processing_bit));
 225     z_btrue(L);
 226 
 227     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 228     // address of the same-named entrypoint in the generated interpreter code.
 229     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 230     // The above call should (as its only effect) return the contents of the field
 231     // _remove_activation_preserving_args_entry in Z_RET.
 232     // We just jump there to have the work done.
 233     z_br(Z_RET);
 234     // There is no way for control to fall thru here.
 235 
 236     bind(L);
 237     BLOCK_COMMENT("} check_and_handle_popframe");
 238   }
 239 }
 240 
 241 
 242 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 243   Register RjvmtiState = Z_R1_scratch;
 244   int      tos_off     = in_bytes(JvmtiThreadState::earlyret_tos_offset());
 245   int      oop_off     = in_bytes(JvmtiThreadState::earlyret_oop_offset());
 246   int      val_off     = in_bytes(JvmtiThreadState::earlyret_value_offset());
 247   int      state_off   = in_bytes(JavaThread::jvmti_thread_state_offset());
 248 
 249   z_lg(RjvmtiState, state_off, Z_thread);
 250 
 251   switch (state) {
 252     case atos: z_lg(Z_tos, oop_off, RjvmtiState);
 253       store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch);
 254                                                     break;
 255     case ltos: z_lg(Z_tos, val_off, RjvmtiState);   break;
 256     case btos: // fall through
 257     case ztos: // fall through
 258     case ctos: // fall through
 259     case stos: // fall through
 260     case itos: z_llgf(Z_tos, val_off, RjvmtiState); break;
 261     case ftos: z_le(Z_ftos, val_off, RjvmtiState);  break;
 262     case dtos: z_ld(Z_ftos, val_off, RjvmtiState);  break;
 263     case vtos:   /* nothing to do */                break;
 264     default  : ShouldNotReachHere();
 265   }
 266 
 267   // Clean up tos value in the jvmti thread state.
 268   store_const(Address(RjvmtiState, val_off),   0L, 8, 8, Z_R0_scratch);
 269   // Set tos state field to illegal value.
 270   store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch);
 271 }
 272 
 273 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
 274   if (JvmtiExport::can_force_early_return()) {
 275     BLOCK_COMMENT("check_and_handle_earlyret {");
 276     Label L;
 277     // arg regs are save, because we are just behind the call in call_VM_base
 278     Register jvmti_thread_state = Z_ARG2;
 279     Register tmp                = Z_ARG3;
 280     load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset()));
 281     z_bre(L); // if (thread->jvmti_thread_state() == nullptr) exit;
 282 
 283     // Initiate earlyret handling only if it is not already being processed.
 284     // If the flag has the earlyret_processing bit set, it means that this code
 285     // is called *during* earlyret handling - we don't want to reenter.
 286 
 287     assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0),
 288           "must fix this check, when changing the values of the earlyret enum");
 289     assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum");
 290 
 291     load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset()));
 292     z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit;
 293 
 294     // Call Interpreter::remove_activation_early_entry() to get the address of the
 295     // same-named entrypoint in the generated interpreter code.
 296     assert(sizeof(TosState) == 4, "unexpected size");
 297     z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset()));
 298     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1);
 299     // The above call should (as its only effect) return the contents of the field
 300     // _remove_activation_preserving_args_entry in Z_RET.
 301     // We just jump there to have the work done.
 302     z_br(Z_RET);
 303     // There is no way for control to fall thru here.
 304 
 305     bind(L);
 306     BLOCK_COMMENT("} check_and_handle_earlyret");
 307   }
 308 }
 309 
 310 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
 311   lgr_if_needed(Z_ARG1, arg_1);
 312   assert(arg_2 != Z_ARG1, "smashed argument");
 313   lgr_if_needed(Z_ARG2, arg_2);
 314   MacroAssembler::call_VM_leaf_base(entry_point, true);
 315 }
 316 
 317 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) {
 318   Address param(Z_bcp, bcp_offset);
 319 
 320   BLOCK_COMMENT("get_cache_index_at_bcp {");
 321   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 322   if (index_size == sizeof(u2)) {
 323     load_sized_value(index, param, 2, false /*signed*/);
 324   } else if (index_size == sizeof(u4)) {
 325 
 326     load_sized_value(index, param, 4, false);
 327   } else if (index_size == sizeof(u1)) {
 328     z_llgc(index, param);
 329   } else {
 330     ShouldNotReachHere();
 331   }
 332   BLOCK_COMMENT("}");
 333 }
 334 
 335 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
 336   // Get index out of bytecode pointer.
 337   get_cache_index_at_bcp(index, 1, sizeof(u4));
 338 
 339   // Get the address of the ResolvedIndyEntry array
 340   get_constant_pool_cache(cache);
 341   z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
 342 
 343   // Scale the index to form a byte offset into the ResolvedIndyEntry array
 344   size_t entry_size = sizeof(ResolvedIndyEntry);
 345   if (is_power_of_2(entry_size)) {
 346     z_sllg(index, index, exact_log2(entry_size));
 347   } else {
 348     z_mghi(index, entry_size);
 349   }
 350 
 351   // Calculate the final field address.
 352   z_la(cache, Array<ResolvedIndyEntry>::base_offset_in_bytes(), index, cache);
 353 }
 354 
 355 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
 356   // Get field index out of bytecode pointer.
 357   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
 358 
 359   // Get the address of the ResolvedFieldEntry array.
 360   get_constant_pool_cache(cache);
 361   z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::field_entries_offset())));
 362 
 363   // Scale the index to form a byte offset into the ResolvedFieldEntry array
 364   size_t entry_size = sizeof(ResolvedFieldEntry);
 365   if (is_power_of_2(entry_size)) {
 366     z_sllg(index, index, exact_log2(entry_size));
 367   } else {
 368     z_mghi(index, entry_size);
 369   }
 370 
 371   // Calculate the final field address.
 372   z_la(cache, Array<ResolvedFieldEntry>::base_offset_in_bytes(), index, cache);
 373 }
 374 
 375 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
 376   // Get field index out of bytecode pointer.
 377   get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
 378 
 379   // Get the address of the ResolvedMethodEntry array.
 380   get_constant_pool_cache(cache);
 381   z_lg(cache, Address(cache, in_bytes(ConstantPoolCache::method_entries_offset())));
 382 
 383   // Scale the index to form a byte offset into the ResolvedMethodEntry array
 384   size_t entry_size = sizeof(ResolvedMethodEntry);
 385   if (is_power_of_2(entry_size)) {
 386     z_sllg(index, index, exact_log2(entry_size));
 387   } else {
 388     z_mghi(index, entry_size);
 389   }
 390 
 391   // Calculate the final field address.
 392   z_la(cache, Array<ResolvedMethodEntry>::base_offset_in_bytes(), index, cache);
 393 }
 394 
 395 // Load object from cpool->resolved_references(index).
 396 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
 397   assert_different_registers(result, index);
 398   get_constant_pool(result);
 399 
 400   // Convert
 401   //  - from field index to resolved_references() index and
 402   //  - from word index to byte offset.
 403   // Since this is a java object, it is potentially compressed.
 404   Register tmp = index;  // reuse
 405   z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array.
 406   // Load pointer for resolved_references[] objArray.
 407   z_lg(result, in_bytes(ConstantPool::cache_offset()), result);
 408   z_lg(result, in_bytes(ConstantPoolCache::resolved_references_offset()), result);
 409   resolve_oop_handle(result); // Load resolved references array itself.
 410 #ifdef ASSERT
 411   NearLabel index_ok;
 412   z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes()));
 413   z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop);
 414   compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok);
 415   stop("resolved reference index out of bounds", 0x09256);
 416   bind(index_ok);
 417 #endif
 418   z_agr(result, index);    // Address of indexed array element.
 419   load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp, noreg);
 420 }
 421 
 422 // load cpool->resolved_klass_at(index)
 423 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) {
 424   // int value = *(Rcpool->int_at_addr(which));
 425   // int resolved_klass_index = extract_low_short_from_int(value);
 426   z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian)
 427   z_sllg(offset, offset, LogBytesPerWord);                          // Convert 'index' to 'offset'
 428   z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset())); // iklass = cpool->_resolved_klasses
 429   z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes()));
 430 }
 431 
 432 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
 433 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
 434 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 435                                                   Register Rsuper_klass,
 436                                                   Register Rtmp1,
 437                                                   Register Rtmp2,
 438                                                   Label &ok_is_subtype) {
 439   // Profile the not-null value's klass.
 440   profile_typecheck(Rtmp1, Rsub_klass, Rtmp2);
 441 
 442   // Do the check.
 443   check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
 444 }
 445 
 446 // Pop topmost element from stack. It just disappears.
 447 // Useful if consumed previously by access via stackTop().
 448 void InterpreterMacroAssembler::popx(int len) {
 449   add2reg(Z_esp, len*Interpreter::stackElementSize);
 450   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 451 }
 452 
 453 // Get Address object of stack top. No checks. No pop.
 454 // Purpose: - Provide address of stack operand to exploit reg-mem operations.
 455 //          - Avoid RISC-like mem2reg - reg-reg-op sequence.
 456 Address InterpreterMacroAssembler::stackTop() {
 457   return Address(Z_esp, Interpreter::expr_offset_in_bytes(0));
 458 }
 459 
 460 void InterpreterMacroAssembler::pop_i(Register r) {
 461   z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 462   add2reg(Z_esp, Interpreter::stackElementSize);
 463   assert_different_registers(r, Z_R1_scratch);
 464   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 465 }
 466 
 467 void InterpreterMacroAssembler::pop_ptr(Register r) {
 468   z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 469   add2reg(Z_esp, Interpreter::stackElementSize);
 470   assert_different_registers(r, Z_R1_scratch);
 471   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 472 }
 473 
 474 void InterpreterMacroAssembler::pop_l(Register r) {
 475   z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 476   add2reg(Z_esp, 2*Interpreter::stackElementSize);
 477   assert_different_registers(r, Z_R1_scratch);
 478   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 479 }
 480 
 481 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
 482   mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false);
 483   add2reg(Z_esp, Interpreter::stackElementSize);
 484   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 485 }
 486 
 487 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
 488   mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true);
 489   add2reg(Z_esp, 2*Interpreter::stackElementSize);
 490   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 491 }
 492 
 493 void InterpreterMacroAssembler::push_i(Register r) {
 494   assert_different_registers(r, Z_R1_scratch);
 495   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 496   z_st(r, Address(Z_esp));
 497   add2reg(Z_esp, -Interpreter::stackElementSize);
 498 }
 499 
 500 void InterpreterMacroAssembler::push_ptr(Register r) {
 501   z_stg(r, Address(Z_esp));
 502   add2reg(Z_esp, -Interpreter::stackElementSize);
 503 }
 504 
 505 void InterpreterMacroAssembler::push_l(Register r) {
 506   assert_different_registers(r, Z_R1_scratch);
 507   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 508   int offset = -Interpreter::stackElementSize;
 509   z_stg(r, Address(Z_esp, offset));
 510   clear_mem(Address(Z_esp), Interpreter::stackElementSize);
 511   add2reg(Z_esp, 2 * offset);
 512 }
 513 
 514 void InterpreterMacroAssembler::push_f(FloatRegister f) {
 515   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 516   freg2mem_opt(f, Address(Z_esp), false);
 517   add2reg(Z_esp, -Interpreter::stackElementSize);
 518 }
 519 
 520 void InterpreterMacroAssembler::push_d(FloatRegister d) {
 521   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 522   int offset = -Interpreter::stackElementSize;
 523   freg2mem_opt(d, Address(Z_esp, offset));
 524   add2reg(Z_esp, 2 * offset);
 525 }
 526 
 527 void InterpreterMacroAssembler::push(TosState state) {
 528   verify_oop(Z_tos, state);
 529   switch (state) {
 530     case atos: push_ptr();           break;
 531     case btos: push_i();             break;
 532     case ztos:
 533     case ctos:
 534     case stos: push_i();             break;
 535     case itos: push_i();             break;
 536     case ltos: push_l();             break;
 537     case ftos: push_f();             break;
 538     case dtos: push_d();             break;
 539     case vtos: /* nothing to do */   break;
 540     default  : ShouldNotReachHere();
 541   }
 542 }
 543 
 544 void InterpreterMacroAssembler::pop(TosState state) {
 545   switch (state) {
 546     case atos: pop_ptr(Z_tos);       break;
 547     case btos: pop_i(Z_tos);         break;
 548     case ztos:
 549     case ctos:
 550     case stos: pop_i(Z_tos);         break;
 551     case itos: pop_i(Z_tos);         break;
 552     case ltos: pop_l(Z_tos);         break;
 553     case ftos: pop_f(Z_ftos);        break;
 554     case dtos: pop_d(Z_ftos);        break;
 555     case vtos: /* nothing to do */   break;
 556     default  : ShouldNotReachHere();
 557   }
 558   verify_oop(Z_tos, state);
 559 }
 560 
 561 // Helpers for swap and dup.
 562 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 563   z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
 564 }
 565 
 566 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 567   z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
 568 }
 569 
 570 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) {
 571   // Satisfy interpreter calling convention (see generate_normal_entry()).
 572   z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp).
 573   // Record top_frame_sp, because the callee might modify it, if it's compiled.
 574   z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp);
 575   save_bcp();
 576   save_esp();
 577   z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
 578 }
 579 
 580 // Jump to from_interpreted entry of a call unless single stepping is possible
 581 // in this thread in which case we must call the i2i entry.
 582 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 583   assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
 584   prepare_to_jump_from_interpreted(method);
 585 
 586   if (JvmtiExport::can_post_interpreter_events()) {
 587     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 588     // compiled code in threads for which the event is enabled. Check here for
 589     // interp_only_mode if these events CAN be enabled.
 590     z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
 591     MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
 592     z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
 593     // Run interpreted.
 594     z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
 595     z_br(Z_R1_scratch);
 596   } else {
 597     // Run compiled code.
 598     z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
 599     z_br(Z_R1_scratch);
 600   }
 601 }
 602 
 603 #ifdef ASSERT
 604 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
 605   // About to read or write Resp[0].
 606   // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
 607   address reentry = nullptr;
 608 
 609   {
 610     // Check if the frame pointer in Z_fp is correct.
 611     NearLabel OK;
 612     z_cg(Z_fp, 0, Z_SP);
 613     z_bre(OK);
 614     reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
 615     bind(OK);
 616   }
 617   {
 618     // Resp must not point into or below the operand stack,
 619     // i.e. IJAVA_STATE.monitors > Resp.
 620     NearLabel OK;
 621     Register Rmonitors = Rtemp;
 622     z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp);
 623     compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
 624     reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
 625     bind(OK);
 626   }
 627   {
 628     // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
 629     // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
 630     NearLabel OK;
 631     Register Rabi_bottom = Rtemp;
 632     add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
 633     compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
 634     reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
 635     bind(OK);
 636   }
 637 }
 638 
 639 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
 640   Label magic_ok;
 641   load_const_optimized(tmp, frame::z_istate_magic_number);
 642   z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
 643   z_bre(magic_ok);
 644   stop_static("error: wrong magic number in ijava_state access");
 645   bind(magic_ok);
 646 }
 647 #endif // ASSERT
 648 
 649 void InterpreterMacroAssembler::save_bcp() {
 650   z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
 651   asm_assert_ijava_state_magic(Z_bcp);
 652   NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
 653 }
 654 
 655 void InterpreterMacroAssembler::restore_bcp() {
 656   asm_assert_ijava_state_magic(Z_bcp);
 657   z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
 658 }
 659 
 660 void InterpreterMacroAssembler::save_esp() {
 661   z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
 662 }
 663 
 664 void InterpreterMacroAssembler::restore_esp() {
 665   asm_assert_ijava_state_magic(Z_esp);
 666   z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
 667 }
 668 
 669 void InterpreterMacroAssembler::get_monitors(Register reg) {
 670   asm_assert_ijava_state_magic(reg);
 671   mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
 672 }
 673 
 674 void InterpreterMacroAssembler::save_monitors(Register reg) {
 675   reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
 676 }
 677 
 678 void InterpreterMacroAssembler::get_mdp(Register mdp) {
 679   z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
 680 }
 681 
 682 void InterpreterMacroAssembler::save_mdp(Register mdp) {
 683   z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
 684 }
 685 
 686 // Values that are only read (besides initialization).
 687 void InterpreterMacroAssembler::restore_locals() {
 688   asm_assert_ijava_state_magic(Z_locals);
 689   z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
 690 }
 691 
 692 void InterpreterMacroAssembler::get_method(Register reg) {
 693   asm_assert_ijava_state_magic(reg);
 694   z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
 695 }
 696 
 697 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
 698                                                           signedOrNot is_signed) {
 699   // Rdst is an 8-byte return value!!!
 700 
 701   // Unaligned loads incur only a small penalty on z/Architecture. The penalty
 702   // is a few (2..3) ticks, even when the load crosses a cache line
 703   // boundary. In case of a cache miss, the stall could, of course, be
 704   // much longer.
 705 
 706   switch (is_signed) {
 707     case Signed:
 708       z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
 709      break;
 710    case Unsigned:
 711      z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
 712      break;
 713    default:
 714      ShouldNotReachHere();
 715   }
 716 }
 717 
 718 
 719 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
 720                                                           setCCOrNot set_cc) {
 721   // Rdst is an 8-byte return value!!!
 722 
 723   // Unaligned loads incur only a small penalty on z/Architecture. The penalty
 724   // is a few (2..3) ticks, even when the load crosses a cache line
 725   // boundary. In case of a cache miss, the stall could, of course, be
 726   // much longer.
 727 
 728   // Both variants implement a sign-extending int2long load.
 729   if (set_cc == set_CC) {
 730     load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
 731   } else {
 732     mem2reg_signed_opt(    Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
 733   }
 734 }
 735 
 736 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
 737   get_method(Rdst);
 738   mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
 739   mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
 740 }
 741 
 742 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
 743   get_constant_pool(Rdst);
 744   mem2reg_opt(Rdst, Address(Rdst, ConstantPool::cache_offset()));
 745 }
 746 
 747 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
 748   get_constant_pool(Rcpool);
 749   mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset()));
 750 }
 751 
 752 // Unlock if synchronized method.
 753 //
 754 // Unlock the receiver if this is a synchronized method.
 755 // Unlock any Java monitors from synchronized blocks.
 756 //
 757 // If there are locked Java monitors
 758 //   If throw_monitor_exception
 759 //     throws IllegalMonitorStateException
 760 //   Else if install_monitor_exception
 761 //     installs IllegalMonitorStateException
 762 //   Else
 763 //     no error processing
 764 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
 765                                                               bool throw_monitor_exception,
 766                                                               bool install_monitor_exception) {
 767   NearLabel unlocked, unlock, no_unlock;
 768 
 769   {
 770     Register R_method = Z_ARG2;
 771     Register R_do_not_unlock_if_synchronized = Z_ARG3;
 772 
 773     // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
 774     const Address do_not_unlock_if_synchronized(Z_thread,
 775                                                 JavaThread::do_not_unlock_if_synchronized_offset());
 776     load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
 777     z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
 778 
 779     // Check if synchronized method.
 780     get_method(R_method);
 781     verify_oop(Z_tos, state);
 782     push(state); // Save tos/result.
 783     testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 784     z_bfalse(unlocked);
 785 
 786     // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 787     // is set.
 788     compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
 789   }
 790 
 791   // unlock monitor
 792 
 793   // BasicObjectLock will be first in list, since this is a
 794   // synchronized method. However, need to check that the object has
 795   // not been unlocked by an explicit monitorexit bytecode.
 796   const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
 797   // We use Z_ARG2 so that if we go slow path it will be the correct
 798   // register for unlock_object to pass to VM directly.
 799   load_address(Z_ARG2, monitor); // Address of first monitor.
 800   z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset()));
 801   compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
 802 
 803   if (throw_monitor_exception) {
 804     // Entry already unlocked need to throw an exception.
 805     MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
 806     should_not_reach_here();
 807   } else {
 808     // Monitor already unlocked during a stack unroll.
 809     // If requested, install an illegal_monitor_state_exception.
 810     // Continue with stack unrolling.
 811     if (install_monitor_exception) {
 812       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
 813     }
 814    z_bru(unlocked);
 815   }
 816 
 817   bind(unlock);
 818 
 819   unlock_object(Z_ARG2);
 820 
 821   bind(unlocked);
 822 
 823   // I0, I1: Might contain return value
 824 
 825   // Check that all monitors are unlocked.
 826   {
 827     NearLabel loop, exception, entry, restart;
 828     const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 829     // We use Z_ARG2 so that if we go slow path it will be the correct
 830     // register for unlock_object to pass to VM directly.
 831     Register R_current_monitor = Z_ARG2;
 832     Register R_monitor_block_bot = Z_ARG1;
 833     const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors));
 834     const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
 835 
 836     bind(restart);
 837     // Starting with top-most entry.
 838     z_lg(R_current_monitor, monitor_block_top);
 839     // Points to word before bottom of monitor block.
 840     load_address(R_monitor_block_bot, monitor_block_bot);
 841     z_bru(entry);
 842 
 843     // Entry already locked, need to throw exception.
 844     bind(exception);
 845 
 846     if (throw_monitor_exception) {
 847       // Throw exception.
 848       MacroAssembler::call_VM(noreg,
 849                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 850                                                throw_illegal_monitor_state_exception));
 851       should_not_reach_here();
 852     } else {
 853       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 854       // Unlock does not block, so don't have to worry about the frame.
 855       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 856       unlock_object(R_current_monitor);
 857       if (install_monitor_exception) {
 858         call_VM(noreg, CAST_FROM_FN_PTR(address,
 859                                         InterpreterRuntime::
 860                                         new_illegal_monitor_state_exception));
 861       }
 862       z_bru(restart);
 863     }
 864 
 865     bind(loop);
 866     // Check if current entry is used.
 867     load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset()));
 868     z_brne(exception);
 869 
 870     add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
 871     bind(entry);
 872     compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
 873   }
 874 
 875   bind(no_unlock);
 876   pop(state);
 877   verify_oop(Z_tos, state);
 878 }
 879 
 880 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) {
 881   get_method(ret_type);
 882   z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset())));
 883   z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset())));
 884 
 885   Label notBool, notByte, notChar, done;
 886 
 887   // common case first
 888   compareU32_and_branch(ret_type, T_INT, bcondEqual, done);
 889 
 890   compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool);
 891   z_nilf(result, 0x1);
 892   z_bru(done);
 893 
 894   bind(notBool);
 895   compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte);
 896   z_lbr(result, result);
 897   z_bru(done);
 898 
 899   bind(notByte);
 900   compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar);
 901   z_nilf(result, 0xffff);
 902   z_bru(done);
 903 
 904   bind(notChar);
 905   // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort);
 906   z_lhr(result, result);
 907 
 908   // Nothing to do for T_INT
 909   bind(done);
 910 }
 911 
 912 // remove activation
 913 //
 914 // Unlock the receiver if this is a synchronized method.
 915 // Unlock any Java monitors from synchronized blocks.
 916 // Remove the activation from the stack.
 917 //
 918 // If there are locked Java monitors
 919 //   If throw_monitor_exception
 920 //     throws IllegalMonitorStateException
 921 //   Else if install_monitor_exception
 922 //     installs IllegalMonitorStateException
 923 //   Else
 924 //     no error processing
 925 void InterpreterMacroAssembler::remove_activation(TosState state,
 926                                                   Register return_pc,
 927                                                   bool throw_monitor_exception,
 928                                                   bool install_monitor_exception,
 929                                                   bool notify_jvmti) {
 930   BLOCK_COMMENT("remove_activation {");
 931   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
 932 
 933   // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
 934   notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
 935 
 936   if (StackReservedPages > 0) {
 937     BLOCK_COMMENT("reserved_stack_check:");
 938     // Test if reserved zone needs to be enabled.
 939     Label no_reserved_zone_enabling;
 940 
 941     // check if already enabled - if so no re-enabling needed
 942     assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
 943     z_ly(Z_R0, Address(Z_thread, JavaThread::stack_guard_state_offset()));
 944     compare32_and_branch(Z_R0, StackOverflow::stack_guard_enabled, bcondEqual, no_reserved_zone_enabling);
 945 
 946     // Compare frame pointers. There is no good stack pointer, as with stack
 947     // frame compression we can get different SPs when we do calls. A subsequent
 948     // call could have a smaller SP, so that this compare succeeds for an
 949     // inner call of the method annotated with ReservedStack.
 950     z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
 951     z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
 952     z_brl(no_reserved_zone_enabling);
 953 
 954     // Enable reserved zone again, throw stack overflow exception.
 955     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
 956     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
 957 
 958     should_not_reach_here();
 959 
 960     bind(no_reserved_zone_enabling);
 961   }
 962 
 963   verify_oop(Z_tos, state);
 964 
 965   pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
 966   BLOCK_COMMENT("} remove_activation");
 967 }
 968 
 969 // lock object
 970 //
 971 // Registers alive
 972 //   monitor (Z_R10) - Address of the BasicObjectLock to be used for locking,
 973 //             which must be initialized with the object to lock.
 974 //   object  (Z_R11, Z_R2) - Address of the object to be locked.
 975 //  templateTable (monitorenter) is using Z_R2 for object
 976 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
 977 
 978   if (LockingMode == LM_MONITOR) {
 979     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor);
 980     return;
 981   }
 982 
 983   // template code: (for LM_LEGACY)
 984   //
 985   // markWord displaced_header = obj->mark().set_unlocked();
 986   // monitor->lock()->set_displaced_header(displaced_header);
 987   // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
 988   //   // We stored the monitor address into the object's mark word.
 989   // } else if (THREAD->is_lock_owned((address)displaced_header))
 990   //   // Simple recursive case.
 991   //   monitor->lock()->set_displaced_header(nullptr);
 992   // } else {
 993   //   // Slow path.
 994   //   InterpreterRuntime::monitorenter(THREAD, monitor);
 995   // }
 996 
 997   const int hdr_offset = oopDesc::mark_offset_in_bytes();
 998 
 999   const Register header           = Z_ARG5;
1000   const Register object_mark_addr = Z_ARG4;
1001   const Register current_header   = Z_ARG5;
1002   const Register tmp              = Z_R1_scratch;
1003 
1004   NearLabel done, slow_case;
1005 
1006   // markWord header = obj->mark().set_unlocked();
1007 
1008   // Load markWord from object into header.
1009   z_lg(header, hdr_offset, object);
1010 
1011   if (DiagnoseSyncOnValueBasedClasses != 0) {
1012     load_klass(tmp, object);
1013     testbit(Address(tmp, Klass::access_flags_offset()), exact_log2(JVM_ACC_IS_VALUE_BASED_CLASS));
1014     z_btrue(slow_case);
1015   }
1016 
1017   if (LockingMode == LM_LIGHTWEIGHT) {
1018     lightweight_lock(object, /* mark word */ header, tmp, slow_case);
1019   } else if (LockingMode == LM_LEGACY) {
1020 
1021     // Set header to be (markWord of object | UNLOCK_VALUE).
1022     // This will not change anything if it was unlocked before.
1023     z_oill(header, markWord::unlocked_value);
1024 
1025     // monitor->lock()->set_displaced_header(displaced_header);
1026     const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1027     const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
1028 
1029     // Initialize the box (Must happen before we update the object mark!).
1030     z_stg(header, mark_offset, monitor);
1031 
1032     // if (Atomic::cmpxchg(/*addr*/obj->mark_addr(), /*cmp*/displaced_header, /*ex=*/monitor) == displaced_header) {
1033 
1034     // not necessary, use offset in instruction directly.
1035     // add2reg(object_mark_addr, hdr_offset, object);
1036 
1037     // Store stack address of the BasicObjectLock (this is monitor) into object.
1038     z_csg(header, monitor, hdr_offset, object);
1039     assert(current_header == header,
1040            "must be same register"); // Identified two registers from z/Architecture.
1041 
1042     z_bre(done);
1043 
1044     // } else if (THREAD->is_lock_owned((address)displaced_header))
1045     //   // Simple recursive case.
1046     //   monitor->lock()->set_displaced_header(nullptr);
1047 
1048     // We did not see an unlocked object so try the fast recursive case.
1049 
1050     // Check if owner is self by comparing the value in the markWord of object
1051     // (current_header) with the stack pointer.
1052     z_sgr(current_header, Z_SP);
1053 
1054     assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1055 
1056     // The prior sequence "LGR, NGR, LTGR" can be done better
1057     // (Z_R1 is temp and not used after here).
1058     load_const_optimized(Z_R0, (~(os::vm_page_size() - 1) | markWord::lock_mask_in_place));
1059     z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0)
1060 
1061     // If condition is true we are done and hence we can store 0 in the displaced
1062     // header indicating it is a recursive lock and be done.
1063     z_brne(slow_case);
1064     z_release();  // Member unnecessary on zarch AND because the above csg does a sync before and after.
1065     z_stg(Z_R0/*==0!*/, mark_offset, monitor);
1066   }
1067   z_bru(done);
1068   // } else {
1069   //   // Slow path.
1070   //   InterpreterRuntime::monitorenter(THREAD, monitor);
1071 
1072   // None of the above fast optimizations worked so we have to get into the
1073   // slow case of monitor enter.
1074   bind(slow_case);
1075   if (LockingMode == LM_LIGHTWEIGHT) {
1076     // for lightweight locking we need to use monitorenter_obj, see interpreterRuntime.cpp
1077     call_VM(noreg,
1078             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
1079             object);
1080   } else {
1081     call_VM(noreg,
1082             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1083             monitor);
1084   }
1085   // }
1086 
1087   bind(done);
1088 }
1089 
1090 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1091 //
1092 // Registers alive
1093 //   monitor - address of the BasicObjectLock to be used for locking,
1094 //             which must be initialized with the object to lock.
1095 //
1096 // Throw IllegalMonitorException if object is not locked by current thread.
1097 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1098 
1099   if (LockingMode == LM_MONITOR) {
1100     call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1101     return;
1102   }
1103 
1104 // else {
1105   // template code: (for LM_LEGACY):
1106   //
1107   // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1108   //   // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1109   //   monitor->set_obj(nullptr);
1110   // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1111   //   // We swapped the unlocked mark in displaced_header into the object's mark word.
1112   //   monitor->set_obj(nullptr);
1113   // } else {
1114   //   // Slow path.
1115   //   InterpreterRuntime::monitorexit(monitor);
1116   // }
1117 
1118   const int hdr_offset = oopDesc::mark_offset_in_bytes();
1119 
1120   const Register header         = Z_ARG4;
1121   const Register current_header = Z_R1_scratch;
1122   Address obj_entry(monitor, BasicObjectLock::obj_offset());
1123   Label done, slow_case;
1124 
1125   if (object == noreg) {
1126     // In the template interpreter, we must assure that the object
1127     // entry in the monitor is cleared on all paths. Thus we move
1128     // loading up to here, and clear the entry afterwards.
1129     object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1130     z_lg(object, obj_entry);
1131   }
1132 
1133   assert_different_registers(monitor, object, header, current_header);
1134 
1135   // if ((displaced_header = monitor->displaced_header()) == nullptr) {
1136   //   // Recursive unlock. Mark the monitor unlocked by setting the object field to null.
1137   //   monitor->set_obj(nullptr);
1138 
1139   // monitor->lock()->set_displaced_header(displaced_header);
1140   const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1141   const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
1142 
1143   clear_mem(obj_entry, sizeof(oop));
1144   if (LockingMode != LM_LIGHTWEIGHT) {
1145     // Test first if we are in the fast recursive case.
1146     MacroAssembler::load_and_test_long(header, Address(monitor, mark_offset));
1147     z_bre(done); // header == 0 -> goto done
1148   }
1149 
1150   // } else if (Atomic::cmpxchg(obj->mark_addr(), monitor, displaced_header) == monitor) {
1151   //   // We swapped the unlocked mark in displaced_header into the object's mark word.
1152   //   monitor->set_obj(nullptr);
1153 
1154   // If we still have a lightweight lock, unlock the object and be done.
1155   if (LockingMode == LM_LIGHTWEIGHT) {
1156     // Check for non-symmetric locking. This is allowed by the spec and the interpreter
1157     // must handle it.
1158 
1159     Register tmp = current_header;
1160 
1161     // First check for lock-stack underflow.
1162     z_lgf(tmp, Address(Z_thread, JavaThread::lock_stack_top_offset()));
1163     compareU32_and_branch(tmp, (unsigned)LockStack::start_offset(), Assembler::bcondNotHigh, slow_case);
1164 
1165     // Then check if the top of the lock-stack matches the unlocked object.
1166     z_aghi(tmp, -oopSize);
1167     z_lg(tmp, Address(Z_thread, tmp));
1168     compare64_and_branch(tmp, object, Assembler::bcondNotEqual, slow_case);
1169 
1170     z_lg(header, Address(object, hdr_offset));
1171     z_lgr(tmp, header);
1172     z_nill(tmp, markWord::monitor_value);
1173     z_brne(slow_case);
1174 
1175     lightweight_unlock(object, header, tmp, slow_case);
1176 
1177     z_bru(done);
1178   } else {
1179     // The markword is expected to be at offset 0.
1180     // This is not required on s390, at least not here.
1181     assert(hdr_offset == 0, "unlock_object: review code below");
1182 
1183     // We have the displaced header in header. If the lock is still
1184     // lightweight, it will contain the monitor address and we'll store the
1185     // displaced header back into the object's mark word.
1186     z_lgr(current_header, monitor);
1187     z_csg(current_header, header, hdr_offset, object);
1188     z_bre(done);
1189   }
1190 
1191   // } else {
1192   //   // Slow path.
1193   //   InterpreterRuntime::monitorexit(monitor);
1194 
1195   // The lock has been converted into a heavy lock and hence
1196   // we need to get into the slow case.
1197   bind(slow_case);
1198   z_stg(object, obj_entry);   // Restore object entry, has been cleared above.
1199   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), monitor);
1200 
1201   // }
1202 
1203   bind(done);
1204 }
1205 
1206 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1207   assert(ProfileInterpreter, "must be profiling interpreter");
1208   load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1209   z_brz(zero_continue);
1210 }
1211 
1212 // Set the method data pointer for the current bcp.
1213 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1214   assert(ProfileInterpreter, "must be profiling interpreter");
1215   Label    set_mdp;
1216   Register mdp    = Z_ARG4;
1217   Register method = Z_ARG5;
1218 
1219   get_method(method);
1220   // Test MDO to avoid the call if it is null.
1221   load_and_test_long(mdp, method2_(method, method_data));
1222   z_brz(set_mdp);
1223 
1224   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1225   // Z_RET: mdi
1226   // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1227   assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1228   z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1229   add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1230 
1231   bind(set_mdp);
1232   save_mdp(mdp);
1233 }
1234 
1235 void InterpreterMacroAssembler::verify_method_data_pointer() {
1236   assert(ProfileInterpreter, "must be profiling interpreter");
1237 #ifdef ASSERT
1238   NearLabel verify_continue;
1239   Register bcp_expected = Z_ARG3;
1240   Register mdp    = Z_ARG4;
1241   Register method = Z_ARG5;
1242 
1243   test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1244   get_method(method);
1245 
1246   // If the mdp is valid, it will point to a DataLayout header which is
1247   // consistent with the bcp. The converse is highly probable also.
1248   load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1249   z_ag(bcp_expected, Address(method, Method::const_offset()));
1250   load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1251   compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1252   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1253   bind(verify_continue);
1254 #endif // ASSERT
1255 }
1256 
1257 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1258   assert(ProfileInterpreter, "must be profiling interpreter");
1259   z_stg(value, constant, mdp_in);
1260 }
1261 
1262 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1263                                                       int constant,
1264                                                       Register tmp,
1265                                                       bool decrement) {
1266   assert_different_registers(mdp_in, tmp);
1267   // counter address
1268   Address data(mdp_in, constant);
1269   const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1270   add2mem_64(Address(mdp_in, constant), delta, tmp);
1271 }
1272 
1273 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1274                                                 int flag_byte_constant) {
1275   assert(ProfileInterpreter, "must be profiling interpreter");
1276   // Set the flag.
1277   z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1278 }
1279 
1280 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1281                                                  int offset,
1282                                                  Register value,
1283                                                  Register test_value_out,
1284                                                  Label& not_equal_continue) {
1285   assert(ProfileInterpreter, "must be profiling interpreter");
1286   if (test_value_out == noreg) {
1287     z_cg(value, Address(mdp_in, offset));
1288     z_brne(not_equal_continue);
1289   } else {
1290     // Put the test value into a register, so caller can use it:
1291     z_lg(test_value_out, Address(mdp_in, offset));
1292     compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1293   }
1294 }
1295 
1296 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1297   update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1298 }
1299 
1300 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1301                                                      Register dataidx,
1302                                                      int offset_of_disp) {
1303   assert(ProfileInterpreter, "must be profiling interpreter");
1304   Address disp_address(mdp_in, dataidx, offset_of_disp);
1305   Assembler::z_ag(mdp_in, disp_address);
1306   save_mdp(mdp_in);
1307 }
1308 
1309 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1310   assert(ProfileInterpreter, "must be profiling interpreter");
1311   add2reg(mdp_in, constant);
1312   save_mdp(mdp_in);
1313 }
1314 
1315 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1316   assert(ProfileInterpreter, "must be profiling interpreter");
1317   assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1318   call_VM(noreg,
1319           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1320           return_bci);
1321 }
1322 
1323 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1324   if (ProfileInterpreter) {
1325     Label profile_continue;
1326 
1327     // If no method data exists, go to profile_continue.
1328     // Otherwise, assign to mdp.
1329     test_method_data_pointer(mdp, profile_continue);
1330 
1331     // We are taking a branch. Increment the taken count.
1332     // We inline increment_mdp_data_at to return bumped_count in a register
1333     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1334     Address data(mdp, JumpData::taken_offset());
1335     z_lg(bumped_count, data);
1336     // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1337     // performed when reading the counts.
1338     add2reg(bumped_count, DataLayout::counter_increment);
1339     z_stg(bumped_count, data); // Store back out
1340 
1341     // The method data pointer needs to be updated to reflect the new target.
1342     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1343     bind(profile_continue);
1344   }
1345 }
1346 
1347 // Kills Z_R1_scratch.
1348 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1349   if (ProfileInterpreter) {
1350     Label profile_continue;
1351 
1352     // If no method data exists, go to profile_continue.
1353     test_method_data_pointer(mdp, profile_continue);
1354 
1355     // We are taking a branch. Increment the not taken count.
1356     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1357 
1358     // The method data pointer needs to be updated to correspond to
1359     // the next bytecode.
1360     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1361     bind(profile_continue);
1362   }
1363 }
1364 
1365 // Kills: Z_R1_scratch.
1366 void InterpreterMacroAssembler::profile_call(Register mdp) {
1367   if (ProfileInterpreter) {
1368     Label profile_continue;
1369 
1370     // If no method data exists, go to profile_continue.
1371     test_method_data_pointer(mdp, profile_continue);
1372 
1373     // We are making a call. Increment the count.
1374     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1375 
1376     // The method data pointer needs to be updated to reflect the new target.
1377     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1378     bind(profile_continue);
1379   }
1380 }
1381 
1382 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1383   if (ProfileInterpreter) {
1384     Label profile_continue;
1385 
1386     // If no method data exists, go to profile_continue.
1387     test_method_data_pointer(mdp, profile_continue);
1388 
1389     // We are making a call. Increment the count.
1390     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1391 
1392     // The method data pointer needs to be updated to reflect the new target.
1393     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1394     bind(profile_continue);
1395   }
1396 }
1397 
1398 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1399                                                      Register mdp,
1400                                                      Register reg2,
1401                                                      bool receiver_can_be_null) {
1402   if (ProfileInterpreter) {
1403     NearLabel profile_continue;
1404 
1405     // If no method data exists, go to profile_continue.
1406     test_method_data_pointer(mdp, profile_continue);
1407 
1408     NearLabel skip_receiver_profile;
1409     if (receiver_can_be_null) {
1410       NearLabel not_null;
1411       compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1412       // We are making a call. Increment the count for null receiver.
1413       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1414       z_bru(skip_receiver_profile);
1415       bind(not_null);
1416     }
1417 
1418     // Record the receiver type.
1419     record_klass_in_profile(receiver, mdp, reg2);
1420     bind(skip_receiver_profile);
1421 
1422     // The method data pointer needs to be updated to reflect the new target.
1423     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1424     bind(profile_continue);
1425   }
1426 }
1427 
1428 // This routine creates a state machine for updating the multi-row
1429 // type profile at a virtual call site (or other type-sensitive bytecode).
1430 // The machine visits each row (of receiver/count) until the receiver type
1431 // is found, or until it runs out of rows. At the same time, it remembers
1432 // the location of the first empty row. (An empty row records null for its
1433 // receiver, and can be allocated for a newly-observed receiver type.)
1434 // Because there are two degrees of freedom in the state, a simple linear
1435 // search will not work; it must be a decision tree. Hence this helper
1436 // function is recursive, to generate the required tree structured code.
1437 // It's the interpreter, so we are trading off code space for speed.
1438 // See below for example code.
1439 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1440                                         Register receiver, Register mdp,
1441                                         Register reg2, int start_row,
1442                                         Label& done) {
1443   if (TypeProfileWidth == 0) {
1444     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1445     return;
1446   }
1447 
1448   int last_row = VirtualCallData::row_limit() - 1;
1449   assert(start_row <= last_row, "must be work left to do");
1450   // Test this row for both the receiver and for null.
1451   // Take any of three different outcomes:
1452   //   1. found receiver => increment count and goto done
1453   //   2. found null => keep looking for case 1, maybe allocate this cell
1454   //   3. found something else => keep looking for cases 1 and 2
1455   // Case 3 is handled by a recursive call.
1456   for (int row = start_row; row <= last_row; row++) {
1457     NearLabel next_test;
1458     bool test_for_null_also = (row == start_row);
1459 
1460     // See if the receiver is receiver[n].
1461     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1462     test_mdp_data_at(mdp, recvr_offset, receiver,
1463                      (test_for_null_also ? reg2 : noreg),
1464                      next_test);
1465     // (Reg2 now contains the receiver from the CallData.)
1466 
1467     // The receiver is receiver[n]. Increment count[n].
1468     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1469     increment_mdp_data_at(mdp, count_offset);
1470     z_bru(done);
1471     bind(next_test);
1472 
1473     if (test_for_null_also) {
1474       Label found_null;
1475       // Failed the equality check on receiver[n]... Test for null.
1476       z_ltgr(reg2, reg2);
1477       if (start_row == last_row) {
1478         // The only thing left to do is handle the null case.
1479         z_brz(found_null);
1480         // Receiver did not match any saved receiver and there is no empty row for it.
1481         // Increment total counter to indicate polymorphic case.
1482         increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1483         z_bru(done);
1484         bind(found_null);
1485         break;
1486       }
1487       // Since null is rare, make it be the branch-taken case.
1488       z_brz(found_null);
1489 
1490       // Put all the "Case 3" tests here.
1491       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done);
1492 
1493       // Found a null. Keep searching for a matching receiver,
1494       // but remember that this is an empty (unused) slot.
1495       bind(found_null);
1496     }
1497   }
1498 
1499   // In the fall-through case, we found no matching receiver, but we
1500   // observed the receiver[start_row] is null.
1501 
1502   // Fill in the receiver field and increment the count.
1503   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1504   set_mdp_data_at(mdp, recvr_offset, receiver);
1505   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1506   load_const_optimized(reg2, DataLayout::counter_increment);
1507   set_mdp_data_at(mdp, count_offset, reg2);
1508   if (start_row > 0) {
1509     z_bru(done);
1510   }
1511 }
1512 
1513 // Example state machine code for three profile rows:
1514 //   // main copy of decision tree, rooted at row[1]
1515 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1516 //   if (row[0].rec != nullptr) {
1517 //     // inner copy of decision tree, rooted at row[1]
1518 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1519 //     if (row[1].rec != nullptr) {
1520 //       // degenerate decision tree, rooted at row[2]
1521 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1522 //       if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1523 //       row[2].init(rec); goto done;
1524 //     } else {
1525 //       // remember row[1] is empty
1526 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1527 //       row[1].init(rec); goto done;
1528 //     }
1529 //   } else {
1530 //     // remember row[0] is empty
1531 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1532 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1533 //     row[0].init(rec); goto done;
1534 //   }
1535 //   done:
1536 
1537 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1538                                                         Register mdp, Register reg2) {
1539   assert(ProfileInterpreter, "must be profiling");
1540   Label done;
1541 
1542   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1543 
1544   bind (done);
1545 }
1546 
1547 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1548   if (ProfileInterpreter) {
1549     NearLabel profile_continue;
1550     uint row;
1551 
1552     // If no method data exists, go to profile_continue.
1553     test_method_data_pointer(mdp, profile_continue);
1554 
1555     // Update the total ret count.
1556     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1557 
1558     for (row = 0; row < RetData::row_limit(); row++) {
1559       NearLabel next_test;
1560 
1561       // See if return_bci is equal to bci[n]:
1562       test_mdp_data_at(mdp,
1563                        in_bytes(RetData::bci_offset(row)),
1564                        return_bci, noreg,
1565                        next_test);
1566 
1567       // Return_bci is equal to bci[n]. Increment the count.
1568       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1569 
1570       // The method data pointer needs to be updated to reflect the new target.
1571       update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1572       z_bru(profile_continue);
1573       bind(next_test);
1574     }
1575 
1576     update_mdp_for_ret(return_bci);
1577 
1578     bind(profile_continue);
1579   }
1580 }
1581 
1582 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1583   if (ProfileInterpreter) {
1584     Label profile_continue;
1585 
1586     // If no method data exists, go to profile_continue.
1587     test_method_data_pointer(mdp, profile_continue);
1588 
1589     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1590 
1591     // The method data pointer needs to be updated.
1592     int mdp_delta = in_bytes(BitData::bit_data_size());
1593     if (TypeProfileCasts) {
1594       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1595     }
1596     update_mdp_by_constant(mdp, mdp_delta);
1597 
1598     bind(profile_continue);
1599   }
1600 }
1601 
1602 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1603   if (ProfileInterpreter) {
1604     Label profile_continue;
1605 
1606     // If no method data exists, go to profile_continue.
1607     test_method_data_pointer(mdp, profile_continue);
1608 
1609     // The method data pointer needs to be updated.
1610     int mdp_delta = in_bytes(BitData::bit_data_size());
1611     if (TypeProfileCasts) {
1612       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1613 
1614       // Record the object type.
1615       record_klass_in_profile(klass, mdp, reg2);
1616     }
1617     update_mdp_by_constant(mdp, mdp_delta);
1618 
1619     bind(profile_continue);
1620   }
1621 }
1622 
1623 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1624   if (ProfileInterpreter) {
1625     Label profile_continue;
1626 
1627     // If no method data exists, go to profile_continue.
1628     test_method_data_pointer(mdp, profile_continue);
1629 
1630     // Update the default case count.
1631     increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1632 
1633     // The method data pointer needs to be updated.
1634     update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1635 
1636     bind(profile_continue);
1637   }
1638 }
1639 
1640 // Kills: index, scratch1, scratch2.
1641 void InterpreterMacroAssembler::profile_switch_case(Register index,
1642                                                     Register mdp,
1643                                                     Register scratch1,
1644                                                     Register scratch2) {
1645   if (ProfileInterpreter) {
1646     Label profile_continue;
1647     assert_different_registers(index, mdp, scratch1, scratch2);
1648 
1649     // If no method data exists, go to profile_continue.
1650     test_method_data_pointer(mdp, profile_continue);
1651 
1652     // Build the base (index * per_case_size_in_bytes()) +
1653     // case_array_offset_in_bytes().
1654     z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1655     add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1656 
1657     // Add the calculated base to the mdp -> address of the case' data.
1658     Address case_data_addr(mdp, index);
1659     Register case_data = scratch1;
1660     load_address(case_data, case_data_addr);
1661 
1662     // Update the case count.
1663     increment_mdp_data_at(case_data,
1664                           in_bytes(MultiBranchData::relative_count_offset()),
1665                           scratch2);
1666 
1667     // The method data pointer needs to be updated.
1668     update_mdp_by_offset(mdp,
1669                          index,
1670                          in_bytes(MultiBranchData::relative_displacement_offset()));
1671 
1672     bind(profile_continue);
1673   }
1674 }
1675 
1676 // kills: R0, R1, flags, loads klass from obj (if not null)
1677 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1678   NearLabel null_seen, init_klass, do_nothing, do_update;
1679 
1680   // Klass = obj is allowed.
1681   const Register tmp = Z_R1;
1682   assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1683   assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1684 
1685   z_lg(tmp, mdo_addr);
1686   if (cmp_done) {
1687     z_brz(null_seen);
1688   } else {
1689     compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1690   }
1691 
1692   MacroAssembler::verify_oop(obj, FILE_AND_LINE);
1693   load_klass(klass, obj);
1694 
1695   // Klass seen before, nothing to do (regardless of unknown bit).
1696   z_lgr(Z_R0, tmp);
1697   assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1698   z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1699   compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1700 
1701   // Already unknown. Nothing to do anymore.
1702   z_tmll(tmp, TypeEntries::type_unknown);
1703   z_brc(Assembler::bcondAllOne, do_nothing);
1704 
1705   z_lgr(Z_R0, tmp);
1706   assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1707   z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1708   compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1709 
1710   // Different than before. Cannot keep accurate profile.
1711   z_oill(tmp, TypeEntries::type_unknown);
1712   z_bru(do_update);
1713 
1714   bind(init_klass);
1715   // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1716   z_ogr(tmp, klass);
1717   z_bru(do_update);
1718 
1719   bind(null_seen);
1720   // Set null_seen if obj is 0.
1721   z_oill(tmp, TypeEntries::null_seen);
1722   // fallthru: z_bru(do_update);
1723 
1724   bind(do_update);
1725   z_stg(tmp, mdo_addr);
1726 
1727   bind(do_nothing);
1728 }
1729 
1730 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1731   if (!ProfileInterpreter) {
1732     return;
1733   }
1734 
1735   assert_different_registers(mdp, callee, tmp);
1736 
1737   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1738     Label profile_continue;
1739 
1740     test_method_data_pointer(mdp, profile_continue);
1741 
1742     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1743 
1744     z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1745            is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1746     z_brne(profile_continue);
1747 
1748     if (MethodData::profile_arguments()) {
1749       NearLabel done;
1750       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1751       add2reg(mdp, off_to_args);
1752 
1753       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1754         if (i > 0 || MethodData::profile_return()) {
1755           // If return value type is profiled we may have no argument to profile.
1756           z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1757           add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1758           compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1759         }
1760         z_lg(tmp, Address(callee, Method::const_offset()));
1761         z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1762         // Stack offset o (zero based) from the start of the argument
1763         // list. For n arguments translates into offset n - o - 1 from
1764         // the end of the argument list. But there is an extra slot at
1765         // the top of the stack. So the offset is n - o from Lesp.
1766         z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1767         z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1768         Address stack_slot_addr(tmp, Z_esp);
1769         z_ltg(tmp, stack_slot_addr);
1770 
1771         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1772         profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1773 
1774         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1775         add2reg(mdp, to_add);
1776         off_to_args += to_add;
1777       }
1778 
1779       if (MethodData::profile_return()) {
1780         z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1781         add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1782       }
1783 
1784       bind(done);
1785 
1786       if (MethodData::profile_return()) {
1787         // We're right after the type profile for the last
1788         // argument. Tmp is the number of cells left in the
1789         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1790         // if there's a return to profile.
1791         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1792         z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1793         z_agr(mdp, tmp);
1794       }
1795       z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1796     } else {
1797       assert(MethodData::profile_return(), "either profile call args or call ret");
1798       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1799     }
1800 
1801     // Mdp points right after the end of the
1802     // CallTypeData/VirtualCallTypeData, right after the cells for the
1803     // return value type if there's one.
1804     bind(profile_continue);
1805   }
1806 }
1807 
1808 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1809   assert_different_registers(mdp, ret, tmp);
1810   if (ProfileInterpreter && MethodData::profile_return()) {
1811     Label profile_continue;
1812 
1813     test_method_data_pointer(mdp, profile_continue);
1814 
1815     if (MethodData::profile_return_jsr292_only()) {
1816       // If we don't profile all invoke bytecodes we must make sure
1817       // it's a bytecode we indeed profile. We can't go back to the
1818       // beginning of the ProfileData we intend to update to check its
1819       // type because we're right after it and we don't known its
1820       // length.
1821       NearLabel do_profile;
1822       Address bc(Z_bcp);
1823       z_lb(tmp, bc);
1824       compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1825       compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1826       get_method(tmp);
1827       // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1828       if (Method::intrinsic_id_size_in_bytes() == 1) {
1829         z_cli(in_bytes(Method::intrinsic_id_offset()), tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1830       } else {
1831         assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1832         z_lh(tmp, in_bytes(Method::intrinsic_id_offset()), Z_R0, tmp);
1833         z_chi(tmp, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1834       }
1835       z_brne(profile_continue);
1836 
1837       bind(do_profile);
1838     }
1839 
1840     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1841     profile_obj_type(ret, mdo_ret_addr, tmp);
1842 
1843     bind(profile_continue);
1844   }
1845 }
1846 
1847 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1848   if (ProfileInterpreter && MethodData::profile_parameters()) {
1849     Label profile_continue, done;
1850 
1851     test_method_data_pointer(mdp, profile_continue);
1852 
1853     // Load the offset of the area within the MDO used for
1854     // parameters. If it's negative we're not profiling any parameters.
1855     Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1856     load_and_test_int2long(tmp1, parm_di_addr);
1857     z_brl(profile_continue);
1858 
1859     // Compute a pointer to the area for parameters from the offset
1860     // and move the pointer to the slot for the last
1861     // parameters. Collect profiling from last parameter down.
1862     // mdo start + parameters offset + array length - 1
1863 
1864     // Pointer to the parameter area in the MDO.
1865     z_agr(mdp, tmp1);
1866 
1867     // Offset of the current profile entry to update.
1868     const Register entry_offset = tmp1;
1869     // entry_offset = array len in number of cells.
1870     z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1871     // entry_offset (number of cells) = array len - size of 1 entry
1872     add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1873     // entry_offset in bytes
1874     z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1875 
1876     Label loop;
1877     bind(loop);
1878 
1879     Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1880     Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1881 
1882     // Load offset on the stack from the slot for this parameter.
1883     z_lg(tmp2, arg_off);
1884     z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1885     z_lcgr(tmp2); // Negate.
1886 
1887     // Profile the parameter.
1888     z_ltg(tmp2, Address(Z_locals, tmp2));
1889     profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1890 
1891     // Go to next parameter.
1892     z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1893     z_brnl(loop);
1894 
1895     bind(profile_continue);
1896   }
1897 }
1898 
1899 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1900 void InterpreterMacroAssembler::increment_mask_and_jump(Address          counter_addr,
1901                                                         int              increment,
1902                                                         Address          mask,
1903                                                         Register         scratch,
1904                                                         bool             preloaded,
1905                                                         branch_condition cond,
1906                                                         Label           *where) {
1907   assert_different_registers(counter_addr.base(), scratch);
1908   if (preloaded) {
1909     add2reg(scratch, increment);
1910     reg2mem_opt(scratch, counter_addr, false);
1911   } else {
1912     if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1913       z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1914       mem2reg_signed_opt(scratch, counter_addr);
1915     } else {
1916       mem2reg_signed_opt(scratch, counter_addr);
1917       add2reg(scratch, increment);
1918       reg2mem_opt(scratch, counter_addr, false);
1919     }
1920   }
1921   z_n(scratch, mask);
1922   if (where) { z_brc(cond, *where); }
1923 }
1924 
1925 // Get MethodCounters object for given method. Lazily allocated if necessary.
1926 //   method    - Ptr to Method object.
1927 //   Rcounters - Ptr to MethodCounters object associated with Method object.
1928 //   skip      - Exit point if MethodCounters object can't be created (OOM condition).
1929 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1930                                                     Register Rcounters,
1931                                                     Label& skip) {
1932   assert_different_registers(Rmethod, Rcounters);
1933 
1934   BLOCK_COMMENT("get MethodCounters object {");
1935 
1936   Label has_counters;
1937   load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1938   z_brnz(has_counters);
1939 
1940   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod);
1941   z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1942   z_brz(skip); // No MethodCounters, out of memory.
1943 
1944   bind(has_counters);
1945 
1946   BLOCK_COMMENT("} get MethodCounters object");
1947 }
1948 
1949 // Increment invocation counter in MethodCounters object.
1950 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1951 // Counter values are all unsigned.
1952 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1953   assert(UseCompiler, "incrementing must be useful");
1954   assert_different_registers(Rcounters, RctrSum);
1955 
1956   int increment          = InvocationCounter::count_increment;
1957   int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1958   int be_counter_offset  = in_bytes(MethodCounters::backedge_counter_offset()   + InvocationCounter::counter_offset());
1959 
1960   BLOCK_COMMENT("Increment invocation counter {");
1961 
1962   if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1963     // Increment the invocation counter in place,
1964     // then add the incremented value to the backedge counter.
1965     z_l(RctrSum, be_counter_offset, Rcounters);
1966     z_alsi(inv_counter_offset, Rcounters, increment);     // Atomic increment @no extra cost!
1967     z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1968     z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1969   } else {
1970     // This path is optimized for low register consumption
1971     // at the cost of somewhat higher operand delays.
1972     // It does not need an extra temp register.
1973 
1974     // Update the invocation counter.
1975     z_l(RctrSum, inv_counter_offset, Rcounters);
1976     if (RctrSum == Z_R0) {
1977       z_ahi(RctrSum, increment);
1978     } else {
1979       add2reg(RctrSum, increment);
1980     }
1981     z_st(RctrSum, inv_counter_offset, Rcounters);
1982 
1983     // Mask off the state bits.
1984     z_nilf(RctrSum, InvocationCounter::count_mask_value);
1985 
1986     // Add the backedge counter to the updated invocation counter to
1987     // form the result.
1988     z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1989   }
1990 
1991   BLOCK_COMMENT("} Increment invocation counter");
1992 
1993   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1994 }
1995 
1996 
1997 // increment backedge counter in MethodCounters object.
1998 // return (invocation_counter+backedge_counter) as "result" in RctrSum
1999 // counter values are all unsigned!
2000 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
2001   assert(UseCompiler, "incrementing must be useful");
2002   assert_different_registers(Rcounters, RctrSum);
2003 
2004   int increment          = InvocationCounter::count_increment;
2005   int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
2006   int be_counter_offset  = in_bytes(MethodCounters::backedge_counter_offset()   + InvocationCounter::counter_offset());
2007 
2008   BLOCK_COMMENT("Increment backedge counter {");
2009 
2010   if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
2011     // Increment the invocation counter in place,
2012     // then add the incremented value to the backedge counter.
2013     z_l(RctrSum, inv_counter_offset, Rcounters);
2014     z_alsi(be_counter_offset, Rcounters, increment);      // Atomic increment @no extra cost!
2015     z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
2016     z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
2017   } else {
2018     // This path is optimized for low register consumption
2019     // at the cost of somewhat higher operand delays.
2020     // It does not need an extra temp register.
2021 
2022     // Update the invocation counter.
2023     z_l(RctrSum, be_counter_offset, Rcounters);
2024     if (RctrSum == Z_R0) {
2025       z_ahi(RctrSum, increment);
2026     } else {
2027       add2reg(RctrSum, increment);
2028     }
2029     z_st(RctrSum, be_counter_offset, Rcounters);
2030 
2031     // Mask off the state bits.
2032     z_nilf(RctrSum, InvocationCounter::count_mask_value);
2033 
2034     // Add the backedge counter to the updated invocation counter to
2035     // form the result.
2036     z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
2037   }
2038 
2039   BLOCK_COMMENT("} Increment backedge counter");
2040 
2041   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2042 }
2043 
2044 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
2045 void InterpreterMacroAssembler::add_monitor_to_stack(bool     stack_is_empty,
2046                                                      Register Rtemp1,
2047                                                      Register Rtemp2,
2048                                                      Register Rtemp3) {
2049 
2050   const Register Rcurr_slot = Rtemp1;
2051   const Register Rlimit     = Rtemp2;
2052   const jint delta = -frame::interpreter_frame_monitor_size_in_bytes();
2053 
2054   assert((delta & LongAlignmentMask) == 0,
2055          "sizeof BasicObjectLock must be even number of doublewords");
2056   assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
2057   assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
2058   assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
2059 
2060   get_monitors(Rlimit);
2061 
2062   // Adjust stack pointer for additional monitor entry.
2063   resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
2064 
2065   if (!stack_is_empty) {
2066     // Must copy stack contents down.
2067     NearLabel next, done;
2068 
2069     // Rtemp := addr(Tos), Z_esp is pointing below it!
2070     add2reg(Rcurr_slot, wordSize, Z_esp);
2071 
2072     // Nothing to do, if already at monitor area.
2073     compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
2074 
2075     bind(next);
2076 
2077     // Move one stack slot.
2078     mem2reg_opt(Rtemp3, Address(Rcurr_slot));
2079     reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
2080     add2reg(Rcurr_slot, wordSize);
2081     compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
2082 
2083     bind(done);
2084     // Done copying stack.
2085   }
2086 
2087   // Adjust expression stack and monitor pointers.
2088   add2reg(Z_esp, delta);
2089   add2reg(Rlimit, delta);
2090   save_monitors(Rlimit);
2091 }
2092 
2093 // Note: Index holds the offset in bytes afterwards.
2094 // You can use this to store a new value (with Llocals as the base).
2095 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
2096   z_sllg(index, index, LogBytesPerWord);
2097   mem2reg_opt(dst, Address(Z_locals, index), false);
2098 }
2099 
2100 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2101   if (state == atos) { MacroAssembler::verify_oop(reg, FILE_AND_LINE); }
2102 }
2103 
2104 // Inline assembly for:
2105 //
2106 // if (thread is in interp_only_mode) {
2107 //   InterpreterRuntime::post_method_entry();
2108 // }
2109 
2110 void InterpreterMacroAssembler::notify_method_entry() {
2111 
2112   // JVMTI
2113   // Whenever JVMTI puts a thread in interp_only_mode, method
2114   // entry/exit events are sent for that thread to track stack
2115   // depth. If it is possible to enter interp_only_mode we add
2116   // the code to check if the event should be sent.
2117   if (JvmtiExport::can_post_interpreter_events()) {
2118     Label jvmti_post_done;
2119     MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2120     z_bre(jvmti_post_done);
2121     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2122     bind(jvmti_post_done);
2123   }
2124 }
2125 
2126 // Inline assembly for:
2127 //
2128 // if (thread is in interp_only_mode) {
2129 //   if (!native_method) save result
2130 //   InterpreterRuntime::post_method_exit();
2131 //   if (!native_method) restore result
2132 // }
2133 // if (DTraceMethodProbes) {
2134 //   SharedRuntime::dtrace_method_exit(thread, method);
2135 // }
2136 //
2137 // For native methods their result is stored in z_ijava_state.lresult
2138 // and z_ijava_state.fresult before coming here.
2139 // Java methods have their result stored in the expression stack.
2140 //
2141 // Notice the dependency to frame::interpreter_frame_result().
2142 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2143                                                    TosState state,
2144                                                    NotifyMethodExitMode mode) {
2145   // JVMTI
2146   // Whenever JVMTI puts a thread in interp_only_mode, method
2147   // entry/exit events are sent for that thread to track stack
2148   // depth. If it is possible to enter interp_only_mode we add
2149   // the code to check if the event should be sent.
2150   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2151     Label jvmti_post_done;
2152     MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2153     z_bre(jvmti_post_done);
2154     if (!native_method) push(state); // see frame::interpreter_frame_result()
2155     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2156     if (!native_method) pop(state);
2157     bind(jvmti_post_done);
2158   }
2159 
2160 #if 0
2161   // Dtrace currently not supported on z/Architecture.
2162   {
2163     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2164     push(state);
2165     get_method(c_rarg1);
2166     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2167                  r15_thread, c_rarg1);
2168     pop(state);
2169   }
2170 #endif
2171 }
2172 
2173 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2174   if (!JvmtiExport::can_post_interpreter_events()) {
2175     return;
2176   }
2177 
2178   load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2179   z_brnz(Lskip);
2180 
2181 }
2182 
2183 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2184 // The return pc is loaded into the register return_pc.
2185 //
2186 // Registers updated:
2187 //     return_pc  - The return pc of the calling frame.
2188 //     tmp1, tmp2 - scratch
2189 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2190   // F0  Z_SP -> caller_sp (F1's)
2191   //             ...
2192   //             sender_sp (F1's)
2193   //             ...
2194   // F1  Z_fp -> caller_sp (F2's)
2195   //             return_pc (Continuation after return from F0.)
2196   //             ...
2197   // F2          caller_sp
2198 
2199   // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2200   // (a) by a c2i adapter and (b) by generate_fixed_frame().
2201   // In case (a) the new top frame F1 is an unextended compiled frame.
2202   // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2203 
2204   // Case (b) seems to be redundant when returning to a interpreted caller,
2205   // because then the caller's top_frame_sp is installed as sp (see
2206   // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2207   // pop_interpreter_frame() is also used in exception handling and there the
2208   // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2209   // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2210 
2211   Register R_f1_sender_sp = tmp1;
2212   Register R_f2_sp = tmp2;
2213 
2214   // First check for the interpreter frame's magic.
2215   asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2216   z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2217   z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2218   if (return_pc->is_valid())
2219     z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2220   // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2221   resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2222 
2223 #ifdef ASSERT
2224   // The return_pc in the new top frame is dead... at least that's my
2225   // current understanding; to assert this I overwrite it.
2226   load_const_optimized(Z_ARG3, 0xb00b1);
2227   z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2228 #endif
2229 }
2230 
2231 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2232   if (VerifyFPU) {
2233     unimplemented("verifyFPU");
2234   }
2235 }