1 /*
   2  * Copyright (c) 2016, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016 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 "interpreter/abstractInterpreter.hpp"
  29 #include "interpreter/bytecodeHistogram.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateInterpreterGenerator.hpp"
  34 #include "interpreter/templateTable.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 
  49 
  50 // Size of interpreter code.  Increase if too small.  Interpreter will
  51 // fail with a guarantee ("not enough space for interpreter generation");
  52 // if too small.
  53 // Run with +PrintInterpreter to get the VM to print out the size.
  54 // Max size with JVMTI
  55 int TemplateInterpreter::InterpreterCodeSize = 320*K;
  56 
  57 #undef  __
  58 #ifdef PRODUCT
  59   #define __ _masm->
  60 #else
  61   #define __ _masm->
  62 //  #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
  63 #endif
  64 
  65 #define BLOCK_COMMENT(str) __ block_comment(str)
  66 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  67 
  68 #define oop_tmp_offset     _z_ijava_state_neg(oop_tmp)
  69 
  70 //-----------------------------------------------------------------------------
  71 
  72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  73   //
  74   // New slow_signature handler that respects the z/Architecture
  75   // C calling conventions.
  76   //
  77   // We get called by the native entry code with our output register
  78   // area == 8. First we call InterpreterRuntime::get_result_handler
  79   // to copy the pointer to the signature string temporarily to the
  80   // first C-argument and to return the result_handler in
  81   // Z_RET. Since native_entry will copy the jni-pointer to the
  82   // first C-argument slot later on, it's OK to occupy this slot
  83   // temporarily. Then we copy the argument list on the java
  84   // expression stack into native varargs format on the native stack
  85   // and load arguments into argument registers. Integer arguments in
  86   // the varargs vector will be sign-extended to 8 bytes.
  87   //
  88   // On entry:
  89   //   Z_ARG1  - intptr_t*       Address of java argument list in memory.
  90   //   Z_state - cppInterpreter* Address of interpreter state for
  91   //                               this method
  92   //   Z_method
  93   //
  94   // On exit (just before return instruction):
  95   //   Z_RET contains the address of the result_handler.
  96   //   Z_ARG2 is not updated for static methods and contains "this" otherwise.
  97   //   Z_ARG3-Z_ARG5 contain the first 3 arguments of types other than float and double.
  98   //   Z_FARG1-Z_FARG4 contain the first 4 arguments of type float or double.
  99 
 100   const int LogSizeOfCase = 3;
 101 
 102   const int max_fp_register_arguments   = Argument::n_float_register_parameters;
 103   const int max_int_register_arguments  = Argument::n_register_parameters - 2;  // First 2 are reserved.
 104 
 105   const Register arg_java       = Z_tmp_2;
 106   const Register arg_c          = Z_tmp_3;
 107   const Register signature      = Z_R1_scratch; // Is a string.
 108   const Register fpcnt          = Z_R0_scratch;
 109   const Register argcnt         = Z_tmp_4;
 110   const Register intSlot        = Z_tmp_1;
 111   const Register sig_end        = Z_tmp_1; // Assumed end of signature (only used in do_object).
 112   const Register target_sp      = Z_tmp_1;
 113   const FloatRegister floatSlot = Z_F1;
 114 
 115   const int d_signature         = _z_abi(gpr6); // Only spill space, register contents not affected.
 116   const int d_fpcnt             = _z_abi(gpr7); // Only spill space, register contents not affected.
 117 
 118   unsigned int entry_offset = __ offset();
 119 
 120   BLOCK_COMMENT("slow_signature_handler {");
 121 
 122   // We use target_sp for storing arguments in the C frame.
 123   __ save_return_pc();
 124 
 125   __ z_stmg(Z_R10,Z_R13,-32,Z_SP);
 126   __ push_frame_abi160(32);
 127 
 128   __ z_lgr(arg_java, Z_ARG1);
 129 
 130   Register   method = Z_ARG2; // Directly load into correct argument register.
 131 
 132   __ get_method(method);
 133   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), Z_thread, method);
 134 
 135   // Move signature to callee saved register.
 136   // Don't directly write to stack. Frame is used by VM call.
 137   __ z_lgr(Z_tmp_1, Z_RET);
 138 
 139   // Reload method. Register may have been altered by VM call.
 140   __ get_method(method);
 141 
 142   // Get address of result handler.
 143   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), Z_thread, method);
 144 
 145   // Save signature address to stack.
 146   __ z_stg(Z_tmp_1, d_signature, Z_SP);
 147 
 148   // Don't overwrite return value (Z_RET, Z_ARG1) in rest of the method !
 149 
 150   {
 151     Label   isStatic;
 152 
 153     // Test if static.
 154     // We can test the bit directly.
 155     // Path is Z_method->_access_flags._flags.
 156     // We only support flag bits in the least significant byte (assert !).
 157     // Therefore add 3 to address that byte within "_flags".
 158     // Reload method. VM call above may have destroyed register contents
 159     __ get_method(method);
 160     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 161     method = noreg;  // end of life
 162     __ z_btrue(isStatic);
 163 
 164     // For non-static functions, pass "this" in Z_ARG2 and copy it to 2nd C-arg slot.
 165     // Need to box the Java object here, so we use arg_java
 166     // (address of current Java stack slot) as argument and
 167     // don't dereference it as in case of ints, floats, etc..
 168     __ z_lgr(Z_ARG2, arg_java);
 169     __ add2reg(arg_java, -BytesPerWord);
 170     __ bind(isStatic);
 171   }
 172 
 173   // argcnt == 0 corresponds to 3rd C argument.
 174   //   arg #1 (result handler) and
 175   //   arg #2 (this, for non-statics), unused else
 176   // are reserved and pre-filled above.
 177   // arg_java points to the corresponding Java argument here. It
 178   // has been decremented by one argument (this) in case of non-static.
 179   __ clear_reg(argcnt, true, false);  // Don't set CC.
 180   __ z_lg(target_sp, 0, Z_SP);
 181   __ add2reg(arg_c, _z_abi(remaining_cargs), target_sp);
 182   // No floating-point args parsed so far.
 183   __ clear_mem(Address(Z_SP, d_fpcnt), 8);
 184 
 185   NearLabel   move_intSlot_to_ARG, move_floatSlot_to_FARG;
 186   NearLabel   loop_start, loop_start_restore, loop_end;
 187   NearLabel   do_int, do_long, do_float, do_double;
 188   NearLabel   do_dontreachhere, do_object, do_array, do_boxed;
 189 
 190 #ifdef ASSERT
 191   // Signature needs to point to '(' (== 0x28) at entry.
 192   __ z_lg(signature, d_signature, Z_SP);
 193   __ z_cli(0, signature, (int) '(');
 194   __ z_brne(do_dontreachhere);
 195 #endif
 196 
 197   __ bind(loop_start_restore);
 198   __ z_lg(signature, d_signature, Z_SP);  // Restore signature ptr, destroyed by move_XX_to_ARG.
 199 
 200   BIND(loop_start);
 201   // Advance to next argument type token from the signature.
 202   __ add2reg(signature, 1);
 203 
 204   // Use CLI, works well on all CPU versions.
 205     __ z_cli(0, signature, (int) ')');
 206     __ z_bre(loop_end);                // end of signature
 207     __ z_cli(0, signature, (int) 'L');
 208     __ z_bre(do_object);               // object     #9
 209     __ z_cli(0, signature, (int) 'F');
 210     __ z_bre(do_float);                // float      #7
 211     __ z_cli(0, signature, (int) 'J');
 212     __ z_bre(do_long);                 // long       #6
 213     __ z_cli(0, signature, (int) 'B');
 214     __ z_bre(do_int);                  // byte       #1
 215     __ z_cli(0, signature, (int) 'Z');
 216     __ z_bre(do_int);                  // boolean    #2
 217     __ z_cli(0, signature, (int) 'C');
 218     __ z_bre(do_int);                  // char       #3
 219     __ z_cli(0, signature, (int) 'S');
 220     __ z_bre(do_int);                  // short      #4
 221     __ z_cli(0, signature, (int) 'I');
 222     __ z_bre(do_int);                  // int        #5
 223     __ z_cli(0, signature, (int) 'D');
 224     __ z_bre(do_double);               // double     #8
 225     __ z_cli(0, signature, (int) '[');
 226     __ z_bre(do_array);                // array      #10
 227 
 228   __ bind(do_dontreachhere);
 229 
 230   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 231 
 232   // Array argument
 233   BIND(do_array);
 234 
 235   {
 236     Label   start_skip, end_skip;
 237 
 238     __ bind(start_skip);
 239 
 240     // Advance to next type tag from signature.
 241     __ add2reg(signature, 1);
 242 
 243     // Use CLI, works well on all CPU versions.
 244     __ z_cli(0, signature, (int) '[');
 245     __ z_bre(start_skip);               // Skip further brackets.
 246 
 247     __ z_cli(0, signature, (int) '9');
 248     __ z_brh(end_skip);                 // no optional size
 249 
 250     __ z_cli(0, signature, (int) '0');
 251     __ z_brnl(start_skip);              // Skip optional size.
 252 
 253     __ bind(end_skip);
 254 
 255     __ z_cli(0, signature, (int) 'L');
 256     __ z_brne(do_boxed);                // If not array of objects: go directly to do_boxed.
 257   }
 258 
 259   //  OOP argument
 260   BIND(do_object);
 261   // Pass by an object's type name.
 262   {
 263     Label   L;
 264 
 265     __ add2reg(sig_end, 4095, signature);     // Assume object type name is shorter than 4k.
 266     __ load_const_optimized(Z_R0, (int) ';'); // Type name terminator (must be in Z_R0!).
 267     __ MacroAssembler::search_string(sig_end, signature);
 268     __ z_brl(L);
 269     __ z_illtrap();  // No semicolon found: internal error or object name too long.
 270     __ bind(L);
 271     __ z_lgr(signature, sig_end);
 272     // fallthru to do_boxed
 273   }
 274 
 275   // Need to box the Java object here, so we use arg_java
 276   // (address of current Java stack slot) as argument and
 277   // don't dereference it as in case of ints, floats, etc..
 278 
 279   // UNBOX argument
 280   // Load reference and check for NULL.
 281   Label  do_int_Entry4Boxed;
 282   __ bind(do_boxed);
 283   {
 284     __ load_and_test_long(intSlot, Address(arg_java));
 285     __ z_bre(do_int_Entry4Boxed);
 286     __ z_lgr(intSlot, arg_java);
 287     __ z_bru(do_int_Entry4Boxed);
 288   }
 289 
 290   // INT argument
 291 
 292   // (also for byte, boolean, char, short)
 293   // Use lgf for load (sign-extend) and stg for store.
 294   BIND(do_int);
 295   __ z_lgf(intSlot, 0, arg_java);
 296 
 297   __ bind(do_int_Entry4Boxed);
 298   __ add2reg(arg_java, -BytesPerWord);
 299   // If argument fits into argument register, go and handle it, otherwise continue.
 300   __ compare32_and_branch(argcnt, max_int_register_arguments,
 301                           Assembler::bcondLow, move_intSlot_to_ARG);
 302   __ z_stg(intSlot, 0, arg_c);
 303   __ add2reg(arg_c, BytesPerWord);
 304   __ z_bru(loop_start);
 305 
 306   // LONG argument
 307 
 308   BIND(do_long);
 309   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 310   __ z_lg(intSlot, BytesPerWord, arg_java);
 311   // If argument fits into argument register, go and handle it, otherwise continue.
 312   __ compare32_and_branch(argcnt, max_int_register_arguments,
 313                           Assembler::bcondLow, move_intSlot_to_ARG);
 314   __ z_stg(intSlot, 0, arg_c);
 315   __ add2reg(arg_c, BytesPerWord);
 316   __ z_bru(loop_start);
 317 
 318   // FLOAT argumen
 319 
 320   BIND(do_float);
 321   __ z_le(floatSlot, 0, arg_java);
 322   __ add2reg(arg_java, -BytesPerWord);
 323   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 324   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 325   __ z_brl(move_floatSlot_to_FARG);
 326   __ z_ste(floatSlot, 4, arg_c);
 327   __ add2reg(arg_c, BytesPerWord);
 328   __ z_bru(loop_start);
 329 
 330   // DOUBLE argument
 331 
 332   BIND(do_double);
 333   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 334   __ z_ld(floatSlot, BytesPerWord, arg_java);
 335   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 336   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 337   __ z_brl(move_floatSlot_to_FARG);
 338   __ z_std(floatSlot, 0, arg_c);
 339   __ add2reg(arg_c, BytesPerWord);
 340   __ z_bru(loop_start);
 341 
 342   // Method exit, all arguments proocessed.
 343   __ bind(loop_end);
 344   __ pop_frame();
 345   __ restore_return_pc();
 346   __ z_lmg(Z_R10,Z_R13,-32,Z_SP);
 347   __ z_br(Z_R14);
 348 
 349   // Copy int arguments.
 350 
 351   Label  iarg_caselist;   // Distance between each case has to be a power of 2
 352                           // (= 1 << LogSizeOfCase).
 353   __ align(16);
 354   BIND(iarg_caselist);
 355   __ z_lgr(Z_ARG3, intSlot);    // 4 bytes
 356   __ z_bru(loop_start_restore); // 4 bytes
 357 
 358   __ z_lgr(Z_ARG4, intSlot);
 359   __ z_bru(loop_start_restore);
 360 
 361   __ z_lgr(Z_ARG5, intSlot);
 362   __ z_bru(loop_start_restore);
 363 
 364   __ align(16);
 365   __ bind(move_intSlot_to_ARG);
 366   __ z_stg(signature, d_signature, Z_SP);       // Spill since signature == Z_R1_scratch.
 367   __ z_larl(Z_R1_scratch, iarg_caselist);
 368   __ z_sllg(Z_R0_scratch, argcnt, LogSizeOfCase);
 369   __ add2reg(argcnt, 1);
 370   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 371   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 372 
 373   // Copy float arguments.
 374 
 375   Label  farg_caselist;   // Distance between each case has to be a power of 2
 376                           // (= 1 << logSizeOfCase, padded with nop.
 377   __ align(16);
 378   BIND(farg_caselist);
 379   __ z_ldr(Z_FARG1, floatSlot); // 2 bytes
 380   __ z_bru(loop_start_restore); // 4 bytes
 381   __ z_nop();                   // 2 bytes
 382 
 383   __ z_ldr(Z_FARG2, floatSlot);
 384   __ z_bru(loop_start_restore);
 385   __ z_nop();
 386 
 387   __ z_ldr(Z_FARG3, floatSlot);
 388   __ z_bru(loop_start_restore);
 389   __ z_nop();
 390 
 391   __ z_ldr(Z_FARG4, floatSlot);
 392   __ z_bru(loop_start_restore);
 393   __ z_nop();
 394 
 395   __ align(16);
 396   __ bind(move_floatSlot_to_FARG);
 397   __ z_stg(signature, d_signature, Z_SP);        // Spill since signature == Z_R1_scratch.
 398   __ z_lg(Z_R0_scratch, d_fpcnt, Z_SP);          // Need old value for indexing.
 399   __ add2mem_64(Address(Z_SP, d_fpcnt), 1, Z_R1_scratch); // Increment index.
 400   __ z_larl(Z_R1_scratch, farg_caselist);
 401   __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogSizeOfCase);
 402   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 403   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 404 
 405   BLOCK_COMMENT("} slow_signature_handler");
 406 
 407   return __ addr_at(entry_offset);
 408 }
 409 
 410 address TemplateInterpreterGenerator::generate_result_handler_for (BasicType type) {
 411   address entry = __ pc();
 412 
 413   assert(Z_tos == Z_RET, "Result handler: must move result!");
 414   assert(Z_ftos == Z_FRET, "Result handler: must move float result!");
 415 
 416   switch (type) {
 417     case T_BOOLEAN:
 418       __ c2bool(Z_tos);
 419       break;
 420     case T_CHAR:
 421       __ and_imm(Z_tos, 0xffff);
 422       break;
 423     case T_BYTE:
 424       __ z_lbr(Z_tos, Z_tos);
 425       break;
 426     case T_SHORT:
 427       __ z_lhr(Z_tos, Z_tos);
 428       break;
 429     case T_INT:
 430     case T_LONG:
 431     case T_VOID:
 432     case T_FLOAT:
 433     case T_DOUBLE:
 434       break;
 435     case T_OBJECT:
 436       // Retrieve result from frame...
 437       __ mem2reg_opt(Z_tos, Address(Z_fp, oop_tmp_offset));
 438       // and verify it.
 439       __ verify_oop(Z_tos);
 440       break;
 441     default:
 442       ShouldNotReachHere();
 443   }
 444   __ z_br(Z_R14);      // Return from result handler.
 445   return entry;
 446 }
 447 
 448 // Abstract method entry.
 449 // Attempt to execute abstract method. Throw exception.
 450 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 451   unsigned int entry_offset = __ offset();
 452 
 453   // Caller could be the call_stub or a compiled method (x86 version is wrong!).
 454 
 455   BLOCK_COMMENT("abstract_entry {");
 456 
 457   // Implement call of InterpreterRuntime::throw_AbstractMethodError.
 458   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1);
 459   __ save_return_pc();       // Save Z_R14.
 460   __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
 461 
 462   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), Z_thread);
 463 
 464   __ pop_frame();
 465   __ restore_return_pc();    // Restore Z_R14.
 466   __ reset_last_Java_frame();
 467 
 468   // Restore caller sp for c2i case.
 469   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 470 
 471   // branch to SharedRuntime::generate_forward_exception() which handles all possible callers,
 472   // i.e. call stub, compiled method, interpreted method.
 473   __ load_absolute_address(Z_tmp_1, StubRoutines::forward_exception_entry());
 474   __ z_br(Z_tmp_1);
 475 
 476   BLOCK_COMMENT("} abstract_entry");
 477 
 478   return __ addr_at(entry_offset);
 479 }
 480 
 481 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 482 #if INCLUDE_ALL_GCS
 483   if (UseG1GC) {
 484     // Inputs:
 485     //  Z_ARG1 - receiver
 486     //
 487     // What we do:
 488     //  - Load the referent field address.
 489     //  - Load the value in the referent field.
 490     //  - Pass that value to the pre-barrier.
 491     //
 492     // In the case of G1 this will record the value of the
 493     // referent in an SATB buffer if marking is active.
 494     // This will cause concurrent marking to mark the referent
 495     // field as live.
 496 
 497     Register  scratch1 = Z_tmp_2;
 498     Register  scratch2 = Z_tmp_3;
 499     Register  pre_val  = Z_RET;   // return value
 500     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
 501     Register  Rargp    = Z_esp;
 502 
 503     Label     slow_path;
 504     address   entry = __ pc();
 505 
 506     const int referent_offset = java_lang_ref_Reference::referent_offset;
 507     guarantee(referent_offset > 0, "referent offset not initialized");
 508 
 509     BLOCK_COMMENT("Reference_get {");
 510 
 511     //  If the receiver is null then it is OK to jump to the slow path.
 512     __ load_and_test_long(pre_val, Address(Rargp, Interpreter::stackElementSize)); // Get receiver.
 513     __ z_bre(slow_path);
 514 
 515     //  Load the value of the referent field.
 516     __ load_heap_oop(pre_val, referent_offset, pre_val);
 517 
 518     // Restore caller sp for c2i case.
 519     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 520 
 521     // Generate the G1 pre-barrier code to log the value of
 522     // the referent field in an SATB buffer.
 523     // Note:
 524     //   With these parameters the write_barrier_pre does not
 525     //   generate instructions to load the previous value.
 526     __ g1_write_barrier_pre(noreg,      // obj
 527                             noreg,      // offset
 528                             pre_val,    // pre_val
 529                             noreg,      // no new val to preserve
 530                             scratch1,   // tmp
 531                             scratch2,   // tmp
 532                             true);      // pre_val_needed
 533 
 534     __ z_br(Z_R14);
 535 
 536     // Branch to previously generated regular method entry.
 537     __ bind(slow_path);
 538 
 539     address meth_entry = Interpreter::entry_for_kind(Interpreter::zerolocals);
 540     __ jump_to_entry(meth_entry, Z_R1);
 541 
 542     BLOCK_COMMENT("} Reference_get");
 543 
 544     return entry;
 545   }
 546 #endif // INCLUDE_ALL_GCS
 547 
 548   return NULL;
 549 }
 550 
 551 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 552   address entry = __ pc();
 553 
 554   DEBUG_ONLY(__ verify_esp(Z_esp, Z_ARG5));
 555 
 556   // Restore bcp under the assumption that the current frame is still
 557   // interpreted.
 558   __ restore_bcp();
 559 
 560   // Expression stack must be empty before entering the VM if an
 561   // exception happened.
 562   __ empty_expression_stack();
 563   // Throw exception.
 564   __ call_VM(noreg,
 565              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
 566   return entry;
 567 }
 568 
 569 //
 570 // Args:
 571 //   Z_ARG3: aberrant index
 572 //
 573 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char * name) {
 574   address entry = __ pc();
 575   address excp = CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException);
 576 
 577   // Expression stack must be empty before entering the VM if an
 578   // exception happened.
 579   __ empty_expression_stack();
 580 
 581   // Setup parameters.
 582   // Leave out the name and use register for array to create more detailed exceptions.
 583   __ load_absolute_address(Z_ARG2, (address) name);
 584   __ call_VM(noreg, excp, Z_ARG2, Z_ARG3);
 585   return entry;
 586 }
 587 
 588 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 589   address entry = __ pc();
 590 
 591   // Object is at TOS.
 592   __ pop_ptr(Z_ARG2);
 593 
 594   // Expression stack must be empty before entering the VM if an
 595   // exception happened.
 596   __ empty_expression_stack();
 597 
 598   __ call_VM(Z_ARG1,
 599              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException),
 600              Z_ARG2);
 601 
 602   DEBUG_ONLY(__ should_not_reach_here();)
 603 
 604   return entry;
 605 }
 606 
 607 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 608   assert(!pass_oop || message == NULL, "either oop or message but not both");
 609   address entry = __ pc();
 610 
 611   BLOCK_COMMENT("exception_handler_common {");
 612 
 613   // Expression stack must be empty before entering the VM if an
 614   // exception happened.
 615   __ empty_expression_stack();
 616   if (name != NULL) {
 617     __ load_absolute_address(Z_ARG2, (address)name);
 618   } else {
 619     __ clear_reg(Z_ARG2, true, false);
 620   }
 621 
 622   if (pass_oop) {
 623     __ call_VM(Z_tos,
 624                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception),
 625                Z_ARG2, Z_tos /*object (see TT::aastore())*/);
 626   } else {
 627     if (message != NULL) {
 628       __ load_absolute_address(Z_ARG3, (address)message);
 629     } else {
 630       __ clear_reg(Z_ARG3, true, false);
 631     }
 632     __ call_VM(Z_tos,
 633                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 634                Z_ARG2, Z_ARG3);
 635   }
 636   // Throw exception.
 637   __ load_absolute_address(Z_R1_scratch, Interpreter::throw_exception_entry());
 638   __ z_br(Z_R1_scratch);
 639 
 640   BLOCK_COMMENT("} exception_handler_common");
 641 
 642   return entry;
 643 }
 644 
 645 address TemplateInterpreterGenerator::generate_return_entry_for (TosState state, int step, size_t index_size) {
 646   address entry = __ pc();
 647 
 648   BLOCK_COMMENT("return_entry {");
 649 
 650   // Pop i2c extension or revert top-2-parent-resize done by interpreted callees.
 651   Register sp_before_i2c_extension = Z_bcp;
 652   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 653   __ z_lg(sp_before_i2c_extension, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
 654   __ resize_frame_absolute(sp_before_i2c_extension, Z_locals/*tmp*/, true/*load_fp*/);
 655 
 656   // TODO(ZASM): necessary??
 657   //  // and NULL it as marker that esp is now tos until next java call
 658   //  __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 659 
 660   __ restore_bcp();
 661   __ restore_locals();
 662   __ restore_esp();
 663 
 664   if (state == atos) {
 665     __ profile_return_type(Z_tmp_1, Z_tos, Z_tmp_2);
 666   }
 667 
 668   Register cache  = Z_tmp_1;
 669   Register size   = Z_tmp_1;
 670   Register offset = Z_tmp_2;
 671   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
 672                                     ConstantPoolCacheEntry::flags_offset());
 673   __ get_cache_and_index_at_bcp(cache, offset, 1, index_size);
 674 
 675   // #args is in rightmost byte of the _flags field.
 676   __ z_llgc(size, Address(cache, offset, flags_offset+(sizeof(size_t)-1)));
 677   __ z_sllg(size, size, Interpreter::logStackElementSize); // Each argument size in bytes.
 678   __ z_agr(Z_esp, size);                                   // Pop arguments.
 679 
 680   __ check_and_handle_popframe(Z_thread);
 681   __ check_and_handle_earlyret(Z_thread);
 682 
 683   __ dispatch_next(state, step);
 684 
 685   BLOCK_COMMENT("} return_entry");
 686 
 687   return entry;
 688 }
 689 
 690 address TemplateInterpreterGenerator::generate_deopt_entry_for (TosState state,
 691                                                                int step) {
 692   address entry = __ pc();
 693 
 694   BLOCK_COMMENT("deopt_entry {");
 695 
 696   // TODO(ZASM): necessary? NULL last_sp until next java call
 697   // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 698   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 699   __ restore_bcp();
 700   __ restore_locals();
 701   __ restore_esp();
 702 
 703   // Handle exceptions.
 704   {
 705     Label L;
 706     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
 707     __ z_bre(L);
 708     __ call_VM(noreg,
 709                CAST_FROM_FN_PTR(address,
 710                                 InterpreterRuntime::throw_pending_exception));
 711     __ should_not_reach_here();
 712     __ bind(L);
 713   }
 714   __ dispatch_next(state, step);
 715 
 716   BLOCK_COMMENT("} deopt_entry");
 717 
 718   return entry;
 719 }
 720 
 721 address TemplateInterpreterGenerator::generate_safept_entry_for (TosState state,
 722                                                                 address runtime_entry) {
 723   address entry = __ pc();
 724   __ push(state);
 725   __ call_VM(noreg, runtime_entry);
 726   __ dispatch_via(vtos, Interpreter::_normal_table.table_for (vtos));
 727   return entry;
 728 }
 729 
 730 //
 731 // Helpers for commoning out cases in the various type of method entries.
 732 //
 733 
 734 // Increment invocation count & check for overflow.
 735 //
 736 // Note: checking for negative value instead of overflow
 737 // so we have a 'sticky' overflow test.
 738 //
 739 // Z_ARG2: method (see generate_fixed_frame())
 740 //
 741 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 742   Label done;
 743   Register method = Z_ARG2; // Generate_fixed_frame() copies Z_method into Z_ARG2.
 744   Register m_counters = Z_ARG4;
 745 
 746   BLOCK_COMMENT("counter_incr {");
 747 
 748   // Note: In tiered we increment either counters in method or in MDO depending
 749   // if we are profiling or not.
 750   if (TieredCompilation) {
 751     int increment = InvocationCounter::count_increment;
 752     if (ProfileInterpreter) {
 753       NearLabel no_mdo;
 754       Register mdo = m_counters;
 755       // Are we profiling?
 756       __ load_and_test_long(mdo, method2_(method, method_data));
 757       __ branch_optimized(Assembler::bcondZero, no_mdo);
 758       // Increment counter in the MDO.
 759       const Address mdo_invocation_counter(mdo, MethodData::invocation_counter_offset() +
 760                                            InvocationCounter::counter_offset());
 761       const Address mask(mdo, MethodData::invoke_mask_offset());
 762       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 763                                  Z_R1_scratch, false, Assembler::bcondZero,
 764                                  overflow);
 765       __ z_bru(done);
 766       __ bind(no_mdo);
 767     }
 768 
 769     // Increment counter in MethodCounters.
 770     const Address invocation_counter(m_counters,
 771                                      MethodCounters::invocation_counter_offset() +
 772                                      InvocationCounter::counter_offset());
 773     // Get address of MethodCounters object.
 774     __ get_method_counters(method, m_counters, done);
 775     const Address mask(m_counters, MethodCounters::invoke_mask_offset());
 776     __ increment_mask_and_jump(invocation_counter,
 777                                increment, mask,
 778                                Z_R1_scratch, false, Assembler::bcondZero,
 779                                overflow);
 780   } else {
 781     Register counter_sum = Z_ARG3; // The result of this piece of code.
 782     Register tmp         = Z_R1_scratch;
 783 #ifdef ASSERT
 784     {
 785       NearLabel ok;
 786       __ get_method(tmp);
 787       __ compare64_and_branch(method, tmp, Assembler::bcondEqual, ok);
 788       __ z_illtrap(0x66);
 789       __ bind(ok);
 790     }
 791 #endif
 792 
 793     // Get address of MethodCounters object.
 794     __ get_method_counters(method, m_counters, done);
 795     // Update standard invocation counters.
 796     __ increment_invocation_counter(m_counters, counter_sum);
 797     if (ProfileInterpreter) {
 798       __ add2mem_32(Address(m_counters, MethodCounters::interpreter_invocation_counter_offset()), 1, tmp);
 799       if (profile_method != NULL) {
 800         const Address profile_limit(m_counters, MethodCounters::interpreter_profile_limit_offset());
 801         __ z_cl(counter_sum, profile_limit);
 802         __ branch_optimized(Assembler::bcondLow, *profile_method_continue);
 803         // If no method data exists, go to profile_method.
 804         __ test_method_data_pointer(tmp, *profile_method);
 805       }
 806     }
 807 
 808     const Address invocation_limit(m_counters, MethodCounters::interpreter_invocation_limit_offset());
 809     __ z_cl(counter_sum, invocation_limit);
 810     __ branch_optimized(Assembler::bcondNotLow, *overflow);
 811   }
 812 
 813   __ bind(done);
 814 
 815   BLOCK_COMMENT("} counter_incr");
 816 }
 817 
 818 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 819   // InterpreterRuntime::frequency_counter_overflow takes two
 820   // arguments, the first (thread) is passed by call_VM, the second
 821   // indicates if the counter overflow occurs at a backwards branch
 822   // (NULL bcp). We pass zero for it. The call returns the address
 823   // of the verified entry point for the method or NULL if the
 824   // compilation did not complete (either went background or bailed
 825   // out).
 826   __ clear_reg(Z_ARG2);
 827   __ call_VM(noreg,
 828              CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow),
 829              Z_ARG2);
 830   __ z_bru(do_continue);
 831 }
 832 
 833 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register frame_size, Register tmp1) {
 834   Register tmp2 = Z_R1_scratch;
 835   const int page_size = os::vm_page_size();
 836   NearLabel after_frame_check;
 837 
 838   BLOCK_COMMENT("counter_overflow {");
 839 
 840   assert_different_registers(frame_size, tmp1);
 841 
 842   // Stack banging is sufficient overflow check if frame_size < page_size.
 843   if (Immediate::is_uimm(page_size, 15)) {
 844     __ z_chi(frame_size, page_size);
 845     __ z_brl(after_frame_check);
 846   } else {
 847     __ load_const_optimized(tmp1, page_size);
 848     __ compareU32_and_branch(frame_size, tmp1, Assembler::bcondLow, after_frame_check);
 849   }
 850 
 851   // Get the stack base, and in debug, verify it is non-zero.
 852   __ z_lg(tmp1, thread_(stack_base));
 853 #ifdef ASSERT
 854   address reentry = NULL;
 855   NearLabel base_not_zero;
 856   __ compareU64_and_branch(tmp1, (intptr_t)0L, Assembler::bcondNotEqual, base_not_zero);
 857   reentry = __ stop_chain_static(reentry, "stack base is zero in generate_stack_overflow_check");
 858   __ bind(base_not_zero);
 859 #endif
 860 
 861   // Get the stack size, and in debug, verify it is non-zero.
 862   assert(sizeof(size_t) == sizeof(intptr_t), "wrong load size");
 863   __ z_lg(tmp2, thread_(stack_size));
 864 #ifdef ASSERT
 865   NearLabel size_not_zero;
 866   __ compareU64_and_branch(tmp2, (intptr_t)0L, Assembler::bcondNotEqual, size_not_zero);
 867   reentry = __ stop_chain_static(reentry, "stack size is zero in generate_stack_overflow_check");
 868   __ bind(size_not_zero);
 869 #endif
 870 
 871   // Compute the beginning of the protected zone minus the requested frame size.
 872   __ z_sgr(tmp1, tmp2);
 873   __ add2reg(tmp1, JavaThread::stack_guard_zone_size());
 874 
 875   // Add in the size of the frame (which is the same as subtracting it from the
 876   // SP, which would take another register.
 877   __ z_agr(tmp1, frame_size);
 878 
 879   // The frame is greater than one page in size, so check against
 880   // the bottom of the stack.
 881   __ compareU64_and_branch(Z_SP, tmp1, Assembler::bcondHigh, after_frame_check);
 882 
 883   // The stack will overflow, throw an exception.
 884 
 885   // Restore SP to sender's sp. This is necessary if the sender's frame is an
 886   // extended compiled frame (see gen_c2i_adapter()) and safer anyway in case of
 887   // JSR292 adaptations.
 888   __ resize_frame_absolute(Z_R10, tmp1, true/*load_fp*/);
 889 
 890   // Note also that the restored frame is not necessarily interpreted.
 891   // Use the shared runtime version of the StackOverflowError.
 892   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 893   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 894   __ load_absolute_address(tmp1, StubRoutines::throw_StackOverflowError_entry());
 895   __ z_br(tmp1);
 896 
 897   // If you get to here, then there is enough stack space.
 898   __ bind(after_frame_check);
 899 
 900   BLOCK_COMMENT("} counter_overflow");
 901 }
 902 
 903 // Allocate monitor and lock method (asm interpreter).
 904 //
 905 // Args:
 906 //   Z_locals: locals
 907 
 908 void TemplateInterpreterGenerator::lock_method(void) {
 909 
 910   BLOCK_COMMENT("lock_method {");
 911 
 912   // Synchronize method.
 913   const Register method = Z_tmp_2;
 914   __ get_method(method);
 915 
 916 #ifdef ASSERT
 917   address reentry = NULL;
 918   {
 919     Label L;
 920     __ testbit(method2_(method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 921     __ z_btrue(L);
 922     reentry = __ stop_chain_static(reentry, "method doesn't need synchronization");
 923     __ bind(L);
 924   }
 925 #endif // ASSERT
 926 
 927   // Get synchronization object.
 928   const Register object = Z_tmp_2;
 929 
 930   {
 931     Label     done;
 932     Label     static_method;
 933 
 934     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 935     __ z_btrue(static_method);
 936 
 937     // non-static method: Load receiver obj from stack.
 938     __ mem2reg_opt(object, Address(Z_locals, Interpreter::local_offset_in_bytes(0)));
 939     __ z_bru(done);
 940 
 941     __ bind(static_method);
 942 
 943     // Lock the java mirror.
 944     __ load_mirror(object, method);
 945 #ifdef ASSERT
 946     {
 947       NearLabel L;
 948       __ compare64_and_branch(object, (intptr_t) 0, Assembler::bcondNotEqual, L);
 949       reentry = __ stop_chain_static(reentry, "synchronization object is NULL");
 950       __ bind(L);
 951     }
 952 #endif // ASSERT
 953 
 954     __ bind(done);
 955   }
 956 
 957   __ add_monitor_to_stack(true, Z_ARG3, Z_ARG4, Z_ARG5); // Allocate monitor elem.
 958   // Store object and lock it.
 959   __ get_monitors(Z_tmp_1);
 960   __ reg2mem_opt(object, Address(Z_tmp_1, BasicObjectLock::obj_offset_in_bytes()));
 961   __ lock_object(Z_tmp_1, object);
 962 
 963   BLOCK_COMMENT("} lock_method");
 964 }
 965 
 966 // Generate a fixed interpreter frame. This is identical setup for
 967 // interpreted methods and for native methods hence the shared code.
 968 //
 969 // Registers alive
 970 //   Z_thread   - JavaThread*
 971 //   Z_SP       - old stack pointer
 972 //   Z_method   - callee's method
 973 //   Z_esp      - parameter list (slot 'above' last param)
 974 //   Z_R14      - return pc, to be stored in caller's frame
 975 //   Z_R10      - sender sp, note: Z_tmp_1 is Z_R10!
 976 //
 977 // Registers updated
 978 //   Z_SP       - new stack pointer
 979 //   Z_esp      - callee's operand stack pointer
 980 //                points to the slot above the value on top
 981 //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
 982 //   Z_bcp      - the bytecode pointer
 983 //   Z_fp       - the frame pointer, thereby killing Z_method
 984 //   Z_ARG2     - copy of Z_method
 985 //
 986 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 987 
 988   //  stack layout
 989   //
 990   //   F1 [TOP_IJAVA_FRAME_ABI]              <-- Z_SP, Z_R10 (see note below)
 991   //      [F1's operand stack (unused)]
 992   //      [F1's outgoing Java arguments]     <-- Z_esp
 993   //      [F1's operand stack (non args)]
 994   //      [monitors]      (optional)
 995   //      [IJAVA_STATE]
 996   //
 997   //   F2 [PARENT_IJAVA_FRAME_ABI]
 998   //      ...
 999   //
1000   //  0x000
1001   //
1002   // Note: Z_R10, the sender sp, will be below Z_SP if F1 was extended by a c2i adapter.
1003 
1004   //=============================================================================
1005   // Allocate space for locals other than the parameters, the
1006   // interpreter state, monitors, and the expression stack.
1007 
1008   const Register local_count     = Z_ARG5;
1009   const Register fp              = Z_tmp_2;
1010 
1011   BLOCK_COMMENT("generate_fixed_frame {");
1012 
1013   {
1014   // local registers
1015   const Register top_frame_size  = Z_ARG2;
1016   const Register sp_after_resize = Z_ARG3;
1017   const Register max_stack       = Z_ARG4;
1018 
1019   // local_count = method->constMethod->max_locals();
1020   __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1021   __ z_llgh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1022 
1023   if (native_call) {
1024     // If we're calling a native method, we replace max_stack (which is
1025     // zero) with space for the worst-case signature handler varargs
1026     // vector, which is:
1027     //   max_stack = max(Argument::n_register_parameters, parameter_count+2);
1028     //
1029     // We add two slots to the parameter_count, one for the jni
1030     // environment and one for a possible native mirror. We allocate
1031     // space for at least the number of ABI registers, even though
1032     // InterpreterRuntime::slow_signature_handler won't write more than
1033     // parameter_count+2 words when it creates the varargs vector at the
1034     // top of the stack. The generated slow signature handler will just
1035     // load trash into registers beyond the necessary number. We're
1036     // still going to cut the stack back by the ABI register parameter
1037     // count so as to get SP+16 pointing at the ABI outgoing parameter
1038     // area, so we need to allocate at least that much even though we're
1039     // going to throw it away.
1040     //
1041 
1042     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1043     __ z_llgh(max_stack,  Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1044     __ add2reg(max_stack, 2);
1045 
1046     NearLabel passing_args_on_stack;
1047 
1048     // max_stack in bytes
1049     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1050 
1051     int argument_registers_in_bytes = Argument::n_register_parameters << LogBytesPerWord;
1052     __ compare64_and_branch(max_stack, argument_registers_in_bytes, Assembler::bcondNotLow, passing_args_on_stack);
1053 
1054     __ load_const_optimized(max_stack, argument_registers_in_bytes);
1055 
1056     __ bind(passing_args_on_stack);
1057   } else {
1058     // !native_call
1059     __ z_lg(max_stack, method_(const));
1060 
1061     // Calculate number of non-parameter locals (in slots):
1062     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1063     __ z_sh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1064 
1065     // max_stack = method->max_stack();
1066     __ z_llgh(max_stack, Address(max_stack, ConstMethod::max_stack_offset()));
1067     // max_stack in bytes
1068     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1069   }
1070 
1071   // Resize (i.e. normally shrink) the top frame F1 ...
1072   //   F1      [TOP_IJAVA_FRAME_ABI]          <-- Z_SP, Z_R10
1073   //           F1's operand stack (free)
1074   //           ...
1075   //           F1's operand stack (free)      <-- Z_esp
1076   //           F1's outgoing Java arg m
1077   //           ...
1078   //           F1's outgoing Java arg 0
1079   //           ...
1080   //
1081   //  ... into a parent frame (Z_R10 holds F1's SP before any modification, see also above)
1082   //
1083   //           +......................+
1084   //           :                      :        <-- Z_R10, saved below as F0's z_ijava_state.sender_sp
1085   //           :                      :
1086   //   F1      [PARENT_IJAVA_FRAME_ABI]        <-- Z_SP       \
1087   //           F0's non arg local                             | = delta
1088   //           ...                                            |
1089   //           F0's non arg local              <-- Z_esp      /
1090   //           F1's outgoing Java arg m
1091   //           ...
1092   //           F1's outgoing Java arg 0
1093   //           ...
1094   //
1095   // then push the new top frame F0.
1096   //
1097   //   F0      [TOP_IJAVA_FRAME_ABI]    = frame::z_top_ijava_frame_abi_size \
1098   //           [operand stack]          = max_stack                          | = top_frame_size
1099   //           [IJAVA_STATE]            = frame::z_ijava_state_size         /
1100 
1101   // sp_after_resize = Z_esp - delta
1102   //
1103   // delta = PARENT_IJAVA_FRAME_ABI + (locals_count - params_count)
1104 
1105   __ add2reg(sp_after_resize, (Interpreter::stackElementSize) - (frame::z_parent_ijava_frame_abi_size), Z_esp);
1106   __ z_sllg(Z_R0_scratch, local_count, LogBytesPerWord); // Params have already been subtracted from local_count.
1107   __ z_slgr(sp_after_resize, Z_R0_scratch);
1108 
1109   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
1110   __ add2reg(top_frame_size,
1111              frame::z_top_ijava_frame_abi_size +
1112                frame::z_ijava_state_size +
1113                frame::interpreter_frame_monitor_size() * wordSize,
1114              max_stack);
1115 
1116   // Check if there's room for the new frame...
1117   Register frame_size = max_stack; // Reuse the regiser for max_stack.
1118   __ z_lgr(frame_size, Z_SP);
1119   __ z_sgr(frame_size, sp_after_resize);
1120   __ z_agr(frame_size, top_frame_size);
1121   generate_stack_overflow_check(frame_size, fp/*tmp1*/);
1122 
1123   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
1124   __ asm_assert_eq("killed Z_R14", 0);
1125   __ resize_frame_absolute(sp_after_resize, fp, true);
1126   __ save_return_pc(Z_R14);
1127 
1128   // ... and push the new frame F0.
1129   __ push_frame(top_frame_size, fp, true /*copy_sp*/, false);
1130   }
1131 
1132   //=============================================================================
1133   // Initialize the new frame F0: initialize interpreter state.
1134 
1135   {
1136   // locals
1137   const Register local_addr = Z_ARG4;
1138 
1139   BLOCK_COMMENT("generate_fixed_frame: initialize interpreter state {");
1140 
1141 #ifdef ASSERT
1142   // Set the magic number (using local_addr as tmp register).
1143   __ load_const_optimized(local_addr, frame::z_istate_magic_number);
1144   __ z_stg(local_addr, _z_ijava_state_neg(magic), fp);
1145 #endif
1146 
1147   // Save sender SP from F1 (i.e. before it was potentially modified by an
1148   // adapter) into F0's interpreter state. We us it as well to revert
1149   // resizing the frame above.
1150   __ z_stg(Z_R10, _z_ijava_state_neg(sender_sp), fp);
1151 
1152   // Load cp cache and save it at the and of this block.
1153   __ z_lg(Z_R1_scratch, Address(Z_method,    Method::const_offset()));
1154   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstMethod::constants_offset()));
1155   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstantPool::cache_offset_in_bytes()));
1156 
1157   // z_ijava_state->method = method;
1158   __ z_stg(Z_method, _z_ijava_state_neg(method), fp);
1159 
1160   // Point locals at the first argument. Method's locals are the
1161   // parameters on top of caller's expression stack.
1162   // Tos points past last Java argument.
1163 
1164   __ z_lg(Z_locals, Address(Z_method, Method::const_offset()));
1165   __ z_llgh(Z_locals /*parameter_count words*/,
1166             Address(Z_locals, ConstMethod::size_of_parameters_offset()));
1167   __ z_sllg(Z_locals /*parameter_count bytes*/, Z_locals /*parameter_count*/, LogBytesPerWord);
1168   __ z_agr(Z_locals, Z_esp);
1169   // z_ijava_state->locals - i*BytesPerWord points to i-th Java local (i starts at 0)
1170   // z_ijava_state->locals = Z_esp + parameter_count bytes
1171   __ z_stg(Z_locals, _z_ijava_state_neg(locals), fp);
1172 
1173   // z_ijava_state->oop_temp = NULL;
1174   __ store_const(Address(fp, oop_tmp_offset), 0);
1175 
1176   // Initialize z_ijava_state->mdx.
1177   Register Rmdp = Z_bcp;
1178   // native_call: assert that mdo == NULL
1179   const bool check_for_mdo = !native_call DEBUG_ONLY(|| native_call);
1180   if (ProfileInterpreter && check_for_mdo) {
1181 #ifdef FAST_DISPATCH
1182     // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
1183     // they both use I2.
1184     assert(0, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
1185 #endif // FAST_DISPATCH
1186     Label get_continue;
1187 
1188     __ load_and_test_long(Rmdp, method_(method_data));
1189     __ z_brz(get_continue);
1190     DEBUG_ONLY(if (native_call) __ stop("native methods don't have a mdo"));
1191     __ add2reg(Rmdp, in_bytes(MethodData::data_offset()));
1192     __ bind(get_continue);
1193   }
1194   __ z_stg(Rmdp, _z_ijava_state_neg(mdx), fp);
1195 
1196   // Initialize z_ijava_state->bcp and Z_bcp.
1197   if (native_call) {
1198     __ clear_reg(Z_bcp); // Must initialize. Will get written into frame where GC reads it.
1199   } else {
1200     __ z_lg(Z_bcp, method_(const));
1201     __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset()));
1202   }
1203   __ z_stg(Z_bcp, _z_ijava_state_neg(bcp), fp);
1204 
1205   // no monitors and empty operand stack
1206   // => z_ijava_state->monitors points to the top slot in IJAVA_STATE.
1207   // => Z_ijava_state->esp points one slot above into the operand stack.
1208   // z_ijava_state->monitors = fp - frame::z_ijava_state_size - Interpreter::stackElementSize;
1209   // z_ijava_state->esp = Z_esp = z_ijava_state->monitors;
1210   __ add2reg(Z_esp, -frame::z_ijava_state_size, fp);
1211   __ z_stg(Z_esp, _z_ijava_state_neg(monitors), fp);
1212   __ add2reg(Z_esp, -Interpreter::stackElementSize);
1213   __ z_stg(Z_esp, _z_ijava_state_neg(esp), fp);
1214 
1215   // z_ijava_state->cpoolCache = Z_R1_scratch (see load above);
1216   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(cpoolCache), fp);
1217 
1218   // Get mirror and store it in the frame as GC root for this Method*.
1219   __ load_mirror(Z_R1_scratch, Z_method);
1220   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(mirror), fp);
1221 
1222   BLOCK_COMMENT("} generate_fixed_frame: initialize interpreter state");
1223 
1224   //=============================================================================
1225   if (!native_call) {
1226     // Fill locals with 0x0s.
1227     NearLabel locals_zeroed;
1228     NearLabel doXC;
1229 
1230     // Local_count is already num_locals_slots - num_param_slots.
1231     __ compare64_and_branch(local_count, (intptr_t)0L, Assembler::bcondNotHigh, locals_zeroed);
1232 
1233     // Advance local_addr to point behind locals (creates positive incr. in loop).
1234     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1235     __ z_llgh(Z_R0_scratch,
1236               Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1237     if (Z_R0_scratch == Z_R0) {
1238       __ z_aghi(Z_R0_scratch, -1);
1239     } else {
1240       __ add2reg(Z_R0_scratch, -1);
1241     }
1242     __ z_lgr(local_addr/*locals*/, Z_locals);
1243     __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogBytesPerWord);
1244     __ z_sllg(local_count, local_count, LogBytesPerWord); // Local_count are non param locals.
1245     __ z_sgr(local_addr, Z_R0_scratch);
1246 
1247     if (VM_Version::has_Prefetch()) {
1248       __ z_pfd(0x02, 0, Z_R0, local_addr);
1249       __ z_pfd(0x02, 256, Z_R0, local_addr);
1250     }
1251 
1252     // Can't optimise for Z10 using "compare and branch" (immediate value is too big).
1253     __ z_cghi(local_count, 256);
1254     __ z_brnh(doXC);
1255 
1256     // MVCLE: Initialize if quite a lot locals.
1257     //  __ bind(doMVCLE);
1258     __ z_lgr(Z_R0_scratch, local_addr);
1259     __ z_lgr(Z_R1_scratch, local_count);
1260     __ clear_reg(Z_ARG2);        // Src len of MVCLE is zero.
1261 
1262     __ MacroAssembler::move_long_ext(Z_R0_scratch, Z_ARG1, 0);
1263     __ z_bru(locals_zeroed);
1264 
1265     Label  XC_template;
1266     __ bind(XC_template);
1267     __ z_xc(0, 0, local_addr, 0, local_addr);
1268 
1269     __ bind(doXC);
1270     __ z_bctgr(local_count, Z_R0);                  // Get #bytes-1 for EXECUTE.
1271     if (VM_Version::has_ExecuteExtensions()) {
1272       __ z_exrl(local_count, XC_template);          // Execute XC with variable length.
1273     } else {
1274       __ z_larl(Z_R1_scratch, XC_template);
1275       __ z_ex(local_count, 0, Z_R0, Z_R1_scratch);  // Execute XC with variable length.
1276     }
1277 
1278     __ bind(locals_zeroed);
1279   }
1280 
1281   }
1282   // Finally set the frame pointer, destroying Z_method.
1283   assert(Z_fp == Z_method, "maybe set Z_fp earlier if other register than Z_method");
1284   // Oprofile analysis suggests to keep a copy in a register to be used by
1285   // generate_counter_incr().
1286   __ z_lgr(Z_ARG2, Z_method);
1287   __ z_lgr(Z_fp, fp);
1288 
1289   BLOCK_COMMENT("} generate_fixed_frame");
1290 }
1291 
1292 // Various method entries
1293 
1294 // Math function, frame manager must set up an interpreter state, etc.
1295 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1296 
1297   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1298   bool use_instruction = false;
1299   address runtime_entry = NULL;
1300   int num_args = 1;
1301   bool double_precision = true;
1302 
1303   // s390 specific:
1304   switch (kind) {
1305     case Interpreter::java_lang_math_sqrt:
1306     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1307     case Interpreter::java_lang_math_fmaF:
1308     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1309     default: break; // Fall back to runtime call.
1310   }
1311 
1312   switch (kind) {
1313     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1314     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1315     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1316     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1317     case Interpreter::java_lang_math_sqrt : /* runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); not available */ break;
1318     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1319     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1320     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1321     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1322     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1323     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1324     default: ShouldNotReachHere();
1325   }
1326 
1327   // Use normal entry if neither instruction nor runtime call is used.
1328   if (!use_instruction && runtime_entry == NULL) return NULL;
1329 
1330   address entry = __ pc();
1331 
1332   if (use_instruction) {
1333     switch (kind) {
1334       case Interpreter::java_lang_math_sqrt:
1335         // Can use memory operand directly.
1336         __ z_sqdb(Z_FRET, Interpreter::stackElementSize, Z_esp);
1337         break;
1338       case Interpreter::java_lang_math_abs:
1339         // Load operand from stack.
1340         __ mem2freg_opt(Z_FRET, Address(Z_esp, Interpreter::stackElementSize));
1341         __ z_lpdbr(Z_FRET);
1342         break;
1343       case Interpreter::java_lang_math_fmaF:
1344         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1345         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize)); // arg1
1346         __ z_maeb(Z_FRET, Z_FARG2, Address(Z_esp, 2 * Interpreter::stackElementSize));
1347         break;
1348       case Interpreter::java_lang_math_fmaD:
1349         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1350         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 5 * Interpreter::stackElementSize)); // arg1
1351         __ z_madb(Z_FRET, Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize));
1352         break;
1353       default: ShouldNotReachHere();
1354     }
1355   } else {
1356     // Load arguments
1357     assert(num_args <= 4, "passed in registers");
1358     if (double_precision) {
1359       int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1360       for (int i = 0; i < num_args; ++i) {
1361         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1362         offset -= 2 * Interpreter::stackElementSize;
1363       }
1364     } else {
1365       int offset = num_args * Interpreter::stackElementSize;
1366       for (int i = 0; i < num_args; ++i) {
1367         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1368         offset -= Interpreter::stackElementSize;
1369       }
1370     }
1371     // Call runtime
1372     __ save_return_pc();       // Save Z_R14.
1373     __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
1374 
1375     __ call_VM_leaf(runtime_entry);
1376 
1377     __ pop_frame();
1378     __ restore_return_pc();    // Restore Z_R14.
1379   }
1380 
1381   // Pop c2i arguments (if any) off when we return.
1382   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1383 
1384   __ z_br(Z_R14);
1385 
1386   return entry;
1387 }
1388 
1389 // Interpreter stub for calling a native method. (asm interpreter).
1390 // This sets up a somewhat different looking stack for calling the
1391 // native method than the typical interpreter frame setup.
1392 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1393   // Determine code generation flags.
1394   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1395 
1396   // Interpreter entry for ordinary Java methods.
1397   //
1398   // Registers alive
1399   //   Z_SP          - stack pointer
1400   //   Z_thread      - JavaThread*
1401   //   Z_method      - callee's method (method to be invoked)
1402   //   Z_esp         - operand (or expression) stack pointer of caller. one slot above last arg.
1403   //   Z_R10         - sender sp (before modifications, e.g. by c2i adapter
1404   //                   and as well by generate_fixed_frame below)
1405   //   Z_R14         - return address to caller (call_stub or c2i_adapter)
1406   //
1407   // Registers updated
1408   //   Z_SP          - stack pointer
1409   //   Z_fp          - callee's framepointer
1410   //   Z_esp         - callee's operand stack pointer
1411   //                   points to the slot above the value on top
1412   //   Z_locals      - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1413   //   Z_tos         - integer result, if any
1414   //   z_ftos        - floating point result, if any
1415   //
1416   // Stack layout at this point:
1417   //
1418   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1419   //                                                          frame was extended by c2i adapter)
1420   //           [outgoing Java arguments]     <-- Z_esp
1421   //           ...
1422   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1423   //           ...
1424   //
1425 
1426   address entry_point = __ pc();
1427 
1428   // Make sure registers are different!
1429   assert_different_registers(Z_thread, Z_method, Z_esp);
1430 
1431   BLOCK_COMMENT("native_entry {");
1432 
1433   // Make sure method is native and not abstract.
1434 #ifdef ASSERT
1435   address reentry = NULL;
1436   { Label L;
1437     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1438     __ z_btrue(L);
1439     reentry = __ stop_chain_static(reentry, "tried to execute non-native method as native");
1440     __ bind(L);
1441   }
1442   { Label L;
1443     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1444     __ z_bfalse(L);
1445     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1446     __ bind(L);
1447   }
1448 #endif // ASSERT
1449 
1450 #ifdef ASSERT
1451   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1452   __ save_return_pc(Z_R14);
1453 #endif
1454 
1455   // Generate the code to allocate the interpreter stack frame.
1456   generate_fixed_frame(true);
1457 
1458   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1459   // Since at this point in the method invocation the exception handler
1460   // would try to exit the monitor of synchronized methods which hasn't
1461   // been entered yet, we set the thread local variable
1462   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1463   // runtime, exception handling i.e. unlock_if_synchronized_method will
1464   // check this thread local flag.
1465   __ z_mvi(do_not_unlock_if_synchronized, true);
1466 
1467   // Increment invocation count and check for overflow.
1468   NearLabel invocation_counter_overflow;
1469   if (inc_counter) {
1470     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1471   }
1472 
1473   Label continue_after_compile;
1474   __ bind(continue_after_compile);
1475 
1476   bang_stack_shadow_pages(true);
1477 
1478   // Reset the _do_not_unlock_if_synchronized flag.
1479   __ z_mvi(do_not_unlock_if_synchronized, false);
1480 
1481   // Check for synchronized methods.
1482   // This mst happen AFTER invocation_counter check and stack overflow check,
1483   // so method is not locked if overflows.
1484   if (synchronized) {
1485     lock_method();
1486   } else {
1487     // No synchronization necessary.
1488 #ifdef ASSERT
1489     { Label L;
1490       __ get_method(Z_R1_scratch);
1491       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1492       __ z_bfalse(L);
1493       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1494       __ bind(L);
1495     }
1496 #endif // ASSERT
1497   }
1498 
1499   // start execution
1500 
1501   // jvmti support
1502   __ notify_method_entry();
1503 
1504   //=============================================================================
1505   // Get and call the signature handler.
1506   const Register Rmethod                 = Z_tmp_2;
1507   const Register signature_handler_entry = Z_tmp_1;
1508   const Register Rresult_handler         = Z_tmp_3;
1509   Label call_signature_handler;
1510 
1511   assert_different_registers(Z_fp, Rmethod, signature_handler_entry, Rresult_handler);
1512   assert(Rresult_handler->is_nonvolatile(), "Rresult_handler must be in a non-volatile register");
1513 
1514   // Reload method.
1515   __ get_method(Rmethod);
1516 
1517   // Check for signature handler.
1518   __ load_and_test_long(signature_handler_entry, method2_(Rmethod, signature_handler));
1519   __ z_brne(call_signature_handler);
1520 
1521   // Method has never been called. Either generate a specialized
1522   // handler or point to the slow one.
1523   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call),
1524              Rmethod);
1525 
1526   // Reload method.
1527   __ get_method(Rmethod);
1528 
1529   // Reload signature handler, it must have been created/assigned in the meantime.
1530   __ z_lg(signature_handler_entry, method2_(Rmethod, signature_handler));
1531 
1532   __ bind(call_signature_handler);
1533 
1534   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1535   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1/*tmp*/);
1536 
1537   // Call signature handler and pass locals address in Z_ARG1.
1538   __ z_lgr(Z_ARG1, Z_locals);
1539   __ call_stub(signature_handler_entry);
1540   // Save result handler returned by signature handler.
1541   __ z_lgr(Rresult_handler, Z_RET);
1542 
1543   // Reload method (the slow signature handler may block for GC).
1544   __ get_method(Rmethod);
1545 
1546   // Pass mirror handle if static call.
1547   {
1548     Label method_is_not_static;
1549     __ testbit(method2_(Rmethod, access_flags), JVM_ACC_STATIC_BIT);
1550     __ z_bfalse(method_is_not_static);
1551     // Get mirror.
1552     __ load_mirror(Z_R1, Rmethod);
1553     // z_ijava_state.oop_temp = pool_holder->klass_part()->java_mirror();
1554     __ z_stg(Z_R1, oop_tmp_offset, Z_fp);
1555     // Pass handle to mirror as 2nd argument to JNI method.
1556     __ add2reg(Z_ARG2, oop_tmp_offset, Z_fp);
1557     __ bind(method_is_not_static);
1558   }
1559 
1560   // Pass JNIEnv address as first parameter.
1561   __ add2reg(Z_ARG1, in_bytes(JavaThread::jni_environment_offset()), Z_thread);
1562 
1563   // Note: last java frame has been set above already. The pc from there
1564   // is precise enough.
1565 
1566   // Get native function entry point before we change the thread state.
1567   __ z_lg(Z_R1/*native_method_entry*/, method2_(Rmethod, native_function));
1568 
1569   //=============================================================================
1570   // Transition from _thread_in_Java to _thread_in_native. As soon as
1571   // we make this change the safepoint code needs to be certain that
1572   // the last Java frame we established is good. The pc in that frame
1573   // just need to be near here not an actual return address.
1574 #ifdef ASSERT
1575   {
1576     NearLabel L;
1577     __ mem2reg_opt(Z_R14, Address(Z_thread, JavaThread::thread_state_offset()), false /*32 bits*/);
1578     __ compareU32_and_branch(Z_R14, _thread_in_Java, Assembler::bcondEqual, L);
1579     reentry = __ stop_chain_static(reentry, "Wrong thread state in native stub");
1580     __ bind(L);
1581   }
1582 #endif
1583 
1584   // Memory ordering: Z does not reorder store/load with subsequent load. That's strong enough.
1585   __ set_thread_state(_thread_in_native);
1586 
1587   //=============================================================================
1588   // Call the native method. Argument registers must not have been
1589   // overwritten since "__ call_stub(signature_handler);" (except for
1590   // ARG1 and ARG2 for static methods).
1591 
1592   __ call_c(Z_R1/*native_method_entry*/);
1593 
1594   // NOTE: frame::interpreter_frame_result() depends on these stores.
1595   __ z_stg(Z_RET, _z_ijava_state_neg(lresult), Z_fp);
1596   __ freg2mem_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1597   const Register Rlresult = signature_handler_entry;
1598   assert(Rlresult->is_nonvolatile(), "Rlresult must be in a non-volatile register");
1599   __ z_lgr(Rlresult, Z_RET);
1600 
1601   // Z_method may no longer be valid, because of GC.
1602 
1603   // Block, if necessary, before resuming in _thread_in_Java state.
1604   // In order for GC to work, don't clear the last_Java_sp until after
1605   // blocking.
1606 
1607   //=============================================================================
1608   // Switch thread to "native transition" state before reading the
1609   // synchronization state. This additional state is necessary
1610   // because reading and testing the synchronization state is not
1611   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1612   // in _thread_in_native state, loads _not_synchronized and is
1613   // preempted. VM thread changes sync state to synchronizing and
1614   // suspends threads for GC. Thread A is resumed to finish this
1615   // native method, but doesn't block here since it didn't see any
1616   // synchronization is progress, and escapes.
1617 
1618   __ set_thread_state(_thread_in_native_trans);
1619   if (UseMembar) {
1620     __ z_fence();
1621   } else {
1622     // Write serialization page so VM thread can do a pseudo remote
1623     // membar. We use the current thread pointer to calculate a thread
1624     // specific offset to write to within the page. This minimizes bus
1625     // traffic due to cache line collision.
1626     __ serialize_memory(Z_thread, Z_R1, Z_R0);
1627   }
1628   // Now before we return to java we must look for a current safepoint
1629   // (a new safepoint can not start since we entered native_trans).
1630   // We must check here because a current safepoint could be modifying
1631   // the callers registers right this moment.
1632 
1633   // Check for safepoint operation in progress and/or pending suspend requests.
1634   {
1635     Label Continue, do_safepoint;
1636     __ generate_safepoint_check(do_safepoint, Z_R1, true);
1637     // Check for suspend.
1638     __ load_and_test_int(Z_R0/*suspend_flags*/, thread_(suspend_flags));
1639     __ z_bre(Continue); // 0 -> no flag set -> not suspended
1640     __ bind(do_safepoint);
1641     __ z_lgr(Z_ARG1, Z_thread);
1642     __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1643     __ bind(Continue);
1644   }
1645 
1646   //=============================================================================
1647   // Back in Interpreter Frame.
1648 
1649   // We are in thread_in_native_trans here and back in the normal
1650   // interpreter frame. We don't have to do anything special about
1651   // safepoints and we can switch to Java mode anytime we are ready.
1652 
1653   // Note: frame::interpreter_frame_result has a dependency on how the
1654   // method result is saved across the call to post_method_exit. For
1655   // native methods it assumes that the non-FPU/non-void result is
1656   // saved in z_ijava_state.lresult and a FPU result in z_ijava_state.fresult. If
1657   // this changes then the interpreter_frame_result implementation
1658   // will need to be updated too.
1659 
1660   //=============================================================================
1661   // Back in Java.
1662 
1663   // Memory ordering: Z does not reorder store/load with subsequent
1664   // load. That's strong enough.
1665   __ set_thread_state(_thread_in_Java);
1666 
1667   __ reset_last_Java_frame();
1668 
1669   // We reset the JNI handle block only after unboxing the result; see below.
1670 
1671   // The method register is junk from after the thread_in_native transition
1672   // until here. Also can't call_VM until the bcp has been
1673   // restored. Need bcp for throwing exception below so get it now.
1674   __ get_method(Rmethod);
1675 
1676   // Restore Z_bcp to have legal interpreter frame,
1677   // i.e., bci == 0 <=> Z_bcp == code_base().
1678   __ z_lg(Z_bcp, Address(Rmethod, Method::const_offset())); // get constMethod
1679   __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1680 
1681   if (CheckJNICalls) {
1682     // clear_pending_jni_exception_check
1683     __ clear_mem(Address(Z_thread, JavaThread::pending_jni_exception_check_fn_offset()), sizeof(oop));
1684   }
1685 
1686   // Check if the native method returns an oop, and if so, move it
1687   // from the jni handle to z_ijava_state.oop_temp. This is
1688   // necessary, because we reset the jni handle block below.
1689   // NOTE: frame::interpreter_frame_result() depends on this, too.
1690   { NearLabel no_oop_result, store_oop_result;
1691   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1692   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_result);
1693   __ compareU64_and_branch(Rlresult, (intptr_t)0L, Assembler::bcondEqual, store_oop_result);
1694   __ z_lg(Rlresult, 0, Rlresult);  // unbox
1695   __ bind(store_oop_result);
1696   __ z_stg(Rlresult, oop_tmp_offset, Z_fp);
1697   __ verify_oop(Rlresult);
1698   __ bind(no_oop_result);
1699   }
1700 
1701   // Reset handle block.
1702   __ z_lg(Z_R1/*active_handles*/, thread_(active_handles));
1703   __ clear_mem(Address(Z_R1, JNIHandleBlock::top_offset_in_bytes()), 4);
1704 
1705   // Bandle exceptions (exception handling will handle unlocking!).
1706   {
1707     Label L;
1708     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
1709     __ z_bre(L);
1710     __ MacroAssembler::call_VM(noreg,
1711                                CAST_FROM_FN_PTR(address,
1712                                InterpreterRuntime::throw_pending_exception));
1713     __ should_not_reach_here();
1714     __ bind(L);
1715   }
1716 
1717   if (synchronized) {
1718     Register Rfirst_monitor = Z_ARG2;
1719     __ add2reg(Rfirst_monitor, -(frame::z_ijava_state_size + (int)sizeof(BasicObjectLock)), Z_fp);
1720 #ifdef ASSERT
1721     NearLabel ok;
1722     __ z_lg(Z_R1, _z_ijava_state_neg(monitors), Z_fp);
1723     __ compareU64_and_branch(Rfirst_monitor, Z_R1, Assembler::bcondEqual, ok);
1724     reentry = __ stop_chain_static(reentry, "native_entry:unlock: inconsistent z_ijava_state.monitors");
1725     __ bind(ok);
1726 #endif
1727     __ unlock_object(Rfirst_monitor);
1728   }
1729 
1730   // JVMTI support. Result has already been saved above to the frame.
1731   __ notify_method_exit(true/*native_method*/, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1732 
1733   // Move native method result back into proper registers and return.
1734   // C++ interpreter does not use result handler. So do we need to here? TODO(ZASM): check if correct.
1735   { NearLabel no_oop_or_null;
1736   __ mem2freg_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1737   __ load_and_test_long(Z_RET, Address(Z_fp, _z_ijava_state_neg(lresult)));
1738   __ z_bre(no_oop_or_null); // No unboxing if the result is NULL.
1739   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1740   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_or_null);
1741   __ z_lg(Z_RET, oop_tmp_offset, Z_fp);
1742   __ verify_oop(Z_RET);
1743   __ bind(no_oop_or_null);
1744   }
1745 
1746   // Pop the native method's interpreter frame.
1747   __ pop_interpreter_frame(Z_R14 /*return_pc*/, Z_ARG2/*tmp1*/, Z_ARG3/*tmp2*/);
1748 
1749   // Return to caller.
1750   __ z_br(Z_R14);
1751 
1752   if (inc_counter) {
1753     // Handle overflow of counter and compile method.
1754     __ bind(invocation_counter_overflow);
1755     generate_counter_overflow(continue_after_compile);
1756   }
1757 
1758   BLOCK_COMMENT("} native_entry");
1759 
1760   return entry_point;
1761 }
1762 
1763 //
1764 // Generic interpreted method entry to template interpreter.
1765 //
1766 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1767   address entry_point = __ pc();
1768 
1769   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1770 
1771   // Interpreter entry for ordinary Java methods.
1772   //
1773   // Registers alive
1774   //   Z_SP       - stack pointer
1775   //   Z_thread   - JavaThread*
1776   //   Z_method   - callee's method (method to be invoked)
1777   //   Z_esp      - operand (or expression) stack pointer of caller. one slot above last arg.
1778   //   Z_R10      - sender sp (before modifications, e.g. by c2i adapter
1779   //                           and as well by generate_fixed_frame below)
1780   //   Z_R14      - return address to caller (call_stub or c2i_adapter)
1781   //
1782   // Registers updated
1783   //   Z_SP       - stack pointer
1784   //   Z_fp       - callee's framepointer
1785   //   Z_esp      - callee's operand stack pointer
1786   //                points to the slot above the value on top
1787   //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1788   //   Z_tos      - integer result, if any
1789   //   z_ftos     - floating point result, if any
1790   //
1791   //
1792   // stack layout at this point:
1793   //
1794   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1795   //                                                          frame was extended by c2i adapter)
1796   //           [outgoing Java arguments]     <-- Z_esp
1797   //           ...
1798   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1799   //           ...
1800   //
1801   // stack layout before dispatching the first bytecode:
1802   //
1803   //   F0      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP
1804   //           [operand stack]               <-- Z_esp
1805   //           monitor (optional, can grow)
1806   //           [IJAVA_STATE]
1807   //   F1      [PARENT_IJAVA_FRAME_ABI]      <-- Z_fp (== *Z_SP)
1808   //           [F0's locals]                 <-- Z_locals
1809   //           [F1's operand stack]
1810   //           [F1's monitors] (optional)
1811   //           [IJAVA_STATE]
1812 
1813   // Make sure registers are different!
1814   assert_different_registers(Z_thread, Z_method, Z_esp);
1815 
1816   BLOCK_COMMENT("normal_entry {");
1817 
1818   // Make sure method is not native and not abstract.
1819   // Rethink these assertions - they can be simplified and shared.
1820 #ifdef ASSERT
1821   address reentry = NULL;
1822   { Label L;
1823     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1824     __ z_bfalse(L);
1825     reentry = __ stop_chain_static(reentry, "tried to execute native method as non-native");
1826     __ bind(L);
1827   }
1828   { Label L;
1829     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1830     __ z_bfalse(L);
1831     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1832     __ bind(L);
1833   }
1834 #endif // ASSERT
1835 
1836 #ifdef ASSERT
1837   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1838   __ save_return_pc(Z_R14);
1839 #endif
1840 
1841   // Generate the code to allocate the interpreter stack frame.
1842   generate_fixed_frame(false);
1843 
1844   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1845   // Since at this point in the method invocation the exception handler
1846   // would try to exit the monitor of synchronized methods which hasn't
1847   // been entered yet, we set the thread local variable
1848   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1849   // runtime, exception handling i.e. unlock_if_synchronized_method will
1850   // check this thread local flag.
1851   __ z_mvi(do_not_unlock_if_synchronized, true);
1852 
1853   __ profile_parameters_type(Z_tmp_2, Z_ARG3, Z_ARG4);
1854 
1855   // Increment invocation counter and check for overflow.
1856   //
1857   // Note: checking for negative value instead of overflow so we have a 'sticky'
1858   // overflow test (may be of importance as soon as we have true MT/MP).
1859   NearLabel invocation_counter_overflow;
1860   NearLabel profile_method;
1861   NearLabel profile_method_continue;
1862   NearLabel Lcontinue;
1863   if (inc_counter) {
1864     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1865     if (ProfileInterpreter) {
1866       __ bind(profile_method_continue);
1867     }
1868   }
1869   __ bind(Lcontinue);
1870 
1871   bang_stack_shadow_pages(false);
1872 
1873   // Reset the _do_not_unlock_if_synchronized flag.
1874   __ z_mvi(do_not_unlock_if_synchronized, false);
1875 
1876   // Check for synchronized methods.
1877   // Must happen AFTER invocation_counter check and stack overflow check,
1878   // so method is not locked if overflows.
1879   if (synchronized) {
1880     // Allocate monitor and lock method.
1881     lock_method();
1882   } else {
1883 #ifdef ASSERT
1884     { Label L;
1885       __ get_method(Z_R1_scratch);
1886       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1887       __ z_bfalse(L);
1888       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1889       __ bind(L);
1890     }
1891 #endif // ASSERT
1892   }
1893 
1894   // start execution
1895 
1896 #ifdef ASSERT
1897   __ verify_esp(Z_esp, Z_R1_scratch);
1898 
1899   __ verify_thread();
1900 #endif
1901 
1902   // jvmti support
1903   __ notify_method_entry();
1904 
1905   // Start executing instructions.
1906   __ dispatch_next(vtos);
1907   // Dispatch_next does not return.
1908   DEBUG_ONLY(__ should_not_reach_here());
1909 
1910   // Invocation counter overflow.
1911   if (inc_counter) {
1912     if (ProfileInterpreter) {
1913       // We have decided to profile this method in the interpreter.
1914       __ bind(profile_method);
1915 
1916       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1917       __ set_method_data_pointer_for_bcp();
1918       __ z_bru(profile_method_continue);
1919     }
1920 
1921     // Handle invocation counter overflow.
1922     __ bind(invocation_counter_overflow);
1923     generate_counter_overflow(Lcontinue);
1924   }
1925 
1926   BLOCK_COMMENT("} normal_entry");
1927 
1928   return entry_point;
1929 }
1930 
1931 // Method entry for static native methods:
1932 //   int java.util.zip.CRC32.update(int crc, int b)
1933 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1934 
1935   if (UseCRC32Intrinsics) {
1936     uint64_t entry_off = __ offset();
1937     Label    slow_path;
1938 
1939     // If we need a safepoint check, generate full interpreter entry.
1940     __ generate_safepoint_check(slow_path, Z_R1, false);
1941 
1942     BLOCK_COMMENT("CRC32_update {");
1943 
1944     // We don't generate local frame and don't align stack because
1945     // we not even call stub code (we generate the code inline)
1946     // and there is no safepoint on this path.
1947 
1948     // Load java parameters.
1949     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1950     const Register argP    = Z_esp;
1951     const Register crc     = Z_ARG1;  // crc value
1952     const Register data    = Z_ARG2;  // address of java byte value (kernel_crc32 needs address)
1953     const Register dataLen = Z_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1954     const Register table   = Z_ARG4;  // address of crc32 table
1955 
1956     // Arguments are reversed on java expression stack.
1957     __ z_la(data, 3+1*wordSize, argP);  // byte value (stack address).
1958                                         // Being passed as an int, the single byte is at offset +3.
1959     __ z_llgf(crc, 2 * wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1960 
1961     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1962     __ kernel_crc32_singleByte(crc, data, dataLen, table, Z_R1);
1963 
1964     // Restore caller sp for c2i case.
1965     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1966 
1967     __ z_br(Z_R14);
1968 
1969     BLOCK_COMMENT("} CRC32_update");
1970 
1971     // Use a previously generated vanilla native entry as the slow path.
1972     BIND(slow_path);
1973     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1974     return __ addr_at(entry_off);
1975   }
1976 
1977   return NULL;
1978 }
1979 
1980 
1981 // Method entry for static native methods:
1982 //   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1983 //   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1984 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1985 
1986   if (UseCRC32Intrinsics) {
1987     uint64_t entry_off = __ offset();
1988     Label    slow_path;
1989 
1990     // If we need a safepoint check, generate full interpreter entry.
1991     __ generate_safepoint_check(slow_path, Z_R1, false);
1992 
1993     // We don't generate local frame and don't align stack because
1994     // we call stub code and there is no safepoint on this path.
1995 
1996     // Load parameters.
1997     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1998     const Register argP    = Z_esp;
1999     const Register crc     = Z_ARG1;  // crc value
2000     const Register data    = Z_ARG2;  // address of java byte array
2001     const Register dataLen = Z_ARG3;  // source data len
2002     const Register table   = Z_ARG4;  // address of crc32 table
2003     const Register t0      = Z_R10;   // work reg for kernel* emitters
2004     const Register t1      = Z_R11;   // work reg for kernel* emitters
2005     const Register t2      = Z_R12;   // work reg for kernel* emitters
2006     const Register t3      = Z_R13;   // work reg for kernel* emitters
2007 
2008     // Arguments are reversed on java expression stack.
2009     // Calculate address of start element.
2010     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
2011       // crc     @ (SP + 5W) (32bit)
2012       // buf     @ (SP + 3W) (64bit ptr to long array)
2013       // off     @ (SP + 2W) (32bit)
2014       // dataLen @ (SP + 1W) (32bit)
2015       // data = buf + off
2016       BLOCK_COMMENT("CRC32_updateByteBuffer {");
2017       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
2018       __ z_lg(data,    3*wordSize, argP);   // start of byte buffer
2019       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2020       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2021     } else {                         // Used for "updateBytes update".
2022       // crc     @ (SP + 4W) (32bit)
2023       // buf     @ (SP + 3W) (64bit ptr to byte array)
2024       // off     @ (SP + 2W) (32bit)
2025       // dataLen @ (SP + 1W) (32bit)
2026       // data = buf + off + base_offset
2027       BLOCK_COMMENT("CRC32_updateBytes {");
2028       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
2029       __ z_lg(data,    3*wordSize, argP);   // start of byte buffer
2030       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2031       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2032       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2033     }
2034 
2035     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
2036 
2037     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2038     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
2039     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3);
2040     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
2041 
2042     // Restore caller sp for c2i case.
2043     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2044 
2045     __ z_br(Z_R14);
2046 
2047     BLOCK_COMMENT("} CRC32_update{Bytes|ByteBuffer}");
2048 
2049     // Use a previously generated vanilla native entry as the slow path.
2050     BIND(slow_path);
2051     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
2052     return __ addr_at(entry_off);
2053   }
2054 
2055   return NULL;
2056 }
2057 
2058 // Not supported
2059 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
2060   return NULL;
2061 }
2062 
2063 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
2064   // Quick & dirty stack overflow checking: bang the stack & handle trap.
2065   // Note that we do the banging after the frame is setup, since the exception
2066   // handling code expects to find a valid interpreter frame on the stack.
2067   // Doing the banging earlier fails if the caller frame is not an interpreter
2068   // frame.
2069   // (Also, the exception throwing code expects to unlock any synchronized
2070   // method receiver, so do the banging after locking the receiver.)
2071 
2072   // Bang each page in the shadow zone. We can't assume it's been done for
2073   // an interpreter frame with greater than a page of locals, so each page
2074   // needs to be checked. Only true for non-native. For native, we only bang the last page.
2075   if (UseStackBanging) {
2076     const int page_size      = os::vm_page_size();
2077     const int n_shadow_pages = (int)(JavaThread::stack_shadow_zone_size()/page_size);
2078     const int start_page_num = native_call ? n_shadow_pages : 1;
2079     for (int pages = start_page_num; pages <= n_shadow_pages; pages++) {
2080       __ bang_stack_with_offset(pages*page_size);
2081     }
2082   }
2083 }
2084 
2085 //-----------------------------------------------------------------------------
2086 // Exceptions
2087 
2088 void TemplateInterpreterGenerator::generate_throw_exception() {
2089 
2090   BLOCK_COMMENT("throw_exception {");
2091 
2092   // Entry point in previous activation (i.e., if the caller was interpreted).
2093   Interpreter::_rethrow_exception_entry = __ pc();
2094   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Frame accessors use Z_fp.
2095   // Z_ARG1 (==Z_tos): exception
2096   // Z_ARG2          : Return address/pc that threw exception.
2097   __ restore_bcp();    // R13 points to call/send.
2098   __ restore_locals();
2099 
2100   // Fallthrough, no need to restore Z_esp.
2101 
2102   // Entry point for exceptions thrown within interpreter code.
2103   Interpreter::_throw_exception_entry = __ pc();
2104   // Expression stack is undefined here.
2105   // Z_ARG1 (==Z_tos): exception
2106   // Z_bcp: exception bcp
2107   __ verify_oop(Z_ARG1);
2108   __ z_lgr(Z_ARG2, Z_ARG1);
2109 
2110   // Expression stack must be empty before entering the VM in case of
2111   // an exception.
2112   __ empty_expression_stack();
2113   // Find exception handler address and preserve exception oop.
2114   const Register Rpreserved_exc_oop = Z_tmp_1;
2115   __ call_VM(Rpreserved_exc_oop,
2116              CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception),
2117              Z_ARG2);
2118   // Z_RET: exception handler entry point
2119   // Z_bcp: bcp for exception handler
2120   __ push_ptr(Rpreserved_exc_oop); // Push exception which is now the only value on the stack.
2121   __ z_br(Z_RET); // Jump to exception handler (may be _remove_activation_entry!).
2122 
2123   // If the exception is not handled in the current frame the frame is
2124   // removed and the exception is rethrown (i.e. exception
2125   // continuation is _rethrow_exception).
2126   //
2127   // Note: At this point the bci is still the bci for the instruction
2128   // which caused the exception and the expression stack is
2129   // empty. Thus, for any VM calls at this point, GC will find a legal
2130   // oop map (with empty expression stack).
2131 
2132   //
2133   // JVMTI PopFrame support
2134   //
2135 
2136   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2137   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2138   __ empty_expression_stack();
2139   // Set the popframe_processing bit in pending_popframe_condition
2140   // indicating that we are currently handling popframe, so that
2141   // call_VMs that may happen later do not trigger new popframe
2142   // handling cycles.
2143   __ load_sized_value(Z_tmp_1, Address(Z_thread, JavaThread::popframe_condition_offset()), 4, false /*signed*/);
2144   __ z_oill(Z_tmp_1, JavaThread::popframe_processing_bit);
2145   __ z_sty(Z_tmp_1, thread_(popframe_condition));
2146 
2147   {
2148     // Check to see whether we are returning to a deoptimized frame.
2149     // (The PopFrame call ensures that the caller of the popped frame is
2150     // either interpreted or compiled and deoptimizes it if compiled.)
2151     // In this case, we can't call dispatch_next() after the frame is
2152     // popped, but instead must save the incoming arguments and restore
2153     // them after deoptimization has occurred.
2154     //
2155     // Note that we don't compare the return PC against the
2156     // deoptimization blob's unpack entry because of the presence of
2157     // adapter frames in C2.
2158     NearLabel caller_not_deoptimized;
2159     __ z_lg(Z_ARG1, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2160     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), Z_ARG1);
2161     __ compareU64_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, caller_not_deoptimized);
2162 
2163     // Compute size of arguments for saving when returning to
2164     // deoptimized caller.
2165     __ get_method(Z_ARG2);
2166     __ z_lg(Z_ARG2, Address(Z_ARG2, Method::const_offset()));
2167     __ z_llgh(Z_ARG2, Address(Z_ARG2, ConstMethod::size_of_parameters_offset()));
2168     __ z_sllg(Z_ARG2, Z_ARG2, Interpreter::logStackElementSize); // slots 2 bytes
2169     __ restore_locals();
2170     // Compute address of args to be saved.
2171     __ z_lgr(Z_ARG3, Z_locals);
2172     __ z_slgr(Z_ARG3, Z_ARG2);
2173     __ add2reg(Z_ARG3, wordSize);
2174     // Save these arguments.
2175     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args),
2176                     Z_thread, Z_ARG2, Z_ARG3);
2177 
2178     __ remove_activation(vtos, Z_R14,
2179                          /* throw_monitor_exception */ false,
2180                          /* install_monitor_exception */ false,
2181                          /* notify_jvmdi */ false);
2182 
2183     // Inform deoptimization that it is responsible for restoring
2184     // these arguments.
2185     __ store_const(thread_(popframe_condition),
2186                    JavaThread::popframe_force_deopt_reexecution_bit,
2187                    Z_tmp_1, false);
2188 
2189     // Continue in deoptimization handler.
2190     __ z_br(Z_R14);
2191 
2192     __ bind(caller_not_deoptimized);
2193   }
2194 
2195   // Clear the popframe condition flag.
2196   __ clear_mem(thread_(popframe_condition), sizeof(int));
2197 
2198   __ remove_activation(vtos,
2199                        noreg,  // Retaddr is not used.
2200                        false,  // throw_monitor_exception
2201                        false,  // install_monitor_exception
2202                        false); // notify_jvmdi
2203   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2204   __ restore_bcp();
2205   __ restore_locals();
2206   __ restore_esp();
2207   // The method data pointer was incremented already during
2208   // call profiling. We have to restore the mdp for the current bcp.
2209   if (ProfileInterpreter) {
2210     __ set_method_data_pointer_for_bcp();
2211   }
2212 #if INCLUDE_JVMTI
2213   {
2214     Label L_done;
2215 
2216     __ z_cli(0, Z_bcp, Bytecodes::_invokestatic);
2217     __ z_brc(Assembler::bcondNotEqual, L_done);
2218 
2219     // The member name argument must be restored if _invokestatic is
2220     // re-executed after a PopFrame call.  Detect such a case in the
2221     // InterpreterRuntime function and return the member name
2222     // argument, or NULL.
2223     __ z_lg(Z_ARG2, Address(Z_locals));
2224     __ get_method(Z_ARG3);
2225     __ call_VM(Z_tmp_1,
2226                CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
2227                Z_ARG2, Z_ARG3, Z_bcp);
2228 
2229     __ z_ltgr(Z_tmp_1, Z_tmp_1);
2230     __ z_brc(Assembler::bcondEqual, L_done);
2231 
2232     __ z_stg(Z_tmp_1, Address(Z_esp, wordSize));
2233     __ bind(L_done);
2234   }
2235 #endif // INCLUDE_JVMTI
2236   __ dispatch_next(vtos);
2237   // End of PopFrame support.
2238   Interpreter::_remove_activation_entry = __ pc();
2239 
2240   // In between activations - previous activation type unknown yet
2241   // compute continuation point - the continuation point expects the
2242   // following registers set up:
2243   //
2244   // Z_ARG1 (==Z_tos): exception
2245   // Z_ARG2          : return address/pc that threw exception
2246 
2247   Register return_pc = Z_tmp_1;
2248   Register handler   = Z_tmp_2;
2249    assert(return_pc->is_nonvolatile(), "use non-volatile reg. to preserve exception pc");
2250    assert(handler->is_nonvolatile(),   "use non-volatile reg. to handler pc");
2251   __ asm_assert_ijava_state_magic(return_pc/*tmp*/); // The top frame should be an interpreter frame.
2252   __ z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2253 
2254   // Moved removing the activation after VM call, because the new top
2255   // frame does not necessarily have the z_abi_160 required for a VM
2256   // call (e.g. if it is compiled).
2257 
2258   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2259                                          SharedRuntime::exception_handler_for_return_address),
2260                         Z_thread, return_pc);
2261   __ z_lgr(handler, Z_RET); // Save exception handler.
2262 
2263   // Preserve exception over this code sequence.
2264   __ pop_ptr(Z_ARG1);
2265   __ set_vm_result(Z_ARG1);
2266   // Remove the activation (without doing throws on illegalMonitorExceptions).
2267   __ remove_activation(vtos, noreg/*ret.pc already loaded*/, false/*throw exc*/, true/*install exc*/, false/*notify jvmti*/);
2268   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2269 
2270   __ get_vm_result(Z_ARG1);     // Restore exception.
2271   __ verify_oop(Z_ARG1);
2272   __ z_lgr(Z_ARG2, return_pc);  // Restore return address.
2273 
2274 #ifdef ASSERT
2275   // The return_pc in the new top frame is dead... at least that's my
2276   // current understanding. To assert this I overwrite it.
2277   // Note: for compiled frames the handler is the deopt blob
2278   // which writes Z_ARG2 into the return_pc slot.
2279   __ load_const_optimized(return_pc, 0xb00b1);
2280   __ z_stg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2281 #endif
2282 
2283   // Z_ARG1 (==Z_tos): exception
2284   // Z_ARG2          : return address/pc that threw exception
2285 
2286   // Note that an "issuing PC" is actually the next PC after the call.
2287   __ z_br(handler);         // Jump to exception handler of caller.
2288 
2289   BLOCK_COMMENT("} throw_exception");
2290 }
2291 
2292 //
2293 // JVMTI ForceEarlyReturn support
2294 //
2295 address TemplateInterpreterGenerator::generate_earlyret_entry_for (TosState state) {
2296   address entry = __ pc();
2297 
2298   BLOCK_COMMENT("earlyret_entry {");
2299 
2300   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2301   __ restore_bcp();
2302   __ restore_locals();
2303   __ restore_esp();
2304   __ empty_expression_stack();
2305   __ load_earlyret_value(state);
2306 
2307   Register RjvmtiState = Z_tmp_1;
2308   __ z_lg(RjvmtiState, thread_(jvmti_thread_state));
2309   __ store_const(Address(RjvmtiState, JvmtiThreadState::earlyret_state_offset()),
2310                  JvmtiThreadState::earlyret_inactive, 4, 4, Z_R0_scratch);
2311 
2312   __ remove_activation(state,
2313                        Z_tmp_1, // retaddr
2314                        false,   // throw_monitor_exception
2315                        false,   // install_monitor_exception
2316                        true);   // notify_jvmdi
2317   __ z_br(Z_tmp_1);
2318 
2319   BLOCK_COMMENT("} earlyret_entry");
2320 
2321   return entry;
2322 }
2323 
2324 //-----------------------------------------------------------------------------
2325 // Helper for vtos entry point generation.
2326 
2327 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2328                                                          address& bep,
2329                                                          address& cep,
2330                                                          address& sep,
2331                                                          address& aep,
2332                                                          address& iep,
2333                                                          address& lep,
2334                                                          address& fep,
2335                                                          address& dep,
2336                                                          address& vep) {
2337   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2338   Label L;
2339   aep = __ pc(); __ push_ptr(); __ z_bru(L);
2340   fep = __ pc(); __ push_f();   __ z_bru(L);
2341   dep = __ pc(); __ push_d();   __ z_bru(L);
2342   lep = __ pc(); __ push_l();   __ z_bru(L);
2343   bep = cep = sep =
2344   iep = __ pc(); __ push_i();
2345   vep = __ pc();
2346   __ bind(L);
2347   generate_and_dispatch(t);
2348 }
2349 
2350 //-----------------------------------------------------------------------------
2351 
2352 #ifndef PRODUCT
2353 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2354   address entry = __ pc();
2355   NearLabel counter_below_trace_threshold;
2356 
2357   if (TraceBytecodesAt > 0) {
2358     // Skip runtime call, if the trace threshold is not yet reached.
2359     __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2360     __ load_absolute_address(Z_tmp_2, (address)&TraceBytecodesAt);
2361     __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2362     __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2363     __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, counter_below_trace_threshold);
2364   }
2365 
2366   int offset2 = state == ltos || state == dtos ? 2 : 1;
2367 
2368   __ push(state);
2369   // Preserved return pointer is in Z_R14.
2370   // InterpreterRuntime::trace_bytecode() preserved and returns the value passed as second argument.
2371   __ z_lgr(Z_ARG2, Z_R14);
2372   __ z_lg(Z_ARG3, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)));
2373   if (WizardMode) {
2374     __ z_lgr(Z_ARG4, Z_esp); // Trace Z_esp in WizardMode.
2375   } else {
2376     __ z_lg(Z_ARG4, Address(Z_esp, Interpreter::expr_offset_in_bytes(offset2)));
2377   }
2378   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), Z_ARG2, Z_ARG3, Z_ARG4);
2379   __ z_lgr(Z_R14, Z_RET); // Estore return address (see above).
2380   __ pop(state);
2381 
2382   __ bind(counter_below_trace_threshold);
2383   __ z_br(Z_R14); // return
2384 
2385   return entry;
2386 }
2387 
2388 // Make feasible for old CPUs.
2389 void TemplateInterpreterGenerator::count_bytecode() {
2390   __ load_absolute_address(Z_R1_scratch, (address) &BytecodeCounter::_counter_value);
2391   __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2392 }
2393 
2394 void TemplateInterpreterGenerator::histogram_bytecode(Template * t) {
2395   __ load_absolute_address(Z_R1_scratch, (address)&BytecodeHistogram::_counters[ t->bytecode() ]);
2396   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2397 }
2398 
2399 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template * t) {
2400   Address  index_addr(Z_tmp_1, (intptr_t) 0);
2401   Register index = Z_tmp_2;
2402 
2403   // Load previous index.
2404   __ load_absolute_address(Z_tmp_1, (address) &BytecodePairHistogram::_index);
2405   __ mem2reg_opt(index, index_addr, false);
2406 
2407   // Mask with current bytecode and store as new previous index.
2408   __ z_srl(index, BytecodePairHistogram::log2_number_of_codes);
2409   __ load_const_optimized(Z_R0_scratch,
2410                           (int)t->bytecode() << BytecodePairHistogram::log2_number_of_codes);
2411   __ z_or(index, Z_R0_scratch);
2412   __ reg2mem_opt(index, index_addr, false);
2413 
2414   // Load counter array's address.
2415   __ z_lgfr(index, index);   // Sign extend for addressing.
2416   __ z_sllg(index, index, LogBytesPerInt);  // index2bytes
2417   __ load_absolute_address(Z_R1_scratch,
2418                            (address) &BytecodePairHistogram::_counters);
2419   // Add index and increment counter.
2420   __ z_agr(Z_R1_scratch, index);
2421   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2422 }
2423 
2424 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2425   // Call a little run-time stub to avoid blow-up for each bytecode.
2426   // The run-time runtime saves the right registers, depending on
2427   // the tosca in-state for the given template.
2428   address entry = Interpreter::trace_code(t->tos_in());
2429   guarantee(entry != NULL, "entry must have been generated");
2430   __ call_stub(entry);
2431 }
2432 
2433 void TemplateInterpreterGenerator::stop_interpreter_at() {
2434   NearLabel L;
2435 
2436   __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2437   __ load_absolute_address(Z_tmp_2, (address)&StopInterpreterAt);
2438   __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2439   __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2440   __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, L);
2441   assert(Z_tmp_1->is_nonvolatile(), "must be nonvolatile to preserve Z_tos");
2442   assert(Z_F8->is_nonvolatile(), "must be nonvolatile to preserve Z_ftos");
2443   __ z_lgr(Z_tmp_1, Z_tos);      // Save tos.
2444   __ z_lgr(Z_tmp_2, Z_bytecode); // Save Z_bytecode.
2445   __ z_ldr(Z_F8, Z_ftos);        // Save ftos.
2446   // Use -XX:StopInterpreterAt=<num> to set the limit
2447   // and break at breakpoint().
2448   __ call_VM(noreg, CAST_FROM_FN_PTR(address, breakpoint), false);
2449   __ z_lgr(Z_tos, Z_tmp_1);      // Restore tos.
2450   __ z_lgr(Z_bytecode, Z_tmp_2); // Save Z_bytecode.
2451   __ z_ldr(Z_ftos, Z_F8);        // Restore ftos.
2452   __ bind(L);
2453 }
2454 
2455 #endif // !PRODUCT