1 /*
   2  * Copyright (c) 2016, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2017, SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #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   __ push_frame_abi160(4*BytesPerWord);                 // Reserve space to save the tmp_[1..4] registers.
 125   __ z_stmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // Save registers only after frame is pushed.
 126 
 127   __ z_lgr(arg_java, Z_ARG1);
 128 
 129   Register   method = Z_ARG2; // Directly load into correct argument register.
 130 
 131   __ get_method(method);
 132   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), Z_thread, method);
 133 
 134   // Move signature to callee saved register.
 135   // Don't directly write to stack. Frame is used by VM call.
 136   __ z_lgr(Z_tmp_1, Z_RET);
 137 
 138   // Reload method. Register may have been altered by VM call.
 139   __ get_method(method);
 140 
 141   // Get address of result handler.
 142   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), Z_thread, method);
 143 
 144   // Save signature address to stack.
 145   __ z_stg(Z_tmp_1, d_signature, Z_SP);
 146 
 147   // Don't overwrite return value (Z_RET, Z_ARG1) in rest of the method !
 148 
 149   {
 150     Label   isStatic;
 151 
 152     // Test if static.
 153     // We can test the bit directly.
 154     // Path is Z_method->_access_flags._flags.
 155     // We only support flag bits in the least significant byte (assert !).
 156     // Therefore add 3 to address that byte within "_flags".
 157     // Reload method. VM call above may have destroyed register contents
 158     __ get_method(method);
 159     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 160     method = noreg;  // end of life
 161     __ z_btrue(isStatic);
 162 
 163     // For non-static functions, pass "this" in Z_ARG2 and copy it to 2nd C-arg slot.
 164     // Need to box the Java object here, so we use arg_java
 165     // (address of current Java stack slot) as argument and
 166     // don't dereference it as in case of ints, floats, etc..
 167     __ z_lgr(Z_ARG2, arg_java);
 168     __ add2reg(arg_java, -BytesPerWord);
 169     __ bind(isStatic);
 170   }
 171 
 172   // argcnt == 0 corresponds to 3rd C argument.
 173   //   arg #1 (result handler) and
 174   //   arg #2 (this, for non-statics), unused else
 175   // are reserved and pre-filled above.
 176   // arg_java points to the corresponding Java argument here. It
 177   // has been decremented by one argument (this) in case of non-static.
 178   __ clear_reg(argcnt, true, false);  // Don't set CC.
 179   __ z_lg(target_sp, 0, Z_SP);
 180   __ add2reg(arg_c, _z_abi(remaining_cargs), target_sp);
 181   // No floating-point args parsed so far.
 182   __ clear_mem(Address(Z_SP, d_fpcnt), 8);
 183 
 184   NearLabel   move_intSlot_to_ARG, move_floatSlot_to_FARG;
 185   NearLabel   loop_start, loop_start_restore, loop_end;
 186   NearLabel   do_int, do_long, do_float, do_double;
 187   NearLabel   do_dontreachhere, do_object, do_array, do_boxed;
 188 
 189 #ifdef ASSERT
 190   // Signature needs to point to '(' (== 0x28) at entry.
 191   __ z_lg(signature, d_signature, Z_SP);
 192   __ z_cli(0, signature, (int) '(');
 193   __ z_brne(do_dontreachhere);
 194 #endif
 195 
 196   __ bind(loop_start_restore);
 197   __ z_lg(signature, d_signature, Z_SP);  // Restore signature ptr, destroyed by move_XX_to_ARG.
 198 
 199   BIND(loop_start);
 200   // Advance to next argument type token from the signature.
 201   __ add2reg(signature, 1);
 202 
 203   // Use CLI, works well on all CPU versions.
 204     __ z_cli(0, signature, (int) ')');
 205     __ z_bre(loop_end);                // end of signature
 206     __ z_cli(0, signature, (int) 'L');
 207     __ z_bre(do_object);               // object     #9
 208     __ z_cli(0, signature, (int) 'F');
 209     __ z_bre(do_float);                // float      #7
 210     __ z_cli(0, signature, (int) 'J');
 211     __ z_bre(do_long);                 // long       #6
 212     __ z_cli(0, signature, (int) 'B');
 213     __ z_bre(do_int);                  // byte       #1
 214     __ z_cli(0, signature, (int) 'Z');
 215     __ z_bre(do_int);                  // boolean    #2
 216     __ z_cli(0, signature, (int) 'C');
 217     __ z_bre(do_int);                  // char       #3
 218     __ z_cli(0, signature, (int) 'S');
 219     __ z_bre(do_int);                  // short      #4
 220     __ z_cli(0, signature, (int) 'I');
 221     __ z_bre(do_int);                  // int        #5
 222     __ z_cli(0, signature, (int) 'D');
 223     __ z_bre(do_double);               // double     #8
 224     __ z_cli(0, signature, (int) '[');
 225     __ z_bre(do_array);                // array      #10
 226 
 227   __ bind(do_dontreachhere);
 228 
 229   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 230 
 231   // Array argument
 232   BIND(do_array);
 233 
 234   {
 235     Label   start_skip, end_skip;
 236 
 237     __ bind(start_skip);
 238 
 239     // Advance to next type tag from signature.
 240     __ add2reg(signature, 1);
 241 
 242     // Use CLI, works well on all CPU versions.
 243     __ z_cli(0, signature, (int) '[');
 244     __ z_bre(start_skip);               // Skip further brackets.
 245 
 246     __ z_cli(0, signature, (int) '9');
 247     __ z_brh(end_skip);                 // no optional size
 248 
 249     __ z_cli(0, signature, (int) '0');
 250     __ z_brnl(start_skip);              // Skip optional size.
 251 
 252     __ bind(end_skip);
 253 
 254     __ z_cli(0, signature, (int) 'L');
 255     __ z_brne(do_boxed);                // If not array of objects: go directly to do_boxed.
 256   }
 257 
 258   //  OOP argument
 259   BIND(do_object);
 260   // Pass by an object's type name.
 261   {
 262     Label   L;
 263 
 264     __ add2reg(sig_end, 4095, signature);     // Assume object type name is shorter than 4k.
 265     __ load_const_optimized(Z_R0, (int) ';'); // Type name terminator (must be in Z_R0!).
 266     __ MacroAssembler::search_string(sig_end, signature);
 267     __ z_brl(L);
 268     __ z_illtrap();  // No semicolon found: internal error or object name too long.
 269     __ bind(L);
 270     __ z_lgr(signature, sig_end);
 271     // fallthru to do_boxed
 272   }
 273 
 274   // Need to box the Java object here, so we use arg_java
 275   // (address of current Java stack slot) as argument and
 276   // don't dereference it as in case of ints, floats, etc..
 277 
 278   // UNBOX argument
 279   // Load reference and check for NULL.
 280   Label  do_int_Entry4Boxed;
 281   __ bind(do_boxed);
 282   {
 283     __ load_and_test_long(intSlot, Address(arg_java));
 284     __ z_bre(do_int_Entry4Boxed);
 285     __ z_lgr(intSlot, arg_java);
 286     __ z_bru(do_int_Entry4Boxed);
 287   }
 288 
 289   // INT argument
 290 
 291   // (also for byte, boolean, char, short)
 292   // Use lgf for load (sign-extend) and stg for store.
 293   BIND(do_int);
 294   __ z_lgf(intSlot, 0, arg_java);
 295 
 296   __ bind(do_int_Entry4Boxed);
 297   __ add2reg(arg_java, -BytesPerWord);
 298   // If argument fits into argument register, go and handle it, otherwise continue.
 299   __ compare32_and_branch(argcnt, max_int_register_arguments,
 300                           Assembler::bcondLow, move_intSlot_to_ARG);
 301   __ z_stg(intSlot, 0, arg_c);
 302   __ add2reg(arg_c, BytesPerWord);
 303   __ z_bru(loop_start);
 304 
 305   // LONG argument
 306 
 307   BIND(do_long);
 308   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 309   __ z_lg(intSlot, BytesPerWord, arg_java);
 310   // If argument fits into argument register, go and handle it, otherwise continue.
 311   __ compare32_and_branch(argcnt, max_int_register_arguments,
 312                           Assembler::bcondLow, move_intSlot_to_ARG);
 313   __ z_stg(intSlot, 0, arg_c);
 314   __ add2reg(arg_c, BytesPerWord);
 315   __ z_bru(loop_start);
 316 
 317   // FLOAT argumen
 318 
 319   BIND(do_float);
 320   __ z_le(floatSlot, 0, arg_java);
 321   __ add2reg(arg_java, -BytesPerWord);
 322   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 323   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 324   __ z_brl(move_floatSlot_to_FARG);
 325   __ z_ste(floatSlot, 4, arg_c);
 326   __ add2reg(arg_c, BytesPerWord);
 327   __ z_bru(loop_start);
 328 
 329   // DOUBLE argument
 330 
 331   BIND(do_double);
 332   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 333   __ z_ld(floatSlot, BytesPerWord, arg_java);
 334   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 335   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 336   __ z_brl(move_floatSlot_to_FARG);
 337   __ z_std(floatSlot, 0, arg_c);
 338   __ add2reg(arg_c, BytesPerWord);
 339   __ z_bru(loop_start);
 340 
 341   // Method exit, all arguments proocessed.
 342   __ bind(loop_end);
 343   __ z_lmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // restore registers before frame is popped.
 344   __ pop_frame();
 345   __ restore_return_pc();
 346   __ z_br(Z_R14);
 347 
 348   // Copy int arguments.
 349 
 350   Label  iarg_caselist;   // Distance between each case has to be a power of 2
 351                           // (= 1 << LogSizeOfCase).
 352   __ align(16);
 353   BIND(iarg_caselist);
 354   __ z_lgr(Z_ARG3, intSlot);    // 4 bytes
 355   __ z_bru(loop_start_restore); // 4 bytes
 356 
 357   __ z_lgr(Z_ARG4, intSlot);
 358   __ z_bru(loop_start_restore);
 359 
 360   __ z_lgr(Z_ARG5, intSlot);
 361   __ z_bru(loop_start_restore);
 362 
 363   __ align(16);
 364   __ bind(move_intSlot_to_ARG);
 365   __ z_stg(signature, d_signature, Z_SP);       // Spill since signature == Z_R1_scratch.
 366   __ z_larl(Z_R1_scratch, iarg_caselist);
 367   __ z_sllg(Z_R0_scratch, argcnt, LogSizeOfCase);
 368   __ add2reg(argcnt, 1);
 369   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 370   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 371 
 372   // Copy float arguments.
 373 
 374   Label  farg_caselist;   // Distance between each case has to be a power of 2
 375                           // (= 1 << logSizeOfCase, padded with nop.
 376   __ align(16);
 377   BIND(farg_caselist);
 378   __ z_ldr(Z_FARG1, floatSlot); // 2 bytes
 379   __ z_bru(loop_start_restore); // 4 bytes
 380   __ z_nop();                   // 2 bytes
 381 
 382   __ z_ldr(Z_FARG2, floatSlot);
 383   __ z_bru(loop_start_restore);
 384   __ z_nop();
 385 
 386   __ z_ldr(Z_FARG3, floatSlot);
 387   __ z_bru(loop_start_restore);
 388   __ z_nop();
 389 
 390   __ z_ldr(Z_FARG4, floatSlot);
 391   __ z_bru(loop_start_restore);
 392   __ z_nop();
 393 
 394   __ align(16);
 395   __ bind(move_floatSlot_to_FARG);
 396   __ z_stg(signature, d_signature, Z_SP);        // Spill since signature == Z_R1_scratch.
 397   __ z_lg(Z_R0_scratch, d_fpcnt, Z_SP);          // Need old value for indexing.
 398   __ add2mem_64(Address(Z_SP, d_fpcnt), 1, Z_R1_scratch); // Increment index.
 399   __ z_larl(Z_R1_scratch, farg_caselist);
 400   __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogSizeOfCase);
 401   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 402   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 403 
 404   BLOCK_COMMENT("} slow_signature_handler");
 405 
 406   return __ addr_at(entry_offset);
 407 }
 408 
 409 address TemplateInterpreterGenerator::generate_result_handler_for (BasicType type) {
 410   address entry = __ pc();
 411 
 412   assert(Z_tos == Z_RET, "Result handler: must move result!");
 413   assert(Z_ftos == Z_FRET, "Result handler: must move float result!");
 414 
 415   switch (type) {
 416     case T_BOOLEAN:
 417       __ c2bool(Z_tos);
 418       break;
 419     case T_CHAR:
 420       __ and_imm(Z_tos, 0xffff);
 421       break;
 422     case T_BYTE:
 423       __ z_lbr(Z_tos, Z_tos);
 424       break;
 425     case T_SHORT:
 426       __ z_lhr(Z_tos, Z_tos);
 427       break;
 428     case T_INT:
 429     case T_LONG:
 430     case T_VOID:
 431     case T_FLOAT:
 432     case T_DOUBLE:
 433       break;
 434     case T_OBJECT:
 435       // Retrieve result from frame...
 436       __ mem2reg_opt(Z_tos, Address(Z_fp, oop_tmp_offset));
 437       // and verify it.
 438       __ verify_oop(Z_tos);
 439       break;
 440     default:
 441       ShouldNotReachHere();
 442   }
 443   __ z_br(Z_R14);      // Return from result handler.
 444   return entry;
 445 }
 446 
 447 // Abstract method entry.
 448 // Attempt to execute abstract method. Throw exception.
 449 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 450   unsigned int entry_offset = __ offset();
 451 
 452   // Caller could be the call_stub or a compiled method (x86 version is wrong!).
 453 
 454   BLOCK_COMMENT("abstract_entry {");
 455 
 456   // Implement call of InterpreterRuntime::throw_AbstractMethodError.
 457   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1);
 458   __ save_return_pc();       // Save Z_R14.
 459   __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
 460 
 461   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 462                   Z_thread, Z_method);
 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 continuation) {
 693   address entry = __ pc();
 694 
 695   BLOCK_COMMENT("deopt_entry {");
 696 
 697   // TODO(ZASM): necessary? NULL last_sp until next java call
 698   // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 699   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 700   __ restore_bcp();
 701   __ restore_locals();
 702   __ restore_esp();
 703 
 704   // Handle exceptions.
 705   {
 706     Label L;
 707     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
 708     __ z_bre(L);
 709     __ call_VM(noreg,
 710                CAST_FROM_FN_PTR(address,
 711                                 InterpreterRuntime::throw_pending_exception));
 712     __ should_not_reach_here();
 713     __ bind(L);
 714   }
 715   if (continuation == NULL) {
 716     __ dispatch_next(state, step);
 717   } else {
 718     __ jump_to_entry(continuation, Z_R1_scratch);
 719   }
 720 
 721   BLOCK_COMMENT("} deopt_entry");
 722 
 723   return entry;
 724 }
 725 
 726 address TemplateInterpreterGenerator::generate_safept_entry_for (TosState state,
 727                                                                 address runtime_entry) {
 728   address entry = __ pc();
 729   __ push(state);
 730   __ call_VM(noreg, runtime_entry);
 731   __ dispatch_via(vtos, Interpreter::_normal_table.table_for (vtos));
 732   return entry;
 733 }
 734 
 735 //
 736 // Helpers for commoning out cases in the various type of method entries.
 737 //
 738 
 739 // Increment invocation count & check for overflow.
 740 //
 741 // Note: checking for negative value instead of overflow
 742 // so we have a 'sticky' overflow test.
 743 //
 744 // Z_ARG2: method (see generate_fixed_frame())
 745 //
 746 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 747   Label done;
 748   Register method = Z_ARG2; // Generate_fixed_frame() copies Z_method into Z_ARG2.
 749   Register m_counters = Z_ARG4;
 750 
 751   BLOCK_COMMENT("counter_incr {");
 752 
 753   // Note: In tiered we increment either counters in method or in MDO depending
 754   // if we are profiling or not.
 755   if (TieredCompilation) {
 756     int increment = InvocationCounter::count_increment;
 757     if (ProfileInterpreter) {
 758       NearLabel no_mdo;
 759       Register mdo = m_counters;
 760       // Are we profiling?
 761       __ load_and_test_long(mdo, method2_(method, method_data));
 762       __ branch_optimized(Assembler::bcondZero, no_mdo);
 763       // Increment counter in the MDO.
 764       const Address mdo_invocation_counter(mdo, MethodData::invocation_counter_offset() +
 765                                            InvocationCounter::counter_offset());
 766       const Address mask(mdo, MethodData::invoke_mask_offset());
 767       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 768                                  Z_R1_scratch, false, Assembler::bcondZero,
 769                                  overflow);
 770       __ z_bru(done);
 771       __ bind(no_mdo);
 772     }
 773 
 774     // Increment counter in MethodCounters.
 775     const Address invocation_counter(m_counters,
 776                                      MethodCounters::invocation_counter_offset() +
 777                                      InvocationCounter::counter_offset());
 778     // Get address of MethodCounters object.
 779     __ get_method_counters(method, m_counters, done);
 780     const Address mask(m_counters, MethodCounters::invoke_mask_offset());
 781     __ increment_mask_and_jump(invocation_counter,
 782                                increment, mask,
 783                                Z_R1_scratch, false, Assembler::bcondZero,
 784                                overflow);
 785   } else {
 786     Register counter_sum = Z_ARG3; // The result of this piece of code.
 787     Register tmp         = Z_R1_scratch;
 788 #ifdef ASSERT
 789     {
 790       NearLabel ok;
 791       __ get_method(tmp);
 792       __ compare64_and_branch(method, tmp, Assembler::bcondEqual, ok);
 793       __ z_illtrap(0x66);
 794       __ bind(ok);
 795     }
 796 #endif
 797 
 798     // Get address of MethodCounters object.
 799     __ get_method_counters(method, m_counters, done);
 800     // Update standard invocation counters.
 801     __ increment_invocation_counter(m_counters, counter_sum);
 802     if (ProfileInterpreter) {
 803       __ add2mem_32(Address(m_counters, MethodCounters::interpreter_invocation_counter_offset()), 1, tmp);
 804       if (profile_method != NULL) {
 805         const Address profile_limit(m_counters, MethodCounters::interpreter_profile_limit_offset());
 806         __ z_cl(counter_sum, profile_limit);
 807         __ branch_optimized(Assembler::bcondLow, *profile_method_continue);
 808         // If no method data exists, go to profile_method.
 809         __ test_method_data_pointer(tmp, *profile_method);
 810       }
 811     }
 812 
 813     const Address invocation_limit(m_counters, MethodCounters::interpreter_invocation_limit_offset());
 814     __ z_cl(counter_sum, invocation_limit);
 815     __ branch_optimized(Assembler::bcondNotLow, *overflow);
 816   }
 817 
 818   __ bind(done);
 819 
 820   BLOCK_COMMENT("} counter_incr");
 821 }
 822 
 823 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 824   // InterpreterRuntime::frequency_counter_overflow takes two
 825   // arguments, the first (thread) is passed by call_VM, the second
 826   // indicates if the counter overflow occurs at a backwards branch
 827   // (NULL bcp). We pass zero for it. The call returns the address
 828   // of the verified entry point for the method or NULL if the
 829   // compilation did not complete (either went background or bailed
 830   // out).
 831   __ clear_reg(Z_ARG2);
 832   __ call_VM(noreg,
 833              CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow),
 834              Z_ARG2);
 835   __ z_bru(do_continue);
 836 }
 837 
 838 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register frame_size, Register tmp1) {
 839   Register tmp2 = Z_R1_scratch;
 840   const int page_size = os::vm_page_size();
 841   NearLabel after_frame_check;
 842 
 843   BLOCK_COMMENT("counter_overflow {");
 844 
 845   assert_different_registers(frame_size, tmp1);
 846 
 847   // Stack banging is sufficient overflow check if frame_size < page_size.
 848   if (Immediate::is_uimm(page_size, 15)) {
 849     __ z_chi(frame_size, page_size);
 850     __ z_brl(after_frame_check);
 851   } else {
 852     __ load_const_optimized(tmp1, page_size);
 853     __ compareU32_and_branch(frame_size, tmp1, Assembler::bcondLow, after_frame_check);
 854   }
 855 
 856   // Get the stack base, and in debug, verify it is non-zero.
 857   __ z_lg(tmp1, thread_(stack_base));
 858 #ifdef ASSERT
 859   address reentry = NULL;
 860   NearLabel base_not_zero;
 861   __ compareU64_and_branch(tmp1, (intptr_t)0L, Assembler::bcondNotEqual, base_not_zero);
 862   reentry = __ stop_chain_static(reentry, "stack base is zero in generate_stack_overflow_check");
 863   __ bind(base_not_zero);
 864 #endif
 865 
 866   // Get the stack size, and in debug, verify it is non-zero.
 867   assert(sizeof(size_t) == sizeof(intptr_t), "wrong load size");
 868   __ z_lg(tmp2, thread_(stack_size));
 869 #ifdef ASSERT
 870   NearLabel size_not_zero;
 871   __ compareU64_and_branch(tmp2, (intptr_t)0L, Assembler::bcondNotEqual, size_not_zero);
 872   reentry = __ stop_chain_static(reentry, "stack size is zero in generate_stack_overflow_check");
 873   __ bind(size_not_zero);
 874 #endif
 875 
 876   // Compute the beginning of the protected zone minus the requested frame size.
 877   __ z_sgr(tmp1, tmp2);
 878   __ add2reg(tmp1, JavaThread::stack_guard_zone_size());
 879 
 880   // Add in the size of the frame (which is the same as subtracting it from the
 881   // SP, which would take another register.
 882   __ z_agr(tmp1, frame_size);
 883 
 884   // The frame is greater than one page in size, so check against
 885   // the bottom of the stack.
 886   __ compareU64_and_branch(Z_SP, tmp1, Assembler::bcondHigh, after_frame_check);
 887 
 888   // The stack will overflow, throw an exception.
 889 
 890   // Restore SP to sender's sp. This is necessary if the sender's frame is an
 891   // extended compiled frame (see gen_c2i_adapter()) and safer anyway in case of
 892   // JSR292 adaptations.
 893   __ resize_frame_absolute(Z_R10, tmp1, true/*load_fp*/);
 894 
 895   // Note also that the restored frame is not necessarily interpreted.
 896   // Use the shared runtime version of the StackOverflowError.
 897   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 898   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 899   __ load_absolute_address(tmp1, StubRoutines::throw_StackOverflowError_entry());
 900   __ z_br(tmp1);
 901 
 902   // If you get to here, then there is enough stack space.
 903   __ bind(after_frame_check);
 904 
 905   BLOCK_COMMENT("} counter_overflow");
 906 }
 907 
 908 // Allocate monitor and lock method (asm interpreter).
 909 //
 910 // Args:
 911 //   Z_locals: locals
 912 
 913 void TemplateInterpreterGenerator::lock_method(void) {
 914 
 915   BLOCK_COMMENT("lock_method {");
 916 
 917   // Synchronize method.
 918   const Register method = Z_tmp_2;
 919   __ get_method(method);
 920 
 921 #ifdef ASSERT
 922   address reentry = NULL;
 923   {
 924     Label L;
 925     __ testbit(method2_(method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 926     __ z_btrue(L);
 927     reentry = __ stop_chain_static(reentry, "method doesn't need synchronization");
 928     __ bind(L);
 929   }
 930 #endif // ASSERT
 931 
 932   // Get synchronization object.
 933   const Register object = Z_tmp_2;
 934 
 935   {
 936     Label     done;
 937     Label     static_method;
 938 
 939     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 940     __ z_btrue(static_method);
 941 
 942     // non-static method: Load receiver obj from stack.
 943     __ mem2reg_opt(object, Address(Z_locals, Interpreter::local_offset_in_bytes(0)));
 944     __ z_bru(done);
 945 
 946     __ bind(static_method);
 947 
 948     // Lock the java mirror.
 949     __ load_mirror(object, method);
 950 #ifdef ASSERT
 951     {
 952       NearLabel L;
 953       __ compare64_and_branch(object, (intptr_t) 0, Assembler::bcondNotEqual, L);
 954       reentry = __ stop_chain_static(reentry, "synchronization object is NULL");
 955       __ bind(L);
 956     }
 957 #endif // ASSERT
 958 
 959     __ bind(done);
 960   }
 961 
 962   __ add_monitor_to_stack(true, Z_ARG3, Z_ARG4, Z_ARG5); // Allocate monitor elem.
 963   // Store object and lock it.
 964   __ get_monitors(Z_tmp_1);
 965   __ reg2mem_opt(object, Address(Z_tmp_1, BasicObjectLock::obj_offset_in_bytes()));
 966   __ lock_object(Z_tmp_1, object);
 967 
 968   BLOCK_COMMENT("} lock_method");
 969 }
 970 
 971 // Generate a fixed interpreter frame. This is identical setup for
 972 // interpreted methods and for native methods hence the shared code.
 973 //
 974 // Registers alive
 975 //   Z_thread   - JavaThread*
 976 //   Z_SP       - old stack pointer
 977 //   Z_method   - callee's method
 978 //   Z_esp      - parameter list (slot 'above' last param)
 979 //   Z_R14      - return pc, to be stored in caller's frame
 980 //   Z_R10      - sender sp, note: Z_tmp_1 is Z_R10!
 981 //
 982 // Registers updated
 983 //   Z_SP       - new stack pointer
 984 //   Z_esp      - callee's operand stack pointer
 985 //                points to the slot above the value on top
 986 //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
 987 //   Z_bcp      - the bytecode pointer
 988 //   Z_fp       - the frame pointer, thereby killing Z_method
 989 //   Z_ARG2     - copy of Z_method
 990 //
 991 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 992 
 993   //  stack layout
 994   //
 995   //   F1 [TOP_IJAVA_FRAME_ABI]              <-- Z_SP, Z_R10 (see note below)
 996   //      [F1's operand stack (unused)]
 997   //      [F1's outgoing Java arguments]     <-- Z_esp
 998   //      [F1's operand stack (non args)]
 999   //      [monitors]      (optional)
1000   //      [IJAVA_STATE]
1001   //
1002   //   F2 [PARENT_IJAVA_FRAME_ABI]
1003   //      ...
1004   //
1005   //  0x000
1006   //
1007   // Note: Z_R10, the sender sp, will be below Z_SP if F1 was extended by a c2i adapter.
1008 
1009   //=============================================================================
1010   // Allocate space for locals other than the parameters, the
1011   // interpreter state, monitors, and the expression stack.
1012 
1013   const Register local_count     = Z_ARG5;
1014   const Register fp              = Z_tmp_2;
1015 
1016   BLOCK_COMMENT("generate_fixed_frame {");
1017 
1018   {
1019   // local registers
1020   const Register top_frame_size  = Z_ARG2;
1021   const Register sp_after_resize = Z_ARG3;
1022   const Register max_stack       = Z_ARG4;
1023 
1024   // local_count = method->constMethod->max_locals();
1025   __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1026   __ z_llgh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1027 
1028   if (native_call) {
1029     // If we're calling a native method, we replace max_stack (which is
1030     // zero) with space for the worst-case signature handler varargs
1031     // vector, which is:
1032     //   max_stack = max(Argument::n_register_parameters, parameter_count+2);
1033     //
1034     // We add two slots to the parameter_count, one for the jni
1035     // environment and one for a possible native mirror. We allocate
1036     // space for at least the number of ABI registers, even though
1037     // InterpreterRuntime::slow_signature_handler won't write more than
1038     // parameter_count+2 words when it creates the varargs vector at the
1039     // top of the stack. The generated slow signature handler will just
1040     // load trash into registers beyond the necessary number. We're
1041     // still going to cut the stack back by the ABI register parameter
1042     // count so as to get SP+16 pointing at the ABI outgoing parameter
1043     // area, so we need to allocate at least that much even though we're
1044     // going to throw it away.
1045     //
1046 
1047     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1048     __ z_llgh(max_stack,  Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1049     __ add2reg(max_stack, 2);
1050 
1051     NearLabel passing_args_on_stack;
1052 
1053     // max_stack in bytes
1054     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1055 
1056     int argument_registers_in_bytes = Argument::n_register_parameters << LogBytesPerWord;
1057     __ compare64_and_branch(max_stack, argument_registers_in_bytes, Assembler::bcondNotLow, passing_args_on_stack);
1058 
1059     __ load_const_optimized(max_stack, argument_registers_in_bytes);
1060 
1061     __ bind(passing_args_on_stack);
1062   } else {
1063     // !native_call
1064     __ z_lg(max_stack, method_(const));
1065 
1066     // Calculate number of non-parameter locals (in slots):
1067     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1068     __ z_sh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1069 
1070     // max_stack = method->max_stack();
1071     __ z_llgh(max_stack, Address(max_stack, ConstMethod::max_stack_offset()));
1072     // max_stack in bytes
1073     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1074   }
1075 
1076   // Resize (i.e. normally shrink) the top frame F1 ...
1077   //   F1      [TOP_IJAVA_FRAME_ABI]          <-- Z_SP, Z_R10
1078   //           F1's operand stack (free)
1079   //           ...
1080   //           F1's operand stack (free)      <-- Z_esp
1081   //           F1's outgoing Java arg m
1082   //           ...
1083   //           F1's outgoing Java arg 0
1084   //           ...
1085   //
1086   //  ... into a parent frame (Z_R10 holds F1's SP before any modification, see also above)
1087   //
1088   //           +......................+
1089   //           :                      :        <-- Z_R10, saved below as F0's z_ijava_state.sender_sp
1090   //           :                      :
1091   //   F1      [PARENT_IJAVA_FRAME_ABI]        <-- Z_SP       \
1092   //           F0's non arg local                             | = delta
1093   //           ...                                            |
1094   //           F0's non arg local              <-- Z_esp      /
1095   //           F1's outgoing Java arg m
1096   //           ...
1097   //           F1's outgoing Java arg 0
1098   //           ...
1099   //
1100   // then push the new top frame F0.
1101   //
1102   //   F0      [TOP_IJAVA_FRAME_ABI]    = frame::z_top_ijava_frame_abi_size \
1103   //           [operand stack]          = max_stack                          | = top_frame_size
1104   //           [IJAVA_STATE]            = frame::z_ijava_state_size         /
1105 
1106   // sp_after_resize = Z_esp - delta
1107   //
1108   // delta = PARENT_IJAVA_FRAME_ABI + (locals_count - params_count)
1109 
1110   __ add2reg(sp_after_resize, (Interpreter::stackElementSize) - (frame::z_parent_ijava_frame_abi_size), Z_esp);
1111   __ z_sllg(Z_R0_scratch, local_count, LogBytesPerWord); // Params have already been subtracted from local_count.
1112   __ z_slgr(sp_after_resize, Z_R0_scratch);
1113 
1114   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
1115   __ add2reg(top_frame_size,
1116              frame::z_top_ijava_frame_abi_size +
1117              frame::z_ijava_state_size +
1118              frame::interpreter_frame_monitor_size() * wordSize,
1119              max_stack);
1120 
1121   if (!native_call) {
1122     // Stack overflow check.
1123     // Native calls don't need the stack size check since they have no
1124     // expression stack and the arguments are already on the stack and
1125     // we only add a handful of words to the stack.
1126     Register frame_size = max_stack; // Reuse the regiser for max_stack.
1127     __ z_lgr(frame_size, Z_SP);
1128     __ z_sgr(frame_size, sp_after_resize);
1129     __ z_agr(frame_size, top_frame_size);
1130     generate_stack_overflow_check(frame_size, fp/*tmp1*/);
1131   }
1132 
1133   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
1134   __ asm_assert_eq("killed Z_R14", 0);
1135   __ resize_frame_absolute(sp_after_resize, fp, true);
1136   __ save_return_pc(Z_R14);
1137 
1138   // ... and push the new frame F0.
1139   __ push_frame(top_frame_size, fp, true /*copy_sp*/, false);
1140   }
1141 
1142   //=============================================================================
1143   // Initialize the new frame F0: initialize interpreter state.
1144 
1145   {
1146   // locals
1147   const Register local_addr = Z_ARG4;
1148 
1149   BLOCK_COMMENT("generate_fixed_frame: initialize interpreter state {");
1150 
1151 #ifdef ASSERT
1152   // Set the magic number (using local_addr as tmp register).
1153   __ load_const_optimized(local_addr, frame::z_istate_magic_number);
1154   __ z_stg(local_addr, _z_ijava_state_neg(magic), fp);
1155 #endif
1156 
1157   // Save sender SP from F1 (i.e. before it was potentially modified by an
1158   // adapter) into F0's interpreter state. We us it as well to revert
1159   // resizing the frame above.
1160   __ z_stg(Z_R10, _z_ijava_state_neg(sender_sp), fp);
1161 
1162   // Load cp cache and save it at the and of this block.
1163   __ z_lg(Z_R1_scratch, Address(Z_method,    Method::const_offset()));
1164   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstMethod::constants_offset()));
1165   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstantPool::cache_offset_in_bytes()));
1166 
1167   // z_ijava_state->method = method;
1168   __ z_stg(Z_method, _z_ijava_state_neg(method), fp);
1169 
1170   // Point locals at the first argument. Method's locals are the
1171   // parameters on top of caller's expression stack.
1172   // Tos points past last Java argument.
1173 
1174   __ z_lg(Z_locals, Address(Z_method, Method::const_offset()));
1175   __ z_llgh(Z_locals /*parameter_count words*/,
1176             Address(Z_locals, ConstMethod::size_of_parameters_offset()));
1177   __ z_sllg(Z_locals /*parameter_count bytes*/, Z_locals /*parameter_count*/, LogBytesPerWord);
1178   __ z_agr(Z_locals, Z_esp);
1179   // z_ijava_state->locals - i*BytesPerWord points to i-th Java local (i starts at 0)
1180   // z_ijava_state->locals = Z_esp + parameter_count bytes
1181   __ z_stg(Z_locals, _z_ijava_state_neg(locals), fp);
1182 
1183   // z_ijava_state->oop_temp = NULL;
1184   __ store_const(Address(fp, oop_tmp_offset), 0);
1185 
1186   // Initialize z_ijava_state->mdx.
1187   Register Rmdp = Z_bcp;
1188   // native_call: assert that mdo == NULL
1189   const bool check_for_mdo = !native_call DEBUG_ONLY(|| native_call);
1190   if (ProfileInterpreter && check_for_mdo) {
1191     Label get_continue;
1192 
1193     __ load_and_test_long(Rmdp, method_(method_data));
1194     __ z_brz(get_continue);
1195     DEBUG_ONLY(if (native_call) __ stop("native methods don't have a mdo"));
1196     __ add2reg(Rmdp, in_bytes(MethodData::data_offset()));
1197     __ bind(get_continue);
1198   }
1199   __ z_stg(Rmdp, _z_ijava_state_neg(mdx), fp);
1200 
1201   // Initialize z_ijava_state->bcp and Z_bcp.
1202   if (native_call) {
1203     __ clear_reg(Z_bcp); // Must initialize. Will get written into frame where GC reads it.
1204   } else {
1205     __ z_lg(Z_bcp, method_(const));
1206     __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset()));
1207   }
1208   __ z_stg(Z_bcp, _z_ijava_state_neg(bcp), fp);
1209 
1210   // no monitors and empty operand stack
1211   // => z_ijava_state->monitors points to the top slot in IJAVA_STATE.
1212   // => Z_ijava_state->esp points one slot above into the operand stack.
1213   // z_ijava_state->monitors = fp - frame::z_ijava_state_size - Interpreter::stackElementSize;
1214   // z_ijava_state->esp = Z_esp = z_ijava_state->monitors;
1215   __ add2reg(Z_esp, -frame::z_ijava_state_size, fp);
1216   __ z_stg(Z_esp, _z_ijava_state_neg(monitors), fp);
1217   __ add2reg(Z_esp, -Interpreter::stackElementSize);
1218   __ z_stg(Z_esp, _z_ijava_state_neg(esp), fp);
1219 
1220   // z_ijava_state->cpoolCache = Z_R1_scratch (see load above);
1221   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(cpoolCache), fp);
1222 
1223   // Get mirror and store it in the frame as GC root for this Method*.
1224   __ load_mirror(Z_R1_scratch, Z_method);
1225   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(mirror), fp);
1226 
1227   BLOCK_COMMENT("} generate_fixed_frame: initialize interpreter state");
1228 
1229   //=============================================================================
1230   if (!native_call) {
1231     // Fill locals with 0x0s.
1232     NearLabel locals_zeroed;
1233     NearLabel doXC;
1234 
1235     // Local_count is already num_locals_slots - num_param_slots.
1236     __ compare64_and_branch(local_count, (intptr_t)0L, Assembler::bcondNotHigh, locals_zeroed);
1237 
1238     // Advance local_addr to point behind locals (creates positive incr. in loop).
1239     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1240     __ z_llgh(Z_R0_scratch, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1241     __ add2reg(Z_R0_scratch, -1);
1242 
1243     __ z_lgr(local_addr/*locals*/, Z_locals);
1244     __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogBytesPerWord);
1245     __ z_sllg(local_count, local_count, LogBytesPerWord); // Local_count are non param locals.
1246     __ z_sgr(local_addr, Z_R0_scratch);
1247 
1248     if (VM_Version::has_Prefetch()) {
1249       __ z_pfd(0x02, 0, Z_R0, local_addr);
1250       __ z_pfd(0x02, 256, Z_R0, local_addr);
1251     }
1252 
1253     // Can't optimise for Z10 using "compare and branch" (immediate value is too big).
1254     __ z_cghi(local_count, 256);
1255     __ z_brnh(doXC);
1256 
1257     // MVCLE: Initialize if quite a lot locals.
1258     //  __ bind(doMVCLE);
1259     __ z_lgr(Z_R0_scratch, local_addr);
1260     __ z_lgr(Z_R1_scratch, local_count);
1261     __ clear_reg(Z_ARG2);        // Src len of MVCLE is zero.
1262 
1263     __ MacroAssembler::move_long_ext(Z_R0_scratch, Z_ARG1, 0);
1264     __ z_bru(locals_zeroed);
1265 
1266     Label  XC_template;
1267     __ bind(XC_template);
1268     __ z_xc(0, 0, local_addr, 0, local_addr);
1269 
1270     __ bind(doXC);
1271     __ z_bctgr(local_count, Z_R0);                  // Get #bytes-1 for EXECUTE.
1272     if (VM_Version::has_ExecuteExtensions()) {
1273       __ z_exrl(local_count, XC_template);          // Execute XC with variable length.
1274     } else {
1275       __ z_larl(Z_R1_scratch, XC_template);
1276       __ z_ex(local_count, 0, Z_R0, Z_R1_scratch);  // Execute XC with variable length.
1277     }
1278 
1279     __ bind(locals_zeroed);
1280   }
1281 
1282   }
1283   // Finally set the frame pointer, destroying Z_method.
1284   assert(Z_fp == Z_method, "maybe set Z_fp earlier if other register than Z_method");
1285   // Oprofile analysis suggests to keep a copy in a register to be used by
1286   // generate_counter_incr().
1287   __ z_lgr(Z_ARG2, Z_method);
1288   __ z_lgr(Z_fp, fp);
1289 
1290   BLOCK_COMMENT("} generate_fixed_frame");
1291 }
1292 
1293 // Various method entries
1294 
1295 // Math function, frame manager must set up an interpreter state, etc.
1296 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1297 
1298   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1299   bool use_instruction = false;
1300   address runtime_entry = NULL;
1301   int num_args = 1;
1302   bool double_precision = true;
1303 
1304   // s390 specific:
1305   switch (kind) {
1306     case Interpreter::java_lang_math_sqrt:
1307     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1308     case Interpreter::java_lang_math_fmaF:
1309     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1310     default: break; // Fall back to runtime call.
1311   }
1312 
1313   switch (kind) {
1314     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1315     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1316     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1317     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1318     case Interpreter::java_lang_math_sqrt : /* runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); not available */ break;
1319     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1320     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1321     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1322     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1323     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1324     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1325     default: ShouldNotReachHere();
1326   }
1327 
1328   // Use normal entry if neither instruction nor runtime call is used.
1329   if (!use_instruction && runtime_entry == NULL) return NULL;
1330 
1331   address entry = __ pc();
1332 
1333   if (use_instruction) {
1334     switch (kind) {
1335       case Interpreter::java_lang_math_sqrt:
1336         // Can use memory operand directly.
1337         __ z_sqdb(Z_FRET, Interpreter::stackElementSize, Z_esp);
1338         break;
1339       case Interpreter::java_lang_math_abs:
1340         // Load operand from stack.
1341         __ mem2freg_opt(Z_FRET, Address(Z_esp, Interpreter::stackElementSize));
1342         __ z_lpdbr(Z_FRET);
1343         break;
1344       case Interpreter::java_lang_math_fmaF:
1345         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1346         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize)); // arg1
1347         __ z_maeb(Z_FRET, Z_FARG2, Address(Z_esp, 2 * Interpreter::stackElementSize));
1348         break;
1349       case Interpreter::java_lang_math_fmaD:
1350         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1351         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 5 * Interpreter::stackElementSize)); // arg1
1352         __ z_madb(Z_FRET, Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize));
1353         break;
1354       default: ShouldNotReachHere();
1355     }
1356   } else {
1357     // Load arguments
1358     assert(num_args <= 4, "passed in registers");
1359     if (double_precision) {
1360       int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1361       for (int i = 0; i < num_args; ++i) {
1362         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1363         offset -= 2 * Interpreter::stackElementSize;
1364       }
1365     } else {
1366       int offset = num_args * Interpreter::stackElementSize;
1367       for (int i = 0; i < num_args; ++i) {
1368         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1369         offset -= Interpreter::stackElementSize;
1370       }
1371     }
1372     // Call runtime
1373     __ save_return_pc();       // Save Z_R14.
1374     __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
1375 
1376     __ call_VM_leaf(runtime_entry);
1377 
1378     __ pop_frame();
1379     __ restore_return_pc();    // Restore Z_R14.
1380   }
1381 
1382   // Pop c2i arguments (if any) off when we return.
1383   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1384 
1385   __ z_br(Z_R14);
1386 
1387   return entry;
1388 }
1389 
1390 // Interpreter stub for calling a native method. (asm interpreter).
1391 // This sets up a somewhat different looking stack for calling the
1392 // native method than the typical interpreter frame setup.
1393 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1394   // Determine code generation flags.
1395   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1396 
1397   // Interpreter entry for ordinary Java methods.
1398   //
1399   // Registers alive
1400   //   Z_SP          - stack pointer
1401   //   Z_thread      - JavaThread*
1402   //   Z_method      - callee's method (method to be invoked)
1403   //   Z_esp         - operand (or expression) stack pointer of caller. one slot above last arg.
1404   //   Z_R10         - sender sp (before modifications, e.g. by c2i adapter
1405   //                   and as well by generate_fixed_frame below)
1406   //   Z_R14         - return address to caller (call_stub or c2i_adapter)
1407   //
1408   // Registers updated
1409   //   Z_SP          - stack pointer
1410   //   Z_fp          - callee's framepointer
1411   //   Z_esp         - callee's operand stack pointer
1412   //                   points to the slot above the value on top
1413   //   Z_locals      - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1414   //   Z_tos         - integer result, if any
1415   //   z_ftos        - floating point result, if any
1416   //
1417   // Stack layout at this point:
1418   //
1419   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1420   //                                                          frame was extended by c2i adapter)
1421   //           [outgoing Java arguments]     <-- Z_esp
1422   //           ...
1423   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1424   //           ...
1425   //
1426 
1427   address entry_point = __ pc();
1428 
1429   // Make sure registers are different!
1430   assert_different_registers(Z_thread, Z_method, Z_esp);
1431 
1432   BLOCK_COMMENT("native_entry {");
1433 
1434   // Make sure method is native and not abstract.
1435 #ifdef ASSERT
1436   address reentry = NULL;
1437   { Label L;
1438     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1439     __ z_btrue(L);
1440     reentry = __ stop_chain_static(reentry, "tried to execute non-native method as native");
1441     __ bind(L);
1442   }
1443   { Label L;
1444     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1445     __ z_bfalse(L);
1446     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1447     __ bind(L);
1448   }
1449 #endif // ASSERT
1450 
1451 #ifdef ASSERT
1452   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1453   __ save_return_pc(Z_R14);
1454 #endif
1455 
1456   // Generate the code to allocate the interpreter stack frame.
1457   generate_fixed_frame(true);
1458 
1459   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1460   // Since at this point in the method invocation the exception handler
1461   // would try to exit the monitor of synchronized methods which hasn't
1462   // been entered yet, we set the thread local variable
1463   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1464   // runtime, exception handling i.e. unlock_if_synchronized_method will
1465   // check this thread local flag.
1466   __ z_mvi(do_not_unlock_if_synchronized, true);
1467 
1468   // Increment invocation count and check for overflow.
1469   NearLabel invocation_counter_overflow;
1470   if (inc_counter) {
1471     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1472   }
1473 
1474   Label continue_after_compile;
1475   __ bind(continue_after_compile);
1476 
1477   bang_stack_shadow_pages(true);
1478 
1479   // Reset the _do_not_unlock_if_synchronized flag.
1480   __ z_mvi(do_not_unlock_if_synchronized, false);
1481 
1482   // Check for synchronized methods.
1483   // This mst happen AFTER invocation_counter check and stack overflow check,
1484   // so method is not locked if overflows.
1485   if (synchronized) {
1486     lock_method();
1487   } else {
1488     // No synchronization necessary.
1489 #ifdef ASSERT
1490     { Label L;
1491       __ get_method(Z_R1_scratch);
1492       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1493       __ z_bfalse(L);
1494       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1495       __ bind(L);
1496     }
1497 #endif // ASSERT
1498   }
1499 
1500   // start execution
1501 
1502   // jvmti support
1503   __ notify_method_entry();
1504 
1505   //=============================================================================
1506   // Get and call the signature handler.
1507   const Register Rmethod                 = Z_tmp_2;
1508   const Register signature_handler_entry = Z_tmp_1;
1509   const Register Rresult_handler         = Z_tmp_3;
1510   Label call_signature_handler;
1511 
1512   assert_different_registers(Z_fp, Rmethod, signature_handler_entry, Rresult_handler);
1513   assert(Rresult_handler->is_nonvolatile(), "Rresult_handler must be in a non-volatile register");
1514 
1515   // Reload method.
1516   __ get_method(Rmethod);
1517 
1518   // Check for signature handler.
1519   __ load_and_test_long(signature_handler_entry, method2_(Rmethod, signature_handler));
1520   __ z_brne(call_signature_handler);
1521 
1522   // Method has never been called. Either generate a specialized
1523   // handler or point to the slow one.
1524   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call),
1525              Rmethod);
1526 
1527   // Reload method.
1528   __ get_method(Rmethod);
1529 
1530   // Reload signature handler, it must have been created/assigned in the meantime.
1531   __ z_lg(signature_handler_entry, method2_(Rmethod, signature_handler));
1532 
1533   __ bind(call_signature_handler);
1534 
1535   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1536   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1/*tmp*/);
1537 
1538   // Call signature handler and pass locals address in Z_ARG1.
1539   __ z_lgr(Z_ARG1, Z_locals);
1540   __ call_stub(signature_handler_entry);
1541   // Save result handler returned by signature handler.
1542   __ z_lgr(Rresult_handler, Z_RET);
1543 
1544   // Reload method (the slow signature handler may block for GC).
1545   __ get_method(Rmethod);
1546 
1547   // Pass mirror handle if static call.
1548   {
1549     Label method_is_not_static;
1550     __ testbit(method2_(Rmethod, access_flags), JVM_ACC_STATIC_BIT);
1551     __ z_bfalse(method_is_not_static);
1552     // Get mirror.
1553     __ load_mirror(Z_R1, Rmethod);
1554     // z_ijava_state.oop_temp = pool_holder->klass_part()->java_mirror();
1555     __ z_stg(Z_R1, oop_tmp_offset, Z_fp);
1556     // Pass handle to mirror as 2nd argument to JNI method.
1557     __ add2reg(Z_ARG2, oop_tmp_offset, Z_fp);
1558     __ bind(method_is_not_static);
1559   }
1560 
1561   // Pass JNIEnv address as first parameter.
1562   __ add2reg(Z_ARG1, in_bytes(JavaThread::jni_environment_offset()), Z_thread);
1563 
1564   // Note: last java frame has been set above already. The pc from there
1565   // is precise enough.
1566 
1567   // Get native function entry point before we change the thread state.
1568   __ z_lg(Z_R1/*native_method_entry*/, method2_(Rmethod, native_function));
1569 
1570   //=============================================================================
1571   // Transition from _thread_in_Java to _thread_in_native. As soon as
1572   // we make this change the safepoint code needs to be certain that
1573   // the last Java frame we established is good. The pc in that frame
1574   // just need to be near here not an actual return address.
1575 #ifdef ASSERT
1576   {
1577     NearLabel L;
1578     __ mem2reg_opt(Z_R14, Address(Z_thread, JavaThread::thread_state_offset()), false /*32 bits*/);
1579     __ compareU32_and_branch(Z_R14, _thread_in_Java, Assembler::bcondEqual, L);
1580     reentry = __ stop_chain_static(reentry, "Wrong thread state in native stub");
1581     __ bind(L);
1582   }
1583 #endif
1584 
1585   // Memory ordering: Z does not reorder store/load with subsequent load. That's strong enough.
1586   __ set_thread_state(_thread_in_native);
1587 
1588   //=============================================================================
1589   // Call the native method. Argument registers must not have been
1590   // overwritten since "__ call_stub(signature_handler);" (except for
1591   // ARG1 and ARG2 for static methods).
1592 
1593   __ call_c(Z_R1/*native_method_entry*/);
1594 
1595   // NOTE: frame::interpreter_frame_result() depends on these stores.
1596   __ z_stg(Z_RET, _z_ijava_state_neg(lresult), Z_fp);
1597   __ freg2mem_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1598   const Register Rlresult = signature_handler_entry;
1599   assert(Rlresult->is_nonvolatile(), "Rlresult must be in a non-volatile register");
1600   __ z_lgr(Rlresult, Z_RET);
1601 
1602   // Z_method may no longer be valid, because of GC.
1603 
1604   // Block, if necessary, before resuming in _thread_in_Java state.
1605   // In order for GC to work, don't clear the last_Java_sp until after
1606   // blocking.
1607 
1608   //=============================================================================
1609   // Switch thread to "native transition" state before reading the
1610   // synchronization state. This additional state is necessary
1611   // because reading and testing the synchronization state is not
1612   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1613   // in _thread_in_native state, loads _not_synchronized and is
1614   // preempted. VM thread changes sync state to synchronizing and
1615   // suspends threads for GC. Thread A is resumed to finish this
1616   // native method, but doesn't block here since it didn't see any
1617   // synchronization is progress, and escapes.
1618 
1619   __ set_thread_state(_thread_in_native_trans);
1620   if (UseMembar) {
1621     __ z_fence();
1622   } else {
1623     // Write serialization page so VM thread can do a pseudo remote
1624     // membar. We use the current thread pointer to calculate a thread
1625     // specific offset to write to within the page. This minimizes bus
1626     // traffic due to cache line collision.
1627     __ serialize_memory(Z_thread, Z_R1, Z_R0);
1628   }
1629   // Now before we return to java we must look for a current safepoint
1630   // (a new safepoint can not start since we entered native_trans).
1631   // We must check here because a current safepoint could be modifying
1632   // the callers registers right this moment.
1633 
1634   // Check for safepoint operation in progress and/or pending suspend requests.
1635   {
1636     Label Continue, do_safepoint;
1637     __ safepoint_poll(do_safepoint, Z_R1);
1638     // Check for suspend.
1639     __ load_and_test_int(Z_R0/*suspend_flags*/, thread_(suspend_flags));
1640     __ z_bre(Continue); // 0 -> no flag set -> not suspended
1641     __ bind(do_safepoint);
1642     __ z_lgr(Z_ARG1, Z_thread);
1643     __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1644     __ bind(Continue);
1645   }
1646 
1647   //=============================================================================
1648   // Back in Interpreter Frame.
1649 
1650   // We are in thread_in_native_trans here and back in the normal
1651   // interpreter frame. We don't have to do anything special about
1652   // safepoints and we can switch to Java mode anytime we are ready.
1653 
1654   // Note: frame::interpreter_frame_result has a dependency on how the
1655   // method result is saved across the call to post_method_exit. For
1656   // native methods it assumes that the non-FPU/non-void result is
1657   // saved in z_ijava_state.lresult and a FPU result in z_ijava_state.fresult. If
1658   // this changes then the interpreter_frame_result implementation
1659   // will need to be updated too.
1660 
1661   //=============================================================================
1662   // Back in Java.
1663 
1664   // Memory ordering: Z does not reorder store/load with subsequent
1665   // load. That's strong enough.
1666   __ set_thread_state(_thread_in_Java);
1667 
1668   __ reset_last_Java_frame();
1669 
1670   // We reset the JNI handle block only after unboxing the result; see below.
1671 
1672   // The method register is junk from after the thread_in_native transition
1673   // until here. Also can't call_VM until the bcp has been
1674   // restored. Need bcp for throwing exception below so get it now.
1675   __ get_method(Rmethod);
1676 
1677   // Restore Z_bcp to have legal interpreter frame,
1678   // i.e., bci == 0 <=> Z_bcp == code_base().
1679   __ z_lg(Z_bcp, Address(Rmethod, Method::const_offset())); // get constMethod
1680   __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1681 
1682   if (CheckJNICalls) {
1683     // clear_pending_jni_exception_check
1684     __ clear_mem(Address(Z_thread, JavaThread::pending_jni_exception_check_fn_offset()), sizeof(oop));
1685   }
1686 
1687   // Check if the native method returns an oop, and if so, move it
1688   // from the jni handle to z_ijava_state.oop_temp. This is
1689   // necessary, because we reset the jni handle block below.
1690   // NOTE: frame::interpreter_frame_result() depends on this, too.
1691   { NearLabel no_oop_result;
1692   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1693   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_result);
1694   __ resolve_jobject(Rlresult, /* tmp1 */ Rmethod, /* tmp2 */ Z_R1);
1695   __ z_stg(Rlresult, oop_tmp_offset, Z_fp);
1696   __ bind(no_oop_result);
1697   }
1698 
1699   // Reset handle block.
1700   __ z_lg(Z_R1/*active_handles*/, thread_(active_handles));
1701   __ clear_mem(Address(Z_R1, JNIHandleBlock::top_offset_in_bytes()), 4);
1702 
1703   // Bandle exceptions (exception handling will handle unlocking!).
1704   {
1705     Label L;
1706     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
1707     __ z_bre(L);
1708     __ MacroAssembler::call_VM(noreg,
1709                                CAST_FROM_FN_PTR(address,
1710                                InterpreterRuntime::throw_pending_exception));
1711     __ should_not_reach_here();
1712     __ bind(L);
1713   }
1714 
1715   if (synchronized) {
1716     Register Rfirst_monitor = Z_ARG2;
1717     __ add2reg(Rfirst_monitor, -(frame::z_ijava_state_size + (int)sizeof(BasicObjectLock)), Z_fp);
1718 #ifdef ASSERT
1719     NearLabel ok;
1720     __ z_lg(Z_R1, _z_ijava_state_neg(monitors), Z_fp);
1721     __ compareU64_and_branch(Rfirst_monitor, Z_R1, Assembler::bcondEqual, ok);
1722     reentry = __ stop_chain_static(reentry, "native_entry:unlock: inconsistent z_ijava_state.monitors");
1723     __ bind(ok);
1724 #endif
1725     __ unlock_object(Rfirst_monitor);
1726   }
1727 
1728   // JVMTI support. Result has already been saved above to the frame.
1729   __ notify_method_exit(true/*native_method*/, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1730 
1731   // Move native method result back into proper registers and return.
1732   // C++ interpreter does not use result handler. So do we need to here? TODO(ZASM): check if correct.
1733   { NearLabel no_oop_or_null;
1734   __ mem2freg_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1735   __ load_and_test_long(Z_RET, Address(Z_fp, _z_ijava_state_neg(lresult)));
1736   __ z_bre(no_oop_or_null); // No unboxing if the result is NULL.
1737   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1738   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_or_null);
1739   __ z_lg(Z_RET, oop_tmp_offset, Z_fp);
1740   __ verify_oop(Z_RET);
1741   __ bind(no_oop_or_null);
1742   }
1743 
1744   // Pop the native method's interpreter frame.
1745   __ pop_interpreter_frame(Z_R14 /*return_pc*/, Z_ARG2/*tmp1*/, Z_ARG3/*tmp2*/);
1746 
1747   // Return to caller.
1748   __ z_br(Z_R14);
1749 
1750   if (inc_counter) {
1751     // Handle overflow of counter and compile method.
1752     __ bind(invocation_counter_overflow);
1753     generate_counter_overflow(continue_after_compile);
1754   }
1755 
1756   BLOCK_COMMENT("} native_entry");
1757 
1758   return entry_point;
1759 }
1760 
1761 //
1762 // Generic interpreted method entry to template interpreter.
1763 //
1764 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1765   address entry_point = __ pc();
1766 
1767   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1768 
1769   // Interpreter entry for ordinary Java methods.
1770   //
1771   // Registers alive
1772   //   Z_SP       - stack pointer
1773   //   Z_thread   - JavaThread*
1774   //   Z_method   - callee's method (method to be invoked)
1775   //   Z_esp      - operand (or expression) stack pointer of caller. one slot above last arg.
1776   //   Z_R10      - sender sp (before modifications, e.g. by c2i adapter
1777   //                           and as well by generate_fixed_frame below)
1778   //   Z_R14      - return address to caller (call_stub or c2i_adapter)
1779   //
1780   // Registers updated
1781   //   Z_SP       - stack pointer
1782   //   Z_fp       - callee's framepointer
1783   //   Z_esp      - callee's operand stack pointer
1784   //                points to the slot above the value on top
1785   //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1786   //   Z_tos      - integer result, if any
1787   //   z_ftos     - floating point result, if any
1788   //
1789   //
1790   // stack layout at this point:
1791   //
1792   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1793   //                                                          frame was extended by c2i adapter)
1794   //           [outgoing Java arguments]     <-- Z_esp
1795   //           ...
1796   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1797   //           ...
1798   //
1799   // stack layout before dispatching the first bytecode:
1800   //
1801   //   F0      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP
1802   //           [operand stack]               <-- Z_esp
1803   //           monitor (optional, can grow)
1804   //           [IJAVA_STATE]
1805   //   F1      [PARENT_IJAVA_FRAME_ABI]      <-- Z_fp (== *Z_SP)
1806   //           [F0's locals]                 <-- Z_locals
1807   //           [F1's operand stack]
1808   //           [F1's monitors] (optional)
1809   //           [IJAVA_STATE]
1810 
1811   // Make sure registers are different!
1812   assert_different_registers(Z_thread, Z_method, Z_esp);
1813 
1814   BLOCK_COMMENT("normal_entry {");
1815 
1816   // Make sure method is not native and not abstract.
1817   // Rethink these assertions - they can be simplified and shared.
1818 #ifdef ASSERT
1819   address reentry = NULL;
1820   { Label L;
1821     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1822     __ z_bfalse(L);
1823     reentry = __ stop_chain_static(reentry, "tried to execute native method as non-native");
1824     __ bind(L);
1825   }
1826   { Label L;
1827     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1828     __ z_bfalse(L);
1829     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1830     __ bind(L);
1831   }
1832 #endif // ASSERT
1833 
1834 #ifdef ASSERT
1835   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1836   __ save_return_pc(Z_R14);
1837 #endif
1838 
1839   // Generate the code to allocate the interpreter stack frame.
1840   generate_fixed_frame(false);
1841 
1842   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1843   // Since at this point in the method invocation the exception handler
1844   // would try to exit the monitor of synchronized methods which hasn't
1845   // been entered yet, we set the thread local variable
1846   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1847   // runtime, exception handling i.e. unlock_if_synchronized_method will
1848   // check this thread local flag.
1849   __ z_mvi(do_not_unlock_if_synchronized, true);
1850 
1851   __ profile_parameters_type(Z_tmp_2, Z_ARG3, Z_ARG4);
1852 
1853   // Increment invocation counter and check for overflow.
1854   //
1855   // Note: checking for negative value instead of overflow so we have a 'sticky'
1856   // overflow test (may be of importance as soon as we have true MT/MP).
1857   NearLabel invocation_counter_overflow;
1858   NearLabel profile_method;
1859   NearLabel profile_method_continue;
1860   NearLabel Lcontinue;
1861   if (inc_counter) {
1862     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1863     if (ProfileInterpreter) {
1864       __ bind(profile_method_continue);
1865     }
1866   }
1867   __ bind(Lcontinue);
1868 
1869   bang_stack_shadow_pages(false);
1870 
1871   // Reset the _do_not_unlock_if_synchronized flag.
1872   __ z_mvi(do_not_unlock_if_synchronized, false);
1873 
1874   // Check for synchronized methods.
1875   // Must happen AFTER invocation_counter check and stack overflow check,
1876   // so method is not locked if overflows.
1877   if (synchronized) {
1878     // Allocate monitor and lock method.
1879     lock_method();
1880   } else {
1881 #ifdef ASSERT
1882     { Label L;
1883       __ get_method(Z_R1_scratch);
1884       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1885       __ z_bfalse(L);
1886       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1887       __ bind(L);
1888     }
1889 #endif // ASSERT
1890   }
1891 
1892   // start execution
1893 
1894 #ifdef ASSERT
1895   __ verify_esp(Z_esp, Z_R1_scratch);
1896 
1897   __ verify_thread();
1898 #endif
1899 
1900   // jvmti support
1901   __ notify_method_entry();
1902 
1903   // Start executing instructions.
1904   __ dispatch_next(vtos);
1905   // Dispatch_next does not return.
1906   DEBUG_ONLY(__ should_not_reach_here());
1907 
1908   // Invocation counter overflow.
1909   if (inc_counter) {
1910     if (ProfileInterpreter) {
1911       // We have decided to profile this method in the interpreter.
1912       __ bind(profile_method);
1913 
1914       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1915       __ set_method_data_pointer_for_bcp();
1916       __ z_bru(profile_method_continue);
1917     }
1918 
1919     // Handle invocation counter overflow.
1920     __ bind(invocation_counter_overflow);
1921     generate_counter_overflow(Lcontinue);
1922   }
1923 
1924   BLOCK_COMMENT("} normal_entry");
1925 
1926   return entry_point;
1927 }
1928 
1929 
1930 /**
1931  * Method entry for static native methods:
1932  *   int java.util.zip.CRC32.update(int crc, int b)
1933  */
1934 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1935 
1936   if (UseCRC32Intrinsics) {
1937     uint64_t entry_off = __ offset();
1938     Label    slow_path;
1939 
1940     // If we need a safepoint check, generate full interpreter entry.
1941     __ safepoint_poll(slow_path, Z_R1);
1942 
1943     BLOCK_COMMENT("CRC32_update {");
1944 
1945     // We don't generate local frame and don't align stack because
1946     // we not even call stub code (we generate the code inline)
1947     // and there is no safepoint on this path.
1948 
1949     // Load java parameters.
1950     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1951     const Register argP    = Z_esp;
1952     const Register crc     = Z_ARG1;  // crc value
1953     const Register data    = Z_ARG2;  // address of java byte value (kernel_crc32 needs address)
1954     const Register dataLen = Z_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1955     const Register table   = Z_ARG4;  // address of crc32 table
1956 
1957     // Arguments are reversed on java expression stack.
1958     __ z_la(data, 3+1*wordSize, argP);  // byte value (stack address).
1959                                         // Being passed as an int, the single byte is at offset +3.
1960     __ z_llgf(crc, 2 * wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1961 
1962     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1963     __ kernel_crc32_singleByte(crc, data, dataLen, table, Z_R1, true);
1964 
1965     // Restore caller sp for c2i case.
1966     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1967 
1968     __ z_br(Z_R14);
1969 
1970     BLOCK_COMMENT("} CRC32_update");
1971 
1972     // Use a previously generated vanilla native entry as the slow path.
1973     BIND(slow_path);
1974     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1975     return __ addr_at(entry_off);
1976   }
1977 
1978   return NULL;
1979 }
1980 
1981 
1982 /**
1983  * Method entry for static native methods:
1984  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1985  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1986  */
1987 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1988 
1989   if (UseCRC32Intrinsics) {
1990     uint64_t entry_off = __ offset();
1991     Label    slow_path;
1992 
1993     // If we need a safepoint check, generate full interpreter entry.
1994     __ safepoint_poll(slow_path, Z_R1);
1995 
1996     // We don't generate local frame and don't align stack because
1997     // we call stub code and there is no safepoint on this path.
1998 
1999     // Load parameters.
2000     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
2001     const Register argP    = Z_esp;
2002     const Register crc     = Z_ARG1;  // crc value
2003     const Register data    = Z_ARG2;  // address of java byte array
2004     const Register dataLen = Z_ARG3;  // source data len
2005     const Register table   = Z_ARG4;  // address of crc32 table
2006     const Register t0      = Z_R10;   // work reg for kernel* emitters
2007     const Register t1      = Z_R11;   // work reg for kernel* emitters
2008     const Register t2      = Z_R12;   // work reg for kernel* emitters
2009     const Register t3      = Z_R13;   // work reg for kernel* emitters
2010 
2011     // Arguments are reversed on java expression stack.
2012     // Calculate address of start element.
2013     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
2014       // crc     @ (SP + 5W) (32bit)
2015       // buf     @ (SP + 3W) (64bit ptr to long array)
2016       // off     @ (SP + 2W) (32bit)
2017       // dataLen @ (SP + 1W) (32bit)
2018       // data = buf + off
2019       BLOCK_COMMENT("CRC32_updateByteBuffer {");
2020       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
2021       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2022       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2023       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2024     } else {                                                         // Used for "updateBytes update".
2025       // crc     @ (SP + 4W) (32bit)
2026       // buf     @ (SP + 3W) (64bit ptr to byte array)
2027       // off     @ (SP + 2W) (32bit)
2028       // dataLen @ (SP + 1W) (32bit)
2029       // data = buf + off + base_offset
2030       BLOCK_COMMENT("CRC32_updateBytes {");
2031       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
2032       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2033       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2034       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2035       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2036     }
2037 
2038     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
2039 
2040     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2041     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
2042     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, true);
2043     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
2044 
2045     // Restore caller sp for c2i case.
2046     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2047 
2048     __ z_br(Z_R14);
2049 
2050     BLOCK_COMMENT("} CRC32_update{Bytes|ByteBuffer}");
2051 
2052     // Use a previously generated vanilla native entry as the slow path.
2053     BIND(slow_path);
2054     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
2055     return __ addr_at(entry_off);
2056   }
2057 
2058   return NULL;
2059 }
2060 
2061 
2062 /**
2063  * Method entry for intrinsic-candidate (non-native) methods:
2064  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
2065  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
2066  * Unlike CRC32, CRC32C does not have any methods marked as native
2067  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
2068  */
2069 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
2070 
2071   if (UseCRC32CIntrinsics) {
2072     uint64_t entry_off = __ offset();
2073 
2074     // We don't generate local frame and don't align stack because
2075     // we call stub code and there is no safepoint on this path.
2076 
2077     // Load parameters.
2078     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
2079     const Register argP    = Z_esp;
2080     const Register crc     = Z_ARG1;  // crc value
2081     const Register data    = Z_ARG2;  // address of java byte array
2082     const Register dataLen = Z_ARG3;  // source data len
2083     const Register table   = Z_ARG4;  // address of crc32 table
2084     const Register t0      = Z_R10;   // work reg for kernel* emitters
2085     const Register t1      = Z_R11;   // work reg for kernel* emitters
2086     const Register t2      = Z_R12;   // work reg for kernel* emitters
2087     const Register t3      = Z_R13;   // work reg for kernel* emitters
2088 
2089     // Arguments are reversed on java expression stack.
2090     // Calculate address of start element.
2091     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateByteBuffer direct".
2092       // crc     @ (SP + 5W) (32bit)
2093       // buf     @ (SP + 3W) (64bit ptr to long array)
2094       // off     @ (SP + 2W) (32bit)
2095       // dataLen @ (SP + 1W) (32bit)
2096       // data = buf + off
2097       BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
2098       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
2099       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2100       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2101       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
2102       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
2103     } else {                                                                // Used for "updateBytes update".
2104       // crc     @ (SP + 4W) (32bit)
2105       // buf     @ (SP + 3W) (64bit ptr to byte array)
2106       // off     @ (SP + 2W) (32bit)
2107       // dataLen @ (SP + 1W) (32bit)
2108       // data = buf + off + base_offset
2109       BLOCK_COMMENT("CRC32C_updateBytes {");
2110       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
2111       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2112       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2113       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
2114       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
2115       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2116     }
2117 
2118     StubRoutines::zarch::generate_load_crc32c_table_addr(_masm, table);
2119 
2120     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2121     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
2122     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, false);
2123     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
2124 
2125     // Restore caller sp for c2i case.
2126     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2127 
2128     __ z_br(Z_R14);
2129 
2130     BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
2131     return __ addr_at(entry_off);
2132   }
2133 
2134   return NULL;
2135 }
2136 
2137 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
2138   // Quick & dirty stack overflow checking: bang the stack & handle trap.
2139   // Note that we do the banging after the frame is setup, since the exception
2140   // handling code expects to find a valid interpreter frame on the stack.
2141   // Doing the banging earlier fails if the caller frame is not an interpreter
2142   // frame.
2143   // (Also, the exception throwing code expects to unlock any synchronized
2144   // method receiver, so do the banging after locking the receiver.)
2145 
2146   // Bang each page in the shadow zone. We can't assume it's been done for
2147   // an interpreter frame with greater than a page of locals, so each page
2148   // needs to be checked. Only true for non-native. For native, we only bang the last page.
2149   if (UseStackBanging) {
2150     const int page_size      = os::vm_page_size();
2151     const int n_shadow_pages = (int)(JavaThread::stack_shadow_zone_size()/page_size);
2152     const int start_page_num = native_call ? n_shadow_pages : 1;
2153     for (int pages = start_page_num; pages <= n_shadow_pages; pages++) {
2154       __ bang_stack_with_offset(pages*page_size);
2155     }
2156   }
2157 }
2158 
2159 //-----------------------------------------------------------------------------
2160 // Exceptions
2161 
2162 void TemplateInterpreterGenerator::generate_throw_exception() {
2163 
2164   BLOCK_COMMENT("throw_exception {");
2165 
2166   // Entry point in previous activation (i.e., if the caller was interpreted).
2167   Interpreter::_rethrow_exception_entry = __ pc();
2168   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Frame accessors use Z_fp.
2169   // Z_ARG1 (==Z_tos): exception
2170   // Z_ARG2          : Return address/pc that threw exception.
2171   __ restore_bcp();    // R13 points to call/send.
2172   __ restore_locals();
2173 
2174   // Fallthrough, no need to restore Z_esp.
2175 
2176   // Entry point for exceptions thrown within interpreter code.
2177   Interpreter::_throw_exception_entry = __ pc();
2178   // Expression stack is undefined here.
2179   // Z_ARG1 (==Z_tos): exception
2180   // Z_bcp: exception bcp
2181   __ verify_oop(Z_ARG1);
2182   __ z_lgr(Z_ARG2, Z_ARG1);
2183 
2184   // Expression stack must be empty before entering the VM in case of
2185   // an exception.
2186   __ empty_expression_stack();
2187   // Find exception handler address and preserve exception oop.
2188   const Register Rpreserved_exc_oop = Z_tmp_1;
2189   __ call_VM(Rpreserved_exc_oop,
2190              CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception),
2191              Z_ARG2);
2192   // Z_RET: exception handler entry point
2193   // Z_bcp: bcp for exception handler
2194   __ push_ptr(Rpreserved_exc_oop); // Push exception which is now the only value on the stack.
2195   __ z_br(Z_RET); // Jump to exception handler (may be _remove_activation_entry!).
2196 
2197   // If the exception is not handled in the current frame the frame is
2198   // removed and the exception is rethrown (i.e. exception
2199   // continuation is _rethrow_exception).
2200   //
2201   // Note: At this point the bci is still the bci for the instruction
2202   // which caused the exception and the expression stack is
2203   // empty. Thus, for any VM calls at this point, GC will find a legal
2204   // oop map (with empty expression stack).
2205 
2206   //
2207   // JVMTI PopFrame support
2208   //
2209 
2210   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2211   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2212   __ empty_expression_stack();
2213   // Set the popframe_processing bit in pending_popframe_condition
2214   // indicating that we are currently handling popframe, so that
2215   // call_VMs that may happen later do not trigger new popframe
2216   // handling cycles.
2217   __ load_sized_value(Z_tmp_1, Address(Z_thread, JavaThread::popframe_condition_offset()), 4, false /*signed*/);
2218   __ z_oill(Z_tmp_1, JavaThread::popframe_processing_bit);
2219   __ z_sty(Z_tmp_1, thread_(popframe_condition));
2220 
2221   {
2222     // Check to see whether we are returning to a deoptimized frame.
2223     // (The PopFrame call ensures that the caller of the popped frame is
2224     // either interpreted or compiled and deoptimizes it if compiled.)
2225     // In this case, we can't call dispatch_next() after the frame is
2226     // popped, but instead must save the incoming arguments and restore
2227     // them after deoptimization has occurred.
2228     //
2229     // Note that we don't compare the return PC against the
2230     // deoptimization blob's unpack entry because of the presence of
2231     // adapter frames in C2.
2232     NearLabel caller_not_deoptimized;
2233     __ z_lg(Z_ARG1, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2234     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), Z_ARG1);
2235     __ compareU64_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, caller_not_deoptimized);
2236 
2237     // Compute size of arguments for saving when returning to
2238     // deoptimized caller.
2239     __ get_method(Z_ARG2);
2240     __ z_lg(Z_ARG2, Address(Z_ARG2, Method::const_offset()));
2241     __ z_llgh(Z_ARG2, Address(Z_ARG2, ConstMethod::size_of_parameters_offset()));
2242     __ z_sllg(Z_ARG2, Z_ARG2, Interpreter::logStackElementSize); // slots 2 bytes
2243     __ restore_locals();
2244     // Compute address of args to be saved.
2245     __ z_lgr(Z_ARG3, Z_locals);
2246     __ z_slgr(Z_ARG3, Z_ARG2);
2247     __ add2reg(Z_ARG3, wordSize);
2248     // Save these arguments.
2249     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args),
2250                     Z_thread, Z_ARG2, Z_ARG3);
2251 
2252     __ remove_activation(vtos, Z_R14,
2253                          /* throw_monitor_exception */ false,
2254                          /* install_monitor_exception */ false,
2255                          /* notify_jvmdi */ false);
2256 
2257     // Inform deoptimization that it is responsible for restoring
2258     // these arguments.
2259     __ store_const(thread_(popframe_condition),
2260                    JavaThread::popframe_force_deopt_reexecution_bit,
2261                    Z_tmp_1, false);
2262 
2263     // Continue in deoptimization handler.
2264     __ z_br(Z_R14);
2265 
2266     __ bind(caller_not_deoptimized);
2267   }
2268 
2269   // Clear the popframe condition flag.
2270   __ clear_mem(thread_(popframe_condition), sizeof(int));
2271 
2272   __ remove_activation(vtos,
2273                        noreg,  // Retaddr is not used.
2274                        false,  // throw_monitor_exception
2275                        false,  // install_monitor_exception
2276                        false); // notify_jvmdi
2277   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2278   __ restore_bcp();
2279   __ restore_locals();
2280   __ restore_esp();
2281   // The method data pointer was incremented already during
2282   // call profiling. We have to restore the mdp for the current bcp.
2283   if (ProfileInterpreter) {
2284     __ set_method_data_pointer_for_bcp();
2285   }
2286 #if INCLUDE_JVMTI
2287   {
2288     Label L_done;
2289 
2290     __ z_cli(0, Z_bcp, Bytecodes::_invokestatic);
2291     __ z_brc(Assembler::bcondNotEqual, L_done);
2292 
2293     // The member name argument must be restored if _invokestatic is
2294     // re-executed after a PopFrame call.  Detect such a case in the
2295     // InterpreterRuntime function and return the member name
2296     // argument, or NULL.
2297     __ z_lg(Z_ARG2, Address(Z_locals));
2298     __ get_method(Z_ARG3);
2299     __ call_VM(Z_tmp_1,
2300                CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
2301                Z_ARG2, Z_ARG3, Z_bcp);
2302 
2303     __ z_ltgr(Z_tmp_1, Z_tmp_1);
2304     __ z_brc(Assembler::bcondEqual, L_done);
2305 
2306     __ z_stg(Z_tmp_1, Address(Z_esp, wordSize));
2307     __ bind(L_done);
2308   }
2309 #endif // INCLUDE_JVMTI
2310   __ dispatch_next(vtos);
2311   // End of PopFrame support.
2312   Interpreter::_remove_activation_entry = __ pc();
2313 
2314   // In between activations - previous activation type unknown yet
2315   // compute continuation point - the continuation point expects the
2316   // following registers set up:
2317   //
2318   // Z_ARG1 (==Z_tos): exception
2319   // Z_ARG2          : return address/pc that threw exception
2320 
2321   Register return_pc = Z_tmp_1;
2322   Register handler   = Z_tmp_2;
2323    assert(return_pc->is_nonvolatile(), "use non-volatile reg. to preserve exception pc");
2324    assert(handler->is_nonvolatile(),   "use non-volatile reg. to handler pc");
2325   __ asm_assert_ijava_state_magic(return_pc/*tmp*/); // The top frame should be an interpreter frame.
2326   __ z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2327 
2328   // Moved removing the activation after VM call, because the new top
2329   // frame does not necessarily have the z_abi_160 required for a VM
2330   // call (e.g. if it is compiled).
2331 
2332   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2333                                          SharedRuntime::exception_handler_for_return_address),
2334                         Z_thread, return_pc);
2335   __ z_lgr(handler, Z_RET); // Save exception handler.
2336 
2337   // Preserve exception over this code sequence.
2338   __ pop_ptr(Z_ARG1);
2339   __ set_vm_result(Z_ARG1);
2340   // Remove the activation (without doing throws on illegalMonitorExceptions).
2341   __ remove_activation(vtos, noreg/*ret.pc already loaded*/, false/*throw exc*/, true/*install exc*/, false/*notify jvmti*/);
2342   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2343 
2344   __ get_vm_result(Z_ARG1);     // Restore exception.
2345   __ verify_oop(Z_ARG1);
2346   __ z_lgr(Z_ARG2, return_pc);  // Restore return address.
2347 
2348 #ifdef ASSERT
2349   // The return_pc in the new top frame is dead... at least that's my
2350   // current understanding. To assert this I overwrite it.
2351   // Note: for compiled frames the handler is the deopt blob
2352   // which writes Z_ARG2 into the return_pc slot.
2353   __ load_const_optimized(return_pc, 0xb00b1);
2354   __ z_stg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2355 #endif
2356 
2357   // Z_ARG1 (==Z_tos): exception
2358   // Z_ARG2          : return address/pc that threw exception
2359 
2360   // Note that an "issuing PC" is actually the next PC after the call.
2361   __ z_br(handler);         // Jump to exception handler of caller.
2362 
2363   BLOCK_COMMENT("} throw_exception");
2364 }
2365 
2366 //
2367 // JVMTI ForceEarlyReturn support
2368 //
2369 address TemplateInterpreterGenerator::generate_earlyret_entry_for (TosState state) {
2370   address entry = __ pc();
2371 
2372   BLOCK_COMMENT("earlyret_entry {");
2373 
2374   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2375   __ restore_bcp();
2376   __ restore_locals();
2377   __ restore_esp();
2378   __ empty_expression_stack();
2379   __ load_earlyret_value(state);
2380 
2381   Register RjvmtiState = Z_tmp_1;
2382   __ z_lg(RjvmtiState, thread_(jvmti_thread_state));
2383   __ store_const(Address(RjvmtiState, JvmtiThreadState::earlyret_state_offset()),
2384                  JvmtiThreadState::earlyret_inactive, 4, 4, Z_R0_scratch);
2385 
2386   if (state == itos) {
2387     // Narrow result if state is itos but result type is smaller.
2388     // Need to narrow in the return bytecode rather than in generate_return_entry
2389     // since compiled code callers expect the result to already be narrowed.
2390     __ narrow(Z_tos, Z_tmp_1); /* fall through */
2391   }
2392   __ remove_activation(state,
2393                        Z_tmp_1, // retaddr
2394                        false,   // throw_monitor_exception
2395                        false,   // install_monitor_exception
2396                        true);   // notify_jvmdi
2397   __ z_br(Z_tmp_1);
2398 
2399   BLOCK_COMMENT("} earlyret_entry");
2400 
2401   return entry;
2402 }
2403 
2404 //-----------------------------------------------------------------------------
2405 // Helper for vtos entry point generation.
2406 
2407 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2408                                                          address& bep,
2409                                                          address& cep,
2410                                                          address& sep,
2411                                                          address& aep,
2412                                                          address& iep,
2413                                                          address& lep,
2414                                                          address& fep,
2415                                                          address& dep,
2416                                                          address& vep) {
2417   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2418   Label L;
2419   aep = __ pc(); __ push_ptr(); __ z_bru(L);
2420   fep = __ pc(); __ push_f();   __ z_bru(L);
2421   dep = __ pc(); __ push_d();   __ z_bru(L);
2422   lep = __ pc(); __ push_l();   __ z_bru(L);
2423   bep = cep = sep =
2424   iep = __ pc(); __ push_i();
2425   vep = __ pc();
2426   __ bind(L);
2427   generate_and_dispatch(t);
2428 }
2429 
2430 //-----------------------------------------------------------------------------
2431 
2432 #ifndef PRODUCT
2433 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2434   address entry = __ pc();
2435   NearLabel counter_below_trace_threshold;
2436 
2437   if (TraceBytecodesAt > 0) {
2438     // Skip runtime call, if the trace threshold is not yet reached.
2439     __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2440     __ load_absolute_address(Z_tmp_2, (address)&TraceBytecodesAt);
2441     __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2442     __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2443     __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, counter_below_trace_threshold);
2444   }
2445 
2446   int offset2 = state == ltos || state == dtos ? 2 : 1;
2447 
2448   __ push(state);
2449   // Preserved return pointer is in Z_R14.
2450   // InterpreterRuntime::trace_bytecode() preserved and returns the value passed as second argument.
2451   __ z_lgr(Z_ARG2, Z_R14);
2452   __ z_lg(Z_ARG3, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)));
2453   if (WizardMode) {
2454     __ z_lgr(Z_ARG4, Z_esp); // Trace Z_esp in WizardMode.
2455   } else {
2456     __ z_lg(Z_ARG4, Address(Z_esp, Interpreter::expr_offset_in_bytes(offset2)));
2457   }
2458   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), Z_ARG2, Z_ARG3, Z_ARG4);
2459   __ z_lgr(Z_R14, Z_RET); // Estore return address (see above).
2460   __ pop(state);
2461 
2462   __ bind(counter_below_trace_threshold);
2463   __ z_br(Z_R14); // return
2464 
2465   return entry;
2466 }
2467 
2468 // Make feasible for old CPUs.
2469 void TemplateInterpreterGenerator::count_bytecode() {
2470   __ load_absolute_address(Z_R1_scratch, (address) &BytecodeCounter::_counter_value);
2471   __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2472 }
2473 
2474 void TemplateInterpreterGenerator::histogram_bytecode(Template * t) {
2475   __ load_absolute_address(Z_R1_scratch, (address)&BytecodeHistogram::_counters[ t->bytecode() ]);
2476   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2477 }
2478 
2479 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template * t) {
2480   Address  index_addr(Z_tmp_1, (intptr_t) 0);
2481   Register index = Z_tmp_2;
2482 
2483   // Load previous index.
2484   __ load_absolute_address(Z_tmp_1, (address) &BytecodePairHistogram::_index);
2485   __ mem2reg_opt(index, index_addr, false);
2486 
2487   // Mask with current bytecode and store as new previous index.
2488   __ z_srl(index, BytecodePairHistogram::log2_number_of_codes);
2489   __ load_const_optimized(Z_R0_scratch,
2490                           (int)t->bytecode() << BytecodePairHistogram::log2_number_of_codes);
2491   __ z_or(index, Z_R0_scratch);
2492   __ reg2mem_opt(index, index_addr, false);
2493 
2494   // Load counter array's address.
2495   __ z_lgfr(index, index);   // Sign extend for addressing.
2496   __ z_sllg(index, index, LogBytesPerInt);  // index2bytes
2497   __ load_absolute_address(Z_R1_scratch,
2498                            (address) &BytecodePairHistogram::_counters);
2499   // Add index and increment counter.
2500   __ z_agr(Z_R1_scratch, index);
2501   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2502 }
2503 
2504 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2505   // Call a little run-time stub to avoid blow-up for each bytecode.
2506   // The run-time runtime saves the right registers, depending on
2507   // the tosca in-state for the given template.
2508   address entry = Interpreter::trace_code(t->tos_in());
2509   guarantee(entry != NULL, "entry must have been generated");
2510   __ call_stub(entry);
2511 }
2512 
2513 void TemplateInterpreterGenerator::stop_interpreter_at() {
2514   NearLabel L;
2515 
2516   __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2517   __ load_absolute_address(Z_tmp_2, (address)&StopInterpreterAt);
2518   __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2519   __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2520   __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, L);
2521   assert(Z_tmp_1->is_nonvolatile(), "must be nonvolatile to preserve Z_tos");
2522   assert(Z_F8->is_nonvolatile(), "must be nonvolatile to preserve Z_ftos");
2523   __ z_lgr(Z_tmp_1, Z_tos);      // Save tos.
2524   __ z_lgr(Z_tmp_2, Z_bytecode); // Save Z_bytecode.
2525   __ z_ldr(Z_F8, Z_ftos);        // Save ftos.
2526   // Use -XX:StopInterpreterAt=<num> to set the limit
2527   // and break at breakpoint().
2528   __ call_VM(noreg, CAST_FROM_FN_PTR(address, breakpoint), false);
2529   __ z_lgr(Z_tos, Z_tmp_1);      // Restore tos.
2530   __ z_lgr(Z_bytecode, Z_tmp_2); // Save Z_bytecode.
2531   __ z_ldr(Z_ftos, Z_F8);        // Restore ftos.
2532   __ bind(L);
2533 }
2534 
2535 #endif // !PRODUCT