1 /*
   2  * Copyright (c) 2016, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2017 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 "interp_masm_s390.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "interpreter/interpreterRuntime.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/markOop.hpp"
  35 #include "prims/jvmtiExport.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/safepointMechanism.hpp"
  40 #include "runtime/sharedRuntime.hpp"
  41 #include "runtime/thread.inline.hpp"
  42 
  43 // Implementation of InterpreterMacroAssembler.
  44 // This file specializes the assember with interpreter-specific macros.
  45 
  46 #ifdef PRODUCT
  47 #define BLOCK_COMMENT(str)
  48 #define BIND(label)        bind(label);
  49 #else
  50 #define BLOCK_COMMENT(str) block_comment(str)
  51 #define BIND(label)        bind(label); BLOCK_COMMENT(#label ":")
  52 #endif
  53 
  54 void InterpreterMacroAssembler::jump_to_entry(address entry, Register Rscratch) {
  55   assert(entry != NULL, "Entry must have been generated by now");
  56   assert(Rscratch != Z_R0, "Can't use R0 for addressing");
  57   branch_optimized(Assembler::bcondAlways, entry);
  58 }
  59 
  60 void InterpreterMacroAssembler::empty_expression_stack(void) {
  61   get_monitors(Z_R1_scratch);
  62   add2reg(Z_esp, -Interpreter::stackElementSize, Z_R1_scratch);
  63 }
  64 
  65 // Dispatch code executed in the prolog of a bytecode which does not do it's
  66 // own dispatch.
  67 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
  68   // On z/Architecture we are short on registers, therefore we do not preload the
  69   // dispatch address of the next bytecode.
  70 }
  71 
  72 // Dispatch code executed in the epilog of a bytecode which does not do it's
  73 // own dispatch.
  74 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
  75   dispatch_next(state, step);
  76 }
  77 
  78 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr, bool generate_poll) {
  79   z_llgc(Z_bytecode, bcp_incr, Z_R0, Z_bcp);  // Load next bytecode.
  80   add2reg(Z_bcp, bcp_incr);                   // Advance bcp. Add2reg produces optimal code.
  81   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
  82 }
  83 
  84 // Common code to dispatch and dispatch_only.
  85 // Dispatch value in Lbyte_code and increment Lbcp.
  86 
  87 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table, bool generate_poll) {
  88   verify_FPU(1, state);
  89 
  90 #ifdef ASSERT
  91   address reentry = NULL;
  92   { Label OK;
  93     // Check if the frame pointer in Z_fp is correct.
  94     z_cg(Z_fp, 0, Z_SP);
  95     z_bre(OK);
  96     reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp: " FILE_AND_LINE);
  97     bind(OK);
  98   }
  99   { Label OK;
 100     // check if the locals pointer in Z_locals is correct
 101     z_cg(Z_locals, _z_ijava_state_neg(locals), Z_fp);
 102     z_bre(OK);
 103     reentry = stop_chain_static(reentry, "invalid locals pointer Z_locals: " FILE_AND_LINE);
 104     bind(OK);
 105   }
 106 #endif
 107 
 108   // TODO: Maybe implement +VerifyActivationFrameSize here.
 109   // verify_thread(); // Too slow. We will just verify on method entry & exit.
 110   verify_oop(Z_tos, state);
 111 
 112   // Dispatch table to use.
 113   load_absolute_address(Z_tmp_1, (address)table);  // Z_tmp_1 = table;
 114 
 115   if (SafepointMechanism::uses_thread_local_poll() && generate_poll) {
 116     address *sfpt_tbl = Interpreter::safept_table(state);
 117     if (table != sfpt_tbl) {
 118       Label dispatch;
 119       const Address poll_byte_addr(Z_thread, in_bytes(Thread::polling_page_offset()) + 7 /* Big Endian */);
 120       // Armed page has poll_bit set, if poll bit is cleared just continue.
 121       z_tm(poll_byte_addr, SafepointMechanism::poll_bit());
 122       z_braz(dispatch);
 123       load_absolute_address(Z_tmp_1, (address)sfpt_tbl);  // Z_tmp_1 = table;
 124       bind(dispatch);
 125     }
 126   }
 127 
 128   // 0 <= Z_bytecode < 256 => Use a 32 bit shift, because it is shorter than sllg.
 129   // Z_bytecode must have been loaded zero-extended for this approach to be correct.
 130   z_sll(Z_bytecode, LogBytesPerWord, Z_R0);   // Multiply by wordSize.
 131   z_lg(Z_tmp_1, 0, Z_bytecode, Z_tmp_1);      // Get entry addr.
 132 
 133   z_br(Z_tmp_1);
 134 }
 135 
 136 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 137   dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
 138 }
 139 
 140 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 141   dispatch_base(state, Interpreter::normal_table(state));
 142 }
 143 
 144 void InterpreterMacroAssembler::dispatch_via(TosState state, address *table) {
 145   // Load current bytecode.
 146   z_llgc(Z_bytecode, Address(Z_bcp, (intptr_t)0));
 147   dispatch_base(state, table);
 148 }
 149 
 150 // The following call_VM*_base() methods overload and mask the respective
 151 // declarations/definitions in class MacroAssembler. They are meant as a "detour"
 152 // to perform additional, template interpreter specific tasks before actually
 153 // calling their MacroAssembler counterparts.
 154 
 155 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point) {
 156   bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
 157   // interpreter specific
 158   // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
 159   // saved registers and no blocking/ GC can happen in leaf calls.
 160 
 161   // super call
 162   MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
 163 }
 164 
 165 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, bool allow_relocation) {
 166   // interpreter specific
 167   // Note: No need to save/restore bcp (Z_R13) pointer since these are callee
 168   // saved registers and no blocking/ GC can happen in leaf calls.
 169 
 170   // super call
 171   MacroAssembler::call_VM_leaf_base(entry_point, allow_relocation);
 172 }
 173 
 174 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
 175                                              address entry_point, bool check_exceptions) {
 176   bool allow_relocation = true; // Fenerally valid variant. Assume code is relocated.
 177   // interpreter specific
 178 
 179   save_bcp();
 180   save_esp();
 181   // super call
 182   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 183                                entry_point, allow_relocation, check_exceptions);
 184   restore_bcp();
 185 }
 186 
 187 void InterpreterMacroAssembler::call_VM_base(Register oop_result, Register last_java_sp,
 188                                              address entry_point, bool allow_relocation,
 189                                              bool check_exceptions) {
 190   // interpreter specific
 191 
 192   save_bcp();
 193   save_esp();
 194   // super call
 195   MacroAssembler::call_VM_base(oop_result, last_java_sp,
 196                                entry_point, allow_relocation, check_exceptions);
 197   restore_bcp();
 198 }
 199 
 200 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
 201   if (JvmtiExport::can_pop_frame()) {
 202     BLOCK_COMMENT("check_and_handle_popframe {");
 203     Label L;
 204     // Initiate popframe handling only if it is not already being
 205     // processed. If the flag has the popframe_processing bit set, it
 206     // means that this code is called *during* popframe handling - we
 207     // don't want to reenter.
 208     // TODO: Check if all four state combinations could be visible.
 209     // If (processing and !pending) is an invisible/impossible state,
 210     // there is optimization potential by testing both bits at once.
 211     // Then, All_Zeroes and All_Ones means skip, Mixed means doit.
 212     testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
 213             exact_log2(JavaThread::popframe_pending_bit));
 214     z_bfalse(L);
 215     testbit(Address(Z_thread, JavaThread::popframe_condition_offset()),
 216             exact_log2(JavaThread::popframe_processing_bit));
 217     z_btrue(L);
 218 
 219     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 220     // address of the same-named entrypoint in the generated interpreter code.
 221     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 222     // The above call should (as its only effect) return the contents of the field
 223     // _remove_activation_preserving_args_entry in Z_RET.
 224     // We just jump there to have the work done.
 225     z_br(Z_RET);
 226     // There is no way for control to fall thru here.
 227 
 228     bind(L);
 229     BLOCK_COMMENT("} check_and_handle_popframe");
 230   }
 231 }
 232 
 233 
 234 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 235   Register RjvmtiState = Z_R1_scratch;
 236   int      tos_off     = in_bytes(JvmtiThreadState::earlyret_tos_offset());
 237   int      oop_off     = in_bytes(JvmtiThreadState::earlyret_oop_offset());
 238   int      val_off     = in_bytes(JvmtiThreadState::earlyret_value_offset());
 239   int      state_off   = in_bytes(JavaThread::jvmti_thread_state_offset());
 240 
 241   z_lg(RjvmtiState, state_off, Z_thread);
 242 
 243   switch (state) {
 244     case atos: z_lg(Z_tos, oop_off, RjvmtiState);
 245       store_const(Address(RjvmtiState, oop_off), 0L, 8, 8, Z_R0_scratch);
 246                                                     break;
 247     case ltos: z_lg(Z_tos, val_off, RjvmtiState);   break;
 248     case btos: // fall through
 249     case ztos: // fall through
 250     case ctos: // fall through
 251     case stos: // fall through
 252     case itos: z_llgf(Z_tos, val_off, RjvmtiState); break;
 253     case ftos: z_le(Z_ftos, val_off, RjvmtiState);  break;
 254     case dtos: z_ld(Z_ftos, val_off, RjvmtiState);  break;
 255     case vtos:   /* nothing to do */                break;
 256     default  : ShouldNotReachHere();
 257   }
 258 
 259   // Clean up tos value in the jvmti thread state.
 260   store_const(Address(RjvmtiState, val_off),   0L, 8, 8, Z_R0_scratch);
 261   // Set tos state field to illegal value.
 262   store_const(Address(RjvmtiState, tos_off), ilgl, 4, 1, Z_R0_scratch);
 263 }
 264 
 265 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
 266   if (JvmtiExport::can_force_early_return()) {
 267     BLOCK_COMMENT("check_and_handle_earlyret {");
 268     Label L;
 269     // arg regs are save, because we are just behind the call in call_VM_base
 270     Register jvmti_thread_state = Z_ARG2;
 271     Register tmp                = Z_ARG3;
 272     load_and_test_long(jvmti_thread_state, Address(Z_thread, JavaThread::jvmti_thread_state_offset()));
 273     z_bre(L); // if (thread->jvmti_thread_state() == NULL) exit;
 274 
 275     // Initiate earlyret handling only if it is not already being processed.
 276     // If the flag has the earlyret_processing bit set, it means that this code
 277     // is called *during* earlyret handling - we don't want to reenter.
 278 
 279     assert((JvmtiThreadState::earlyret_pending != 0) && (JvmtiThreadState::earlyret_inactive == 0),
 280           "must fix this check, when changing the values of the earlyret enum");
 281     assert(JvmtiThreadState::earlyret_pending == 1, "must fix this check, when changing the values of the earlyret enum");
 282 
 283     load_and_test_int(tmp, Address(jvmti_thread_state, JvmtiThreadState::earlyret_state_offset()));
 284     z_brz(L); // if (thread->jvmti_thread_state()->_earlyret_state != JvmtiThreadState::earlyret_pending) exit;
 285 
 286     // Call Interpreter::remove_activation_early_entry() to get the address of the
 287     // same-named entrypoint in the generated interpreter code.
 288     assert(sizeof(TosState) == 4, "unexpected size");
 289     z_l(Z_ARG1, Address(jvmti_thread_state, JvmtiThreadState::earlyret_tos_offset()));
 290     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Z_ARG1);
 291     // The above call should (as its only effect) return the contents of the field
 292     // _remove_activation_preserving_args_entry in Z_RET.
 293     // We just jump there to have the work done.
 294     z_br(Z_RET);
 295     // There is no way for control to fall thru here.
 296 
 297     bind(L);
 298     BLOCK_COMMENT("} check_and_handle_earlyret");
 299   }
 300 }
 301 
 302 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2) {
 303   lgr_if_needed(Z_ARG1, arg_1);
 304   assert(arg_2 != Z_ARG1, "smashed argument");
 305   lgr_if_needed(Z_ARG2, arg_2);
 306   MacroAssembler::call_VM_leaf_base(entry_point, true);
 307 }
 308 
 309 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, int bcp_offset, size_t index_size) {
 310   Address param(Z_bcp, bcp_offset);
 311 
 312   BLOCK_COMMENT("get_cache_index_at_bcp {");
 313   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 314   if (index_size == sizeof(u2)) {
 315     load_sized_value(index, param, 2, false /*signed*/);
 316   } else if (index_size == sizeof(u4)) {
 317 
 318     load_sized_value(index, param, 4, false);
 319 
 320     // Check if the secondary index definition is still ~x, otherwise
 321     // we have to change the following assembler code to calculate the
 322     // plain index.
 323     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 324     not_(index);  // Convert to plain index.
 325   } else if (index_size == sizeof(u1)) {
 326     z_llgc(index, param);
 327   } else {
 328     ShouldNotReachHere();
 329   }
 330   BLOCK_COMMENT("}");
 331 }
 332 
 333 
 334 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register cpe_offset,
 335                                                            int bcp_offset, size_t index_size) {
 336   BLOCK_COMMENT("get_cache_and_index_at_bcp {");
 337   assert_different_registers(cache, cpe_offset);
 338   get_cache_index_at_bcp(cpe_offset, bcp_offset, index_size);
 339   z_lg(cache, Address(Z_fp, _z_ijava_state_neg(cpoolCache)));
 340   // Convert from field index to ConstantPoolCache offset in bytes.
 341   z_sllg(cpe_offset, cpe_offset, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord));
 342   BLOCK_COMMENT("}");
 343 }
 344 
 345 // Kills Z_R0_scratch.
 346 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 347                                                                         Register cpe_offset,
 348                                                                         Register bytecode,
 349                                                                         int byte_no,
 350                                                                         int bcp_offset,
 351                                                                         size_t index_size) {
 352   BLOCK_COMMENT("get_cache_and_index_and_bytecode_at_bcp {");
 353   get_cache_and_index_at_bcp(cache, cpe_offset, bcp_offset, index_size);
 354 
 355   // We want to load (from CP cache) the bytecode that corresponds to the passed-in byte_no.
 356   // It is located at (cache + cpe_offset + base_offset + indices_offset + (8-1) (last byte in DW) - (byte_no+1).
 357   // Instead of loading, shifting and masking a DW, we just load that one byte of interest with z_llgc (unsigned).
 358   const int base_ix_off = in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset());
 359   const int off_in_DW   = (8-1) - (1+byte_no);
 360   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 361   assert(ConstantPoolCacheEntry::bytecode_1_mask == 0xff, "");
 362   load_sized_value(bytecode, Address(cache, cpe_offset, base_ix_off+off_in_DW), 1, false /*signed*/);
 363 
 364   BLOCK_COMMENT("}");
 365 }
 366 
 367 // Load object from cpool->resolved_references(index).
 368 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
 369   assert_different_registers(result, index);
 370   get_constant_pool(result);
 371 
 372   // Convert
 373   //  - from field index to resolved_references() index and
 374   //  - from word index to byte offset.
 375   // Since this is a java object, it is potentially compressed.
 376   Register tmp = index;  // reuse
 377   z_sllg(index, index, LogBytesPerHeapOop); // Offset into resolved references array.
 378   // Load pointer for resolved_references[] objArray.
 379   z_lg(result, ConstantPool::cache_offset_in_bytes(), result);
 380   z_lg(result, ConstantPoolCache::resolved_references_offset_in_bytes(), result);
 381   resolve_oop_handle(result); // Load resolved references array itself.
 382 #ifdef ASSERT
 383   NearLabel index_ok;
 384   z_lgf(Z_R0, Address(result, arrayOopDesc::length_offset_in_bytes()));
 385   z_sllg(Z_R0, Z_R0, LogBytesPerHeapOop);
 386   compare64_and_branch(tmp, Z_R0, Assembler::bcondLow, index_ok);
 387   stop("resolved reference index out of bounds", 0x09256);
 388   bind(index_ok);
 389 #endif
 390   z_agr(result, index);    // Address of indexed array element.
 391   load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
 392   // The resulting oop is null if the reference is not yet resolved.
 393   // It is Universe::the_null_sentinel() if the reference resolved to NULL via condy.
 394 }
 395 
 396 // load cpool->resolved_klass_at(index)
 397 void InterpreterMacroAssembler::load_resolved_klass_at_offset(Register cpool, Register offset, Register iklass) {
 398   // int value = *(Rcpool->int_at_addr(which));
 399   // int resolved_klass_index = extract_low_short_from_int(value);
 400   z_llgh(offset, Address(cpool, offset, sizeof(ConstantPool) + 2)); // offset = resolved_klass_index (s390 is big-endian)
 401   z_sllg(offset, offset, LogBytesPerWord);                          // Convert 'index' to 'offset'
 402   z_lg(iklass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // iklass = cpool->_resolved_klasses
 403   z_lg(iklass, Address(iklass, offset, Array<Klass*>::base_offset_in_bytes()));
 404 }
 405 
 406 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 407                                                                Register tmp,
 408                                                                int bcp_offset,
 409                                                                size_t index_size) {
 410   BLOCK_COMMENT("get_cache_entry_pointer_at_bcp {");
 411     get_cache_and_index_at_bcp(cache, tmp, bcp_offset, index_size);
 412     add2reg_with_index(cache, in_bytes(ConstantPoolCache::base_offset()), tmp, cache);
 413   BLOCK_COMMENT("}");
 414 }
 415 
 416 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
 417 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
 418 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 419                                                   Register Rsuper_klass,
 420                                                   Register Rtmp1,
 421                                                   Register Rtmp2,
 422                                                   Label &ok_is_subtype) {
 423   // Profile the not-null value's klass.
 424   profile_typecheck(Rtmp1, Rsub_klass, Rtmp2);
 425 
 426   // Do the check.
 427   check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype);
 428 
 429   // Profile the failure of the check.
 430   profile_typecheck_failed(Rtmp1, Rtmp2);
 431 }
 432 
 433 // Pop topmost element from stack. It just disappears.
 434 // Useful if consumed previously by access via stackTop().
 435 void InterpreterMacroAssembler::popx(int len) {
 436   add2reg(Z_esp, len*Interpreter::stackElementSize);
 437   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 438 }
 439 
 440 // Get Address object of stack top. No checks. No pop.
 441 // Purpose: - Provide address of stack operand to exploit reg-mem operations.
 442 //          - Avoid RISC-like mem2reg - reg-reg-op sequence.
 443 Address InterpreterMacroAssembler::stackTop() {
 444   return Address(Z_esp, Interpreter::expr_offset_in_bytes(0));
 445 }
 446 
 447 void InterpreterMacroAssembler::pop_i(Register r) {
 448   z_l(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 449   add2reg(Z_esp, Interpreter::stackElementSize);
 450   assert_different_registers(r, Z_R1_scratch);
 451   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 452 }
 453 
 454 void InterpreterMacroAssembler::pop_ptr(Register r) {
 455   z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 456   add2reg(Z_esp, Interpreter::stackElementSize);
 457   assert_different_registers(r, Z_R1_scratch);
 458   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 459 }
 460 
 461 void InterpreterMacroAssembler::pop_l(Register r) {
 462   z_lg(r, Interpreter::expr_offset_in_bytes(0), Z_esp);
 463   add2reg(Z_esp, 2*Interpreter::stackElementSize);
 464   assert_different_registers(r, Z_R1_scratch);
 465   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 466 }
 467 
 468 void InterpreterMacroAssembler::pop_f(FloatRegister f) {
 469   mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), false);
 470   add2reg(Z_esp, Interpreter::stackElementSize);
 471   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 472 }
 473 
 474 void InterpreterMacroAssembler::pop_d(FloatRegister f) {
 475   mem2freg_opt(f, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)), true);
 476   add2reg(Z_esp, 2*Interpreter::stackElementSize);
 477   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 478 }
 479 
 480 void InterpreterMacroAssembler::push_i(Register r) {
 481   assert_different_registers(r, Z_R1_scratch);
 482   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 483   z_st(r, Address(Z_esp));
 484   add2reg(Z_esp, -Interpreter::stackElementSize);
 485 }
 486 
 487 void InterpreterMacroAssembler::push_ptr(Register r) {
 488   z_stg(r, Address(Z_esp));
 489   add2reg(Z_esp, -Interpreter::stackElementSize);
 490 }
 491 
 492 void InterpreterMacroAssembler::push_l(Register r) {
 493   assert_different_registers(r, Z_R1_scratch);
 494   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 495   int offset = -Interpreter::stackElementSize;
 496   z_stg(r, Address(Z_esp, offset));
 497   clear_mem(Address(Z_esp), Interpreter::stackElementSize);
 498   add2reg(Z_esp, 2 * offset);
 499 }
 500 
 501 void InterpreterMacroAssembler::push_f(FloatRegister f) {
 502   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 503   freg2mem_opt(f, Address(Z_esp), false);
 504   add2reg(Z_esp, -Interpreter::stackElementSize);
 505 }
 506 
 507 void InterpreterMacroAssembler::push_d(FloatRegister d) {
 508   debug_only(verify_esp(Z_esp, Z_R1_scratch));
 509   int offset = -Interpreter::stackElementSize;
 510   freg2mem_opt(d, Address(Z_esp, offset));
 511   add2reg(Z_esp, 2 * offset);
 512 }
 513 
 514 void InterpreterMacroAssembler::push(TosState state) {
 515   verify_oop(Z_tos, state);
 516   switch (state) {
 517     case atos: push_ptr();           break;
 518     case btos: push_i();             break;
 519     case ztos:
 520     case ctos:
 521     case stos: push_i();             break;
 522     case itos: push_i();             break;
 523     case ltos: push_l();             break;
 524     case ftos: push_f();             break;
 525     case dtos: push_d();             break;
 526     case vtos: /* nothing to do */   break;
 527     default  : ShouldNotReachHere();
 528   }
 529 }
 530 
 531 void InterpreterMacroAssembler::pop(TosState state) {
 532   switch (state) {
 533     case atos: pop_ptr(Z_tos);       break;
 534     case btos: pop_i(Z_tos);         break;
 535     case ztos:
 536     case ctos:
 537     case stos: pop_i(Z_tos);         break;
 538     case itos: pop_i(Z_tos);         break;
 539     case ltos: pop_l(Z_tos);         break;
 540     case ftos: pop_f(Z_ftos);        break;
 541     case dtos: pop_d(Z_ftos);        break;
 542     case vtos: /* nothing to do */   break;
 543     default  : ShouldNotReachHere();
 544   }
 545   verify_oop(Z_tos, state);
 546 }
 547 
 548 // Helpers for swap and dup.
 549 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 550   z_lg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
 551 }
 552 
 553 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 554   z_stg(val, Address(Z_esp, Interpreter::expr_offset_in_bytes(n)));
 555 }
 556 
 557 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted(Register method) {
 558   // Satisfy interpreter calling convention (see generate_normal_entry()).
 559   z_lgr(Z_R10, Z_SP); // Set sender sp (aka initial caller sp, aka unextended sp).
 560   // Record top_frame_sp, because the callee might modify it, if it's compiled.
 561   z_stg(Z_SP, _z_ijava_state_neg(top_frame_sp), Z_fp);
 562   save_bcp();
 563   save_esp();
 564   z_lgr(Z_method, method); // Set Z_method (kills Z_fp!).
 565 }
 566 
 567 // Jump to from_interpreted entry of a call unless single stepping is possible
 568 // in this thread in which case we must call the i2i entry.
 569 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 570   assert_different_registers(method, Z_R10 /*used for initial_caller_sp*/, temp);
 571   prepare_to_jump_from_interpreted(method);
 572 
 573   if (JvmtiExport::can_post_interpreter_events()) {
 574     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 575     // compiled code in threads for which the event is enabled. Check here for
 576     // interp_only_mode if these events CAN be enabled.
 577     z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
 578     MacroAssembler::load_and_test_int(Z_R0_scratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
 579     z_bcr(bcondEqual, Z_R1_scratch); // Run compiled code if zero.
 580     // Run interpreted.
 581     z_lg(Z_R1_scratch, Address(method, Method::interpreter_entry_offset()));
 582     z_br(Z_R1_scratch);
 583   } else {
 584     // Run compiled code.
 585     z_lg(Z_R1_scratch, Address(method, Method::from_interpreted_offset()));
 586     z_br(Z_R1_scratch);
 587   }
 588 }
 589 
 590 #ifdef ASSERT
 591 void InterpreterMacroAssembler::verify_esp(Register Resp, Register Rtemp) {
 592   // About to read or write Resp[0].
 593   // Make sure it is not in the monitors or the TOP_IJAVA_FRAME_ABI.
 594   address reentry = NULL;
 595 
 596   {
 597     // Check if the frame pointer in Z_fp is correct.
 598     NearLabel OK;
 599     z_cg(Z_fp, 0, Z_SP);
 600     z_bre(OK);
 601     reentry = stop_chain_static(reentry, "invalid frame pointer Z_fp");
 602     bind(OK);
 603   }
 604   {
 605     // Resp must not point into or below the operand stack,
 606     // i.e. IJAVA_STATE.monitors > Resp.
 607     NearLabel OK;
 608     Register Rmonitors = Rtemp;
 609     z_lg(Rmonitors, _z_ijava_state_neg(monitors), Z_fp);
 610     compareU64_and_branch(Rmonitors, Resp, bcondHigh, OK);
 611     reentry = stop_chain_static(reentry, "too many pops: Z_esp points into monitor area");
 612     bind(OK);
 613   }
 614   {
 615     // Resp may point to the last word of TOP_IJAVA_FRAME_ABI, but not below
 616     // i.e. !(Z_SP + frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize > Resp).
 617     NearLabel OK;
 618     Register Rabi_bottom = Rtemp;
 619     add2reg(Rabi_bottom, frame::z_top_ijava_frame_abi_size - Interpreter::stackElementSize, Z_SP);
 620     compareU64_and_branch(Rabi_bottom, Resp, bcondNotHigh, OK);
 621     reentry = stop_chain_static(reentry, "too many pushes: Z_esp points into TOP_IJAVA_FRAME_ABI");
 622     bind(OK);
 623   }
 624 }
 625 
 626 void InterpreterMacroAssembler::asm_assert_ijava_state_magic(Register tmp) {
 627   Label magic_ok;
 628   load_const_optimized(tmp, frame::z_istate_magic_number);
 629   z_cg(tmp, Address(Z_fp, _z_ijava_state_neg(magic)));
 630   z_bre(magic_ok);
 631   stop_static("error: wrong magic number in ijava_state access");
 632   bind(magic_ok);
 633 }
 634 #endif // ASSERT
 635 
 636 void InterpreterMacroAssembler::save_bcp() {
 637   z_stg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
 638   asm_assert_ijava_state_magic(Z_bcp);
 639   NOT_PRODUCT(z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp))));
 640 }
 641 
 642 void InterpreterMacroAssembler::restore_bcp() {
 643   asm_assert_ijava_state_magic(Z_bcp);
 644   z_lg(Z_bcp, Address(Z_fp, _z_ijava_state_neg(bcp)));
 645 }
 646 
 647 void InterpreterMacroAssembler::save_esp() {
 648   z_stg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
 649 }
 650 
 651 void InterpreterMacroAssembler::restore_esp() {
 652   asm_assert_ijava_state_magic(Z_esp);
 653   z_lg(Z_esp, Address(Z_fp, _z_ijava_state_neg(esp)));
 654 }
 655 
 656 void InterpreterMacroAssembler::get_monitors(Register reg) {
 657   asm_assert_ijava_state_magic(reg);
 658   mem2reg_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
 659 }
 660 
 661 void InterpreterMacroAssembler::save_monitors(Register reg) {
 662   reg2mem_opt(reg, Address(Z_fp, _z_ijava_state_neg(monitors)));
 663 }
 664 
 665 void InterpreterMacroAssembler::get_mdp(Register mdp) {
 666   z_lg(mdp, _z_ijava_state_neg(mdx), Z_fp);
 667 }
 668 
 669 void InterpreterMacroAssembler::save_mdp(Register mdp) {
 670   z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
 671 }
 672 
 673 // Values that are only read (besides initialization).
 674 void InterpreterMacroAssembler::restore_locals() {
 675   asm_assert_ijava_state_magic(Z_locals);
 676   z_lg(Z_locals, Address(Z_fp, _z_ijava_state_neg(locals)));
 677 }
 678 
 679 void InterpreterMacroAssembler::get_method(Register reg) {
 680   asm_assert_ijava_state_magic(reg);
 681   z_lg(reg, Address(Z_fp, _z_ijava_state_neg(method)));
 682 }
 683 
 684 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(Register Rdst, int bcp_offset,
 685                                                           signedOrNot is_signed) {
 686   // Rdst is an 8-byte return value!!!
 687 
 688   // Unaligned loads incur only a small penalty on z/Architecture. The penalty
 689   // is a few (2..3) ticks, even when the load crosses a cache line
 690   // boundary. In case of a cache miss, the stall could, of course, be
 691   // much longer.
 692 
 693   switch (is_signed) {
 694     case Signed:
 695       z_lgh(Rdst, bcp_offset, Z_R0, Z_bcp);
 696      break;
 697    case Unsigned:
 698      z_llgh(Rdst, bcp_offset, Z_R0, Z_bcp);
 699      break;
 700    default:
 701      ShouldNotReachHere();
 702   }
 703 }
 704 
 705 
 706 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(Register Rdst, int bcp_offset,
 707                                                           setCCOrNot set_cc) {
 708   // Rdst is an 8-byte return value!!!
 709 
 710   // Unaligned loads incur only a small penalty on z/Architecture. The penalty
 711   // is a few (2..3) ticks, even when the load crosses a cache line
 712   // boundary. In case of a cache miss, the stall could, of course, be
 713   // much longer.
 714 
 715   // Both variants implement a sign-extending int2long load.
 716   if (set_cc == set_CC) {
 717     load_and_test_int2long(Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
 718   } else {
 719     mem2reg_signed_opt(    Rdst, Address(Z_bcp, (intptr_t)bcp_offset));
 720   }
 721 }
 722 
 723 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
 724   get_method(Rdst);
 725   mem2reg_opt(Rdst, Address(Rdst, Method::const_offset()));
 726   mem2reg_opt(Rdst, Address(Rdst, ConstMethod::constants_offset()));
 727 }
 728 
 729 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
 730   get_constant_pool(Rcpool);
 731   mem2reg_opt(Rtags, Address(Rcpool, ConstantPool::tags_offset_in_bytes()));
 732 }
 733 
 734 // Unlock if synchronized method.
 735 //
 736 // Unlock the receiver if this is a synchronized method.
 737 // Unlock any Java monitors from syncronized blocks.
 738 //
 739 // If there are locked Java monitors
 740 //   If throw_monitor_exception
 741 //     throws IllegalMonitorStateException
 742 //   Else if install_monitor_exception
 743 //     installs IllegalMonitorStateException
 744 //   Else
 745 //     no error processing
 746 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
 747                                                               bool throw_monitor_exception,
 748                                                               bool install_monitor_exception) {
 749   NearLabel unlocked, unlock, no_unlock;
 750 
 751   {
 752     Register R_method = Z_ARG2;
 753     Register R_do_not_unlock_if_synchronized = Z_ARG3;
 754 
 755     // Get the value of _do_not_unlock_if_synchronized into G1_scratch.
 756     const Address do_not_unlock_if_synchronized(Z_thread,
 757                                                 JavaThread::do_not_unlock_if_synchronized_offset());
 758     load_sized_value(R_do_not_unlock_if_synchronized, do_not_unlock_if_synchronized, 1, false /*unsigned*/);
 759     z_mvi(do_not_unlock_if_synchronized, false); // Reset the flag.
 760 
 761     // Check if synchronized method.
 762     get_method(R_method);
 763     verify_oop(Z_tos, state);
 764     push(state); // Save tos/result.
 765     testbit(method2_(R_method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 766     z_bfalse(unlocked);
 767 
 768     // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 769     // is set.
 770     compareU64_and_branch(R_do_not_unlock_if_synchronized, (intptr_t)0L, bcondNotEqual, no_unlock);
 771   }
 772 
 773   // unlock monitor
 774 
 775   // BasicObjectLock will be first in list, since this is a
 776   // synchronized method. However, need to check that the object has
 777   // not been unlocked by an explicit monitorexit bytecode.
 778   const Address monitor(Z_fp, -(frame::z_ijava_state_size + (int) sizeof(BasicObjectLock)));
 779   // We use Z_ARG2 so that if we go slow path it will be the correct
 780   // register for unlock_object to pass to VM directly.
 781   load_address(Z_ARG2, monitor); // Address of first monitor.
 782   z_lg(Z_ARG3, Address(Z_ARG2, BasicObjectLock::obj_offset_in_bytes()));
 783   compareU64_and_branch(Z_ARG3, (intptr_t)0L, bcondNotEqual, unlock);
 784 
 785   if (throw_monitor_exception) {
 786     // Entry already unlocked need to throw an exception.
 787     MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
 788     should_not_reach_here();
 789   } else {
 790     // Monitor already unlocked during a stack unroll.
 791     // If requested, install an illegal_monitor_state_exception.
 792     // Continue with stack unrolling.
 793     if (install_monitor_exception) {
 794       MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
 795     }
 796    z_bru(unlocked);
 797   }
 798 
 799   bind(unlock);
 800 
 801   unlock_object(Z_ARG2);
 802 
 803   bind(unlocked);
 804 
 805   // I0, I1: Might contain return value
 806 
 807   // Check that all monitors are unlocked.
 808   {
 809     NearLabel loop, exception, entry, restart;
 810     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 811     // We use Z_ARG2 so that if we go slow path it will be the correct
 812     // register for unlock_object to pass to VM directly.
 813     Register R_current_monitor = Z_ARG2;
 814     Register R_monitor_block_bot = Z_ARG1;
 815     const Address monitor_block_top(Z_fp, _z_ijava_state_neg(monitors));
 816     const Address monitor_block_bot(Z_fp, -frame::z_ijava_state_size);
 817 
 818     bind(restart);
 819     // Starting with top-most entry.
 820     z_lg(R_current_monitor, monitor_block_top);
 821     // Points to word before bottom of monitor block.
 822     load_address(R_monitor_block_bot, monitor_block_bot);
 823     z_bru(entry);
 824 
 825     // Entry already locked, need to throw exception.
 826     bind(exception);
 827 
 828     if (throw_monitor_exception) {
 829       // Throw exception.
 830       MacroAssembler::call_VM(noreg,
 831                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
 832                                                throw_illegal_monitor_state_exception));
 833       should_not_reach_here();
 834     } else {
 835       // Stack unrolling. Unlock object and install illegal_monitor_exception.
 836       // Unlock does not block, so don't have to worry about the frame.
 837       // We don't have to preserve c_rarg1 since we are going to throw an exception.
 838       unlock_object(R_current_monitor);
 839       if (install_monitor_exception) {
 840         call_VM(noreg, CAST_FROM_FN_PTR(address,
 841                                         InterpreterRuntime::
 842                                         new_illegal_monitor_state_exception));
 843       }
 844       z_bru(restart);
 845     }
 846 
 847     bind(loop);
 848     // Check if current entry is used.
 849     load_and_test_long(Z_R0_scratch, Address(R_current_monitor, BasicObjectLock::obj_offset_in_bytes()));
 850     z_brne(exception);
 851 
 852     add2reg(R_current_monitor, entry_size); // Otherwise advance to next entry.
 853     bind(entry);
 854     compareU64_and_branch(R_current_monitor, R_monitor_block_bot, bcondNotEqual, loop);
 855   }
 856 
 857   bind(no_unlock);
 858   pop(state);
 859   verify_oop(Z_tos, state);
 860 }
 861 
 862 void InterpreterMacroAssembler::narrow(Register result, Register ret_type) {
 863   get_method(ret_type);
 864   z_lg(ret_type, Address(ret_type, in_bytes(Method::const_offset())));
 865   z_lb(ret_type, Address(ret_type, in_bytes(ConstMethod::result_type_offset())));
 866 
 867   Label notBool, notByte, notChar, done;
 868 
 869   // common case first
 870   compareU32_and_branch(ret_type, T_INT, bcondEqual, done);
 871 
 872   compareU32_and_branch(ret_type, T_BOOLEAN, bcondNotEqual, notBool);
 873   z_nilf(result, 0x1);
 874   z_bru(done);
 875 
 876   bind(notBool);
 877   compareU32_and_branch(ret_type, T_BYTE, bcondNotEqual, notByte);
 878   z_lbr(result, result);
 879   z_bru(done);
 880 
 881   bind(notByte);
 882   compareU32_and_branch(ret_type, T_CHAR, bcondNotEqual, notChar);
 883   z_nilf(result, 0xffff);
 884   z_bru(done);
 885 
 886   bind(notChar);
 887   // compareU32_and_branch(ret_type, T_SHORT, bcondNotEqual, notShort);
 888   z_lhr(result, result);
 889 
 890   // Nothing to do for T_INT
 891   bind(done);
 892 }
 893 
 894 // remove activation
 895 //
 896 // Unlock the receiver if this is a synchronized method.
 897 // Unlock any Java monitors from syncronized blocks.
 898 // Remove the activation from the stack.
 899 //
 900 // If there are locked Java monitors
 901 //   If throw_monitor_exception
 902 //     throws IllegalMonitorStateException
 903 //   Else if install_monitor_exception
 904 //     installs IllegalMonitorStateException
 905 //   Else
 906 //     no error processing
 907 void InterpreterMacroAssembler::remove_activation(TosState state,
 908                                                   Register return_pc,
 909                                                   bool throw_monitor_exception,
 910                                                   bool install_monitor_exception,
 911                                                   bool notify_jvmti) {
 912   BLOCK_COMMENT("remove_activation {");
 913   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
 914 
 915   // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
 916   notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
 917 
 918   if (StackReservedPages > 0) {
 919     BLOCK_COMMENT("reserved_stack_check:");
 920     // Test if reserved zone needs to be enabled.
 921     Label no_reserved_zone_enabling;
 922 
 923     // Compare frame pointers. There is no good stack pointer, as with stack
 924     // frame compression we can get different SPs when we do calls. A subsequent
 925     // call could have a smaller SP, so that this compare succeeds for an
 926     // inner call of the method annotated with ReservedStack.
 927     z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
 928     z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
 929     z_brl(no_reserved_zone_enabling);
 930 
 931     // Enable reserved zone again, throw stack overflow exception.
 932     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
 933     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
 934 
 935     should_not_reach_here();
 936 
 937     bind(no_reserved_zone_enabling);
 938   }
 939 
 940   verify_oop(Z_tos, state);
 941   verify_thread();
 942 
 943   pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
 944   BLOCK_COMMENT("} remove_activation");
 945 }
 946 
 947 // lock object
 948 //
 949 // Registers alive
 950 //   monitor - Address of the BasicObjectLock to be used for locking,
 951 //             which must be initialized with the object to lock.
 952 //   object  - Address of the object to be locked.
 953 void InterpreterMacroAssembler::lock_object(Register monitor, Register object) {
 954 
 955   if (UseHeavyMonitors) {
 956     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
 957             monitor, /*check_for_exceptions=*/false);
 958     return;
 959   }
 960 
 961   // template code:
 962   //
 963   // markOop displaced_header = obj->mark().set_unlocked();
 964   // monitor->lock()->set_displaced_header(displaced_header);
 965   // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
 966   //   // We stored the monitor address into the object's mark word.
 967   // } else if (THREAD->is_lock_owned((address)displaced_header))
 968   //   // Simple recursive case.
 969   //   monitor->lock()->set_displaced_header(NULL);
 970   // } else {
 971   //   // Slow path.
 972   //   InterpreterRuntime::monitorenter(THREAD, monitor);
 973   // }
 974 
 975   const Register displaced_header = Z_ARG5;
 976   const Register object_mark_addr = Z_ARG4;
 977   const Register current_header   = Z_ARG5;
 978 
 979   NearLabel done;
 980   NearLabel slow_case;
 981 
 982   // markOop displaced_header = obj->mark().set_unlocked();
 983 
 984   // Load markOop from object into displaced_header.
 985   z_lg(displaced_header, oopDesc::mark_offset_in_bytes(), object);
 986 
 987   if (UseBiasedLocking) {
 988     biased_locking_enter(object, displaced_header, Z_R1, Z_R0, done, &slow_case);
 989   }
 990 
 991   // Set displaced_header to be (markOop of object | UNLOCK_VALUE).
 992   z_oill(displaced_header, markOopDesc::unlocked_value);
 993 
 994   // monitor->lock()->set_displaced_header(displaced_header);
 995 
 996   // Initialize the box (Must happen before we update the object mark!).
 997   z_stg(displaced_header, BasicObjectLock::lock_offset_in_bytes() +
 998                           BasicLock::displaced_header_offset_in_bytes(), monitor);
 999 
1000   // if (Atomic::cmpxchg(/*ex=*/monitor, /*addr*/obj->mark_addr(), /*cmp*/displaced_header) == displaced_header) {
1001 
1002   // Store stack address of the BasicObjectLock (this is monitor) into object.
1003   add2reg(object_mark_addr, oopDesc::mark_offset_in_bytes(), object);
1004 
1005   z_csg(displaced_header, monitor, 0, object_mark_addr);
1006   assert(current_header==displaced_header, "must be same register"); // Identified two registers from z/Architecture.
1007 
1008   z_bre(done);
1009 
1010   // } else if (THREAD->is_lock_owned((address)displaced_header))
1011   //   // Simple recursive case.
1012   //   monitor->lock()->set_displaced_header(NULL);
1013 
1014   // We did not see an unlocked object so try the fast recursive case.
1015 
1016   // Check if owner is self by comparing the value in the markOop of object
1017   // (current_header) with the stack pointer.
1018   z_sgr(current_header, Z_SP);
1019 
1020   assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1021 
1022   // The prior sequence "LGR, NGR, LTGR" can be done better
1023   // (Z_R1 is temp and not used after here).
1024   load_const_optimized(Z_R0, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place));
1025   z_ngr(Z_R0, current_header); // AND sets CC (result eq/ne 0)
1026 
1027   // If condition is true we are done and hence we can store 0 in the displaced
1028   // header indicating it is a recursive lock and be done.
1029   z_brne(slow_case);
1030   z_release();  // Membar unnecessary on zarch AND because the above csg does a sync before and after.
1031   z_stg(Z_R0/*==0!*/, BasicObjectLock::lock_offset_in_bytes() +
1032                       BasicLock::displaced_header_offset_in_bytes(), monitor);
1033   z_bru(done);
1034 
1035   // } else {
1036   //   // Slow path.
1037   //   InterpreterRuntime::monitorenter(THREAD, monitor);
1038 
1039   // None of the above fast optimizations worked so we have to get into the
1040   // slow case of monitor enter.
1041   bind(slow_case);
1042 
1043   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1044           monitor, /*check_for_exceptions=*/false);
1045 
1046   // }
1047 
1048   bind(done);
1049 }
1050 
1051 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1052 //
1053 // Registers alive
1054 //   monitor - address of the BasicObjectLock to be used for locking,
1055 //             which must be initialized with the object to lock.
1056 //
1057 // Throw IllegalMonitorException if object is not locked by current thread.
1058 void InterpreterMacroAssembler::unlock_object(Register monitor, Register object) {
1059 
1060   if (UseHeavyMonitors) {
1061     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1062             monitor, /*check_for_exceptions=*/ true);
1063     return;
1064   }
1065 
1066 // else {
1067   // template code:
1068   //
1069   // if ((displaced_header = monitor->displaced_header()) == NULL) {
1070   //   // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1071   //   monitor->set_obj(NULL);
1072   // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1073   //   // We swapped the unlocked mark in displaced_header into the object's mark word.
1074   //   monitor->set_obj(NULL);
1075   // } else {
1076   //   // Slow path.
1077   //   InterpreterRuntime::monitorexit(THREAD, monitor);
1078   // }
1079 
1080   const Register displaced_header = Z_ARG4;
1081   const Register current_header   = Z_R1;
1082   Address obj_entry(monitor, BasicObjectLock::obj_offset_in_bytes());
1083   Label done;
1084 
1085   if (object == noreg) {
1086     // In the template interpreter, we must assure that the object
1087     // entry in the monitor is cleared on all paths. Thus we move
1088     // loading up to here, and clear the entry afterwards.
1089     object = Z_ARG3; // Use Z_ARG3 if caller didn't pass object.
1090     z_lg(object, obj_entry);
1091   }
1092 
1093   assert_different_registers(monitor, object, displaced_header, current_header);
1094 
1095   // if ((displaced_header = monitor->displaced_header()) == NULL) {
1096   //   // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL.
1097   //   monitor->set_obj(NULL);
1098 
1099   clear_mem(obj_entry, sizeof(oop));
1100 
1101   if (UseBiasedLocking) {
1102     // The object address from the monitor is in object.
1103     assert(oopDesc::mark_offset_in_bytes() == 0, "offset of _mark is not 0");
1104     biased_locking_exit(object, displaced_header, done);
1105   }
1106 
1107   // Test first if we are in the fast recursive case.
1108   MacroAssembler::load_and_test_long(displaced_header,
1109                                      Address(monitor, BasicObjectLock::lock_offset_in_bytes() +
1110                                                       BasicLock::displaced_header_offset_in_bytes()));
1111   z_bre(done); // displaced_header == 0 -> goto done
1112 
1113   // } else if (Atomic::cmpxchg(displaced_header, obj->mark_addr(), monitor) == monitor) {
1114   //   // We swapped the unlocked mark in displaced_header into the object's mark word.
1115   //   monitor->set_obj(NULL);
1116 
1117   // If we still have a lightweight lock, unlock the object and be done.
1118 
1119   // The markword is expected to be at offset 0.
1120   assert(oopDesc::mark_offset_in_bytes() == 0, "unlock_object: review code below");
1121 
1122   // We have the displaced header in displaced_header. If the lock is still
1123   // lightweight, it will contain the monitor address and we'll store the
1124   // displaced header back into the object's mark word.
1125   z_lgr(current_header, monitor);
1126   z_csg(current_header, displaced_header, 0, object);
1127   z_bre(done);
1128 
1129   // } else {
1130   //   // Slow path.
1131   //   InterpreterRuntime::monitorexit(THREAD, monitor);
1132 
1133   // The lock has been converted into a heavy lock and hence
1134   // we need to get into the slow case.
1135   z_stg(object, obj_entry);   // Restore object entry, has been cleared above.
1136   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1137           monitor,  /*check_for_exceptions=*/false);
1138 
1139   // }
1140 
1141   bind(done);
1142 }
1143 
1144 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
1145   assert(ProfileInterpreter, "must be profiling interpreter");
1146   load_and_test_long(mdp, Address(Z_fp, _z_ijava_state_neg(mdx)));
1147   z_brz(zero_continue);
1148 }
1149 
1150 // Set the method data pointer for the current bcp.
1151 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1152   assert(ProfileInterpreter, "must be profiling interpreter");
1153   Label    set_mdp;
1154   Register mdp    = Z_ARG4;
1155   Register method = Z_ARG5;
1156 
1157   get_method(method);
1158   // Test MDO to avoid the call if it is NULL.
1159   load_and_test_long(mdp, method2_(method, method_data));
1160   z_brz(set_mdp);
1161 
1162   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), method, Z_bcp);
1163   // Z_RET: mdi
1164   // Mdo is guaranteed to be non-zero here, we checked for it before the call.
1165   assert(method->is_nonvolatile(), "choose nonvolatile reg or reload from frame");
1166   z_lg(mdp, method2_(method, method_data)); // Must reload, mdp is volatile reg.
1167   add2reg_with_index(mdp, in_bytes(MethodData::data_offset()), Z_RET, mdp);
1168 
1169   bind(set_mdp);
1170   save_mdp(mdp);
1171 }
1172 
1173 void InterpreterMacroAssembler::verify_method_data_pointer() {
1174   assert(ProfileInterpreter, "must be profiling interpreter");
1175 #ifdef ASSERT
1176   NearLabel verify_continue;
1177   Register bcp_expected = Z_ARG3;
1178   Register mdp    = Z_ARG4;
1179   Register method = Z_ARG5;
1180 
1181   test_method_data_pointer(mdp, verify_continue); // If mdp is zero, continue
1182   get_method(method);
1183 
1184   // If the mdp is valid, it will point to a DataLayout header which is
1185   // consistent with the bcp. The converse is highly probable also.
1186   load_sized_value(bcp_expected, Address(mdp, DataLayout::bci_offset()), 2, false /*signed*/);
1187   z_ag(bcp_expected, Address(method, Method::const_offset()));
1188   load_address(bcp_expected, Address(bcp_expected, ConstMethod::codes_offset()));
1189   compareU64_and_branch(bcp_expected, Z_bcp, bcondEqual, verify_continue);
1190   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), method, Z_bcp, mdp);
1191   bind(verify_continue);
1192 #endif // ASSERT
1193 }
1194 
1195 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
1196   assert(ProfileInterpreter, "must be profiling interpreter");
1197   z_stg(value, constant, mdp_in);
1198 }
1199 
1200 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1201                                                       int constant,
1202                                                       Register tmp,
1203                                                       bool decrement) {
1204   assert_different_registers(mdp_in, tmp);
1205   // counter address
1206   Address data(mdp_in, constant);
1207   const int delta = decrement ? -DataLayout::counter_increment : DataLayout::counter_increment;
1208   add2mem_64(Address(mdp_in, constant), delta, tmp);
1209 }
1210 
1211 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1212                                                 int flag_byte_constant) {
1213   assert(ProfileInterpreter, "must be profiling interpreter");
1214   // Set the flag.
1215   z_oi(Address(mdp_in, DataLayout::flags_offset()), flag_byte_constant);
1216 }
1217 
1218 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1219                                                  int offset,
1220                                                  Register value,
1221                                                  Register test_value_out,
1222                                                  Label& not_equal_continue) {
1223   assert(ProfileInterpreter, "must be profiling interpreter");
1224   if (test_value_out == noreg) {
1225     z_cg(value, Address(mdp_in, offset));
1226     z_brne(not_equal_continue);
1227   } else {
1228     // Put the test value into a register, so caller can use it:
1229     z_lg(test_value_out, Address(mdp_in, offset));
1230     compareU64_and_branch(test_value_out, value, bcondNotEqual, not_equal_continue);
1231   }
1232 }
1233 
1234 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
1235   update_mdp_by_offset(mdp_in, noreg, offset_of_disp);
1236 }
1237 
1238 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1239                                                      Register dataidx,
1240                                                      int offset_of_disp) {
1241   assert(ProfileInterpreter, "must be profiling interpreter");
1242   Address disp_address(mdp_in, dataidx, offset_of_disp);
1243   Assembler::z_ag(mdp_in, disp_address);
1244   save_mdp(mdp_in);
1245 }
1246 
1247 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
1248   assert(ProfileInterpreter, "must be profiling interpreter");
1249   add2reg(mdp_in, constant);
1250   save_mdp(mdp_in);
1251 }
1252 
1253 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1254   assert(ProfileInterpreter, "must be profiling interpreter");
1255   assert(return_bci->is_nonvolatile(), "choose nonvolatile reg or save/restore");
1256   call_VM(noreg,
1257           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1258           return_bci);
1259 }
1260 
1261 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
1262   if (ProfileInterpreter) {
1263     Label profile_continue;
1264 
1265     // If no method data exists, go to profile_continue.
1266     // Otherwise, assign to mdp.
1267     test_method_data_pointer(mdp, profile_continue);
1268 
1269     // We are taking a branch. Increment the taken count.
1270     // We inline increment_mdp_data_at to return bumped_count in a register
1271     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1272     Address data(mdp, JumpData::taken_offset());
1273     z_lg(bumped_count, data);
1274     // 64-bit overflow is very unlikely. Saturation to 32-bit values is
1275     // performed when reading the counts.
1276     add2reg(bumped_count, DataLayout::counter_increment);
1277     z_stg(bumped_count, data); // Store back out
1278 
1279     // The method data pointer needs to be updated to reflect the new target.
1280     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1281     bind(profile_continue);
1282   }
1283 }
1284 
1285 // Kills Z_R1_scratch.
1286 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1287   if (ProfileInterpreter) {
1288     Label profile_continue;
1289 
1290     // If no method data exists, go to profile_continue.
1291     test_method_data_pointer(mdp, profile_continue);
1292 
1293     // We are taking a branch. Increment the not taken count.
1294     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()), Z_R1_scratch);
1295 
1296     // The method data pointer needs to be updated to correspond to
1297     // the next bytecode.
1298     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1299     bind(profile_continue);
1300   }
1301 }
1302 
1303 // Kills: Z_R1_scratch.
1304 void InterpreterMacroAssembler::profile_call(Register mdp) {
1305   if (ProfileInterpreter) {
1306     Label profile_continue;
1307 
1308     // If no method data exists, go to profile_continue.
1309     test_method_data_pointer(mdp, profile_continue);
1310 
1311     // We are making a call. Increment the count.
1312     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1313 
1314     // The method data pointer needs to be updated to reflect the new target.
1315     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1316     bind(profile_continue);
1317   }
1318 }
1319 
1320 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1321   if (ProfileInterpreter) {
1322     Label profile_continue;
1323 
1324     // If no method data exists, go to profile_continue.
1325     test_method_data_pointer(mdp, profile_continue);
1326 
1327     // We are making a call. Increment the count.
1328     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1329 
1330     // The method data pointer needs to be updated to reflect the new target.
1331     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1332     bind(profile_continue);
1333   }
1334 }
1335 
1336 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1337                                                      Register mdp,
1338                                                      Register reg2,
1339                                                      bool receiver_can_be_null) {
1340   if (ProfileInterpreter) {
1341     NearLabel profile_continue;
1342 
1343     // If no method data exists, go to profile_continue.
1344     test_method_data_pointer(mdp, profile_continue);
1345 
1346     NearLabel skip_receiver_profile;
1347     if (receiver_can_be_null) {
1348       NearLabel not_null;
1349       compareU64_and_branch(receiver, (intptr_t)0L, bcondNotEqual, not_null);
1350       // We are making a call. Increment the count for null receiver.
1351       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1352       z_bru(skip_receiver_profile);
1353       bind(not_null);
1354     }
1355 
1356     // Record the receiver type.
1357     record_klass_in_profile(receiver, mdp, reg2, true);
1358     bind(skip_receiver_profile);
1359 
1360     // The method data pointer needs to be updated to reflect the new target.
1361     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1362     bind(profile_continue);
1363   }
1364 }
1365 
1366 // This routine creates a state machine for updating the multi-row
1367 // type profile at a virtual call site (or other type-sensitive bytecode).
1368 // The machine visits each row (of receiver/count) until the receiver type
1369 // is found, or until it runs out of rows. At the same time, it remembers
1370 // the location of the first empty row. (An empty row records null for its
1371 // receiver, and can be allocated for a newly-observed receiver type.)
1372 // Because there are two degrees of freedom in the state, a simple linear
1373 // search will not work; it must be a decision tree. Hence this helper
1374 // function is recursive, to generate the required tree structured code.
1375 // It's the interpreter, so we are trading off code space for speed.
1376 // See below for example code.
1377 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1378                                         Register receiver, Register mdp,
1379                                         Register reg2, int start_row,
1380                                         Label& done, bool is_virtual_call) {
1381   if (TypeProfileWidth == 0) {
1382     if (is_virtual_call) {
1383       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1384     }
1385     return;
1386   }
1387 
1388   int last_row = VirtualCallData::row_limit() - 1;
1389   assert(start_row <= last_row, "must be work left to do");
1390   // Test this row for both the receiver and for null.
1391   // Take any of three different outcomes:
1392   //   1. found receiver => increment count and goto done
1393   //   2. found null => keep looking for case 1, maybe allocate this cell
1394   //   3. found something else => keep looking for cases 1 and 2
1395   // Case 3 is handled by a recursive call.
1396   for (int row = start_row; row <= last_row; row++) {
1397     NearLabel next_test;
1398     bool test_for_null_also = (row == start_row);
1399 
1400     // See if the receiver is receiver[n].
1401     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1402     test_mdp_data_at(mdp, recvr_offset, receiver,
1403                      (test_for_null_also ? reg2 : noreg),
1404                      next_test);
1405     // (Reg2 now contains the receiver from the CallData.)
1406 
1407     // The receiver is receiver[n]. Increment count[n].
1408     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1409     increment_mdp_data_at(mdp, count_offset);
1410     z_bru(done);
1411     bind(next_test);
1412 
1413     if (test_for_null_also) {
1414       Label found_null;
1415       // Failed the equality check on receiver[n]... Test for null.
1416       z_ltgr(reg2, reg2);
1417       if (start_row == last_row) {
1418         // The only thing left to do is handle the null case.
1419         if (is_virtual_call) {
1420           z_brz(found_null);
1421           // Receiver did not match any saved receiver and there is no empty row for it.
1422           // Increment total counter to indicate polymorphic case.
1423           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1424           z_bru(done);
1425           bind(found_null);
1426         } else {
1427           z_brnz(done);
1428         }
1429         break;
1430       }
1431       // Since null is rare, make it be the branch-taken case.
1432       z_brz(found_null);
1433 
1434       // Put all the "Case 3" tests here.
1435       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1436 
1437       // Found a null. Keep searching for a matching receiver,
1438       // but remember that this is an empty (unused) slot.
1439       bind(found_null);
1440     }
1441   }
1442 
1443   // In the fall-through case, we found no matching receiver, but we
1444   // observed the receiver[start_row] is NULL.
1445 
1446   // Fill in the receiver field and increment the count.
1447   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1448   set_mdp_data_at(mdp, recvr_offset, receiver);
1449   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1450   load_const_optimized(reg2, DataLayout::counter_increment);
1451   set_mdp_data_at(mdp, count_offset, reg2);
1452   if (start_row > 0) {
1453     z_bru(done);
1454   }
1455 }
1456 
1457 // Example state machine code for three profile rows:
1458 //   // main copy of decision tree, rooted at row[1]
1459 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1460 //   if (row[0].rec != NULL) {
1461 //     // inner copy of decision tree, rooted at row[1]
1462 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1463 //     if (row[1].rec != NULL) {
1464 //       // degenerate decision tree, rooted at row[2]
1465 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1466 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1467 //       row[2].init(rec); goto done;
1468 //     } else {
1469 //       // remember row[1] is empty
1470 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1471 //       row[1].init(rec); goto done;
1472 //     }
1473 //   } else {
1474 //     // remember row[0] is empty
1475 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1476 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1477 //     row[0].init(rec); goto done;
1478 //   }
1479 //   done:
1480 
1481 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1482                                                         Register mdp, Register reg2,
1483                                                         bool is_virtual_call) {
1484   assert(ProfileInterpreter, "must be profiling");
1485   Label done;
1486 
1487   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1488 
1489   bind (done);
1490 }
1491 
1492 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1493   if (ProfileInterpreter) {
1494     NearLabel profile_continue;
1495     uint row;
1496 
1497     // If no method data exists, go to profile_continue.
1498     test_method_data_pointer(mdp, profile_continue);
1499 
1500     // Update the total ret count.
1501     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1502 
1503     for (row = 0; row < RetData::row_limit(); row++) {
1504       NearLabel next_test;
1505 
1506       // See if return_bci is equal to bci[n]:
1507       test_mdp_data_at(mdp,
1508                        in_bytes(RetData::bci_offset(row)),
1509                        return_bci, noreg,
1510                        next_test);
1511 
1512       // Return_bci is equal to bci[n]. Increment the count.
1513       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1514 
1515       // The method data pointer needs to be updated to reflect the new target.
1516       update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
1517       z_bru(profile_continue);
1518       bind(next_test);
1519     }
1520 
1521     update_mdp_for_ret(return_bci);
1522 
1523     bind(profile_continue);
1524   }
1525 }
1526 
1527 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1528   if (ProfileInterpreter) {
1529     Label profile_continue;
1530 
1531     // If no method data exists, go to profile_continue.
1532     test_method_data_pointer(mdp, profile_continue);
1533 
1534     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1535 
1536     // The method data pointer needs to be updated.
1537     int mdp_delta = in_bytes(BitData::bit_data_size());
1538     if (TypeProfileCasts) {
1539       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1540     }
1541     update_mdp_by_constant(mdp, mdp_delta);
1542 
1543     bind(profile_continue);
1544   }
1545 }
1546 
1547 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp, Register tmp) {
1548   if (ProfileInterpreter && TypeProfileCasts) {
1549     Label profile_continue;
1550 
1551     // If no method data exists, go to profile_continue.
1552     test_method_data_pointer(mdp, profile_continue);
1553 
1554     int count_offset = in_bytes(CounterData::count_offset());
1555     // Back up the address, since we have already bumped the mdp.
1556     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1557 
1558     // *Decrement* the counter. We expect to see zero or small negatives.
1559     increment_mdp_data_at(mdp, count_offset, tmp, true);
1560 
1561     bind (profile_continue);
1562   }
1563 }
1564 
1565 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1566   if (ProfileInterpreter) {
1567     Label profile_continue;
1568 
1569     // If no method data exists, go to profile_continue.
1570     test_method_data_pointer(mdp, profile_continue);
1571 
1572     // The method data pointer needs to be updated.
1573     int mdp_delta = in_bytes(BitData::bit_data_size());
1574     if (TypeProfileCasts) {
1575       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1576 
1577       // Record the object type.
1578       record_klass_in_profile(klass, mdp, reg2, false);
1579     }
1580     update_mdp_by_constant(mdp, mdp_delta);
1581 
1582     bind(profile_continue);
1583   }
1584 }
1585 
1586 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1587   if (ProfileInterpreter) {
1588     Label profile_continue;
1589 
1590     // If no method data exists, go to profile_continue.
1591     test_method_data_pointer(mdp, profile_continue);
1592 
1593     // Update the default case count.
1594     increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
1595 
1596     // The method data pointer needs to be updated.
1597     update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
1598 
1599     bind(profile_continue);
1600   }
1601 }
1602 
1603 // Kills: index, scratch1, scratch2.
1604 void InterpreterMacroAssembler::profile_switch_case(Register index,
1605                                                     Register mdp,
1606                                                     Register scratch1,
1607                                                     Register scratch2) {
1608   if (ProfileInterpreter) {
1609     Label profile_continue;
1610     assert_different_registers(index, mdp, scratch1, scratch2);
1611 
1612     // If no method data exists, go to profile_continue.
1613     test_method_data_pointer(mdp, profile_continue);
1614 
1615     // Build the base (index * per_case_size_in_bytes()) +
1616     // case_array_offset_in_bytes().
1617     z_sllg(index, index, exact_log2(in_bytes(MultiBranchData::per_case_size())));
1618     add2reg(index, in_bytes(MultiBranchData::case_array_offset()));
1619 
1620     // Add the calculated base to the mdp -> address of the case' data.
1621     Address case_data_addr(mdp, index);
1622     Register case_data = scratch1;
1623     load_address(case_data, case_data_addr);
1624 
1625     // Update the case count.
1626     increment_mdp_data_at(case_data,
1627                           in_bytes(MultiBranchData::relative_count_offset()),
1628                           scratch2);
1629 
1630     // The method data pointer needs to be updated.
1631     update_mdp_by_offset(mdp,
1632                          index,
1633                          in_bytes(MultiBranchData::relative_displacement_offset()));
1634 
1635     bind(profile_continue);
1636   }
1637 }
1638 
1639 // kills: R0, R1, flags, loads klass from obj (if not null)
1640 void InterpreterMacroAssembler::profile_obj_type(Register obj, Address mdo_addr, Register klass, bool cmp_done) {
1641   NearLabel null_seen, init_klass, do_nothing, do_update;
1642 
1643   // Klass = obj is allowed.
1644   const Register tmp = Z_R1;
1645   assert_different_registers(obj, mdo_addr.base(), tmp, Z_R0);
1646   assert_different_registers(klass, mdo_addr.base(), tmp, Z_R0);
1647 
1648   z_lg(tmp, mdo_addr);
1649   if (cmp_done) {
1650     z_brz(null_seen);
1651   } else {
1652     compareU64_and_branch(obj, (intptr_t)0, Assembler::bcondEqual, null_seen);
1653   }
1654 
1655   verify_oop(obj);
1656   load_klass(klass, obj);
1657 
1658   // Klass seen before, nothing to do (regardless of unknown bit).
1659   z_lgr(Z_R0, tmp);
1660   assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
1661   z_nill(Z_R0, TypeEntries::type_klass_mask & 0xFFFF);
1662   compareU64_and_branch(Z_R0, klass, Assembler::bcondEqual, do_nothing);
1663 
1664   // Already unknown. Nothing to do anymore.
1665   z_tmll(tmp, TypeEntries::type_unknown);
1666   z_brc(Assembler::bcondAllOne, do_nothing);
1667 
1668   z_lgr(Z_R0, tmp);
1669   assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
1670   z_nill(Z_R0, TypeEntries::type_mask & 0xFFFF);
1671   compareU64_and_branch(Z_R0, (intptr_t)0, Assembler::bcondEqual, init_klass);
1672 
1673   // Different than before. Cannot keep accurate profile.
1674   z_oill(tmp, TypeEntries::type_unknown);
1675   z_bru(do_update);
1676 
1677   bind(init_klass);
1678   // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
1679   z_ogr(tmp, klass);
1680   z_bru(do_update);
1681 
1682   bind(null_seen);
1683   // Set null_seen if obj is 0.
1684   z_oill(tmp, TypeEntries::null_seen);
1685   // fallthru: z_bru(do_update);
1686 
1687   bind(do_update);
1688   z_stg(tmp, mdo_addr);
1689 
1690   bind(do_nothing);
1691 }
1692 
1693 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1694   if (!ProfileInterpreter) {
1695     return;
1696   }
1697 
1698   assert_different_registers(mdp, callee, tmp);
1699 
1700   if (MethodData::profile_arguments() || MethodData::profile_return()) {
1701     Label profile_continue;
1702 
1703     test_method_data_pointer(mdp, profile_continue);
1704 
1705     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1706 
1707     z_cliy(in_bytes(DataLayout::tag_offset()) - off_to_start, mdp,
1708            is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1709     z_brne(profile_continue);
1710 
1711     if (MethodData::profile_arguments()) {
1712       NearLabel done;
1713       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1714       add2reg(mdp, off_to_args);
1715 
1716       for (int i = 0; i < TypeProfileArgsLimit; i++) {
1717         if (i > 0 || MethodData::profile_return()) {
1718           // If return value type is profiled we may have no argument to profile.
1719           z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1720           add2reg(tmp, -i*TypeStackSlotEntries::per_arg_count());
1721           compare64_and_branch(tmp, TypeStackSlotEntries::per_arg_count(), Assembler::bcondLow, done);
1722         }
1723         z_lg(tmp, Address(callee, Method::const_offset()));
1724         z_lgh(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1725         // Stack offset o (zero based) from the start of the argument
1726         // list. For n arguments translates into offset n - o - 1 from
1727         // the end of the argument list. But there is an extra slot at
1728         // the top of the stack. So the offset is n - o from Lesp.
1729         z_sg(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
1730         z_sllg(tmp, tmp, Interpreter::logStackElementSize);
1731         Address stack_slot_addr(tmp, Z_esp);
1732         z_ltg(tmp, stack_slot_addr);
1733 
1734         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1735         profile_obj_type(tmp, mdo_arg_addr, tmp, /*ltg did compare to 0*/ true);
1736 
1737         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1738         add2reg(mdp, to_add);
1739         off_to_args += to_add;
1740       }
1741 
1742       if (MethodData::profile_return()) {
1743         z_lg(tmp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, mdp);
1744         add2reg(tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1745       }
1746 
1747       bind(done);
1748 
1749       if (MethodData::profile_return()) {
1750         // We're right after the type profile for the last
1751         // argument. Tmp is the number of cells left in the
1752         // CallTypeData/VirtualCallTypeData to reach its end. Non null
1753         // if there's a return to profile.
1754         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1755         z_sllg(tmp, tmp, exact_log2(DataLayout::cell_size));
1756         z_agr(mdp, tmp);
1757       }
1758       z_stg(mdp, _z_ijava_state_neg(mdx), Z_fp);
1759     } else {
1760       assert(MethodData::profile_return(), "either profile call args or call ret");
1761       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1762     }
1763 
1764     // Mdp points right after the end of the
1765     // CallTypeData/VirtualCallTypeData, right after the cells for the
1766     // return value type if there's one.
1767     bind(profile_continue);
1768   }
1769 }
1770 
1771 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1772   assert_different_registers(mdp, ret, tmp);
1773   if (ProfileInterpreter && MethodData::profile_return()) {
1774     Label profile_continue;
1775 
1776     test_method_data_pointer(mdp, profile_continue);
1777 
1778     if (MethodData::profile_return_jsr292_only()) {
1779       // If we don't profile all invoke bytecodes we must make sure
1780       // it's a bytecode we indeed profile. We can't go back to the
1781       // beginning of the ProfileData we intend to update to check its
1782       // type because we're right after it and we don't known its
1783       // length.
1784       NearLabel do_profile;
1785       Address bc(Z_bcp);
1786       z_lb(tmp, bc);
1787       compare32_and_branch(tmp, Bytecodes::_invokedynamic, Assembler::bcondEqual, do_profile);
1788       compare32_and_branch(tmp, Bytecodes::_invokehandle, Assembler::bcondEqual, do_profile);
1789       get_method(tmp);
1790       // Supplement to 8139891: _intrinsic_id exceeded 1-byte size limit.
1791       if (Method::intrinsic_id_size_in_bytes() == 1) {
1792         z_cli(Method::intrinsic_id_offset_in_bytes(), tmp, vmIntrinsics::_compiledLambdaForm);
1793       } else {
1794         assert(Method::intrinsic_id_size_in_bytes() == 2, "size error: check Method::_intrinsic_id");
1795         z_lh(tmp, Method::intrinsic_id_offset_in_bytes(), Z_R0, tmp);
1796         z_chi(tmp, vmIntrinsics::_compiledLambdaForm);
1797       }
1798       z_brne(profile_continue);
1799 
1800       bind(do_profile);
1801     }
1802 
1803     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1804     profile_obj_type(ret, mdo_ret_addr, tmp);
1805 
1806     bind(profile_continue);
1807   }
1808 }
1809 
1810 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1811   if (ProfileInterpreter && MethodData::profile_parameters()) {
1812     Label profile_continue, done;
1813 
1814     test_method_data_pointer(mdp, profile_continue);
1815 
1816     // Load the offset of the area within the MDO used for
1817     // parameters. If it's negative we're not profiling any parameters.
1818     Address parm_di_addr(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()));
1819     load_and_test_int2long(tmp1, parm_di_addr);
1820     z_brl(profile_continue);
1821 
1822     // Compute a pointer to the area for parameters from the offset
1823     // and move the pointer to the slot for the last
1824     // parameters. Collect profiling from last parameter down.
1825     // mdo start + parameters offset + array length - 1
1826 
1827     // Pointer to the parameter area in the MDO.
1828     z_agr(mdp, tmp1);
1829 
1830     // Offset of the current profile entry to update.
1831     const Register entry_offset = tmp1;
1832     // entry_offset = array len in number of cells.
1833     z_lg(entry_offset, Address(mdp, ArrayData::array_len_offset()));
1834     // entry_offset (number of cells) = array len - size of 1 entry
1835     add2reg(entry_offset, -TypeStackSlotEntries::per_arg_count());
1836     // entry_offset in bytes
1837     z_sllg(entry_offset, entry_offset, exact_log2(DataLayout::cell_size));
1838 
1839     Label loop;
1840     bind(loop);
1841 
1842     Address arg_off(mdp, entry_offset, ParametersTypeData::stack_slot_offset(0));
1843     Address arg_type(mdp, entry_offset, ParametersTypeData::type_offset(0));
1844 
1845     // Load offset on the stack from the slot for this parameter.
1846     z_lg(tmp2, arg_off);
1847     z_sllg(tmp2, tmp2, Interpreter::logStackElementSize);
1848     z_lcgr(tmp2); // Negate.
1849 
1850     // Profile the parameter.
1851     z_ltg(tmp2, Address(Z_locals, tmp2));
1852     profile_obj_type(tmp2, arg_type, tmp2, /*ltg did compare to 0*/ true);
1853 
1854     // Go to next parameter.
1855     z_aghi(entry_offset, -TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size);
1856     z_brnl(loop);
1857 
1858     bind(profile_continue);
1859   }
1860 }
1861 
1862 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1863 void InterpreterMacroAssembler::increment_mask_and_jump(Address          counter_addr,
1864                                                         int              increment,
1865                                                         Address          mask,
1866                                                         Register         scratch,
1867                                                         bool             preloaded,
1868                                                         branch_condition cond,
1869                                                         Label           *where) {
1870   assert_different_registers(counter_addr.base(), scratch);
1871   if (preloaded) {
1872     add2reg(scratch, increment);
1873     reg2mem_opt(scratch, counter_addr, false);
1874   } else {
1875     if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment) && counter_addr.is_RSYform()) {
1876       z_alsi(counter_addr.disp20(), counter_addr.base(), increment);
1877       mem2reg_signed_opt(scratch, counter_addr);
1878     } else {
1879       mem2reg_signed_opt(scratch, counter_addr);
1880       add2reg(scratch, increment);
1881       reg2mem_opt(scratch, counter_addr, false);
1882     }
1883   }
1884   z_n(scratch, mask);
1885   if (where) { z_brc(cond, *where); }
1886 }
1887 
1888 // Get MethodCounters object for given method. Lazily allocated if necessary.
1889 //   method    - Ptr to Method object.
1890 //   Rcounters - Ptr to MethodCounters object associated with Method object.
1891 //   skip      - Exit point if MethodCounters object can't be created (OOM condition).
1892 void InterpreterMacroAssembler::get_method_counters(Register Rmethod,
1893                                                     Register Rcounters,
1894                                                     Label& skip) {
1895   assert_different_registers(Rmethod, Rcounters);
1896 
1897   BLOCK_COMMENT("get MethodCounters object {");
1898 
1899   Label has_counters;
1900   load_and_test_long(Rcounters, Address(Rmethod, Method::method_counters_offset()));
1901   z_brnz(has_counters);
1902 
1903   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters), Rmethod, false);
1904   z_ltgr(Rcounters, Z_RET); // Runtime call returns MethodCounters object.
1905   z_brz(skip); // No MethodCounters, out of memory.
1906 
1907   bind(has_counters);
1908 
1909   BLOCK_COMMENT("} get MethodCounters object");
1910 }
1911 
1912 // Increment invocation counter in MethodCounters object.
1913 // Return (invocation_counter+backedge_counter) as "result" in RctrSum.
1914 // Counter values are all unsigned.
1915 void InterpreterMacroAssembler::increment_invocation_counter(Register Rcounters, Register RctrSum) {
1916   assert(UseCompiler || LogTouchedMethods, "incrementing must be useful");
1917   assert_different_registers(Rcounters, RctrSum);
1918 
1919   int increment          = InvocationCounter::count_increment;
1920   int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1921   int be_counter_offset  = in_bytes(MethodCounters::backedge_counter_offset()   + InvocationCounter::counter_offset());
1922 
1923   BLOCK_COMMENT("Increment invocation counter {");
1924 
1925   if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1926     // Increment the invocation counter in place,
1927     // then add the incremented value to the backedge counter.
1928     z_l(RctrSum, be_counter_offset, Rcounters);
1929     z_alsi(inv_counter_offset, Rcounters, increment);     // Atomic increment @no extra cost!
1930     z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1931     z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
1932   } else {
1933     // This path is optimized for low register consumption
1934     // at the cost of somewhat higher operand delays.
1935     // It does not need an extra temp register.
1936 
1937     // Update the invocation counter.
1938     z_l(RctrSum, inv_counter_offset, Rcounters);
1939     if (RctrSum == Z_R0) {
1940       z_ahi(RctrSum, increment);
1941     } else {
1942       add2reg(RctrSum, increment);
1943     }
1944     z_st(RctrSum, inv_counter_offset, Rcounters);
1945 
1946     // Mask off the state bits.
1947     z_nilf(RctrSum, InvocationCounter::count_mask_value);
1948 
1949     // Add the backedge counter to the updated invocation counter to
1950     // form the result.
1951     z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1952   }
1953 
1954   BLOCK_COMMENT("} Increment invocation counter");
1955 
1956   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
1957 }
1958 
1959 
1960 // increment backedge counter in MethodCounters object.
1961 // return (invocation_counter+backedge_counter) as "result" in RctrSum
1962 // counter values are all unsigned!
1963 void InterpreterMacroAssembler::increment_backedge_counter(Register Rcounters, Register RctrSum) {
1964   assert(UseCompiler, "incrementing must be useful");
1965   assert_different_registers(Rcounters, RctrSum);
1966 
1967   int increment          = InvocationCounter::count_increment;
1968   int inv_counter_offset = in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset());
1969   int be_counter_offset  = in_bytes(MethodCounters::backedge_counter_offset()   + InvocationCounter::counter_offset());
1970 
1971   BLOCK_COMMENT("Increment backedge counter {");
1972 
1973   if (VM_Version::has_MemWithImmALUOps() && Immediate::is_simm8(increment)) {
1974     // Increment the invocation counter in place,
1975     // then add the incremented value to the backedge counter.
1976     z_l(RctrSum, inv_counter_offset, Rcounters);
1977     z_alsi(be_counter_offset, Rcounters, increment);      // Atomic increment @no extra cost!
1978     z_nilf(RctrSum, InvocationCounter::count_mask_value); // Mask off state bits.
1979     z_al(RctrSum, be_counter_offset, Z_R0, Rcounters);
1980   } else {
1981     // This path is optimized for low register consumption
1982     // at the cost of somewhat higher operand delays.
1983     // It does not need an extra temp register.
1984 
1985     // Update the invocation counter.
1986     z_l(RctrSum, be_counter_offset, Rcounters);
1987     if (RctrSum == Z_R0) {
1988       z_ahi(RctrSum, increment);
1989     } else {
1990       add2reg(RctrSum, increment);
1991     }
1992     z_st(RctrSum, be_counter_offset, Rcounters);
1993 
1994     // Mask off the state bits.
1995     z_nilf(RctrSum, InvocationCounter::count_mask_value);
1996 
1997     // Add the backedge counter to the updated invocation counter to
1998     // form the result.
1999     z_al(RctrSum, inv_counter_offset, Z_R0, Rcounters);
2000   }
2001 
2002   BLOCK_COMMENT("} Increment backedge counter");
2003 
2004   // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2005 }
2006 
2007 // Add an InterpMonitorElem to stack (see frame_s390.hpp).
2008 void InterpreterMacroAssembler::add_monitor_to_stack(bool     stack_is_empty,
2009                                                      Register Rtemp1,
2010                                                      Register Rtemp2,
2011                                                      Register Rtemp3) {
2012 
2013   const Register Rcurr_slot = Rtemp1;
2014   const Register Rlimit     = Rtemp2;
2015   const jint delta = -frame::interpreter_frame_monitor_size() * wordSize;
2016 
2017   assert((delta & LongAlignmentMask) == 0,
2018          "sizeof BasicObjectLock must be even number of doublewords");
2019   assert(2 * wordSize == -delta, "this works only as long as delta == -2*wordSize");
2020   assert(Rcurr_slot != Z_R0, "Register must be usable as base register");
2021   assert_different_registers(Rlimit, Rcurr_slot, Rtemp3);
2022 
2023   get_monitors(Rlimit);
2024 
2025   // Adjust stack pointer for additional monitor entry.
2026   resize_frame(RegisterOrConstant((intptr_t) delta), Z_fp, false);
2027 
2028   if (!stack_is_empty) {
2029     // Must copy stack contents down.
2030     NearLabel next, done;
2031 
2032     // Rtemp := addr(Tos), Z_esp is pointing below it!
2033     add2reg(Rcurr_slot, wordSize, Z_esp);
2034 
2035     // Nothing to do, if already at monitor area.
2036     compareU64_and_branch(Rcurr_slot, Rlimit, bcondNotLow, done);
2037 
2038     bind(next);
2039 
2040     // Move one stack slot.
2041     mem2reg_opt(Rtemp3, Address(Rcurr_slot));
2042     reg2mem_opt(Rtemp3, Address(Rcurr_slot, delta));
2043     add2reg(Rcurr_slot, wordSize);
2044     compareU64_and_branch(Rcurr_slot, Rlimit, bcondLow, next); // Are we done?
2045 
2046     bind(done);
2047     // Done copying stack.
2048   }
2049 
2050   // Adjust expression stack and monitor pointers.
2051   add2reg(Z_esp, delta);
2052   add2reg(Rlimit, delta);
2053   save_monitors(Rlimit);
2054 }
2055 
2056 // Note: Index holds the offset in bytes afterwards.
2057 // You can use this to store a new value (with Llocals as the base).
2058 void InterpreterMacroAssembler::access_local_int(Register index, Register dst) {
2059   z_sllg(index, index, LogBytesPerWord);
2060   mem2reg_opt(dst, Address(Z_locals, index), false);
2061 }
2062 
2063 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2064   if (state == atos) { MacroAssembler::verify_oop(reg); }
2065 }
2066 
2067 // Inline assembly for:
2068 //
2069 // if (thread is in interp_only_mode) {
2070 //   InterpreterRuntime::post_method_entry();
2071 // }
2072 
2073 void InterpreterMacroAssembler::notify_method_entry() {
2074 
2075   // JVMTI
2076   // Whenever JVMTI puts a thread in interp_only_mode, method
2077   // entry/exit events are sent for that thread to track stack
2078   // depth. If it is possible to enter interp_only_mode we add
2079   // the code to check if the event should be sent.
2080   if (JvmtiExport::can_post_interpreter_events()) {
2081     Label jvmti_post_done;
2082     MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2083     z_bre(jvmti_post_done);
2084     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), /*check_exceptions=*/false);
2085     bind(jvmti_post_done);
2086   }
2087 }
2088 
2089 // Inline assembly for:
2090 //
2091 // if (thread is in interp_only_mode) {
2092 //   if (!native_method) save result
2093 //   InterpreterRuntime::post_method_exit();
2094 //   if (!native_method) restore result
2095 // }
2096 // if (DTraceMethodProbes) {
2097 //   SharedRuntime::dtrace_method_exit(thread, method);
2098 // }
2099 //
2100 // For native methods their result is stored in z_ijava_state.lresult
2101 // and z_ijava_state.fresult before coming here.
2102 // Java methods have their result stored in the expression stack.
2103 //
2104 // Notice the dependency to frame::interpreter_frame_result().
2105 void InterpreterMacroAssembler::notify_method_exit(bool native_method,
2106                                                    TosState state,
2107                                                    NotifyMethodExitMode mode) {
2108   // JVMTI
2109   // Whenever JVMTI puts a thread in interp_only_mode, method
2110   // entry/exit events are sent for that thread to track stack
2111   // depth. If it is possible to enter interp_only_mode we add
2112   // the code to check if the event should be sent.
2113   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2114     Label jvmti_post_done;
2115     MacroAssembler::load_and_test_int(Z_R0, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2116     z_bre(jvmti_post_done);
2117     if (!native_method) push(state); // see frame::interpreter_frame_result()
2118     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), /*check_exceptions=*/false);
2119     if (!native_method) pop(state);
2120     bind(jvmti_post_done);
2121   }
2122 
2123 #if 0
2124   // Dtrace currently not supported on z/Architecture.
2125   {
2126     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2127     push(state);
2128     get_method(c_rarg1);
2129     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2130                  r15_thread, c_rarg1);
2131     pop(state);
2132   }
2133 #endif
2134 }
2135 
2136 void InterpreterMacroAssembler::skip_if_jvmti_mode(Label &Lskip, Register Rscratch) {
2137   if (!JvmtiExport::can_post_interpreter_events()) {
2138     return;
2139   }
2140 
2141   load_and_test_int(Rscratch, Address(Z_thread, JavaThread::interp_only_mode_offset()));
2142   z_brnz(Lskip);
2143 
2144 }
2145 
2146 // Pop the topmost TOP_IJAVA_FRAME and set it's sender_sp as new Z_SP.
2147 // The return pc is loaded into the register return_pc.
2148 //
2149 // Registers updated:
2150 //     return_pc  - The return pc of the calling frame.
2151 //     tmp1, tmp2 - scratch
2152 void InterpreterMacroAssembler::pop_interpreter_frame(Register return_pc, Register tmp1, Register tmp2) {
2153   // F0  Z_SP -> caller_sp (F1's)
2154   //             ...
2155   //             sender_sp (F1's)
2156   //             ...
2157   // F1  Z_fp -> caller_sp (F2's)
2158   //             return_pc (Continuation after return from F0.)
2159   //             ...
2160   // F2          caller_sp
2161 
2162   // Remove F0's activation. Restoring Z_SP to sender_sp reverts modifications
2163   // (a) by a c2i adapter and (b) by generate_fixed_frame().
2164   // In case (a) the new top frame F1 is an unextended compiled frame.
2165   // In case (b) F1 is converted from PARENT_IJAVA_FRAME to TOP_IJAVA_FRAME.
2166 
2167   // Case (b) seems to be redundant when returning to a interpreted caller,
2168   // because then the caller's top_frame_sp is installed as sp (see
2169   // TemplateInterpreterGenerator::generate_return_entry_for ()). But
2170   // pop_interpreter_frame() is also used in exception handling and there the
2171   // frame type of the caller is unknown, therefore top_frame_sp cannot be used,
2172   // so it is important that sender_sp is the caller's sp as TOP_IJAVA_FRAME.
2173 
2174   Register R_f1_sender_sp = tmp1;
2175   Register R_f2_sp = tmp2;
2176 
2177   // Tirst check the for the interpreter frame's magic.
2178   asm_assert_ijava_state_magic(R_f2_sp/*tmp*/);
2179   z_lg(R_f2_sp, _z_parent_ijava_frame_abi(callers_sp), Z_fp);
2180   z_lg(R_f1_sender_sp, _z_ijava_state_neg(sender_sp), Z_fp);
2181   if (return_pc->is_valid())
2182     z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2183   // Pop F0 by resizing to R_f1_sender_sp and using R_f2_sp as fp.
2184   resize_frame_absolute(R_f1_sender_sp, R_f2_sp, false/*load fp*/);
2185 
2186 #ifdef ASSERT
2187   // The return_pc in the new top frame is dead... at least that's my
2188   // current understanding; to assert this I overwrite it.
2189   load_const_optimized(Z_ARG3, 0xb00b1);
2190   z_stg(Z_ARG3, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2191 #endif
2192 }
2193 
2194 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2195   if (VerifyFPU) {
2196     unimplemented("verfiyFPU");
2197   }
2198 }
2199