1 /*
   2  * Copyright (c) 2016, 2021, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2020 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 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "classfile/javaClasses.hpp"
  29 #include "gc/shared/barrierSetAssembler.hpp"
  30 #include "interpreter/abstractInterpreter.hpp"
  31 #include "interpreter/bytecodeHistogram.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "interpreter/interp_masm.hpp"
  35 #include "interpreter/templateInterpreterGenerator.hpp"
  36 #include "interpreter/templateTable.hpp"
  37 #include "oops/arrayOop.hpp"
  38 #include "oops/methodData.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "prims/jvmtiExport.hpp"
  41 #include "prims/jvmtiThreadState.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/deoptimization.hpp"
  44 #include "runtime/frame.inline.hpp"
  45 #include "runtime/jniHandles.hpp"
  46 #include "runtime/sharedRuntime.hpp"
  47 #include "runtime/stubRoutines.hpp"
  48 #include "runtime/synchronizer.hpp"
  49 #include "runtime/timer.hpp"
  50 #include "runtime/vframeArray.hpp"
  51 #include "utilities/debug.hpp"
  52 
  53 
  54 // Size of interpreter code.  Increase if too small.  Interpreter will
  55 // fail with a guarantee ("not enough space for interpreter generation");
  56 // if too small.
  57 // Run with +PrintInterpreter to get the VM to print out the size.
  58 // Max size with JVMTI
  59 int TemplateInterpreter::InterpreterCodeSize = 320*K;
  60 
  61 #undef  __
  62 #ifdef PRODUCT
  63   #define __ _masm->
  64 #else
  65   #define __ _masm->
  66 //  #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
  67 #endif
  68 
  69 #define BLOCK_COMMENT(str) __ block_comment(str)
  70 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  71 
  72 #define oop_tmp_offset     _z_ijava_state_neg(oop_tmp)
  73 
  74 //-----------------------------------------------------------------------------
  75 
  76 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  77   //
  78   // New slow_signature handler that respects the z/Architecture
  79   // C calling conventions.
  80   //
  81   // We get called by the native entry code with our output register
  82   // area == 8. First we call InterpreterRuntime::get_result_handler
  83   // to copy the pointer to the signature string temporarily to the
  84   // first C-argument and to return the result_handler in
  85   // Z_RET. Since native_entry will copy the jni-pointer to the
  86   // first C-argument slot later on, it's OK to occupy this slot
  87   // temporarily. Then we copy the argument list on the java
  88   // expression stack into native varargs format on the native stack
  89   // and load arguments into argument registers. Integer arguments in
  90   // the varargs vector will be sign-extended to 8 bytes.
  91   //
  92   // On entry:
  93   //   Z_ARG1  - intptr_t*       Address of java argument list in memory.
  94   //   Z_state - zeroInterpreter* Address of interpreter state for
  95   //                              this method
  96   //   Z_method
  97   //
  98   // On exit (just before return instruction):
  99   //   Z_RET contains the address of the result_handler.
 100   //   Z_ARG2 is not updated for static methods and contains "this" otherwise.
 101   //   Z_ARG3-Z_ARG5 contain the first 3 arguments of types other than float and double.
 102   //   Z_FARG1-Z_FARG4 contain the first 4 arguments of type float or double.
 103 
 104   const int LogSizeOfCase = 3;
 105 
 106   const int max_fp_register_arguments   = Argument::n_float_register_parameters;
 107   const int max_int_register_arguments  = Argument::n_register_parameters - 2;  // First 2 are reserved.
 108 
 109   const Register arg_java       = Z_tmp_2;
 110   const Register arg_c          = Z_tmp_3;
 111   const Register signature      = Z_R1_scratch; // Is a string.
 112   const Register fpcnt          = Z_R0_scratch;
 113   const Register argcnt         = Z_tmp_4;
 114   const Register intSlot        = Z_tmp_1;
 115   const Register sig_end        = Z_tmp_1; // Assumed end of signature (only used in do_object).
 116   const Register target_sp      = Z_tmp_1;
 117   const FloatRegister floatSlot = Z_F1;
 118 
 119   const int d_signature         = _z_abi(gpr6); // Only spill space, register contents not affected.
 120   const int d_fpcnt             = _z_abi(gpr7); // Only spill space, register contents not affected.
 121 
 122   unsigned int entry_offset = __ offset();
 123 
 124   BLOCK_COMMENT("slow_signature_handler {");
 125 
 126   // We use target_sp for storing arguments in the C frame.
 127   __ save_return_pc();
 128   __ push_frame_abi160(4*BytesPerWord);                 // Reserve space to save the tmp_[1..4] registers.
 129   __ z_stmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // Save registers only after frame is pushed.
 130 
 131   __ z_lgr(arg_java, Z_ARG1);
 132 
 133   Register   method = Z_ARG2; // Directly load into correct argument register.
 134 
 135   __ get_method(method);
 136   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), Z_thread, method);
 137 
 138   // Move signature to callee saved register.
 139   // Don't directly write to stack. Frame is used by VM call.
 140   __ z_lgr(Z_tmp_1, Z_RET);
 141 
 142   // Reload method. Register may have been altered by VM call.
 143   __ get_method(method);
 144 
 145   // Get address of result handler.
 146   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), Z_thread, method);
 147 
 148   // Save signature address to stack.
 149   __ z_stg(Z_tmp_1, d_signature, Z_SP);
 150 
 151   // Don't overwrite return value (Z_RET, Z_ARG1) in rest of the method !
 152 
 153   {
 154     Label   isStatic;
 155 
 156     // Test if static.
 157     // We can test the bit directly.
 158     // Path is Z_method->_access_flags._flags.
 159     // We only support flag bits in the least significant byte (assert !).
 160     // Therefore add 3 to address that byte within "_flags".
 161     // Reload method. VM call above may have destroyed register contents
 162     __ get_method(method);
 163     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 164     method = noreg;  // end of life
 165     __ z_btrue(isStatic);
 166 
 167     // For non-static functions, pass "this" in Z_ARG2 and copy it to 2nd C-arg slot.
 168     // Need to box the Java object here, so we use arg_java
 169     // (address of current Java stack slot) as argument and
 170     // don't dereference it as in case of ints, floats, etc..
 171     __ z_lgr(Z_ARG2, arg_java);
 172     __ add2reg(arg_java, -BytesPerWord);
 173     __ bind(isStatic);
 174   }
 175 
 176   // argcnt == 0 corresponds to 3rd C argument.
 177   //   arg #1 (result handler) and
 178   //   arg #2 (this, for non-statics), unused else
 179   // are reserved and pre-filled above.
 180   // arg_java points to the corresponding Java argument here. It
 181   // has been decremented by one argument (this) in case of non-static.
 182   __ clear_reg(argcnt, true, false);  // Don't set CC.
 183   __ z_lg(target_sp, 0, Z_SP);
 184   __ add2reg(arg_c, _z_abi(remaining_cargs), target_sp);
 185   // No floating-point args parsed so far.
 186   __ clear_mem(Address(Z_SP, d_fpcnt), 8);
 187 
 188   NearLabel   move_intSlot_to_ARG, move_floatSlot_to_FARG;
 189   NearLabel   loop_start, loop_start_restore, loop_end;
 190   NearLabel   do_int, do_long, do_float, do_double;
 191   NearLabel   do_dontreachhere, do_object, do_array, do_boxed;
 192 
 193 #ifdef ASSERT
 194   // Signature needs to point to '(' (== 0x28) at entry.
 195   __ z_lg(signature, d_signature, Z_SP);
 196   __ z_cli(0, signature, (int) '(');
 197   __ z_brne(do_dontreachhere);
 198 #endif
 199 
 200   __ bind(loop_start_restore);
 201   __ z_lg(signature, d_signature, Z_SP);  // Restore signature ptr, destroyed by move_XX_to_ARG.
 202 
 203   BIND(loop_start);
 204   // Advance to next argument type token from the signature.
 205   __ add2reg(signature, 1);
 206 
 207   // Use CLI, works well on all CPU versions.
 208     __ z_cli(0, signature, (int) ')');
 209     __ z_bre(loop_end);                // end of signature
 210     __ z_cli(0, signature, (int) 'L');
 211     __ z_bre(do_object);               // object     #9
 212     __ z_cli(0, signature, (int) 'F');
 213     __ z_bre(do_float);                // float      #7
 214     __ z_cli(0, signature, (int) 'J');
 215     __ z_bre(do_long);                 // long       #6
 216     __ z_cli(0, signature, (int) 'B');
 217     __ z_bre(do_int);                  // byte       #1
 218     __ z_cli(0, signature, (int) 'Z');
 219     __ z_bre(do_int);                  // boolean    #2
 220     __ z_cli(0, signature, (int) 'C');
 221     __ z_bre(do_int);                  // char       #3
 222     __ z_cli(0, signature, (int) 'S');
 223     __ z_bre(do_int);                  // short      #4
 224     __ z_cli(0, signature, (int) 'I');
 225     __ z_bre(do_int);                  // int        #5
 226     __ z_cli(0, signature, (int) 'D');
 227     __ z_bre(do_double);               // double     #8
 228     __ z_cli(0, signature, (int) '[');
 229     __ z_bre(do_array);                // array      #10
 230 
 231   __ bind(do_dontreachhere);
 232 
 233   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 234 
 235   // Array argument
 236   BIND(do_array);
 237 
 238   {
 239     Label   start_skip, end_skip;
 240 
 241     __ bind(start_skip);
 242 
 243     // Advance to next type tag from signature.
 244     __ add2reg(signature, 1);
 245 
 246     // Use CLI, works well on all CPU versions.
 247     __ z_cli(0, signature, (int) '[');
 248     __ z_bre(start_skip);               // Skip further brackets.
 249 
 250     __ z_cli(0, signature, (int) '9');
 251     __ z_brh(end_skip);                 // no optional size
 252 
 253     __ z_cli(0, signature, (int) '0');
 254     __ z_brnl(start_skip);              // Skip optional size.
 255 
 256     __ bind(end_skip);
 257 
 258     __ z_cli(0, signature, (int) 'L');
 259     __ z_brne(do_boxed);                // If not array of objects: go directly to do_boxed.
 260   }
 261 
 262   //  OOP argument
 263   BIND(do_object);
 264   // Pass by an object's type name.
 265   {
 266     Label   L;
 267 
 268     __ add2reg(sig_end, 4095, signature);     // Assume object type name is shorter than 4k.
 269     __ load_const_optimized(Z_R0, (int) ';'); // Type name terminator (must be in Z_R0!).
 270     __ MacroAssembler::search_string(sig_end, signature);
 271     __ z_brl(L);
 272     __ z_illtrap();  // No semicolon found: internal error or object name too long.
 273     __ bind(L);
 274     __ z_lgr(signature, sig_end);
 275     // fallthru to do_boxed
 276   }
 277 
 278   // Need to box the Java object here, so we use arg_java
 279   // (address of current Java stack slot) as argument and
 280   // don't dereference it as in case of ints, floats, etc..
 281 
 282   // UNBOX argument
 283   // Load reference and check for NULL.
 284   Label  do_int_Entry4Boxed;
 285   __ bind(do_boxed);
 286   {
 287     __ load_and_test_long(intSlot, Address(arg_java));
 288     __ z_bre(do_int_Entry4Boxed);
 289     __ z_lgr(intSlot, arg_java);
 290     __ z_bru(do_int_Entry4Boxed);
 291   }
 292 
 293   // INT argument
 294 
 295   // (also for byte, boolean, char, short)
 296   // Use lgf for load (sign-extend) and stg for store.
 297   BIND(do_int);
 298   __ z_lgf(intSlot, 0, arg_java);
 299 
 300   __ bind(do_int_Entry4Boxed);
 301   __ add2reg(arg_java, -BytesPerWord);
 302   // If argument fits into argument register, go and handle it, otherwise continue.
 303   __ compare32_and_branch(argcnt, max_int_register_arguments,
 304                           Assembler::bcondLow, move_intSlot_to_ARG);
 305   __ z_stg(intSlot, 0, arg_c);
 306   __ add2reg(arg_c, BytesPerWord);
 307   __ z_bru(loop_start);
 308 
 309   // LONG argument
 310 
 311   BIND(do_long);
 312   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 313   __ z_lg(intSlot, BytesPerWord, arg_java);
 314   // If argument fits into argument register, go and handle it, otherwise continue.
 315   __ compare32_and_branch(argcnt, max_int_register_arguments,
 316                           Assembler::bcondLow, move_intSlot_to_ARG);
 317   __ z_stg(intSlot, 0, arg_c);
 318   __ add2reg(arg_c, BytesPerWord);
 319   __ z_bru(loop_start);
 320 
 321   // FLOAT argumen
 322 
 323   BIND(do_float);
 324   __ z_le(floatSlot, 0, arg_java);
 325   __ add2reg(arg_java, -BytesPerWord);
 326   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 327   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 328   __ z_brl(move_floatSlot_to_FARG);
 329   __ z_ste(floatSlot, 4, arg_c);
 330   __ add2reg(arg_c, BytesPerWord);
 331   __ z_bru(loop_start);
 332 
 333   // DOUBLE argument
 334 
 335   BIND(do_double);
 336   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 337   __ z_ld(floatSlot, BytesPerWord, arg_java);
 338   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 339   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 340   __ z_brl(move_floatSlot_to_FARG);
 341   __ z_std(floatSlot, 0, arg_c);
 342   __ add2reg(arg_c, BytesPerWord);
 343   __ z_bru(loop_start);
 344 
 345   // Method exit, all arguments proocessed.
 346   __ bind(loop_end);
 347   __ z_lmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // restore registers before frame is popped.
 348   __ pop_frame();
 349   __ restore_return_pc();
 350   __ z_br(Z_R14);
 351 
 352   // Copy int arguments.
 353 
 354   Label  iarg_caselist;   // Distance between each case has to be a power of 2
 355                           // (= 1 << LogSizeOfCase).
 356   __ align(16);
 357   BIND(iarg_caselist);
 358   __ z_lgr(Z_ARG3, intSlot);    // 4 bytes
 359   __ z_bru(loop_start_restore); // 4 bytes
 360 
 361   __ z_lgr(Z_ARG4, intSlot);
 362   __ z_bru(loop_start_restore);
 363 
 364   __ z_lgr(Z_ARG5, intSlot);
 365   __ z_bru(loop_start_restore);
 366 
 367   __ align(16);
 368   __ bind(move_intSlot_to_ARG);
 369   __ z_stg(signature, d_signature, Z_SP);       // Spill since signature == Z_R1_scratch.
 370   __ z_larl(Z_R1_scratch, iarg_caselist);
 371   __ z_sllg(Z_R0_scratch, argcnt, LogSizeOfCase);
 372   __ add2reg(argcnt, 1);
 373   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 374   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 375 
 376   // Copy float arguments.
 377 
 378   Label  farg_caselist;   // Distance between each case has to be a power of 2
 379                           // (= 1 << logSizeOfCase, padded with nop.
 380   __ align(16);
 381   BIND(farg_caselist);
 382   __ z_ldr(Z_FARG1, floatSlot); // 2 bytes
 383   __ z_bru(loop_start_restore); // 4 bytes
 384   __ z_nop();                   // 2 bytes
 385 
 386   __ z_ldr(Z_FARG2, floatSlot);
 387   __ z_bru(loop_start_restore);
 388   __ z_nop();
 389 
 390   __ z_ldr(Z_FARG3, floatSlot);
 391   __ z_bru(loop_start_restore);
 392   __ z_nop();
 393 
 394   __ z_ldr(Z_FARG4, floatSlot);
 395   __ z_bru(loop_start_restore);
 396   __ z_nop();
 397 
 398   __ align(16);
 399   __ bind(move_floatSlot_to_FARG);
 400   __ z_stg(signature, d_signature, Z_SP);        // Spill since signature == Z_R1_scratch.
 401   __ z_lg(Z_R0_scratch, d_fpcnt, Z_SP);          // Need old value for indexing.
 402   __ add2mem_64(Address(Z_SP, d_fpcnt), 1, Z_R1_scratch); // Increment index.
 403   __ z_larl(Z_R1_scratch, farg_caselist);
 404   __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogSizeOfCase);
 405   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 406   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 407 
 408   BLOCK_COMMENT("} slow_signature_handler");
 409 
 410   return __ addr_at(entry_offset);
 411 }
 412 
 413 address TemplateInterpreterGenerator::generate_result_handler_for (BasicType type) {
 414   address entry = __ pc();
 415 
 416   assert(Z_tos == Z_RET, "Result handler: must move result!");
 417   assert(Z_ftos == Z_FRET, "Result handler: must move float result!");
 418 
 419   switch (type) {
 420     case T_BOOLEAN:
 421       __ c2bool(Z_tos);
 422       break;
 423     case T_CHAR:
 424       __ and_imm(Z_tos, 0xffff);
 425       break;
 426     case T_BYTE:
 427       __ z_lbr(Z_tos, Z_tos);
 428       break;
 429     case T_SHORT:
 430       __ z_lhr(Z_tos, Z_tos);
 431       break;
 432     case T_INT:
 433     case T_LONG:
 434     case T_VOID:
 435     case T_FLOAT:
 436     case T_DOUBLE:
 437       break;
 438     case T_OBJECT:
 439       // Retrieve result from frame...
 440       __ mem2reg_opt(Z_tos, Address(Z_fp, oop_tmp_offset));
 441       // and verify it.
 442       __ verify_oop(Z_tos);
 443       break;
 444     default:
 445       ShouldNotReachHere();
 446   }
 447   __ z_br(Z_R14);      // Return from result handler.
 448   return entry;
 449 }
 450 
 451 // Abstract method entry.
 452 // Attempt to execute abstract method. Throw exception.
 453 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 454   unsigned int entry_offset = __ offset();
 455 
 456   // Caller could be the call_stub or a compiled method (x86 version is wrong!).
 457 
 458   BLOCK_COMMENT("abstract_entry {");
 459 
 460   // Implement call of InterpreterRuntime::throw_AbstractMethodError.
 461   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1);
 462   __ save_return_pc();       // Save Z_R14.
 463   __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
 464 
 465   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 466                   Z_thread, Z_method);
 467 
 468   __ pop_frame();
 469   __ restore_return_pc();    // Restore Z_R14.
 470   __ reset_last_Java_frame();
 471 
 472   // Restore caller sp for c2i case.
 473   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 474 
 475   // branch to SharedRuntime::generate_forward_exception() which handles all possible callers,
 476   // i.e. call stub, compiled method, interpreted method.
 477   __ load_absolute_address(Z_tmp_1, StubRoutines::forward_exception_entry());
 478   __ z_br(Z_tmp_1);
 479 
 480   BLOCK_COMMENT("} abstract_entry");
 481 
 482   return __ addr_at(entry_offset);
 483 }
 484 
 485 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 486   // Inputs:
 487   //  Z_ARG1 - receiver
 488   //
 489   // What we do:
 490   //  - Load the referent field address.
 491   //  - Load the value in the referent field.
 492   //  - Pass that value to the pre-barrier.
 493   //
 494   // In the case of G1 this will record the value of the
 495   // referent in an SATB buffer if marking is active.
 496   // This will cause concurrent marking to mark the referent
 497   // field as live.
 498 
 499   Register  scratch1 = Z_tmp_2;
 500   Register  scratch2 = Z_tmp_3;
 501   Register  pre_val  = Z_RET;   // return value
 502   // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
 503   Register  Rargp    = Z_esp;
 504 
 505   Label     slow_path;
 506   address   entry = __ pc();
 507 
 508   const int referent_offset = java_lang_ref_Reference::referent_offset();
 509 
 510   BLOCK_COMMENT("Reference_get {");
 511 
 512   //  If the receiver is null then it is OK to jump to the slow path.
 513   __ load_and_test_long(pre_val, Address(Rargp, Interpreter::stackElementSize)); // Get receiver.
 514   __ z_bre(slow_path);
 515 
 516   //  Load the value of the referent field.
 517   __ load_heap_oop(pre_val, Address(pre_val, referent_offset), scratch1, scratch2, ON_WEAK_OOP_REF);
 518 
 519   // Restore caller sp for c2i case.
 520   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 521   __ z_br(Z_R14);
 522 
 523   // Branch to previously generated regular method entry.
 524   __ bind(slow_path);
 525 
 526   address meth_entry = Interpreter::entry_for_kind(Interpreter::zerolocals);
 527   __ jump_to_entry(meth_entry, Z_R1);
 528 
 529   BLOCK_COMMENT("} Reference_get");
 530 
 531   return entry;
 532 }
 533 
 534 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 535   address entry = __ pc();
 536 
 537   DEBUG_ONLY(__ verify_esp(Z_esp, Z_ARG5));
 538 
 539   // Restore bcp under the assumption that the current frame is still
 540   // interpreted.
 541   __ restore_bcp();
 542 
 543   // Expression stack must be empty before entering the VM if an
 544   // exception happened.
 545   __ empty_expression_stack();
 546   // Throw exception.
 547   __ call_VM(noreg,
 548              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
 549   return entry;
 550 }
 551 
 552 //
 553 // Args:
 554 //   Z_ARG2: oop of array
 555 //   Z_ARG3: aberrant index
 556 //
 557 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 558   address entry = __ pc();
 559   address excp = CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException);
 560 
 561   // Expression stack must be empty before entering the VM if an
 562   // exception happened.
 563   __ empty_expression_stack();
 564 
 565   // Setup parameters.
 566   // Pass register with array to create more detailed exceptions.
 567   __ call_VM(noreg, excp, Z_ARG2, Z_ARG3);
 568   return entry;
 569 }
 570 
 571 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 572   address entry = __ pc();
 573 
 574   // Object is at TOS.
 575   __ pop_ptr(Z_ARG2);
 576 
 577   // Expression stack must be empty before entering the VM if an
 578   // exception happened.
 579   __ empty_expression_stack();
 580 
 581   __ call_VM(Z_ARG1,
 582              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException),
 583              Z_ARG2);
 584 
 585   DEBUG_ONLY(__ should_not_reach_here();)
 586 
 587   return entry;
 588 }
 589 
 590 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 591   assert(!pass_oop || message == NULL, "either oop or message but not both");
 592   address entry = __ pc();
 593 
 594   BLOCK_COMMENT("exception_handler_common {");
 595 
 596   // Expression stack must be empty before entering the VM if an
 597   // exception happened.
 598   __ empty_expression_stack();
 599   if (name != NULL) {
 600     __ load_absolute_address(Z_ARG2, (address)name);
 601   } else {
 602     __ clear_reg(Z_ARG2, true, false);
 603   }
 604 
 605   if (pass_oop) {
 606     __ call_VM(Z_tos,
 607                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception),
 608                Z_ARG2, Z_tos /*object (see TT::aastore())*/);
 609   } else {
 610     if (message != NULL) {
 611       __ load_absolute_address(Z_ARG3, (address)message);
 612     } else {
 613       __ clear_reg(Z_ARG3, true, false);
 614     }
 615     __ call_VM(Z_tos,
 616                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 617                Z_ARG2, Z_ARG3);
 618   }
 619   // Throw exception.
 620   __ load_absolute_address(Z_R1_scratch, Interpreter::throw_exception_entry());
 621   __ z_br(Z_R1_scratch);
 622 
 623   BLOCK_COMMENT("} exception_handler_common");
 624 
 625   return entry;
 626 }
 627 
 628 address TemplateInterpreterGenerator::generate_return_entry_for (TosState state, int step, size_t index_size) {
 629   address entry = __ pc();
 630 
 631   BLOCK_COMMENT("return_entry {");
 632 
 633   // Pop i2c extension or revert top-2-parent-resize done by interpreted callees.
 634   Register sp_before_i2c_extension = Z_bcp;
 635   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 636   __ z_lg(sp_before_i2c_extension, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
 637   __ resize_frame_absolute(sp_before_i2c_extension, Z_locals/*tmp*/, true/*load_fp*/);
 638 
 639   // TODO(ZASM): necessary??
 640   //  // and NULL it as marker that esp is now tos until next java call
 641   //  __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 642 
 643   __ restore_bcp();
 644   __ restore_locals();
 645   __ restore_esp();
 646 
 647   if (state == atos) {
 648     __ profile_return_type(Z_tmp_1, Z_tos, Z_tmp_2);
 649   }
 650 
 651   Register cache  = Z_tmp_1;
 652   Register size   = Z_tmp_1;
 653   Register offset = Z_tmp_2;
 654   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
 655                                     ConstantPoolCacheEntry::flags_offset());
 656   __ get_cache_and_index_at_bcp(cache, offset, 1, index_size);
 657 
 658   // #args is in rightmost byte of the _flags field.
 659   __ z_llgc(size, Address(cache, offset, flags_offset+(sizeof(size_t)-1)));
 660   __ z_sllg(size, size, Interpreter::logStackElementSize); // Each argument size in bytes.
 661   __ z_agr(Z_esp, size);                                   // Pop arguments.
 662 
 663   __ check_and_handle_popframe(Z_thread);
 664   __ check_and_handle_earlyret(Z_thread);
 665 
 666   __ dispatch_next(state, step);
 667 
 668   BLOCK_COMMENT("} return_entry");
 669 
 670   return entry;
 671 }
 672 
 673 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 674                                                                int step,
 675                                                                address continuation) {
 676   address entry = __ pc();
 677 
 678   BLOCK_COMMENT("deopt_entry {");
 679 
 680   // TODO(ZASM): necessary? NULL last_sp until next java call
 681   // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 682   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 683   __ restore_bcp();
 684   __ restore_locals();
 685   __ restore_esp();
 686 
 687   // Handle exceptions.
 688   {
 689     Label L;
 690     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
 691     __ z_bre(L);
 692     __ call_VM(noreg,
 693                CAST_FROM_FN_PTR(address,
 694                                 InterpreterRuntime::throw_pending_exception));
 695     __ should_not_reach_here();
 696     __ bind(L);
 697   }
 698   if (continuation == NULL) {
 699     __ dispatch_next(state, step);
 700   } else {
 701     __ jump_to_entry(continuation, Z_R1_scratch);
 702   }
 703 
 704   BLOCK_COMMENT("} deopt_entry");
 705 
 706   return entry;
 707 }
 708 
 709 address TemplateInterpreterGenerator::generate_safept_entry_for (TosState state,
 710                                                                 address runtime_entry) {
 711   address entry = __ pc();
 712   __ push(state);
 713   __ call_VM(noreg, runtime_entry);
 714   __ dispatch_via(vtos, Interpreter::_normal_table.table_for (vtos));
 715   return entry;
 716 }
 717 
 718 //
 719 // Helpers for commoning out cases in the various type of method entries.
 720 //
 721 
 722 // Increment invocation count & check for overflow.
 723 //
 724 // Note: checking for negative value instead of overflow
 725 // so we have a 'sticky' overflow test.
 726 //
 727 // Z_ARG2: method (see generate_fixed_frame())
 728 //
 729 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 730   Label done;
 731   Register method = Z_ARG2; // Generate_fixed_frame() copies Z_method into Z_ARG2.
 732   Register m_counters = Z_ARG4;
 733 
 734   BLOCK_COMMENT("counter_incr {");
 735 
 736   // Note: In tiered we increment either counters in method or in MDO depending
 737   // if we are profiling or not.
 738   int increment = InvocationCounter::count_increment;
 739   if (ProfileInterpreter) {
 740     NearLabel no_mdo;
 741     Register mdo = m_counters;
 742     // Are we profiling?
 743     __ load_and_test_long(mdo, method2_(method, method_data));
 744     __ branch_optimized(Assembler::bcondZero, no_mdo);
 745     // Increment counter in the MDO.
 746     const Address mdo_invocation_counter(mdo, MethodData::invocation_counter_offset() +
 747                                          InvocationCounter::counter_offset());
 748     const Address mask(mdo, MethodData::invoke_mask_offset());
 749     __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 750                                Z_R1_scratch, false, Assembler::bcondZero,
 751                                overflow);
 752     __ z_bru(done);
 753     __ bind(no_mdo);
 754   }
 755 
 756   // Increment counter in MethodCounters.
 757   const Address invocation_counter(m_counters,
 758                                    MethodCounters::invocation_counter_offset() +
 759                                    InvocationCounter::counter_offset());
 760   // Get address of MethodCounters object.
 761   __ get_method_counters(method, m_counters, done);
 762   const Address mask(m_counters, MethodCounters::invoke_mask_offset());
 763   __ increment_mask_and_jump(invocation_counter,
 764                              increment, mask,
 765                              Z_R1_scratch, false, Assembler::bcondZero,
 766                              overflow);
 767 
 768   __ bind(done);
 769 
 770   BLOCK_COMMENT("} counter_incr");
 771 }
 772 
 773 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 774   // InterpreterRuntime::frequency_counter_overflow takes two
 775   // arguments, the first (thread) is passed by call_VM, the second
 776   // indicates if the counter overflow occurs at a backwards branch
 777   // (NULL bcp). We pass zero for it. The call returns the address
 778   // of the verified entry point for the method or NULL if the
 779   // compilation did not complete (either went background or bailed
 780   // out).
 781   __ clear_reg(Z_ARG2);
 782   __ call_VM(noreg,
 783              CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow),
 784              Z_ARG2);
 785   __ z_bru(do_continue);
 786 }
 787 
 788 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register frame_size, Register tmp1) {
 789   Register tmp2 = Z_R1_scratch;
 790   const int page_size = os::vm_page_size();
 791   NearLabel after_frame_check;
 792 
 793   BLOCK_COMMENT("stack_overflow_check {");
 794 
 795   assert_different_registers(frame_size, tmp1);
 796 
 797   // Stack banging is sufficient overflow check if frame_size < page_size.
 798   if (Immediate::is_uimm(page_size, 15)) {
 799     __ z_chi(frame_size, page_size);
 800     __ z_brl(after_frame_check);
 801   } else {
 802     __ load_const_optimized(tmp1, page_size);
 803     __ compareU32_and_branch(frame_size, tmp1, Assembler::bcondLow, after_frame_check);
 804   }
 805 
 806   // Get the stack base, and in debug, verify it is non-zero.
 807   __ z_lg(tmp1, thread_(stack_base));
 808 #ifdef ASSERT
 809   address reentry = NULL;
 810   NearLabel base_not_zero;
 811   __ compareU64_and_branch(tmp1, (intptr_t)0L, Assembler::bcondNotEqual, base_not_zero);
 812   reentry = __ stop_chain_static(reentry, "stack base is zero in generate_stack_overflow_check");
 813   __ bind(base_not_zero);
 814 #endif
 815 
 816   // Get the stack size, and in debug, verify it is non-zero.
 817   assert(sizeof(size_t) == sizeof(intptr_t), "wrong load size");
 818   __ z_lg(tmp2, thread_(stack_size));
 819 #ifdef ASSERT
 820   NearLabel size_not_zero;
 821   __ compareU64_and_branch(tmp2, (intptr_t)0L, Assembler::bcondNotEqual, size_not_zero);
 822   reentry = __ stop_chain_static(reentry, "stack size is zero in generate_stack_overflow_check");
 823   __ bind(size_not_zero);
 824 #endif
 825 
 826   // Compute the beginning of the protected zone minus the requested frame size.
 827   __ z_sgr(tmp1, tmp2);
 828   __ add2reg(tmp1, StackOverflow::stack_guard_zone_size());
 829 
 830   // Add in the size of the frame (which is the same as subtracting it from the
 831   // SP, which would take another register.
 832   __ z_agr(tmp1, frame_size);
 833 
 834   // The frame is greater than one page in size, so check against
 835   // the bottom of the stack.
 836   __ compareU64_and_branch(Z_SP, tmp1, Assembler::bcondHigh, after_frame_check);
 837 
 838   // The stack will overflow, throw an exception.
 839 
 840   // Restore SP to sender's sp. This is necessary if the sender's frame is an
 841   // extended compiled frame (see gen_c2i_adapter()) and safer anyway in case of
 842   // JSR292 adaptations.
 843   __ resize_frame_absolute(Z_R10, tmp1, true/*load_fp*/);
 844 
 845   // Note also that the restored frame is not necessarily interpreted.
 846   // Use the shared runtime version of the StackOverflowError.
 847   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 848   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 849   __ load_absolute_address(tmp1, StubRoutines::throw_StackOverflowError_entry());
 850   __ z_br(tmp1);
 851 
 852   // If you get to here, then there is enough stack space.
 853   __ bind(after_frame_check);
 854 
 855   BLOCK_COMMENT("} stack_overflow_check");
 856 }
 857 
 858 // Allocate monitor and lock method (asm interpreter).
 859 //
 860 // Args:
 861 //   Z_locals: locals
 862 
 863 void TemplateInterpreterGenerator::lock_method(void) {
 864 
 865   BLOCK_COMMENT("lock_method {");
 866 
 867   // Synchronize method.
 868   const Register method = Z_tmp_2;
 869   __ get_method(method);
 870 
 871 #ifdef ASSERT
 872   address reentry = NULL;
 873   {
 874     Label L;
 875     __ testbit(method2_(method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 876     __ z_btrue(L);
 877     reentry = __ stop_chain_static(reentry, "method doesn't need synchronization");
 878     __ bind(L);
 879   }
 880 #endif // ASSERT
 881 
 882   // Get synchronization object.
 883   const Register object = Z_tmp_2;
 884 
 885   {
 886     Label     done;
 887     Label     static_method;
 888 
 889     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 890     __ z_btrue(static_method);
 891 
 892     // non-static method: Load receiver obj from stack.
 893     __ mem2reg_opt(object, Address(Z_locals, Interpreter::local_offset_in_bytes(0)));
 894     __ z_bru(done);
 895 
 896     __ bind(static_method);
 897 
 898     // Lock the java mirror.
 899     // Load mirror from interpreter frame.
 900     __ z_lg(object, _z_ijava_state_neg(mirror), Z_fp);
 901 
 902 #ifdef ASSERT
 903     {
 904       NearLabel L;
 905       __ compare64_and_branch(object, (intptr_t) 0, Assembler::bcondNotEqual, L);
 906       reentry = __ stop_chain_static(reentry, "synchronization object is NULL");
 907       __ bind(L);
 908     }
 909 #endif // ASSERT
 910 
 911     __ bind(done);
 912   }
 913 
 914   __ add_monitor_to_stack(true, Z_ARG3, Z_ARG4, Z_ARG5); // Allocate monitor elem.
 915   // Store object and lock it.
 916   __ get_monitors(Z_tmp_1);
 917   __ reg2mem_opt(object, Address(Z_tmp_1, BasicObjectLock::obj_offset_in_bytes()));
 918   __ lock_object(Z_tmp_1, object);
 919 
 920   BLOCK_COMMENT("} lock_method");
 921 }
 922 
 923 // Generate a fixed interpreter frame. This is identical setup for
 924 // interpreted methods and for native methods hence the shared code.
 925 //
 926 // Registers alive
 927 //   Z_thread   - JavaThread*
 928 //   Z_SP       - old stack pointer
 929 //   Z_method   - callee's method
 930 //   Z_esp      - parameter list (slot 'above' last param)
 931 //   Z_R14      - return pc, to be stored in caller's frame
 932 //   Z_R10      - sender sp, note: Z_tmp_1 is Z_R10!
 933 //
 934 // Registers updated
 935 //   Z_SP       - new stack pointer
 936 //   Z_esp      - callee's operand stack pointer
 937 //                points to the slot above the value on top
 938 //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
 939 //   Z_bcp      - the bytecode pointer
 940 //   Z_fp       - the frame pointer, thereby killing Z_method
 941 //   Z_ARG2     - copy of Z_method
 942 //
 943 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 944 
 945   //  stack layout
 946   //
 947   //   F1 [TOP_IJAVA_FRAME_ABI]              <-- Z_SP, Z_R10 (see note below)
 948   //      [F1's operand stack (unused)]
 949   //      [F1's outgoing Java arguments]     <-- Z_esp
 950   //      [F1's operand stack (non args)]
 951   //      [monitors]      (optional)
 952   //      [IJAVA_STATE]
 953   //
 954   //   F2 [PARENT_IJAVA_FRAME_ABI]
 955   //      ...
 956   //
 957   //  0x000
 958   //
 959   // Note: Z_R10, the sender sp, will be below Z_SP if F1 was extended by a c2i adapter.
 960 
 961   //=============================================================================
 962   // Allocate space for locals other than the parameters, the
 963   // interpreter state, monitors, and the expression stack.
 964 
 965   const Register local_count  = Z_ARG5;
 966   const Register fp           = Z_tmp_2;
 967   const Register const_method = Z_ARG1;
 968 
 969   BLOCK_COMMENT("generate_fixed_frame {");
 970   {
 971   // local registers
 972   const Register top_frame_size  = Z_ARG2;
 973   const Register sp_after_resize = Z_ARG3;
 974   const Register max_stack       = Z_ARG4;
 975 
 976   __ z_lg(const_method, Address(Z_method, Method::const_offset()));
 977   __ z_llgh(max_stack, Address(const_method, ConstMethod::size_of_parameters_offset()));
 978   __ z_sllg(Z_locals /*parameter_count bytes*/, max_stack /*parameter_count*/, LogBytesPerWord);
 979 
 980   if (native_call) {
 981     // If we're calling a native method, we replace max_stack (which is
 982     // zero) with space for the worst-case signature handler varargs
 983     // vector, which is:
 984     //   max_stack = max(Argument::n_register_parameters, parameter_count+2);
 985     //
 986     // We add two slots to the parameter_count, one for the jni
 987     // environment and one for a possible native mirror. We allocate
 988     // space for at least the number of ABI registers, even though
 989     // InterpreterRuntime::slow_signature_handler won't write more than
 990     // parameter_count+2 words when it creates the varargs vector at the
 991     // top of the stack. The generated slow signature handler will just
 992     // load trash into registers beyond the necessary number. We're
 993     // still going to cut the stack back by the ABI register parameter
 994     // count so as to get SP+16 pointing at the ABI outgoing parameter
 995     // area, so we need to allocate at least that much even though we're
 996     // going to throw it away.
 997     //
 998     __ add2reg(max_stack, 2);
 999 
1000     NearLabel passing_args_on_stack;
1001 
1002     // max_stack in bytes
1003     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1004 
1005     int argument_registers_in_bytes = Argument::n_register_parameters << LogBytesPerWord;
1006     __ compare64_and_branch(max_stack, argument_registers_in_bytes, Assembler::bcondNotLow, passing_args_on_stack);
1007 
1008     __ load_const_optimized(max_stack, argument_registers_in_bytes);
1009 
1010     __ bind(passing_args_on_stack);
1011   } else {
1012     // !native_call
1013     // local_count = method->constMethod->max_locals();
1014     __ z_llgh(local_count, Address(const_method, ConstMethod::size_of_locals_offset()));
1015 
1016     // Calculate number of non-parameter locals (in slots):
1017     __ z_sgr(local_count, max_stack);
1018 
1019     // max_stack = method->max_stack();
1020     __ z_llgh(max_stack, Address(const_method, ConstMethod::max_stack_offset()));
1021     // max_stack in bytes
1022     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1023   }
1024 
1025   // Resize (i.e. normally shrink) the top frame F1 ...
1026   //   F1      [TOP_IJAVA_FRAME_ABI]          <-- Z_SP, Z_R10
1027   //           F1's operand stack (free)
1028   //           ...
1029   //           F1's operand stack (free)      <-- Z_esp
1030   //           F1's outgoing Java arg m
1031   //           ...
1032   //           F1's outgoing Java arg 0
1033   //           ...
1034   //
1035   //  ... into a parent frame (Z_R10 holds F1's SP before any modification, see also above)
1036   //
1037   //           +......................+
1038   //           :                      :        <-- Z_R10, saved below as F0's z_ijava_state.sender_sp
1039   //           :                      :
1040   //   F1      [PARENT_IJAVA_FRAME_ABI]        <-- Z_SP       \
1041   //           F0's non arg local                             | = delta
1042   //           ...                                            |
1043   //           F0's non arg local              <-- Z_esp      /
1044   //           F1's outgoing Java arg m
1045   //           ...
1046   //           F1's outgoing Java arg 0
1047   //           ...
1048   //
1049   // then push the new top frame F0.
1050   //
1051   //   F0      [TOP_IJAVA_FRAME_ABI]    = frame::z_top_ijava_frame_abi_size \
1052   //           [operand stack]          = max_stack                          | = top_frame_size
1053   //           [IJAVA_STATE]            = frame::z_ijava_state_size         /
1054 
1055   // sp_after_resize = Z_esp - delta
1056   //
1057   // delta = PARENT_IJAVA_FRAME_ABI + (locals_count - params_count)
1058 
1059   __ add2reg(sp_after_resize, (Interpreter::stackElementSize) - (frame::z_parent_ijava_frame_abi_size), Z_esp);
1060   if (!native_call) {
1061     __ z_sllg(Z_R0_scratch, local_count, LogBytesPerWord); // Params have already been subtracted from local_count.
1062     __ z_slgr(sp_after_resize, Z_R0_scratch);
1063   }
1064 
1065   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
1066   __ add2reg(top_frame_size,
1067              frame::z_top_ijava_frame_abi_size +
1068              frame::z_ijava_state_size,
1069              max_stack);
1070 
1071   if (!native_call) {
1072     // Stack overflow check.
1073     // Native calls don't need the stack size check since they have no
1074     // expression stack and the arguments are already on the stack and
1075     // we only add a handful of words to the stack.
1076     Register frame_size = max_stack; // Reuse the register for max_stack.
1077     __ z_lgr(frame_size, Z_SP);
1078     __ z_sgr(frame_size, sp_after_resize);
1079     __ z_agr(frame_size, top_frame_size);
1080     generate_stack_overflow_check(frame_size, fp/*tmp1*/);
1081   }
1082 
1083   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
1084   __ asm_assert_eq("killed Z_R14", 0);
1085   __ resize_frame_absolute(sp_after_resize, fp, true);
1086   __ save_return_pc(Z_R14);
1087 
1088   // ... and push the new frame F0.
1089   __ push_frame(top_frame_size, fp, true /*copy_sp*/, false);
1090   }
1091 
1092   //=============================================================================
1093   // Initialize the new frame F0: initialize interpreter state.
1094 
1095   {
1096   // locals
1097   const Register local_addr = Z_ARG4;
1098 
1099   BLOCK_COMMENT("generate_fixed_frame: initialize interpreter state {");
1100 
1101 #ifdef ASSERT
1102   // Set the magic number (using local_addr as tmp register).
1103   __ load_const_optimized(local_addr, frame::z_istate_magic_number);
1104   __ z_stg(local_addr, _z_ijava_state_neg(magic), fp);
1105 #endif
1106 
1107   // Save sender SP from F1 (i.e. before it was potentially modified by an
1108   // adapter) into F0's interpreter state. We use it as well to revert
1109   // resizing the frame above.
1110   __ z_stg(Z_R10, _z_ijava_state_neg(sender_sp), fp);
1111 
1112   // Load cp cache and save it at the end of this block.
1113   __ z_lg(Z_R1_scratch, Address(const_method, ConstMethod::constants_offset()));
1114   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstantPool::cache_offset_in_bytes()));
1115 
1116   // z_ijava_state->method = method;
1117   __ z_stg(Z_method, _z_ijava_state_neg(method), fp);
1118 
1119   // Point locals at the first argument. Method's locals are the
1120   // parameters on top of caller's expression stack.
1121   // Tos points past last Java argument.
1122 
1123   __ z_agr(Z_locals, Z_esp);
1124   // z_ijava_state->locals - i*BytesPerWord points to i-th Java local (i starts at 0)
1125   // z_ijava_state->locals = Z_esp + parameter_count bytes
1126   __ z_stg(Z_locals, _z_ijava_state_neg(locals), fp);
1127 
1128   // z_ijava_state->oop_temp = NULL;
1129   __ store_const(Address(fp, oop_tmp_offset), 0);
1130 
1131   // Initialize z_ijava_state->mdx.
1132   Register Rmdp = Z_bcp;
1133   // native_call: assert that mdo == NULL
1134   const bool check_for_mdo = !native_call DEBUG_ONLY(|| native_call);
1135   if (ProfileInterpreter && check_for_mdo) {
1136     Label get_continue;
1137 
1138     __ load_and_test_long(Rmdp, method_(method_data));
1139     __ z_brz(get_continue);
1140     DEBUG_ONLY(if (native_call) __ stop("native methods don't have a mdo"));
1141     __ add2reg(Rmdp, in_bytes(MethodData::data_offset()));
1142     __ bind(get_continue);
1143   }
1144   __ z_stg(Rmdp, _z_ijava_state_neg(mdx), fp);
1145 
1146   // Initialize z_ijava_state->bcp and Z_bcp.
1147   if (native_call) {
1148     __ clear_reg(Z_bcp); // Must initialize. Will get written into frame where GC reads it.
1149   } else {
1150     __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset()), const_method);
1151   }
1152   __ z_stg(Z_bcp, _z_ijava_state_neg(bcp), fp);
1153 
1154   // no monitors and empty operand stack
1155   // => z_ijava_state->monitors points to the top slot in IJAVA_STATE.
1156   // => Z_ijava_state->esp points one slot above into the operand stack.
1157   // z_ijava_state->monitors = fp - frame::z_ijava_state_size - Interpreter::stackElementSize;
1158   // z_ijava_state->esp = Z_esp = z_ijava_state->monitors;
1159   __ add2reg(Z_esp, -frame::z_ijava_state_size, fp);
1160   __ z_stg(Z_esp, _z_ijava_state_neg(monitors), fp);
1161   __ add2reg(Z_esp, -Interpreter::stackElementSize);
1162   __ z_stg(Z_esp, _z_ijava_state_neg(esp), fp);
1163 
1164   // z_ijava_state->cpoolCache = Z_R1_scratch (see load above);
1165   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(cpoolCache), fp);
1166 
1167   // Get mirror and store it in the frame as GC root for this Method*.
1168   __ load_mirror_from_const_method(Z_R1_scratch, const_method);
1169   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(mirror), fp);
1170 
1171   BLOCK_COMMENT("} generate_fixed_frame: initialize interpreter state");
1172 
1173   //=============================================================================
1174   if (!native_call) {
1175     // Local_count is already num_locals_slots - num_param_slots.
1176     // Start of locals: local_addr = Z_locals - locals size + 1 slot
1177     __ z_llgh(Z_R0_scratch, Address(const_method, ConstMethod::size_of_locals_offset()));
1178     __ add2reg(local_addr, BytesPerWord, Z_locals);
1179     __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogBytesPerWord);
1180     __ z_sgr(local_addr, Z_R0_scratch);
1181 
1182     __ Clear_Array(local_count, local_addr, Z_ARG2);
1183   }
1184 
1185   }
1186   // Finally set the frame pointer, destroying Z_method.
1187   assert(Z_fp == Z_method, "maybe set Z_fp earlier if other register than Z_method");
1188   // Oprofile analysis suggests to keep a copy in a register to be used by
1189   // generate_counter_incr().
1190   __ z_lgr(Z_ARG2, Z_method);
1191   __ z_lgr(Z_fp, fp);
1192 
1193   BLOCK_COMMENT("} generate_fixed_frame");
1194 }
1195 
1196 // Various method entries
1197 
1198 // Math function, frame manager must set up an interpreter state, etc.
1199 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1200 
1201   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1202   bool use_instruction = false;
1203   address runtime_entry = NULL;
1204   int num_args = 1;
1205   bool double_precision = true;
1206 
1207   // s390 specific:
1208   switch (kind) {
1209     case Interpreter::java_lang_math_sqrt:
1210     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1211     case Interpreter::java_lang_math_fmaF:
1212     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1213     default: break; // Fall back to runtime call.
1214   }
1215 
1216   switch (kind) {
1217     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1218     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1219     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1220     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1221     case Interpreter::java_lang_math_sqrt : /* runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); not available */ break;
1222     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1223     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1224     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1225     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1226     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1227     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1228     default: ShouldNotReachHere();
1229   }
1230 
1231   // Use normal entry if neither instruction nor runtime call is used.
1232   if (!use_instruction && runtime_entry == NULL) return NULL;
1233 
1234   address entry = __ pc();
1235 
1236   if (use_instruction) {
1237     switch (kind) {
1238       case Interpreter::java_lang_math_sqrt:
1239         // Can use memory operand directly.
1240         __ z_sqdb(Z_FRET, Interpreter::stackElementSize, Z_esp);
1241         break;
1242       case Interpreter::java_lang_math_abs:
1243         // Load operand from stack.
1244         __ mem2freg_opt(Z_FRET, Address(Z_esp, Interpreter::stackElementSize));
1245         __ z_lpdbr(Z_FRET);
1246         break;
1247       case Interpreter::java_lang_math_fmaF:
1248         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1249         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize)); // arg1
1250         __ z_maeb(Z_FRET, Z_FARG2, Address(Z_esp, 2 * Interpreter::stackElementSize));
1251         break;
1252       case Interpreter::java_lang_math_fmaD:
1253         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1254         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 5 * Interpreter::stackElementSize)); // arg1
1255         __ z_madb(Z_FRET, Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize));
1256         break;
1257       default: ShouldNotReachHere();
1258     }
1259   } else {
1260     // Load arguments
1261     assert(num_args <= 4, "passed in registers");
1262     if (double_precision) {
1263       int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1264       for (int i = 0; i < num_args; ++i) {
1265         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1266         offset -= 2 * Interpreter::stackElementSize;
1267       }
1268     } else {
1269       int offset = num_args * Interpreter::stackElementSize;
1270       for (int i = 0; i < num_args; ++i) {
1271         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1272         offset -= Interpreter::stackElementSize;
1273       }
1274     }
1275     // Call runtime
1276     __ save_return_pc();       // Save Z_R14.
1277     __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
1278 
1279     __ call_VM_leaf(runtime_entry);
1280 
1281     __ pop_frame();
1282     __ restore_return_pc();    // Restore Z_R14.
1283   }
1284 
1285   // Pop c2i arguments (if any) off when we return.
1286   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1287 
1288   __ z_br(Z_R14);
1289 
1290   return entry;
1291 }
1292 
1293 // Interpreter stub for calling a native method. (asm interpreter).
1294 // This sets up a somewhat different looking stack for calling the
1295 // native method than the typical interpreter frame setup.
1296 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1297   // Determine code generation flags.
1298   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1299 
1300   // Interpreter entry for ordinary Java methods.
1301   //
1302   // Registers alive
1303   //   Z_SP          - stack pointer
1304   //   Z_thread      - JavaThread*
1305   //   Z_method      - callee's method (method to be invoked)
1306   //   Z_esp         - operand (or expression) stack pointer of caller. one slot above last arg.
1307   //   Z_R10         - sender sp (before modifications, e.g. by c2i adapter
1308   //                   and as well by generate_fixed_frame below)
1309   //   Z_R14         - return address to caller (call_stub or c2i_adapter)
1310   //
1311   // Registers updated
1312   //   Z_SP          - stack pointer
1313   //   Z_fp          - callee's framepointer
1314   //   Z_esp         - callee's operand stack pointer
1315   //                   points to the slot above the value on top
1316   //   Z_locals      - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1317   //   Z_tos         - integer result, if any
1318   //   z_ftos        - floating point result, if any
1319   //
1320   // Stack layout at this point:
1321   //
1322   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1323   //                                                          frame was extended by c2i adapter)
1324   //           [outgoing Java arguments]     <-- Z_esp
1325   //           ...
1326   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1327   //           ...
1328   //
1329 
1330   address entry_point = __ pc();
1331 
1332   // Make sure registers are different!
1333   assert_different_registers(Z_thread, Z_method, Z_esp);
1334 
1335   BLOCK_COMMENT("native_entry {");
1336 
1337   // Make sure method is native and not abstract.
1338 #ifdef ASSERT
1339   address reentry = NULL;
1340   { Label L;
1341     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1342     __ z_btrue(L);
1343     reentry = __ stop_chain_static(reentry, "tried to execute non-native method as native");
1344     __ bind(L);
1345   }
1346   { Label L;
1347     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1348     __ z_bfalse(L);
1349     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1350     __ bind(L);
1351   }
1352 #endif // ASSERT
1353 
1354 #ifdef ASSERT
1355   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1356   __ save_return_pc(Z_R14);
1357 #endif
1358 
1359   // Generate the code to allocate the interpreter stack frame.
1360   generate_fixed_frame(true);
1361 
1362   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1363   // Since at this point in the method invocation the exception handler
1364   // would try to exit the monitor of synchronized methods which hasn't
1365   // been entered yet, we set the thread local variable
1366   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1367   // runtime, exception handling i.e. unlock_if_synchronized_method will
1368   // check this thread local flag.
1369   __ z_mvi(do_not_unlock_if_synchronized, true);
1370 
1371   // Increment invocation count and check for overflow.
1372   NearLabel invocation_counter_overflow;
1373   if (inc_counter) {
1374     generate_counter_incr(&invocation_counter_overflow);
1375   }
1376 
1377   Label continue_after_compile;
1378   __ bind(continue_after_compile);
1379 
1380   bang_stack_shadow_pages(true);
1381 
1382   // Reset the _do_not_unlock_if_synchronized flag.
1383   __ z_mvi(do_not_unlock_if_synchronized, false);
1384 
1385   // Check for synchronized methods.
1386   // This mst happen AFTER invocation_counter check and stack overflow check,
1387   // so method is not locked if overflows.
1388   if (synchronized) {
1389     lock_method();
1390   } else {
1391     // No synchronization necessary.
1392 #ifdef ASSERT
1393     { Label L;
1394       __ get_method(Z_R1_scratch);
1395       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1396       __ z_bfalse(L);
1397       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1398       __ bind(L);
1399     }
1400 #endif // ASSERT
1401   }
1402 
1403   // start execution
1404 
1405   // jvmti support
1406   __ notify_method_entry();
1407 
1408   //=============================================================================
1409   // Get and call the signature handler.
1410   const Register Rmethod                 = Z_tmp_2;
1411   const Register signature_handler_entry = Z_tmp_1;
1412   const Register Rresult_handler         = Z_tmp_3;
1413   Label call_signature_handler;
1414 
1415   assert_different_registers(Z_fp, Rmethod, signature_handler_entry, Rresult_handler);
1416   assert(Rresult_handler->is_nonvolatile(), "Rresult_handler must be in a non-volatile register");
1417 
1418   // Reload method.
1419   __ get_method(Rmethod);
1420 
1421   // Check for signature handler.
1422   __ load_and_test_long(signature_handler_entry, method2_(Rmethod, signature_handler));
1423   __ z_brne(call_signature_handler);
1424 
1425   // Method has never been called. Either generate a specialized
1426   // handler or point to the slow one.
1427   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call),
1428              Rmethod);
1429 
1430   // Reload method.
1431   __ get_method(Rmethod);
1432 
1433   // Reload signature handler, it must have been created/assigned in the meantime.
1434   __ z_lg(signature_handler_entry, method2_(Rmethod, signature_handler));
1435 
1436   __ bind(call_signature_handler);
1437 
1438   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1439   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1/*tmp*/);
1440 
1441   // Call signature handler and pass locals address in Z_ARG1.
1442   __ z_lgr(Z_ARG1, Z_locals);
1443   __ call_stub(signature_handler_entry);
1444   // Save result handler returned by signature handler.
1445   __ z_lgr(Rresult_handler, Z_RET);
1446 
1447   // Reload method (the slow signature handler may block for GC).
1448   __ get_method(Rmethod);
1449 
1450   // Pass mirror handle if static call.
1451   {
1452     Label method_is_not_static;
1453     __ testbit(method2_(Rmethod, access_flags), JVM_ACC_STATIC_BIT);
1454     __ z_bfalse(method_is_not_static);
1455     // Load mirror from interpreter frame.
1456     __ z_lg(Z_R1, _z_ijava_state_neg(mirror), Z_fp);
1457     // z_ijava_state.oop_temp = pool_holder->klass_part()->java_mirror();
1458     __ z_stg(Z_R1, oop_tmp_offset, Z_fp);
1459     // Pass handle to mirror as 2nd argument to JNI method.
1460     __ add2reg(Z_ARG2, oop_tmp_offset, Z_fp);
1461     __ bind(method_is_not_static);
1462   }
1463 
1464   // Pass JNIEnv address as first parameter.
1465   __ add2reg(Z_ARG1, in_bytes(JavaThread::jni_environment_offset()), Z_thread);
1466 
1467   // Note: last java frame has been set above already. The pc from there
1468   // is precise enough.
1469 
1470   // Get native function entry point before we change the thread state.
1471   __ z_lg(Z_R1/*native_method_entry*/, method2_(Rmethod, native_function));
1472 
1473   //=============================================================================
1474   // Transition from _thread_in_Java to _thread_in_native. As soon as
1475   // we make this change the safepoint code needs to be certain that
1476   // the last Java frame we established is good. The pc in that frame
1477   // just need to be near here not an actual return address.
1478 #ifdef ASSERT
1479   {
1480     NearLabel L;
1481     __ mem2reg_opt(Z_R14, Address(Z_thread, JavaThread::thread_state_offset()), false /*32 bits*/);
1482     __ compareU32_and_branch(Z_R14, _thread_in_Java, Assembler::bcondEqual, L);
1483     reentry = __ stop_chain_static(reentry, "Wrong thread state in native stub");
1484     __ bind(L);
1485   }
1486 #endif
1487 
1488   // Memory ordering: Z does not reorder store/load with subsequent load. That's strong enough.
1489   __ set_thread_state(_thread_in_native);
1490 
1491   //=============================================================================
1492   // Call the native method. Argument registers must not have been
1493   // overwritten since "__ call_stub(signature_handler);" (except for
1494   // ARG1 and ARG2 for static methods).
1495 
1496   __ call_c(Z_R1/*native_method_entry*/);
1497 
1498   // NOTE: frame::interpreter_frame_result() depends on these stores.
1499   __ z_stg(Z_RET, _z_ijava_state_neg(lresult), Z_fp);
1500   __ freg2mem_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1501   const Register Rlresult = signature_handler_entry;
1502   assert(Rlresult->is_nonvolatile(), "Rlresult must be in a non-volatile register");
1503   __ z_lgr(Rlresult, Z_RET);
1504 
1505   // Z_method may no longer be valid, because of GC.
1506 
1507   // Block, if necessary, before resuming in _thread_in_Java state.
1508   // In order for GC to work, don't clear the last_Java_sp until after
1509   // blocking.
1510 
1511   //=============================================================================
1512   // Switch thread to "native transition" state before reading the
1513   // synchronization state. This additional state is necessary
1514   // because reading and testing the synchronization state is not
1515   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1516   // in _thread_in_native state, loads _not_synchronized and is
1517   // preempted. VM thread changes sync state to synchronizing and
1518   // suspends threads for GC. Thread A is resumed to finish this
1519   // native method, but doesn't block here since it didn't see any
1520   // synchronization is progress, and escapes.
1521 
1522   __ set_thread_state(_thread_in_native_trans);
1523   __ z_fence();
1524 
1525   // Now before we return to java we must look for a current safepoint
1526   // (a new safepoint can not start since we entered native_trans).
1527   // We must check here because a current safepoint could be modifying
1528   // the callers registers right this moment.
1529 
1530   // Check for safepoint operation in progress and/or pending suspend requests.
1531   {
1532     Label Continue, do_safepoint;
1533     __ safepoint_poll(do_safepoint, Z_R1);
1534     // Check for suspend.
1535     __ load_and_test_int(Z_R0/*suspend_flags*/, thread_(suspend_flags));
1536     __ z_bre(Continue); // 0 -> no flag set -> not suspended
1537     __ bind(do_safepoint);
1538     __ z_lgr(Z_ARG1, Z_thread);
1539     __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1540     __ bind(Continue);
1541   }
1542 
1543   //=============================================================================
1544   // Back in Interpreter Frame.
1545 
1546   // We are in thread_in_native_trans here and back in the normal
1547   // interpreter frame. We don't have to do anything special about
1548   // safepoints and we can switch to Java mode anytime we are ready.
1549 
1550   // Note: frame::interpreter_frame_result has a dependency on how the
1551   // method result is saved across the call to post_method_exit. For
1552   // native methods it assumes that the non-FPU/non-void result is
1553   // saved in z_ijava_state.lresult and a FPU result in z_ijava_state.fresult. If
1554   // this changes then the interpreter_frame_result implementation
1555   // will need to be updated too.
1556 
1557   //=============================================================================
1558   // Back in Java.
1559 
1560   // Memory ordering: Z does not reorder store/load with subsequent
1561   // load. That's strong enough.
1562   __ set_thread_state(_thread_in_Java);
1563 
1564   __ reset_last_Java_frame();
1565 
1566   // We reset the JNI handle block only after unboxing the result; see below.
1567 
1568   // The method register is junk from after the thread_in_native transition
1569   // until here. Also can't call_VM until the bcp has been
1570   // restored. Need bcp for throwing exception below so get it now.
1571   __ get_method(Rmethod);
1572 
1573   // Restore Z_bcp to have legal interpreter frame,
1574   // i.e., bci == 0 <=> Z_bcp == code_base().
1575   __ z_lg(Z_bcp, Address(Rmethod, Method::const_offset())); // get constMethod
1576   __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1577 
1578   if (CheckJNICalls) {
1579     // clear_pending_jni_exception_check
1580     __ clear_mem(Address(Z_thread, JavaThread::pending_jni_exception_check_fn_offset()), sizeof(oop));
1581   }
1582 
1583   // Check if the native method returns an oop, and if so, move it
1584   // from the jni handle to z_ijava_state.oop_temp. This is
1585   // necessary, because we reset the jni handle block below.
1586   // NOTE: frame::interpreter_frame_result() depends on this, too.
1587   { NearLabel no_oop_result;
1588   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1589   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_result);
1590   __ resolve_jobject(Rlresult, /* tmp1 */ Rmethod, /* tmp2 */ Z_R1);
1591   __ z_stg(Rlresult, oop_tmp_offset, Z_fp);
1592   __ bind(no_oop_result);
1593   }
1594 
1595   // Reset handle block.
1596   __ z_lg(Z_R1/*active_handles*/, thread_(active_handles));
1597   __ clear_mem(Address(Z_R1, JNIHandleBlock::top_offset_in_bytes()), 4);
1598 
1599   // Handle exceptions (exception handling will handle unlocking!).
1600   {
1601     Label L;
1602     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
1603     __ z_bre(L);
1604     __ MacroAssembler::call_VM(noreg,
1605                                CAST_FROM_FN_PTR(address,
1606                                InterpreterRuntime::throw_pending_exception));
1607     __ should_not_reach_here();
1608     __ bind(L);
1609   }
1610 
1611   if (synchronized) {
1612     Register Rfirst_monitor = Z_ARG2;
1613     __ add2reg(Rfirst_monitor, -(frame::z_ijava_state_size + (int)sizeof(BasicObjectLock)), Z_fp);
1614 #ifdef ASSERT
1615     NearLabel ok;
1616     __ z_lg(Z_R1, _z_ijava_state_neg(monitors), Z_fp);
1617     __ compareU64_and_branch(Rfirst_monitor, Z_R1, Assembler::bcondEqual, ok);
1618     reentry = __ stop_chain_static(reentry, "native_entry:unlock: inconsistent z_ijava_state.monitors");
1619     __ bind(ok);
1620 #endif
1621     __ unlock_object(Rfirst_monitor);
1622   }
1623 
1624   // JVMTI support. Result has already been saved above to the frame.
1625   __ notify_method_exit(true/*native_method*/, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1626 
1627   // Move native method result back into proper registers and return.
1628   __ mem2freg_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1629   __ mem2reg_opt(Z_RET, Address(Z_fp, _z_ijava_state_neg(lresult)));
1630   __ call_stub(Rresult_handler);
1631 
1632   // Pop the native method's interpreter frame.
1633   __ pop_interpreter_frame(Z_R14 /*return_pc*/, Z_ARG2/*tmp1*/, Z_ARG3/*tmp2*/);
1634 
1635   // Return to caller.
1636   __ z_br(Z_R14);
1637 
1638   if (inc_counter) {
1639     // Handle overflow of counter and compile method.
1640     __ bind(invocation_counter_overflow);
1641     generate_counter_overflow(continue_after_compile);
1642   }
1643 
1644   BLOCK_COMMENT("} native_entry");
1645 
1646   return entry_point;
1647 }
1648 
1649 //
1650 // Generic interpreted method entry to template interpreter.
1651 //
1652 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1653   address entry_point = __ pc();
1654 
1655   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1656 
1657   // Interpreter entry for ordinary Java methods.
1658   //
1659   // Registers alive
1660   //   Z_SP       - stack pointer
1661   //   Z_thread   - JavaThread*
1662   //   Z_method   - callee's method (method to be invoked)
1663   //   Z_esp      - operand (or expression) stack pointer of caller. one slot above last arg.
1664   //   Z_R10      - sender sp (before modifications, e.g. by c2i adapter
1665   //                           and as well by generate_fixed_frame below)
1666   //   Z_R14      - return address to caller (call_stub or c2i_adapter)
1667   //
1668   // Registers updated
1669   //   Z_SP       - stack pointer
1670   //   Z_fp       - callee's framepointer
1671   //   Z_esp      - callee's operand stack pointer
1672   //                points to the slot above the value on top
1673   //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1674   //   Z_tos      - integer result, if any
1675   //   z_ftos     - floating point result, if any
1676   //
1677   //
1678   // stack layout at this point:
1679   //
1680   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1681   //                                                          frame was extended by c2i adapter)
1682   //           [outgoing Java arguments]     <-- Z_esp
1683   //           ...
1684   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1685   //           ...
1686   //
1687   // stack layout before dispatching the first bytecode:
1688   //
1689   //   F0      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP
1690   //           [operand stack]               <-- Z_esp
1691   //           monitor (optional, can grow)
1692   //           [IJAVA_STATE]
1693   //   F1      [PARENT_IJAVA_FRAME_ABI]      <-- Z_fp (== *Z_SP)
1694   //           [F0's locals]                 <-- Z_locals
1695   //           [F1's operand stack]
1696   //           [F1's monitors] (optional)
1697   //           [IJAVA_STATE]
1698 
1699   // Make sure registers are different!
1700   assert_different_registers(Z_thread, Z_method, Z_esp);
1701 
1702   BLOCK_COMMENT("normal_entry {");
1703 
1704   // Make sure method is not native and not abstract.
1705   // Rethink these assertions - they can be simplified and shared.
1706 #ifdef ASSERT
1707   address reentry = NULL;
1708   { Label L;
1709     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1710     __ z_bfalse(L);
1711     reentry = __ stop_chain_static(reentry, "tried to execute native method as non-native");
1712     __ bind(L);
1713   }
1714   { Label L;
1715     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1716     __ z_bfalse(L);
1717     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1718     __ bind(L);
1719   }
1720 #endif // ASSERT
1721 
1722 #ifdef ASSERT
1723   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1724   __ save_return_pc(Z_R14);
1725 #endif
1726 
1727   // Generate the code to allocate the interpreter stack frame.
1728   generate_fixed_frame(false);
1729 
1730   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1731   // Since at this point in the method invocation the exception handler
1732   // would try to exit the monitor of synchronized methods which hasn't
1733   // been entered yet, we set the thread local variable
1734   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1735   // runtime, exception handling i.e. unlock_if_synchronized_method will
1736   // check this thread local flag.
1737   __ z_mvi(do_not_unlock_if_synchronized, true);
1738 
1739   __ profile_parameters_type(Z_tmp_2, Z_ARG3, Z_ARG4);
1740 
1741   // Increment invocation counter and check for overflow.
1742   //
1743   // Note: checking for negative value instead of overflow so we have a 'sticky'
1744   // overflow test (may be of importance as soon as we have true MT/MP).
1745   NearLabel invocation_counter_overflow;
1746   NearLabel Lcontinue;
1747   if (inc_counter) {
1748     generate_counter_incr(&invocation_counter_overflow);
1749   }
1750   __ bind(Lcontinue);
1751 
1752   bang_stack_shadow_pages(false);
1753 
1754   // Reset the _do_not_unlock_if_synchronized flag.
1755   __ z_mvi(do_not_unlock_if_synchronized, false);
1756 
1757   // Check for synchronized methods.
1758   // Must happen AFTER invocation_counter check and stack overflow check,
1759   // so method is not locked if overflows.
1760   if (synchronized) {
1761     // Allocate monitor and lock method.
1762     lock_method();
1763   } else {
1764 #ifdef ASSERT
1765     { Label L;
1766       __ get_method(Z_R1_scratch);
1767       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1768       __ z_bfalse(L);
1769       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1770       __ bind(L);
1771     }
1772 #endif // ASSERT
1773   }
1774 
1775   // start execution
1776 
1777 #ifdef ASSERT
1778   __ verify_esp(Z_esp, Z_R1_scratch);
1779 
1780   __ verify_thread();
1781 #endif
1782 
1783   // jvmti support
1784   __ notify_method_entry();
1785 
1786   // Start executing instructions.
1787   __ dispatch_next(vtos);
1788   // Dispatch_next does not return.
1789   DEBUG_ONLY(__ should_not_reach_here());
1790 
1791   // Invocation counter overflow.
1792   if (inc_counter) {
1793     // Handle invocation counter overflow.
1794     __ bind(invocation_counter_overflow);
1795     generate_counter_overflow(Lcontinue);
1796   }
1797 
1798   BLOCK_COMMENT("} normal_entry");
1799 
1800   return entry_point;
1801 }
1802 
1803 
1804 /**
1805  * Method entry for static native methods:
1806  *   int java.util.zip.CRC32.update(int crc, int b)
1807  */
1808 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1809 
1810   if (UseCRC32Intrinsics) {
1811     uint64_t entry_off = __ offset();
1812     Label    slow_path;
1813 
1814     // If we need a safepoint check, generate full interpreter entry.
1815     __ safepoint_poll(slow_path, Z_R1);
1816 
1817     BLOCK_COMMENT("CRC32_update {");
1818 
1819     // We don't generate local frame and don't align stack because
1820     // we not even call stub code (we generate the code inline)
1821     // and there is no safepoint on this path.
1822 
1823     // Load java parameters.
1824     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1825     const Register argP    = Z_esp;
1826     const Register crc     = Z_ARG1;  // crc value
1827     const Register data    = Z_ARG2;  // address of java byte value (kernel_crc32 needs address)
1828     const Register dataLen = Z_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1829     const Register table   = Z_ARG4;  // address of crc32 table
1830 
1831     // Arguments are reversed on java expression stack.
1832     __ z_la(data, 3+1*wordSize, argP);  // byte value (stack address).
1833                                         // Being passed as an int, the single byte is at offset +3.
1834     __ z_llgf(crc, 2 * wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1835 
1836     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1837     __ kernel_crc32_singleByte(crc, data, dataLen, table, Z_R1, true);
1838 
1839     // Restore caller sp for c2i case.
1840     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1841 
1842     __ z_br(Z_R14);
1843 
1844     BLOCK_COMMENT("} CRC32_update");
1845 
1846     // Use a previously generated vanilla native entry as the slow path.
1847     BIND(slow_path);
1848     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1849     return __ addr_at(entry_off);
1850   }
1851 
1852   return NULL;
1853 }
1854 
1855 
1856 /**
1857  * Method entry for static native methods:
1858  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1859  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1860  */
1861 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1862 
1863   if (UseCRC32Intrinsics) {
1864     uint64_t entry_off = __ offset();
1865     Label    slow_path;
1866 
1867     // If we need a safepoint check, generate full interpreter entry.
1868     __ safepoint_poll(slow_path, Z_R1);
1869 
1870     // We don't generate local frame and don't align stack because
1871     // we call stub code and there is no safepoint on this path.
1872 
1873     // Load parameters.
1874     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1875     const Register argP    = Z_esp;
1876     const Register crc     = Z_ARG1;  // crc value
1877     const Register data    = Z_ARG2;  // address of java byte array
1878     const Register dataLen = Z_ARG3;  // source data len
1879     const Register table   = Z_ARG4;  // address of crc32 table
1880     const Register t0      = Z_R10;   // work reg for kernel* emitters
1881     const Register t1      = Z_R11;   // work reg for kernel* emitters
1882     const Register t2      = Z_R12;   // work reg for kernel* emitters
1883     const Register t3      = Z_R13;   // work reg for kernel* emitters
1884 
1885     // Arguments are reversed on java expression stack.
1886     // Calculate address of start element.
1887     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1888       // crc     @ (SP + 5W) (32bit)
1889       // buf     @ (SP + 3W) (64bit ptr to long array)
1890       // off     @ (SP + 2W) (32bit)
1891       // dataLen @ (SP + 1W) (32bit)
1892       // data = buf + off
1893       BLOCK_COMMENT("CRC32_updateByteBuffer {");
1894       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
1895       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
1896       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
1897       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
1898     } else {                                                         // Used for "updateBytes update".
1899       // crc     @ (SP + 4W) (32bit)
1900       // buf     @ (SP + 3W) (64bit ptr to byte array)
1901       // off     @ (SP + 2W) (32bit)
1902       // dataLen @ (SP + 1W) (32bit)
1903       // data = buf + off + base_offset
1904       BLOCK_COMMENT("CRC32_updateBytes {");
1905       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
1906       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
1907       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
1908       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
1909       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1910     }
1911 
1912     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1913 
1914     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
1915     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
1916     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, true);
1917     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
1918 
1919     // Restore caller sp for c2i case.
1920     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1921 
1922     __ z_br(Z_R14);
1923 
1924     BLOCK_COMMENT("} CRC32_update{Bytes|ByteBuffer}");
1925 
1926     // Use a previously generated vanilla native entry as the slow path.
1927     BIND(slow_path);
1928     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1929     return __ addr_at(entry_off);
1930   }
1931 
1932   return NULL;
1933 }
1934 
1935 
1936 /**
1937  * Method entry for intrinsic-candidate (non-native) methods:
1938  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
1939  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1940  * Unlike CRC32, CRC32C does not have any methods marked as native
1941  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1942  */
1943 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1944 
1945   if (UseCRC32CIntrinsics) {
1946     uint64_t entry_off = __ offset();
1947 
1948     // We don't generate local frame and don't align stack because
1949     // we call stub code and there is no safepoint on this path.
1950 
1951     // Load parameters.
1952     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1953     const Register argP    = Z_esp;
1954     const Register crc     = Z_ARG1;  // crc value
1955     const Register data    = Z_ARG2;  // address of java byte array
1956     const Register dataLen = Z_ARG3;  // source data len
1957     const Register table   = Z_ARG4;  // address of crc32 table
1958     const Register t0      = Z_R10;   // work reg for kernel* emitters
1959     const Register t1      = Z_R11;   // work reg for kernel* emitters
1960     const Register t2      = Z_R12;   // work reg for kernel* emitters
1961     const Register t3      = Z_R13;   // work reg for kernel* emitters
1962 
1963     // Arguments are reversed on java expression stack.
1964     // Calculate address of start element.
1965     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateByteBuffer direct".
1966       // crc     @ (SP + 5W) (32bit)
1967       // buf     @ (SP + 3W) (64bit ptr to long array)
1968       // off     @ (SP + 2W) (32bit)
1969       // dataLen @ (SP + 1W) (32bit)
1970       // data = buf + off
1971       BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1972       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
1973       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
1974       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
1975       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
1976       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
1977     } else {                                                                // Used for "updateBytes update".
1978       // crc     @ (SP + 4W) (32bit)
1979       // buf     @ (SP + 3W) (64bit ptr to byte array)
1980       // off     @ (SP + 2W) (32bit)
1981       // dataLen @ (SP + 1W) (32bit)
1982       // data = buf + off + base_offset
1983       BLOCK_COMMENT("CRC32C_updateBytes {");
1984       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
1985       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
1986       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
1987       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
1988       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
1989       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1990     }
1991 
1992     StubRoutines::zarch::generate_load_crc32c_table_addr(_masm, table);
1993 
1994     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
1995     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
1996     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, false);
1997     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
1998 
1999     // Restore caller sp for c2i case.
2000     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2001 
2002     __ z_br(Z_R14);
2003 
2004     BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
2005     return __ addr_at(entry_off);
2006   }
2007 
2008   return NULL;
2009 }
2010 
2011 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
2012   // Quick & dirty stack overflow checking: bang the stack & handle trap.
2013   // Note that we do the banging after the frame is setup, since the exception
2014   // handling code expects to find a valid interpreter frame on the stack.
2015   // Doing the banging earlier fails if the caller frame is not an interpreter
2016   // frame.
2017   // (Also, the exception throwing code expects to unlock any synchronized
2018   // method receiver, so do the banging after locking the receiver.)
2019 
2020   // Bang each page in the shadow zone. We can't assume it's been done for
2021   // an interpreter frame with greater than a page of locals, so each page
2022   // needs to be checked. Only true for non-native. For native, we only bang the last page.
2023   const int page_size      = os::vm_page_size();
2024   const int n_shadow_pages = (int)(StackOverflow::stack_shadow_zone_size()/page_size);
2025   const int start_page_num = native_call ? n_shadow_pages : 1;
2026   for (int pages = start_page_num; pages <= n_shadow_pages; pages++) {
2027     __ bang_stack_with_offset(pages*page_size);
2028   }
2029 }
2030 
2031 //-----------------------------------------------------------------------------
2032 // Exceptions
2033 
2034 void TemplateInterpreterGenerator::generate_throw_exception() {
2035 
2036   BLOCK_COMMENT("throw_exception {");
2037 
2038   // Entry point in previous activation (i.e., if the caller was interpreted).
2039   Interpreter::_rethrow_exception_entry = __ pc();
2040   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Frame accessors use Z_fp.
2041   // Z_ARG1 (==Z_tos): exception
2042   // Z_ARG2          : Return address/pc that threw exception.
2043   __ restore_bcp();    // R13 points to call/send.
2044   __ restore_locals();
2045 
2046   // Fallthrough, no need to restore Z_esp.
2047 
2048   // Entry point for exceptions thrown within interpreter code.
2049   Interpreter::_throw_exception_entry = __ pc();
2050   // Expression stack is undefined here.
2051   // Z_ARG1 (==Z_tos): exception
2052   // Z_bcp: exception bcp
2053   __ verify_oop(Z_ARG1);
2054   __ z_lgr(Z_ARG2, Z_ARG1);
2055 
2056   // Expression stack must be empty before entering the VM in case of
2057   // an exception.
2058   __ empty_expression_stack();
2059   // Find exception handler address and preserve exception oop.
2060   const Register Rpreserved_exc_oop = Z_tmp_1;
2061   __ call_VM(Rpreserved_exc_oop,
2062              CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception),
2063              Z_ARG2);
2064   // Z_RET: exception handler entry point
2065   // Z_bcp: bcp for exception handler
2066   __ push_ptr(Rpreserved_exc_oop); // Push exception which is now the only value on the stack.
2067   __ z_br(Z_RET); // Jump to exception handler (may be _remove_activation_entry!).
2068 
2069   // If the exception is not handled in the current frame the frame is
2070   // removed and the exception is rethrown (i.e. exception
2071   // continuation is _rethrow_exception).
2072   //
2073   // Note: At this point the bci is still the bci for the instruction
2074   // which caused the exception and the expression stack is
2075   // empty. Thus, for any VM calls at this point, GC will find a legal
2076   // oop map (with empty expression stack).
2077 
2078   //
2079   // JVMTI PopFrame support
2080   //
2081 
2082   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2083   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2084   __ empty_expression_stack();
2085   // Set the popframe_processing bit in pending_popframe_condition
2086   // indicating that we are currently handling popframe, so that
2087   // call_VMs that may happen later do not trigger new popframe
2088   // handling cycles.
2089   __ load_sized_value(Z_tmp_1, Address(Z_thread, JavaThread::popframe_condition_offset()), 4, false /*signed*/);
2090   __ z_oill(Z_tmp_1, JavaThread::popframe_processing_bit);
2091   __ z_sty(Z_tmp_1, thread_(popframe_condition));
2092 
2093   {
2094     // Check to see whether we are returning to a deoptimized frame.
2095     // (The PopFrame call ensures that the caller of the popped frame is
2096     // either interpreted or compiled and deoptimizes it if compiled.)
2097     // In this case, we can't call dispatch_next() after the frame is
2098     // popped, but instead must save the incoming arguments and restore
2099     // them after deoptimization has occurred.
2100     //
2101     // Note that we don't compare the return PC against the
2102     // deoptimization blob's unpack entry because of the presence of
2103     // adapter frames in C2.
2104     NearLabel caller_not_deoptimized;
2105     __ z_lg(Z_ARG1, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2106     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), Z_ARG1);
2107     __ compareU64_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, caller_not_deoptimized);
2108 
2109     // Compute size of arguments for saving when returning to
2110     // deoptimized caller.
2111     __ get_method(Z_ARG2);
2112     __ z_lg(Z_ARG2, Address(Z_ARG2, Method::const_offset()));
2113     __ z_llgh(Z_ARG2, Address(Z_ARG2, ConstMethod::size_of_parameters_offset()));
2114     __ z_sllg(Z_ARG2, Z_ARG2, Interpreter::logStackElementSize); // slots 2 bytes
2115     __ restore_locals();
2116     // Compute address of args to be saved.
2117     __ z_lgr(Z_ARG3, Z_locals);
2118     __ z_slgr(Z_ARG3, Z_ARG2);
2119     __ add2reg(Z_ARG3, wordSize);
2120     // Save these arguments.
2121     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args),
2122                     Z_thread, Z_ARG2, Z_ARG3);
2123 
2124     __ remove_activation(vtos, Z_R14,
2125                          /* throw_monitor_exception */ false,
2126                          /* install_monitor_exception */ false,
2127                          /* notify_jvmdi */ false);
2128 
2129     // Inform deoptimization that it is responsible for restoring
2130     // these arguments.
2131     __ store_const(thread_(popframe_condition),
2132                    JavaThread::popframe_force_deopt_reexecution_bit,
2133                    Z_tmp_1, false);
2134 
2135     // Continue in deoptimization handler.
2136     __ z_br(Z_R14);
2137 
2138     __ bind(caller_not_deoptimized);
2139   }
2140 
2141   // Clear the popframe condition flag.
2142   __ clear_mem(thread_(popframe_condition), sizeof(int));
2143 
2144   __ remove_activation(vtos,
2145                        noreg,  // Retaddr is not used.
2146                        false,  // throw_monitor_exception
2147                        false,  // install_monitor_exception
2148                        false); // notify_jvmdi
2149   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2150   __ restore_bcp();
2151   __ restore_locals();
2152   __ restore_esp();
2153   // The method data pointer was incremented already during
2154   // call profiling. We have to restore the mdp for the current bcp.
2155   if (ProfileInterpreter) {
2156     __ set_method_data_pointer_for_bcp();
2157   }
2158 #if INCLUDE_JVMTI
2159   {
2160     Label L_done;
2161 
2162     __ z_cli(0, Z_bcp, Bytecodes::_invokestatic);
2163     __ z_brc(Assembler::bcondNotEqual, L_done);
2164 
2165     // The member name argument must be restored if _invokestatic is
2166     // re-executed after a PopFrame call.  Detect such a case in the
2167     // InterpreterRuntime function and return the member name
2168     // argument, or NULL.
2169     __ z_lg(Z_ARG2, Address(Z_locals));
2170     __ get_method(Z_ARG3);
2171     __ call_VM(Z_tmp_1,
2172                CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
2173                Z_ARG2, Z_ARG3, Z_bcp);
2174 
2175     __ z_ltgr(Z_tmp_1, Z_tmp_1);
2176     __ z_brc(Assembler::bcondEqual, L_done);
2177 
2178     __ z_stg(Z_tmp_1, Address(Z_esp, wordSize));
2179     __ bind(L_done);
2180   }
2181 #endif // INCLUDE_JVMTI
2182   __ dispatch_next(vtos);
2183   // End of PopFrame support.
2184   Interpreter::_remove_activation_entry = __ pc();
2185 
2186   // In between activations - previous activation type unknown yet
2187   // compute continuation point - the continuation point expects the
2188   // following registers set up:
2189   //
2190   // Z_ARG1 (==Z_tos): exception
2191   // Z_ARG2          : return address/pc that threw exception
2192 
2193   Register return_pc = Z_tmp_1;
2194   Register handler   = Z_tmp_2;
2195    assert(return_pc->is_nonvolatile(), "use non-volatile reg. to preserve exception pc");
2196    assert(handler->is_nonvolatile(),   "use non-volatile reg. to handler pc");
2197   __ asm_assert_ijava_state_magic(return_pc/*tmp*/); // The top frame should be an interpreter frame.
2198   __ z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2199 
2200   // Moved removing the activation after VM call, because the new top
2201   // frame does not necessarily have the z_abi_160 required for a VM
2202   // call (e.g. if it is compiled).
2203 
2204   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2205                                          SharedRuntime::exception_handler_for_return_address),
2206                         Z_thread, return_pc);
2207   __ z_lgr(handler, Z_RET); // Save exception handler.
2208 
2209   // Preserve exception over this code sequence.
2210   __ pop_ptr(Z_ARG1);
2211   __ set_vm_result(Z_ARG1);
2212   // Remove the activation (without doing throws on illegalMonitorExceptions).
2213   __ remove_activation(vtos, noreg/*ret.pc already loaded*/, false/*throw exc*/, true/*install exc*/, false/*notify jvmti*/);
2214   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2215 
2216   __ get_vm_result(Z_ARG1);     // Restore exception.
2217   __ verify_oop(Z_ARG1);
2218   __ z_lgr(Z_ARG2, return_pc);  // Restore return address.
2219 
2220 #ifdef ASSERT
2221   // The return_pc in the new top frame is dead... at least that's my
2222   // current understanding. To assert this I overwrite it.
2223   // Note: for compiled frames the handler is the deopt blob
2224   // which writes Z_ARG2 into the return_pc slot.
2225   __ load_const_optimized(return_pc, 0xb00b1);
2226   __ z_stg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2227 #endif
2228 
2229   // Z_ARG1 (==Z_tos): exception
2230   // Z_ARG2          : return address/pc that threw exception
2231 
2232   // Note that an "issuing PC" is actually the next PC after the call.
2233   __ z_br(handler);         // Jump to exception handler of caller.
2234 
2235   BLOCK_COMMENT("} throw_exception");
2236 }
2237 
2238 //
2239 // JVMTI ForceEarlyReturn support
2240 //
2241 address TemplateInterpreterGenerator::generate_earlyret_entry_for (TosState state) {
2242   address entry = __ pc();
2243 
2244   BLOCK_COMMENT("earlyret_entry {");
2245 
2246   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2247   __ restore_bcp();
2248   __ restore_locals();
2249   __ restore_esp();
2250   __ empty_expression_stack();
2251   __ load_earlyret_value(state);
2252 
2253   Register RjvmtiState = Z_tmp_1;
2254   __ z_lg(RjvmtiState, thread_(jvmti_thread_state));
2255   __ store_const(Address(RjvmtiState, JvmtiThreadState::earlyret_state_offset()),
2256                  JvmtiThreadState::earlyret_inactive, 4, 4, Z_R0_scratch);
2257 
2258   if (state == itos) {
2259     // Narrow result if state is itos but result type is smaller.
2260     // Need to narrow in the return bytecode rather than in generate_return_entry
2261     // since compiled code callers expect the result to already be narrowed.
2262     __ narrow(Z_tos, Z_tmp_1); /* fall through */
2263   }
2264   __ remove_activation(state,
2265                        Z_tmp_1, // retaddr
2266                        false,   // throw_monitor_exception
2267                        false,   // install_monitor_exception
2268                        true);   // notify_jvmdi
2269   __ z_br(Z_tmp_1);
2270 
2271   BLOCK_COMMENT("} earlyret_entry");
2272 
2273   return entry;
2274 }
2275 
2276 //-----------------------------------------------------------------------------
2277 // Helper for vtos entry point generation.
2278 
2279 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2280                                                          address& bep,
2281                                                          address& cep,
2282                                                          address& sep,
2283                                                          address& aep,
2284                                                          address& iep,
2285                                                          address& lep,
2286                                                          address& fep,
2287                                                          address& dep,
2288                                                          address& vep) {
2289   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2290   Label L;
2291   aep = __ pc(); __ push_ptr(); __ z_bru(L);
2292   fep = __ pc(); __ push_f();   __ z_bru(L);
2293   dep = __ pc(); __ push_d();   __ z_bru(L);
2294   lep = __ pc(); __ push_l();   __ z_bru(L);
2295   bep = cep = sep =
2296   iep = __ pc(); __ push_i();
2297   vep = __ pc();
2298   __ bind(L);
2299   generate_and_dispatch(t);
2300 }
2301 
2302 //-----------------------------------------------------------------------------
2303 
2304 #ifndef PRODUCT
2305 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2306   address entry = __ pc();
2307   NearLabel counter_below_trace_threshold;
2308 
2309   if (TraceBytecodesAt > 0) {
2310     // Skip runtime call, if the trace threshold is not yet reached.
2311     __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2312     __ load_absolute_address(Z_tmp_2, (address)&TraceBytecodesAt);
2313     __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2314     __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2315     __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, counter_below_trace_threshold);
2316   }
2317 
2318   int offset2 = state == ltos || state == dtos ? 2 : 1;
2319 
2320   __ push(state);
2321   // Preserved return pointer is in Z_R14.
2322   // InterpreterRuntime::trace_bytecode() preserved and returns the value passed as second argument.
2323   __ z_lgr(Z_ARG2, Z_R14);
2324   __ z_lg(Z_ARG3, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)));
2325   if (WizardMode) {
2326     __ z_lgr(Z_ARG4, Z_esp); // Trace Z_esp in WizardMode.
2327   } else {
2328     __ z_lg(Z_ARG4, Address(Z_esp, Interpreter::expr_offset_in_bytes(offset2)));
2329   }
2330   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), Z_ARG2, Z_ARG3, Z_ARG4);
2331   __ z_lgr(Z_R14, Z_RET); // Estore return address (see above).
2332   __ pop(state);
2333 
2334   __ bind(counter_below_trace_threshold);
2335   __ z_br(Z_R14); // return
2336 
2337   return entry;
2338 }
2339 
2340 // Make feasible for old CPUs.
2341 void TemplateInterpreterGenerator::count_bytecode() {
2342   __ load_absolute_address(Z_R1_scratch, (address) &BytecodeCounter::_counter_value);
2343   __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2344 }
2345 
2346 void TemplateInterpreterGenerator::histogram_bytecode(Template * t) {
2347   __ load_absolute_address(Z_R1_scratch, (address)&BytecodeHistogram::_counters[ t->bytecode() ]);
2348   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2349 }
2350 
2351 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template * t) {
2352   Address  index_addr(Z_tmp_1, (intptr_t) 0);
2353   Register index = Z_tmp_2;
2354 
2355   // Load previous index.
2356   __ load_absolute_address(Z_tmp_1, (address) &BytecodePairHistogram::_index);
2357   __ mem2reg_opt(index, index_addr, false);
2358 
2359   // Mask with current bytecode and store as new previous index.
2360   __ z_srl(index, BytecodePairHistogram::log2_number_of_codes);
2361   __ load_const_optimized(Z_R0_scratch,
2362                           (int)t->bytecode() << BytecodePairHistogram::log2_number_of_codes);
2363   __ z_or(index, Z_R0_scratch);
2364   __ reg2mem_opt(index, index_addr, false);
2365 
2366   // Load counter array's address.
2367   __ z_lgfr(index, index);   // Sign extend for addressing.
2368   __ z_sllg(index, index, LogBytesPerInt);  // index2bytes
2369   __ load_absolute_address(Z_R1_scratch,
2370                            (address) &BytecodePairHistogram::_counters);
2371   // Add index and increment counter.
2372   __ z_agr(Z_R1_scratch, index);
2373   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2374 }
2375 
2376 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2377   // Call a little run-time stub to avoid blow-up for each bytecode.
2378   // The run-time runtime saves the right registers, depending on
2379   // the tosca in-state for the given template.
2380   address entry = Interpreter::trace_code(t->tos_in());
2381   guarantee(entry != NULL, "entry must have been generated");
2382   __ call_stub(entry);
2383 }
2384 
2385 void TemplateInterpreterGenerator::stop_interpreter_at() {
2386   NearLabel L;
2387 
2388   __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2389   __ load_absolute_address(Z_tmp_2, (address)&StopInterpreterAt);
2390   __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2391   __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2392   __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, L);
2393   assert(Z_tmp_1->is_nonvolatile(), "must be nonvolatile to preserve Z_tos");
2394   assert(Z_F8->is_nonvolatile(), "must be nonvolatile to preserve Z_ftos");
2395   __ z_lgr(Z_tmp_1, Z_tos);      // Save tos.
2396   __ z_lgr(Z_tmp_2, Z_bytecode); // Save Z_bytecode.
2397   __ z_ldr(Z_F8, Z_ftos);        // Save ftos.
2398   // Use -XX:StopInterpreterAt=<num> to set the limit
2399   // and break at breakpoint().
2400   __ call_VM(noreg, CAST_FROM_FN_PTR(address, breakpoint), false);
2401   __ z_lgr(Z_tos, Z_tmp_1);      // Restore tos.
2402   __ z_lgr(Z_bytecode, Z_tmp_2); // Save Z_bytecode.
2403   __ z_ldr(Z_ftos, Z_F8);        // Restore ftos.
2404   __ bind(L);
2405 }
2406 
2407 #endif // !PRODUCT