1 /*
   2  * Copyright (c) 2016, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "interpreter/abstractInterpreter.hpp"
  29 #include "interpreter/bytecodeHistogram.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateInterpreterGenerator.hpp"
  34 #include "interpreter/templateTable.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 
  49 
  50 // Size of interpreter code.  Increase if too small.  Interpreter will
  51 // fail with a guarantee ("not enough space for interpreter generation");
  52 // if too small.
  53 // Run with +PrintInterpreter to get the VM to print out the size.
  54 // Max size with JVMTI
  55 int TemplateInterpreter::InterpreterCodeSize = 320*K;
  56 
  57 #undef  __
  58 #ifdef PRODUCT
  59   #define __ _masm->
  60 #else
  61   #define __ _masm->
  62 //  #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
  63 #endif
  64 
  65 #define BLOCK_COMMENT(str) __ block_comment(str)
  66 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  67 
  68 #define oop_tmp_offset     _z_ijava_state_neg(oop_tmp)
  69 
  70 //-----------------------------------------------------------------------------
  71 
  72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  73   //
  74   // New slow_signature handler that respects the z/Architecture
  75   // C calling conventions.
  76   //
  77   // We get called by the native entry code with our output register
  78   // area == 8. First we call InterpreterRuntime::get_result_handler
  79   // to copy the pointer to the signature string temporarily to the
  80   // first C-argument and to return the result_handler in
  81   // Z_RET. Since native_entry will copy the jni-pointer to the
  82   // first C-argument slot later on, it's OK to occupy this slot
  83   // temporarily. Then we copy the argument list on the java
  84   // expression stack into native varargs format on the native stack
  85   // and load arguments into argument registers. Integer arguments in
  86   // the varargs vector will be sign-extended to 8 bytes.
  87   //
  88   // On entry:
  89   //   Z_ARG1  - intptr_t*       Address of java argument list in memory.
  90   //   Z_state - cppInterpreter* Address of interpreter state for
  91   //                               this method
  92   //   Z_method
  93   //
  94   // On exit (just before return instruction):
  95   //   Z_RET contains the address of the result_handler.
  96   //   Z_ARG2 is not updated for static methods and contains "this" otherwise.
  97   //   Z_ARG3-Z_ARG5 contain the first 3 arguments of types other than float and double.
  98   //   Z_FARG1-Z_FARG4 contain the first 4 arguments of type float or double.
  99 
 100   const int LogSizeOfCase = 3;
 101 
 102   const int max_fp_register_arguments   = Argument::n_float_register_parameters;
 103   const int max_int_register_arguments  = Argument::n_register_parameters - 2;  // First 2 are reserved.
 104 
 105   const Register arg_java       = Z_tmp_2;
 106   const Register arg_c          = Z_tmp_3;
 107   const Register signature      = Z_R1_scratch; // Is a string.
 108   const Register fpcnt          = Z_R0_scratch;
 109   const Register argcnt         = Z_tmp_4;
 110   const Register intSlot        = Z_tmp_1;
 111   const Register sig_end        = Z_tmp_1; // Assumed end of signature (only used in do_object).
 112   const Register target_sp      = Z_tmp_1;
 113   const FloatRegister floatSlot = Z_F1;
 114 
 115   const int d_signature         = _z_abi(gpr6); // Only spill space, register contents not affected.
 116   const int d_fpcnt             = _z_abi(gpr7); // Only spill space, register contents not affected.
 117 
 118   unsigned int entry_offset = __ offset();
 119 
 120   BLOCK_COMMENT("slow_signature_handler {");
 121 
 122   // We use target_sp for storing arguments in the C frame.
 123   __ save_return_pc();
 124 
 125   __ z_stmg(Z_R10,Z_R13,-32,Z_SP);
 126   __ push_frame_abi160(32);
 127 
 128   __ z_lgr(arg_java, Z_ARG1);
 129 
 130   Register   method = Z_ARG2; // Directly load into correct argument register.
 131 
 132   __ get_method(method);
 133   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), Z_thread, method);
 134 
 135   // Move signature to callee saved register.
 136   // Don't directly write to stack. Frame is used by VM call.
 137   __ z_lgr(Z_tmp_1, Z_RET);
 138 
 139   // Reload method. Register may have been altered by VM call.
 140   __ get_method(method);
 141 
 142   // Get address of result handler.
 143   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), Z_thread, method);
 144 
 145   // Save signature address to stack.
 146   __ z_stg(Z_tmp_1, d_signature, Z_SP);
 147 
 148   // Don't overwrite return value (Z_RET, Z_ARG1) in rest of the method !
 149 
 150   {
 151     Label   isStatic;
 152 
 153     // Test if static.
 154     // We can test the bit directly.
 155     // Path is Z_method->_access_flags._flags.
 156     // We only support flag bits in the least significant byte (assert !).
 157     // Therefore add 3 to address that byte within "_flags".
 158     // Reload method. VM call above may have destroyed register contents
 159     __ get_method(method);
 160     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 161     method = noreg;  // end of life
 162     __ z_btrue(isStatic);
 163 
 164     // For non-static functions, pass "this" in Z_ARG2 and copy it to 2nd C-arg slot.
 165     // Need to box the Java object here, so we use arg_java
 166     // (address of current Java stack slot) as argument and
 167     // don't dereference it as in case of ints, floats, etc..
 168     __ z_lgr(Z_ARG2, arg_java);
 169     __ add2reg(arg_java, -BytesPerWord);
 170     __ bind(isStatic);
 171   }
 172 
 173   // argcnt == 0 corresponds to 3rd C argument.
 174   //   arg #1 (result handler) and
 175   //   arg #2 (this, for non-statics), unused else
 176   // are reserved and pre-filled above.
 177   // arg_java points to the corresponding Java argument here. It
 178   // has been decremented by one argument (this) in case of non-static.
 179   __ clear_reg(argcnt, true, false);  // Don't set CC.
 180   __ z_lg(target_sp, 0, Z_SP);
 181   __ add2reg(arg_c, _z_abi(remaining_cargs), target_sp);
 182   // No floating-point args parsed so far.
 183   __ clear_mem(Address(Z_SP, d_fpcnt), 8);
 184 
 185   NearLabel   move_intSlot_to_ARG, move_floatSlot_to_FARG;
 186   NearLabel   loop_start, loop_start_restore, loop_end;
 187   NearLabel   do_int, do_long, do_float, do_double;
 188   NearLabel   do_dontreachhere, do_object, do_array, do_boxed;
 189 
 190 #ifdef ASSERT
 191   // Signature needs to point to '(' (== 0x28) at entry.
 192   __ z_lg(signature, d_signature, Z_SP);
 193   __ z_cli(0, signature, (int) '(');
 194   __ z_brne(do_dontreachhere);
 195 #endif
 196 
 197   __ bind(loop_start_restore);
 198   __ z_lg(signature, d_signature, Z_SP);  // Restore signature ptr, destroyed by move_XX_to_ARG.
 199 
 200   BIND(loop_start);
 201   // Advance to next argument type token from the signature.
 202   __ add2reg(signature, 1);
 203 
 204   // Use CLI, works well on all CPU versions.
 205     __ z_cli(0, signature, (int) ')');
 206     __ z_bre(loop_end);                // end of signature
 207     __ z_cli(0, signature, (int) 'L');
 208     __ z_bre(do_object);               // object     #9
 209     __ z_cli(0, signature, (int) 'F');
 210     __ z_bre(do_float);                // float      #7
 211     __ z_cli(0, signature, (int) 'J');
 212     __ z_bre(do_long);                 // long       #6
 213     __ z_cli(0, signature, (int) 'B');
 214     __ z_bre(do_int);                  // byte       #1
 215     __ z_cli(0, signature, (int) 'Z');
 216     __ z_bre(do_int);                  // boolean    #2
 217     __ z_cli(0, signature, (int) 'C');
 218     __ z_bre(do_int);                  // char       #3
 219     __ z_cli(0, signature, (int) 'S');
 220     __ z_bre(do_int);                  // short      #4
 221     __ z_cli(0, signature, (int) 'I');
 222     __ z_bre(do_int);                  // int        #5
 223     __ z_cli(0, signature, (int) 'D');
 224     __ z_bre(do_double);               // double     #8
 225     __ z_cli(0, signature, (int) '[');
 226     __ z_bre(do_array);                // array      #10
 227 
 228   __ bind(do_dontreachhere);
 229 
 230   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 231 
 232   // Array argument
 233   BIND(do_array);
 234 
 235   {
 236     Label   start_skip, end_skip;
 237 
 238     __ bind(start_skip);
 239 
 240     // Advance to next type tag from signature.
 241     __ add2reg(signature, 1);
 242 
 243     // Use CLI, works well on all CPU versions.
 244     __ z_cli(0, signature, (int) '[');
 245     __ z_bre(start_skip);               // Skip further brackets.
 246 
 247     __ z_cli(0, signature, (int) '9');
 248     __ z_brh(end_skip);                 // no optional size
 249 
 250     __ z_cli(0, signature, (int) '0');
 251     __ z_brnl(start_skip);              // Skip optional size.
 252 
 253     __ bind(end_skip);
 254 
 255     __ z_cli(0, signature, (int) 'L');
 256     __ z_brne(do_boxed);                // If not array of objects: go directly to do_boxed.
 257   }
 258 
 259   //  OOP argument
 260   BIND(do_object);
 261   // Pass by an object's type name.
 262   {
 263     Label   L;
 264 
 265     __ add2reg(sig_end, 4095, signature);     // Assume object type name is shorter than 4k.
 266     __ load_const_optimized(Z_R0, (int) ';'); // Type name terminator (must be in Z_R0!).
 267     __ MacroAssembler::search_string(sig_end, signature);
 268     __ z_brl(L);
 269     __ z_illtrap();  // No semicolon found: internal error or object name too long.
 270     __ bind(L);
 271     __ z_lgr(signature, sig_end);
 272     // fallthru to do_boxed
 273   }
 274 
 275   // Need to box the Java object here, so we use arg_java
 276   // (address of current Java stack slot) as argument and
 277   // don't dereference it as in case of ints, floats, etc..
 278 
 279   // UNBOX argument
 280   // Load reference and check for NULL.
 281   Label  do_int_Entry4Boxed;
 282   __ bind(do_boxed);
 283   {
 284     __ load_and_test_long(intSlot, Address(arg_java));
 285     __ z_bre(do_int_Entry4Boxed);
 286     __ z_lgr(intSlot, arg_java);
 287     __ z_bru(do_int_Entry4Boxed);
 288   }
 289 
 290   // INT argument
 291 
 292   // (also for byte, boolean, char, short)
 293   // Use lgf for load (sign-extend) and stg for store.
 294   BIND(do_int);
 295   __ z_lgf(intSlot, 0, arg_java);
 296 
 297   __ bind(do_int_Entry4Boxed);
 298   __ add2reg(arg_java, -BytesPerWord);
 299   // If argument fits into argument register, go and handle it, otherwise continue.
 300   __ compare32_and_branch(argcnt, max_int_register_arguments,
 301                           Assembler::bcondLow, move_intSlot_to_ARG);
 302   __ z_stg(intSlot, 0, arg_c);
 303   __ add2reg(arg_c, BytesPerWord);
 304   __ z_bru(loop_start);
 305 
 306   // LONG argument
 307 
 308   BIND(do_long);
 309   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 310   __ z_lg(intSlot, BytesPerWord, arg_java);
 311   // If argument fits into argument register, go and handle it, otherwise continue.
 312   __ compare32_and_branch(argcnt, max_int_register_arguments,
 313                           Assembler::bcondLow, move_intSlot_to_ARG);
 314   __ z_stg(intSlot, 0, arg_c);
 315   __ add2reg(arg_c, BytesPerWord);
 316   __ z_bru(loop_start);
 317 
 318   // FLOAT argumen
 319 
 320   BIND(do_float);
 321   __ z_le(floatSlot, 0, arg_java);
 322   __ add2reg(arg_java, -BytesPerWord);
 323   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 324   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 325   __ z_brl(move_floatSlot_to_FARG);
 326   __ z_ste(floatSlot, 4, arg_c);
 327   __ add2reg(arg_c, BytesPerWord);
 328   __ z_bru(loop_start);
 329 
 330   // DOUBLE argument
 331 
 332   BIND(do_double);
 333   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 334   __ z_ld(floatSlot, BytesPerWord, arg_java);
 335   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 336   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 337   __ z_brl(move_floatSlot_to_FARG);
 338   __ z_std(floatSlot, 0, arg_c);
 339   __ add2reg(arg_c, BytesPerWord);
 340   __ z_bru(loop_start);
 341 
 342   // Method exit, all arguments proocessed.
 343   __ bind(loop_end);
 344   __ pop_frame();
 345   __ restore_return_pc();
 346   __ z_lmg(Z_R10,Z_R13,-32,Z_SP);
 347   __ z_br(Z_R14);
 348 
 349   // Copy int arguments.
 350 
 351   Label  iarg_caselist;   // Distance between each case has to be a power of 2
 352                           // (= 1 << LogSizeOfCase).
 353   __ align(16);
 354   BIND(iarg_caselist);
 355   __ z_lgr(Z_ARG3, intSlot);    // 4 bytes
 356   __ z_bru(loop_start_restore); // 4 bytes
 357 
 358   __ z_lgr(Z_ARG4, intSlot);
 359   __ z_bru(loop_start_restore);
 360 
 361   __ z_lgr(Z_ARG5, intSlot);
 362   __ z_bru(loop_start_restore);
 363 
 364   __ align(16);
 365   __ bind(move_intSlot_to_ARG);
 366   __ z_stg(signature, d_signature, Z_SP);       // Spill since signature == Z_R1_scratch.
 367   __ z_larl(Z_R1_scratch, iarg_caselist);
 368   __ z_sllg(Z_R0_scratch, argcnt, LogSizeOfCase);
 369   __ add2reg(argcnt, 1);
 370   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 371   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 372 
 373   // Copy float arguments.
 374 
 375   Label  farg_caselist;   // Distance between each case has to be a power of 2
 376                           // (= 1 << logSizeOfCase, padded with nop.
 377   __ align(16);
 378   BIND(farg_caselist);
 379   __ z_ldr(Z_FARG1, floatSlot); // 2 bytes
 380   __ z_bru(loop_start_restore); // 4 bytes
 381   __ z_nop();                   // 2 bytes
 382 
 383   __ z_ldr(Z_FARG2, floatSlot);
 384   __ z_bru(loop_start_restore);
 385   __ z_nop();
 386 
 387   __ z_ldr(Z_FARG3, floatSlot);
 388   __ z_bru(loop_start_restore);
 389   __ z_nop();
 390 
 391   __ z_ldr(Z_FARG4, floatSlot);
 392   __ z_bru(loop_start_restore);
 393   __ z_nop();
 394 
 395   __ align(16);
 396   __ bind(move_floatSlot_to_FARG);
 397   __ z_stg(signature, d_signature, Z_SP);        // Spill since signature == Z_R1_scratch.
 398   __ z_lg(Z_R0_scratch, d_fpcnt, Z_SP);          // Need old value for indexing.
 399   __ add2mem_64(Address(Z_SP, d_fpcnt), 1, Z_R1_scratch); // Increment index.
 400   __ z_larl(Z_R1_scratch, farg_caselist);
 401   __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogSizeOfCase);
 402   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 403   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 404 
 405   BLOCK_COMMENT("} slow_signature_handler");
 406 
 407   return __ addr_at(entry_offset);
 408 }
 409 
 410 address TemplateInterpreterGenerator::generate_result_handler_for (BasicType type) {
 411   address entry = __ pc();
 412 
 413   assert(Z_tos == Z_RET, "Result handler: must move result!");
 414   assert(Z_ftos == Z_FRET, "Result handler: must move float result!");
 415 
 416   switch (type) {
 417     case T_BOOLEAN:
 418       __ c2bool(Z_tos);
 419       break;
 420     case T_CHAR:
 421       __ and_imm(Z_tos, 0xffff);
 422       break;
 423     case T_BYTE:
 424       __ z_lbr(Z_tos, Z_tos);
 425       break;
 426     case T_SHORT:
 427       __ z_lhr(Z_tos, Z_tos);
 428       break;
 429     case T_INT:
 430     case T_LONG:
 431     case T_VOID:
 432     case T_FLOAT:
 433     case T_DOUBLE:
 434       break;
 435     case T_OBJECT:
 436       // Retrieve result from frame...
 437       __ mem2reg_opt(Z_tos, Address(Z_fp, oop_tmp_offset));
 438       // and verify it.
 439       __ verify_oop(Z_tos);
 440       break;
 441     default:
 442       ShouldNotReachHere();
 443   }
 444   __ z_br(Z_R14);      // Return from result handler.
 445   return entry;
 446 }
 447 
 448 // Abstract method entry.
 449 // Attempt to execute abstract method. Throw exception.
 450 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 451   unsigned int entry_offset = __ offset();
 452 
 453   // Caller could be the call_stub or a compiled method (x86 version is wrong!).
 454 
 455   BLOCK_COMMENT("abstract_entry {");
 456 
 457   // Implement call of InterpreterRuntime::throw_AbstractMethodError.
 458   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1);
 459   __ save_return_pc();       // Save Z_R14.
 460   __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
 461 
 462   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), Z_thread);
 463 
 464   __ pop_frame();
 465   __ restore_return_pc();    // Restore Z_R14.
 466   __ reset_last_Java_frame();
 467 
 468   // Restore caller sp for c2i case.
 469   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 470 
 471   // branch to SharedRuntime::generate_forward_exception() which handles all possible callers,
 472   // i.e. call stub, compiled method, interpreted method.
 473   __ load_absolute_address(Z_tmp_1, StubRoutines::forward_exception_entry());
 474   __ z_br(Z_tmp_1);
 475 
 476   BLOCK_COMMENT("} abstract_entry");
 477 
 478   return __ addr_at(entry_offset);
 479 }
 480 
 481 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 482 #if INCLUDE_ALL_GCS
 483   if (UseG1GC) {
 484     // Inputs:
 485     //  Z_ARG1 - receiver
 486     //
 487     // What we do:
 488     //  - Load the referent field address.
 489     //  - Load the value in the referent field.
 490     //  - Pass that value to the pre-barrier.
 491     //
 492     // In the case of G1 this will record the value of the
 493     // referent in an SATB buffer if marking is active.
 494     // This will cause concurrent marking to mark the referent
 495     // field as live.
 496 
 497     Register  scratch1 = Z_tmp_2;
 498     Register  scratch2 = Z_tmp_3;
 499     Register  pre_val  = Z_RET;   // return value
 500     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
 501     Register  Rargp    = Z_esp;
 502 
 503     Label     slow_path;
 504     address   entry = __ pc();
 505 
 506     const int referent_offset = java_lang_ref_Reference::referent_offset;
 507     guarantee(referent_offset > 0, "referent offset not initialized");
 508 
 509     BLOCK_COMMENT("Reference_get {");
 510 
 511     //  If the receiver is null then it is OK to jump to the slow path.
 512     __ load_and_test_long(pre_val, Address(Rargp, Interpreter::stackElementSize)); // Get receiver.
 513     __ z_bre(slow_path);
 514 
 515     //  Load the value of the referent field.
 516     __ load_heap_oop(pre_val, referent_offset, pre_val);
 517 
 518     // Restore caller sp for c2i case.
 519     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 520 
 521     // Generate the G1 pre-barrier code to log the value of
 522     // the referent field in an SATB buffer.
 523     // Note:
 524     //   With these parameters the write_barrier_pre does not
 525     //   generate instructions to load the previous value.
 526     __ g1_write_barrier_pre(noreg,      // obj
 527                             noreg,      // offset
 528                             pre_val,    // pre_val
 529                             noreg,      // no new val to preserve
 530                             scratch1,   // tmp
 531                             scratch2,   // tmp
 532                             true);      // pre_val_needed
 533 
 534     __ z_br(Z_R14);
 535 
 536     // Branch to previously generated regular method entry.
 537     __ bind(slow_path);
 538 
 539     address meth_entry = Interpreter::entry_for_kind(Interpreter::zerolocals);
 540     __ jump_to_entry(meth_entry, Z_R1);
 541 
 542     BLOCK_COMMENT("} Reference_get");
 543 
 544     return entry;
 545   }
 546 #endif // INCLUDE_ALL_GCS
 547 
 548   return NULL;
 549 }
 550 
 551 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 552   address entry = __ pc();
 553 
 554   DEBUG_ONLY(__ verify_esp(Z_esp, Z_ARG5));
 555 
 556   // Restore bcp under the assumption that the current frame is still
 557   // interpreted.
 558   __ restore_bcp();
 559 
 560   // Expression stack must be empty before entering the VM if an
 561   // exception happened.
 562   __ empty_expression_stack();
 563   // Throw exception.
 564   __ call_VM(noreg,
 565              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
 566   return entry;
 567 }
 568 
 569 //
 570 // Args:
 571 //   Z_ARG3: aberrant index
 572 //
 573 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char * name) {
 574   address entry = __ pc();
 575   address excp = CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException);
 576 
 577   // Expression stack must be empty before entering the VM if an
 578   // exception happened.
 579   __ empty_expression_stack();
 580 
 581   // Setup parameters.
 582   // Leave out the name and use register for array to create more detailed exceptions.
 583   __ load_absolute_address(Z_ARG2, (address) name);
 584   __ call_VM(noreg, excp, Z_ARG2, Z_ARG3);
 585   return entry;
 586 }
 587 
 588 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 589   address entry = __ pc();
 590 
 591   // Object is at TOS.
 592   __ pop_ptr(Z_ARG2);
 593 
 594   // Expression stack must be empty before entering the VM if an
 595   // exception happened.
 596   __ empty_expression_stack();
 597 
 598   __ call_VM(Z_ARG1,
 599              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException),
 600              Z_ARG2);
 601 
 602   DEBUG_ONLY(__ should_not_reach_here();)
 603 
 604   return entry;
 605 }
 606 
 607 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 608   assert(!pass_oop || message == NULL, "either oop or message but not both");
 609   address entry = __ pc();
 610 
 611   BLOCK_COMMENT("exception_handler_common {");
 612 
 613   // Expression stack must be empty before entering the VM if an
 614   // exception happened.
 615   __ empty_expression_stack();
 616   if (name != NULL) {
 617     __ load_absolute_address(Z_ARG2, (address)name);
 618   } else {
 619     __ clear_reg(Z_ARG2, true, false);
 620   }
 621 
 622   if (pass_oop) {
 623     __ call_VM(Z_tos,
 624                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception),
 625                Z_ARG2, Z_tos /*object (see TT::aastore())*/);
 626   } else {
 627     if (message != NULL) {
 628       __ load_absolute_address(Z_ARG3, (address)message);
 629     } else {
 630       __ clear_reg(Z_ARG3, true, false);
 631     }
 632     __ call_VM(Z_tos,
 633                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 634                Z_ARG2, Z_ARG3);
 635   }
 636   // Throw exception.
 637   __ load_absolute_address(Z_R1_scratch, Interpreter::throw_exception_entry());
 638   __ z_br(Z_R1_scratch);
 639 
 640   BLOCK_COMMENT("} exception_handler_common");
 641 
 642   return entry;
 643 }
 644 
 645 // Unused, should never pass by.
 646 address TemplateInterpreterGenerator::generate_continuation_for (TosState state) {
 647   address entry = __ pc();
 648   __ should_not_reach_here();
 649   return entry;
 650 }
 651 
 652 address TemplateInterpreterGenerator::generate_return_entry_for (TosState state, int step, size_t index_size) {
 653   address entry = __ pc();
 654 
 655   BLOCK_COMMENT("return_entry {");
 656 
 657   // Pop i2c extension or revert top-2-parent-resize done by interpreted callees.
 658   Register sp_before_i2c_extension = Z_bcp;
 659   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 660   __ z_lg(sp_before_i2c_extension, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
 661   __ resize_frame_absolute(sp_before_i2c_extension, Z_locals/*tmp*/, true/*load_fp*/);
 662 
 663   // TODO(ZASM): necessary??
 664   //  // and NULL it as marker that esp is now tos until next java call
 665   //  __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 666 
 667   __ restore_bcp();
 668   __ restore_locals();
 669   __ restore_esp();
 670 
 671   if (state == atos) {
 672     __ profile_return_type(Z_tmp_1, Z_tos, Z_tmp_2);
 673   }
 674 
 675   Register cache  = Z_tmp_1;
 676   Register size   = Z_tmp_1;
 677   Register offset = Z_tmp_2;
 678   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
 679                                     ConstantPoolCacheEntry::flags_offset());
 680   __ get_cache_and_index_at_bcp(cache, offset, 1, index_size);
 681 
 682   // #args is in rightmost byte of the _flags field.
 683   __ z_llgc(size, Address(cache, offset, flags_offset+(sizeof(size_t)-1)));
 684   __ z_sllg(size, size, Interpreter::logStackElementSize); // Each argument size in bytes.
 685   __ z_agr(Z_esp, size);                                   // Pop arguments.
 686   __ dispatch_next(state, step);
 687 
 688   BLOCK_COMMENT("} return_entry");
 689 
 690   return entry;
 691 }
 692 
 693 address TemplateInterpreterGenerator::generate_deopt_entry_for (TosState state,
 694                                                                int step) {
 695   address entry = __ pc();
 696 
 697   BLOCK_COMMENT("deopt_entry {");
 698 
 699   // TODO(ZASM): necessary? NULL last_sp until next java call
 700   // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 701   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 702   __ restore_bcp();
 703   __ restore_locals();
 704   __ restore_esp();
 705 
 706   // Handle exceptions.
 707   {
 708     Label L;
 709     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
 710     __ z_bre(L);
 711     __ call_VM(noreg,
 712                CAST_FROM_FN_PTR(address,
 713                                 InterpreterRuntime::throw_pending_exception));
 714     __ should_not_reach_here();
 715     __ bind(L);
 716   }
 717   __ dispatch_next(state, step);
 718 
 719   BLOCK_COMMENT("} deopt_entry");
 720 
 721   return entry;
 722 }
 723 
 724 address TemplateInterpreterGenerator::generate_safept_entry_for (TosState state,
 725                                                                 address runtime_entry) {
 726   address entry = __ pc();
 727   __ push(state);
 728   __ call_VM(noreg, runtime_entry);
 729   __ dispatch_via(vtos, Interpreter::_normal_table.table_for (vtos));
 730   return entry;
 731 }
 732 
 733 //
 734 // Helpers for commoning out cases in the various type of method entries.
 735 //
 736 
 737 // Increment invocation count & check for overflow.
 738 //
 739 // Note: checking for negative value instead of overflow
 740 // so we have a 'sticky' overflow test.
 741 //
 742 // Z_ARG2: method (see generate_fixed_frame())
 743 //
 744 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 745   Label done;
 746   Register method = Z_ARG2; // Generate_fixed_frame() copies Z_method into Z_ARG2.
 747   Register m_counters = Z_ARG4;
 748 
 749   BLOCK_COMMENT("counter_incr {");
 750 
 751   // Note: In tiered we increment either counters in method or in MDO depending
 752   // if we are profiling or not.
 753   if (TieredCompilation) {
 754     int increment = InvocationCounter::count_increment;
 755     if (ProfileInterpreter) {
 756       NearLabel no_mdo;
 757       Register mdo = m_counters;
 758       // Are we profiling?
 759       __ load_and_test_long(mdo, method2_(method, method_data));
 760       __ branch_optimized(Assembler::bcondZero, no_mdo);
 761       // Increment counter in the MDO.
 762       const Address mdo_invocation_counter(mdo, MethodData::invocation_counter_offset() +
 763                                            InvocationCounter::counter_offset());
 764       const Address mask(mdo, MethodData::invoke_mask_offset());
 765       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 766                                  Z_R1_scratch, false, Assembler::bcondZero,
 767                                  overflow);
 768       __ z_bru(done);
 769       __ bind(no_mdo);
 770     }
 771 
 772     // Increment counter in MethodCounters.
 773     const Address invocation_counter(m_counters,
 774                                      MethodCounters::invocation_counter_offset() +
 775                                      InvocationCounter::counter_offset());
 776     // Get address of MethodCounters object.
 777     __ get_method_counters(method, m_counters, done);
 778     const Address mask(m_counters, MethodCounters::invoke_mask_offset());
 779     __ increment_mask_and_jump(invocation_counter,
 780                                increment, mask,
 781                                Z_R1_scratch, false, Assembler::bcondZero,
 782                                overflow);
 783   } else {
 784     Register counter_sum = Z_ARG3; // The result of this piece of code.
 785     Register tmp         = Z_R1_scratch;
 786 #ifdef ASSERT
 787     {
 788       NearLabel ok;
 789       __ get_method(tmp);
 790       __ compare64_and_branch(method, tmp, Assembler::bcondEqual, ok);
 791       __ z_illtrap(0x66);
 792       __ bind(ok);
 793     }
 794 #endif
 795 
 796     // Get address of MethodCounters object.
 797     __ get_method_counters(method, m_counters, done);
 798     // Update standard invocation counters.
 799     __ increment_invocation_counter(m_counters, counter_sum);
 800     if (ProfileInterpreter) {
 801       __ add2mem_32(Address(m_counters, MethodCounters::interpreter_invocation_counter_offset()), 1, tmp);
 802       if (profile_method != NULL) {
 803         const Address profile_limit(m_counters, MethodCounters::interpreter_profile_limit_offset());
 804         __ z_cl(counter_sum, profile_limit);
 805         __ branch_optimized(Assembler::bcondLow, *profile_method_continue);
 806         // If no method data exists, go to profile_method.
 807         __ test_method_data_pointer(tmp, *profile_method);
 808       }
 809     }
 810 
 811     const Address invocation_limit(m_counters, MethodCounters::interpreter_invocation_limit_offset());
 812     __ z_cl(counter_sum, invocation_limit);
 813     __ branch_optimized(Assembler::bcondNotLow, *overflow);
 814   }
 815 
 816   __ bind(done);
 817 
 818   BLOCK_COMMENT("} counter_incr");
 819 }
 820 
 821 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 822   // InterpreterRuntime::frequency_counter_overflow takes two
 823   // arguments, the first (thread) is passed by call_VM, the second
 824   // indicates if the counter overflow occurs at a backwards branch
 825   // (NULL bcp). We pass zero for it. The call returns the address
 826   // of the verified entry point for the method or NULL if the
 827   // compilation did not complete (either went background or bailed
 828   // out).
 829   __ clear_reg(Z_ARG2);
 830   __ call_VM(noreg,
 831              CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow),
 832              Z_ARG2);
 833   __ z_bru(do_continue);
 834 }
 835 
 836 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register frame_size, Register tmp1) {
 837   Register tmp2 = Z_R1_scratch;
 838   const int page_size = os::vm_page_size();
 839   NearLabel after_frame_check;
 840 
 841   BLOCK_COMMENT("counter_overflow {");
 842 
 843   assert_different_registers(frame_size, tmp1);
 844 
 845   // Stack banging is sufficient overflow check if frame_size < page_size.
 846   if (Immediate::is_uimm(page_size, 15)) {
 847     __ z_chi(frame_size, page_size);
 848     __ z_brl(after_frame_check);
 849   } else {
 850     __ load_const_optimized(tmp1, page_size);
 851     __ compareU32_and_branch(frame_size, tmp1, Assembler::bcondLow, after_frame_check);
 852   }
 853 
 854   // Get the stack base, and in debug, verify it is non-zero.
 855   __ z_lg(tmp1, thread_(stack_base));
 856 #ifdef ASSERT
 857   address reentry = NULL;
 858   NearLabel base_not_zero;
 859   __ compareU64_and_branch(tmp1, (intptr_t)0L, Assembler::bcondNotEqual, base_not_zero);
 860   reentry = __ stop_chain_static(reentry, "stack base is zero in generate_stack_overflow_check");
 861   __ bind(base_not_zero);
 862 #endif
 863 
 864   // Get the stack size, and in debug, verify it is non-zero.
 865   assert(sizeof(size_t) == sizeof(intptr_t), "wrong load size");
 866   __ z_lg(tmp2, thread_(stack_size));
 867 #ifdef ASSERT
 868   NearLabel size_not_zero;
 869   __ compareU64_and_branch(tmp2, (intptr_t)0L, Assembler::bcondNotEqual, size_not_zero);
 870   reentry = __ stop_chain_static(reentry, "stack size is zero in generate_stack_overflow_check");
 871   __ bind(size_not_zero);
 872 #endif
 873 
 874   // Compute the beginning of the protected zone minus the requested frame size.
 875   __ z_sgr(tmp1, tmp2);
 876   __ add2reg(tmp1, JavaThread::stack_guard_zone_size());
 877 
 878   // Add in the size of the frame (which is the same as subtracting it from the
 879   // SP, which would take another register.
 880   __ z_agr(tmp1, frame_size);
 881 
 882   // The frame is greater than one page in size, so check against
 883   // the bottom of the stack.
 884   __ compareU64_and_branch(Z_SP, tmp1, Assembler::bcondHigh, after_frame_check);
 885 
 886   // The stack will overflow, throw an exception.
 887 
 888   // Restore SP to sender's sp. This is necessary if the sender's frame is an
 889   // extended compiled frame (see gen_c2i_adapter()) and safer anyway in case of
 890   // JSR292 adaptations.
 891   __ resize_frame_absolute(Z_R10, tmp1, true/*load_fp*/);
 892 
 893   // Note also that the restored frame is not necessarily interpreted.
 894   // Use the shared runtime version of the StackOverflowError.
 895   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 896   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 897   __ load_absolute_address(tmp1, StubRoutines::throw_StackOverflowError_entry());
 898   __ z_br(tmp1);
 899 
 900   // If you get to here, then there is enough stack space.
 901   __ bind(after_frame_check);
 902 
 903   BLOCK_COMMENT("} counter_overflow");
 904 }
 905 
 906 // Allocate monitor and lock method (asm interpreter).
 907 //
 908 // Args:
 909 //   Z_locals: locals
 910 
 911 void TemplateInterpreterGenerator::lock_method(void) {
 912 
 913   BLOCK_COMMENT("lock_method {");
 914 
 915   // Synchronize method.
 916   const Register method = Z_tmp_2;
 917   __ get_method(method);
 918 
 919 #ifdef ASSERT
 920   address reentry = NULL;
 921   {
 922     Label L;
 923     __ testbit(method2_(method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 924     __ z_btrue(L);
 925     reentry = __ stop_chain_static(reentry, "method doesn't need synchronization");
 926     __ bind(L);
 927   }
 928 #endif // ASSERT
 929 
 930   // Get synchronization object.
 931   const Register object = Z_tmp_2;
 932 
 933   {
 934     Label     done;
 935     Label     static_method;
 936 
 937     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 938     __ z_btrue(static_method);
 939 
 940     // non-static method: Load receiver obj from stack.
 941     __ mem2reg_opt(object, Address(Z_locals, Interpreter::local_offset_in_bytes(0)));
 942     __ z_bru(done);
 943 
 944     __ bind(static_method);
 945 
 946     // Lock the java mirror.
 947     __ load_mirror(object, method);
 948 #ifdef ASSERT
 949     {
 950       NearLabel L;
 951       __ compare64_and_branch(object, (intptr_t) 0, Assembler::bcondNotEqual, L);
 952       reentry = __ stop_chain_static(reentry, "synchronization object is NULL");
 953       __ bind(L);
 954     }
 955 #endif // ASSERT
 956 
 957     __ bind(done);
 958   }
 959 
 960   __ add_monitor_to_stack(true, Z_ARG3, Z_ARG4, Z_ARG5); // Allocate monitor elem.
 961   // Store object and lock it.
 962   __ get_monitors(Z_tmp_1);
 963   __ reg2mem_opt(object, Address(Z_tmp_1, BasicObjectLock::obj_offset_in_bytes()));
 964   __ lock_object(Z_tmp_1, object);
 965 
 966   BLOCK_COMMENT("} lock_method");
 967 }
 968 
 969 // Generate a fixed interpreter frame. This is identical setup for
 970 // interpreted methods and for native methods hence the shared code.
 971 //
 972 // Registers alive
 973 //   Z_thread   - JavaThread*
 974 //   Z_SP       - old stack pointer
 975 //   Z_method   - callee's method
 976 //   Z_esp      - parameter list (slot 'above' last param)
 977 //   Z_R14      - return pc, to be stored in caller's frame
 978 //   Z_R10      - sender sp, note: Z_tmp_1 is Z_R10!
 979 //
 980 // Registers updated
 981 //   Z_SP       - new stack pointer
 982 //   Z_esp      - callee's operand stack pointer
 983 //                points to the slot above the value on top
 984 //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
 985 //   Z_bcp      - the bytecode pointer
 986 //   Z_fp       - the frame pointer, thereby killing Z_method
 987 //   Z_ARG2     - copy of Z_method
 988 //
 989 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 990 
 991   //  stack layout
 992   //
 993   //   F1 [TOP_IJAVA_FRAME_ABI]              <-- Z_SP, Z_R10 (see note below)
 994   //      [F1's operand stack (unused)]
 995   //      [F1's outgoing Java arguments]     <-- Z_esp
 996   //      [F1's operand stack (non args)]
 997   //      [monitors]      (optional)
 998   //      [IJAVA_STATE]
 999   //
1000   //   F2 [PARENT_IJAVA_FRAME_ABI]
1001   //      ...
1002   //
1003   //  0x000
1004   //
1005   // Note: Z_R10, the sender sp, will be below Z_SP if F1 was extended by a c2i adapter.
1006 
1007   //=============================================================================
1008   // Allocate space for locals other than the parameters, the
1009   // interpreter state, monitors, and the expression stack.
1010 
1011   const Register local_count     = Z_ARG5;
1012   const Register fp              = Z_tmp_2;
1013 
1014   BLOCK_COMMENT("generate_fixed_frame {");
1015 
1016   {
1017   // local registers
1018   const Register top_frame_size  = Z_ARG2;
1019   const Register sp_after_resize = Z_ARG3;
1020   const Register max_stack       = Z_ARG4;
1021 
1022   // local_count = method->constMethod->max_locals();
1023   __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1024   __ z_llgh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1025 
1026   if (native_call) {
1027     // If we're calling a native method, we replace max_stack (which is
1028     // zero) with space for the worst-case signature handler varargs
1029     // vector, which is:
1030     //   max_stack = max(Argument::n_register_parameters, parameter_count+2);
1031     //
1032     // We add two slots to the parameter_count, one for the jni
1033     // environment and one for a possible native mirror. We allocate
1034     // space for at least the number of ABI registers, even though
1035     // InterpreterRuntime::slow_signature_handler won't write more than
1036     // parameter_count+2 words when it creates the varargs vector at the
1037     // top of the stack. The generated slow signature handler will just
1038     // load trash into registers beyond the necessary number. We're
1039     // still going to cut the stack back by the ABI register parameter
1040     // count so as to get SP+16 pointing at the ABI outgoing parameter
1041     // area, so we need to allocate at least that much even though we're
1042     // going to throw it away.
1043     //
1044 
1045     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1046     __ z_llgh(max_stack,  Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1047     __ add2reg(max_stack, 2);
1048 
1049     NearLabel passing_args_on_stack;
1050 
1051     // max_stack in bytes
1052     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1053 
1054     int argument_registers_in_bytes = Argument::n_register_parameters << LogBytesPerWord;
1055     __ compare64_and_branch(max_stack, argument_registers_in_bytes, Assembler::bcondNotLow, passing_args_on_stack);
1056 
1057     __ load_const_optimized(max_stack, argument_registers_in_bytes);
1058 
1059     __ bind(passing_args_on_stack);
1060   } else {
1061     // !native_call
1062     __ z_lg(max_stack, method_(const));
1063 
1064     // Calculate number of non-parameter locals (in slots):
1065     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1066     __ z_sh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1067 
1068     // max_stack = method->max_stack();
1069     __ z_llgh(max_stack, Address(max_stack, ConstMethod::max_stack_offset()));
1070     // max_stack in bytes
1071     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1072   }
1073 
1074   // Resize (i.e. normally shrink) the top frame F1 ...
1075   //   F1      [TOP_IJAVA_FRAME_ABI]          <-- Z_SP, Z_R10
1076   //           F1's operand stack (free)
1077   //           ...
1078   //           F1's operand stack (free)      <-- Z_esp
1079   //           F1's outgoing Java arg m
1080   //           ...
1081   //           F1's outgoing Java arg 0
1082   //           ...
1083   //
1084   //  ... into a parent frame (Z_R10 holds F1's SP before any modification, see also above)
1085   //
1086   //           +......................+
1087   //           :                      :        <-- Z_R10, saved below as F0's z_ijava_state.sender_sp
1088   //           :                      :
1089   //   F1      [PARENT_IJAVA_FRAME_ABI]        <-- Z_SP       \
1090   //           F0's non arg local                             | = delta
1091   //           ...                                            |
1092   //           F0's non arg local              <-- Z_esp      /
1093   //           F1's outgoing Java arg m
1094   //           ...
1095   //           F1's outgoing Java arg 0
1096   //           ...
1097   //
1098   // then push the new top frame F0.
1099   //
1100   //   F0      [TOP_IJAVA_FRAME_ABI]    = frame::z_top_ijava_frame_abi_size \
1101   //           [operand stack]          = max_stack                          | = top_frame_size
1102   //           [IJAVA_STATE]            = frame::z_ijava_state_size         /
1103 
1104   // sp_after_resize = Z_esp - delta
1105   //
1106   // delta = PARENT_IJAVA_FRAME_ABI + (locals_count - params_count)
1107 
1108   __ add2reg(sp_after_resize, (Interpreter::stackElementSize) - (frame::z_parent_ijava_frame_abi_size), Z_esp);
1109   __ z_sllg(Z_R0_scratch, local_count, LogBytesPerWord); // Params have already been subtracted from local_count.
1110   __ z_slgr(sp_after_resize, Z_R0_scratch);
1111 
1112   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
1113   __ add2reg(top_frame_size,
1114              frame::z_top_ijava_frame_abi_size +
1115                frame::z_ijava_state_size +
1116                frame::interpreter_frame_monitor_size() * wordSize,
1117              max_stack);
1118 
1119   // Check if there's room for the new frame...
1120   Register frame_size = max_stack; // Reuse the regiser for max_stack.
1121   __ z_lgr(frame_size, Z_SP);
1122   __ z_sgr(frame_size, sp_after_resize);
1123   __ z_agr(frame_size, top_frame_size);
1124   generate_stack_overflow_check(frame_size, fp/*tmp1*/);
1125 
1126   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
1127   __ asm_assert_eq("killed Z_R14", 0);
1128   __ resize_frame_absolute(sp_after_resize, fp, true);
1129   __ save_return_pc(Z_R14);
1130 
1131   // ... and push the new frame F0.
1132   __ push_frame(top_frame_size, fp, true /*copy_sp*/, false);
1133   }
1134 
1135   //=============================================================================
1136   // Initialize the new frame F0: initialize interpreter state.
1137 
1138   {
1139   // locals
1140   const Register local_addr = Z_ARG4;
1141 
1142   BLOCK_COMMENT("generate_fixed_frame: initialize interpreter state {");
1143 
1144 #ifdef ASSERT
1145   // Set the magic number (using local_addr as tmp register).
1146   __ load_const_optimized(local_addr, frame::z_istate_magic_number);
1147   __ z_stg(local_addr, _z_ijava_state_neg(magic), fp);
1148 #endif
1149 
1150   // Save sender SP from F1 (i.e. before it was potentially modified by an
1151   // adapter) into F0's interpreter state. We us it as well to revert
1152   // resizing the frame above.
1153   __ z_stg(Z_R10, _z_ijava_state_neg(sender_sp), fp);
1154 
1155   // Load cp cache and save it at the and of this block.
1156   __ z_lg(Z_R1_scratch, Address(Z_method,    Method::const_offset()));
1157   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstMethod::constants_offset()));
1158   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstantPool::cache_offset_in_bytes()));
1159 
1160   // z_ijava_state->method = method;
1161   __ z_stg(Z_method, _z_ijava_state_neg(method), fp);
1162 
1163   // Point locals at the first argument. Method's locals are the
1164   // parameters on top of caller's expression stack.
1165   // Tos points past last Java argument.
1166 
1167   __ z_lg(Z_locals, Address(Z_method, Method::const_offset()));
1168   __ z_llgh(Z_locals /*parameter_count words*/,
1169             Address(Z_locals, ConstMethod::size_of_parameters_offset()));
1170   __ z_sllg(Z_locals /*parameter_count bytes*/, Z_locals /*parameter_count*/, LogBytesPerWord);
1171   __ z_agr(Z_locals, Z_esp);
1172   // z_ijava_state->locals - i*BytesPerWord points to i-th Java local (i starts at 0)
1173   // z_ijava_state->locals = Z_esp + parameter_count bytes
1174   __ z_stg(Z_locals, _z_ijava_state_neg(locals), fp);
1175 
1176   // z_ijava_state->oop_temp = NULL;
1177   __ store_const(Address(fp, oop_tmp_offset), 0);
1178 
1179   // Initialize z_ijava_state->mdx.
1180   Register Rmdp = Z_bcp;
1181   // native_call: assert that mdo == NULL
1182   const bool check_for_mdo = !native_call DEBUG_ONLY(|| native_call);
1183   if (ProfileInterpreter && check_for_mdo) {
1184 #ifdef FAST_DISPATCH
1185     // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
1186     // they both use I2.
1187     assert(0, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
1188 #endif // FAST_DISPATCH
1189     Label get_continue;
1190 
1191     __ load_and_test_long(Rmdp, method_(method_data));
1192     __ z_brz(get_continue);
1193     DEBUG_ONLY(if (native_call) __ stop("native methods don't have a mdo"));
1194     __ add2reg(Rmdp, in_bytes(MethodData::data_offset()));
1195     __ bind(get_continue);
1196   }
1197   __ z_stg(Rmdp, _z_ijava_state_neg(mdx), fp);
1198 
1199   // Initialize z_ijava_state->bcp and Z_bcp.
1200   if (native_call) {
1201     __ clear_reg(Z_bcp); // Must initialize. Will get written into frame where GC reads it.
1202   } else {
1203     __ z_lg(Z_bcp, method_(const));
1204     __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset()));
1205   }
1206   __ z_stg(Z_bcp, _z_ijava_state_neg(bcp), fp);
1207 
1208   // no monitors and empty operand stack
1209   // => z_ijava_state->monitors points to the top slot in IJAVA_STATE.
1210   // => Z_ijava_state->esp points one slot above into the operand stack.
1211   // z_ijava_state->monitors = fp - frame::z_ijava_state_size - Interpreter::stackElementSize;
1212   // z_ijava_state->esp = Z_esp = z_ijava_state->monitors;
1213   __ add2reg(Z_esp, -frame::z_ijava_state_size, fp);
1214   __ z_stg(Z_esp, _z_ijava_state_neg(monitors), fp);
1215   __ add2reg(Z_esp, -Interpreter::stackElementSize);
1216   __ z_stg(Z_esp, _z_ijava_state_neg(esp), fp);
1217 
1218   // z_ijava_state->cpoolCache = Z_R1_scratch (see load above);
1219   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(cpoolCache), fp);
1220 
1221   // Get mirror and store it in the frame as GC root for this Method*.
1222   __ load_mirror(Z_R1_scratch, Z_method);
1223   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(mirror), fp);
1224 
1225   BLOCK_COMMENT("} generate_fixed_frame: initialize interpreter state");
1226 
1227   //=============================================================================
1228   if (!native_call) {
1229     // Fill locals with 0x0s.
1230     NearLabel locals_zeroed;
1231     NearLabel doXC;
1232 
1233     // Local_count is already num_locals_slots - num_param_slots.
1234     __ compare64_and_branch(local_count, (intptr_t)0L, Assembler::bcondNotHigh, locals_zeroed);
1235 
1236     // Advance local_addr to point behind locals (creates positive incr. in loop).
1237     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1238     __ z_llgh(Z_R0_scratch,
1239               Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1240     if (Z_R0_scratch == Z_R0) {
1241       __ z_aghi(Z_R0_scratch, -1);
1242     } else {
1243       __ add2reg(Z_R0_scratch, -1);
1244     }
1245     __ z_lgr(local_addr/*locals*/, Z_locals);
1246     __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogBytesPerWord);
1247     __ z_sllg(local_count, local_count, LogBytesPerWord); // Local_count are non param locals.
1248     __ z_sgr(local_addr, Z_R0_scratch);
1249 
1250     if (VM_Version::has_Prefetch()) {
1251       __ z_pfd(0x02, 0, Z_R0, local_addr);
1252       __ z_pfd(0x02, 256, Z_R0, local_addr);
1253     }
1254 
1255     // Can't optimise for Z10 using "compare and branch" (immediate value is too big).
1256     __ z_cghi(local_count, 256);
1257     __ z_brnh(doXC);
1258 
1259     // MVCLE: Initialize if quite a lot locals.
1260     //  __ bind(doMVCLE);
1261     __ z_lgr(Z_R0_scratch, local_addr);
1262     __ z_lgr(Z_R1_scratch, local_count);
1263     __ clear_reg(Z_ARG2);        // Src len of MVCLE is zero.
1264 
1265     __ MacroAssembler::move_long_ext(Z_R0_scratch, Z_ARG1, 0);
1266     __ z_bru(locals_zeroed);
1267 
1268     Label  XC_template;
1269     __ bind(XC_template);
1270     __ z_xc(0, 0, local_addr, 0, local_addr);
1271 
1272     __ bind(doXC);
1273     __ z_bctgr(local_count, Z_R0);                  // Get #bytes-1 for EXECUTE.
1274     if (VM_Version::has_ExecuteExtensions()) {
1275       __ z_exrl(local_count, XC_template);          // Execute XC with variable length.
1276     } else {
1277       __ z_larl(Z_R1_scratch, XC_template);
1278       __ z_ex(local_count, 0, Z_R0, Z_R1_scratch);  // Execute XC with variable length.
1279     }
1280 
1281     __ bind(locals_zeroed);
1282   }
1283 
1284   }
1285   // Finally set the frame pointer, destroying Z_method.
1286   assert(Z_fp == Z_method, "maybe set Z_fp earlier if other register than Z_method");
1287   // Oprofile analysis suggests to keep a copy in a register to be used by
1288   // generate_counter_incr().
1289   __ z_lgr(Z_ARG2, Z_method);
1290   __ z_lgr(Z_fp, fp);
1291 
1292   BLOCK_COMMENT("} generate_fixed_frame");
1293 }
1294 
1295 // Various method entries
1296 
1297 // Math function, frame manager must set up an interpreter state, etc.
1298 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1299 
1300   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1301   bool use_instruction = false;
1302   address runtime_entry = NULL;
1303   int num_args = 1;
1304   bool double_precision = true;
1305 
1306   // s390 specific:
1307   switch (kind) {
1308     case Interpreter::java_lang_math_sqrt:
1309     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1310     case Interpreter::java_lang_math_fmaF:
1311     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1312     default: break; // Fall back to runtime call.
1313   }
1314 
1315   switch (kind) {
1316     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1317     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1318     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1319     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1320     case Interpreter::java_lang_math_sqrt : /* runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); not available */ break;
1321     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1322     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1323     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1324     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1325     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1326     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1327     default: ShouldNotReachHere();
1328   }
1329 
1330   // Use normal entry if neither instruction nor runtime call is used.
1331   if (!use_instruction && runtime_entry == NULL) return NULL;
1332 
1333   address entry = __ pc();
1334 
1335   if (use_instruction) {
1336     switch (kind) {
1337       case Interpreter::java_lang_math_sqrt:
1338         // Can use memory operand directly.
1339         __ z_sqdb(Z_FRET, Interpreter::stackElementSize, Z_esp);
1340         break;
1341       case Interpreter::java_lang_math_abs:
1342         // Load operand from stack.
1343         __ mem2freg_opt(Z_FRET, Address(Z_esp, Interpreter::stackElementSize));
1344         __ z_lpdbr(Z_FRET);
1345         break;
1346       case Interpreter::java_lang_math_fmaF:
1347         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1348         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize)); // arg1
1349         __ z_maeb(Z_FRET, Z_FARG2, Address(Z_esp, 2 * Interpreter::stackElementSize));
1350         break;
1351       case Interpreter::java_lang_math_fmaD:
1352         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1353         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 5 * Interpreter::stackElementSize)); // arg1
1354         __ z_madb(Z_FRET, Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize));
1355         break;
1356       default: ShouldNotReachHere();
1357     }
1358   } else {
1359     // Load arguments
1360     assert(num_args <= 4, "passed in registers");
1361     if (double_precision) {
1362       int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1363       for (int i = 0; i < num_args; ++i) {
1364         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1365         offset -= 2 * Interpreter::stackElementSize;
1366       }
1367     } else {
1368       int offset = num_args * Interpreter::stackElementSize;
1369       for (int i = 0; i < num_args; ++i) {
1370         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1371         offset -= Interpreter::stackElementSize;
1372       }
1373     }
1374     // Call runtime
1375     __ save_return_pc();       // Save Z_R14.
1376     __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
1377 
1378     __ call_VM_leaf(runtime_entry);
1379 
1380     __ pop_frame();
1381     __ restore_return_pc();    // Restore Z_R14.
1382   }
1383 
1384   // Pop c2i arguments (if any) off when we return.
1385   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1386 
1387   __ z_br(Z_R14);
1388 
1389   return entry;
1390 }
1391 
1392 // Interpreter stub for calling a native method. (asm interpreter).
1393 // This sets up a somewhat different looking stack for calling the
1394 // native method than the typical interpreter frame setup.
1395 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1396   // Determine code generation flags.
1397   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1398 
1399   // Interpreter entry for ordinary Java methods.
1400   //
1401   // Registers alive
1402   //   Z_SP          - stack pointer
1403   //   Z_thread      - JavaThread*
1404   //   Z_method      - callee's method (method to be invoked)
1405   //   Z_esp         - operand (or expression) stack pointer of caller. one slot above last arg.
1406   //   Z_R10         - sender sp (before modifications, e.g. by c2i adapter
1407   //                   and as well by generate_fixed_frame below)
1408   //   Z_R14         - return address to caller (call_stub or c2i_adapter)
1409   //
1410   // Registers updated
1411   //   Z_SP          - stack pointer
1412   //   Z_fp          - callee's framepointer
1413   //   Z_esp         - callee's operand stack pointer
1414   //                   points to the slot above the value on top
1415   //   Z_locals      - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1416   //   Z_tos         - integer result, if any
1417   //   z_ftos        - floating point result, if any
1418   //
1419   // Stack layout at this point:
1420   //
1421   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1422   //                                                          frame was extended by c2i adapter)
1423   //           [outgoing Java arguments]     <-- Z_esp
1424   //           ...
1425   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1426   //           ...
1427   //
1428 
1429   address entry_point = __ pc();
1430 
1431   // Make sure registers are different!
1432   assert_different_registers(Z_thread, Z_method, Z_esp);
1433 
1434   BLOCK_COMMENT("native_entry {");
1435 
1436   // Make sure method is native and not abstract.
1437 #ifdef ASSERT
1438   address reentry = NULL;
1439   { Label L;
1440     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1441     __ z_btrue(L);
1442     reentry = __ stop_chain_static(reentry, "tried to execute non-native method as native");
1443     __ bind(L);
1444   }
1445   { Label L;
1446     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1447     __ z_bfalse(L);
1448     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1449     __ bind(L);
1450   }
1451 #endif // ASSERT
1452 
1453 #ifdef ASSERT
1454   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1455   __ save_return_pc(Z_R14);
1456 #endif
1457 
1458   // Generate the code to allocate the interpreter stack frame.
1459   generate_fixed_frame(true);
1460 
1461   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1462   // Since at this point in the method invocation the exception handler
1463   // would try to exit the monitor of synchronized methods which hasn't
1464   // been entered yet, we set the thread local variable
1465   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1466   // runtime, exception handling i.e. unlock_if_synchronized_method will
1467   // check this thread local flag.
1468   __ z_mvi(do_not_unlock_if_synchronized, true);
1469 
1470   // Increment invocation count and check for overflow.
1471   NearLabel invocation_counter_overflow;
1472   if (inc_counter) {
1473     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1474   }
1475 
1476   Label continue_after_compile;
1477   __ bind(continue_after_compile);
1478 
1479   bang_stack_shadow_pages(true);
1480 
1481   // Reset the _do_not_unlock_if_synchronized flag.
1482   __ z_mvi(do_not_unlock_if_synchronized, false);
1483 
1484   // Check for synchronized methods.
1485   // This mst happen AFTER invocation_counter check and stack overflow check,
1486   // so method is not locked if overflows.
1487   if (synchronized) {
1488     lock_method();
1489   } else {
1490     // No synchronization necessary.
1491 #ifdef ASSERT
1492     { Label L;
1493       __ get_method(Z_R1_scratch);
1494       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1495       __ z_bfalse(L);
1496       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1497       __ bind(L);
1498     }
1499 #endif // ASSERT
1500   }
1501 
1502   // start execution
1503 
1504   // jvmti support
1505   __ notify_method_entry();
1506 
1507   //=============================================================================
1508   // Get and call the signature handler.
1509   const Register Rmethod                 = Z_tmp_2;
1510   const Register signature_handler_entry = Z_tmp_1;
1511   const Register Rresult_handler         = Z_tmp_3;
1512   Label call_signature_handler;
1513 
1514   assert_different_registers(Z_fp, Rmethod, signature_handler_entry, Rresult_handler);
1515   assert(Rresult_handler->is_nonvolatile(), "Rresult_handler must be in a non-volatile register");
1516 
1517   // Reload method.
1518   __ get_method(Rmethod);
1519 
1520   // Check for signature handler.
1521   __ load_and_test_long(signature_handler_entry, method2_(Rmethod, signature_handler));
1522   __ z_brne(call_signature_handler);
1523 
1524   // Method has never been called. Either generate a specialized
1525   // handler or point to the slow one.
1526   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call),
1527              Rmethod);
1528 
1529   // Reload method.
1530   __ get_method(Rmethod);
1531 
1532   // Reload signature handler, it must have been created/assigned in the meantime.
1533   __ z_lg(signature_handler_entry, method2_(Rmethod, signature_handler));
1534 
1535   __ bind(call_signature_handler);
1536 
1537   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1538   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1/*tmp*/);
1539 
1540   // Call signature handler and pass locals address in Z_ARG1.
1541   __ z_lgr(Z_ARG1, Z_locals);
1542   __ call_stub(signature_handler_entry);
1543   // Save result handler returned by signature handler.
1544   __ z_lgr(Rresult_handler, Z_RET);
1545 
1546   // Reload method (the slow signature handler may block for GC).
1547   __ get_method(Rmethod);
1548 
1549   // Pass mirror handle if static call.
1550   {
1551     Label method_is_not_static;
1552     __ testbit(method2_(Rmethod, access_flags), JVM_ACC_STATIC_BIT);
1553     __ z_bfalse(method_is_not_static);
1554     // Get mirror.
1555     __ load_mirror(Z_R1, Rmethod);
1556     // z_ijava_state.oop_temp = pool_holder->klass_part()->java_mirror();
1557     __ z_stg(Z_R1, oop_tmp_offset, Z_fp);
1558     // Pass handle to mirror as 2nd argument to JNI method.
1559     __ add2reg(Z_ARG2, oop_tmp_offset, Z_fp);
1560     __ bind(method_is_not_static);
1561   }
1562 
1563   // Pass JNIEnv address as first parameter.
1564   __ add2reg(Z_ARG1, in_bytes(JavaThread::jni_environment_offset()), Z_thread);
1565 
1566   // Note: last java frame has been set above already. The pc from there
1567   // is precise enough.
1568 
1569   // Get native function entry point before we change the thread state.
1570   __ z_lg(Z_R1/*native_method_entry*/, method2_(Rmethod, native_function));
1571 
1572   //=============================================================================
1573   // Transition from _thread_in_Java to _thread_in_native. As soon as
1574   // we make this change the safepoint code needs to be certain that
1575   // the last Java frame we established is good. The pc in that frame
1576   // just need to be near here not an actual return address.
1577 #ifdef ASSERT
1578   {
1579     NearLabel L;
1580     __ mem2reg_opt(Z_R14, Address(Z_thread, JavaThread::thread_state_offset()), false /*32 bits*/);
1581     __ compareU32_and_branch(Z_R14, _thread_in_Java, Assembler::bcondEqual, L);
1582     reentry = __ stop_chain_static(reentry, "Wrong thread state in native stub");
1583     __ bind(L);
1584   }
1585 #endif
1586 
1587   // Memory ordering: Z does not reorder store/load with subsequent load. That's strong enough.
1588   __ set_thread_state(_thread_in_native);
1589 
1590   //=============================================================================
1591   // Call the native method. Argument registers must not have been
1592   // overwritten since "__ call_stub(signature_handler);" (except for
1593   // ARG1 and ARG2 for static methods).
1594 
1595   __ call_c(Z_R1/*native_method_entry*/);
1596 
1597   // NOTE: frame::interpreter_frame_result() depends on these stores.
1598   __ z_stg(Z_RET, _z_ijava_state_neg(lresult), Z_fp);
1599   __ freg2mem_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1600   const Register Rlresult = signature_handler_entry;
1601   assert(Rlresult->is_nonvolatile(), "Rlresult must be in a non-volatile register");
1602   __ z_lgr(Rlresult, Z_RET);
1603 
1604   // Z_method may no longer be valid, because of GC.
1605 
1606   // Block, if necessary, before resuming in _thread_in_Java state.
1607   // In order for GC to work, don't clear the last_Java_sp until after
1608   // blocking.
1609 
1610   //=============================================================================
1611   // Switch thread to "native transition" state before reading the
1612   // synchronization state. This additional state is necessary
1613   // because reading and testing the synchronization state is not
1614   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1615   // in _thread_in_native state, loads _not_synchronized and is
1616   // preempted. VM thread changes sync state to synchronizing and
1617   // suspends threads for GC. Thread A is resumed to finish this
1618   // native method, but doesn't block here since it didn't see any
1619   // synchronization is progress, and escapes.
1620 
1621   __ set_thread_state(_thread_in_native_trans);
1622   if (UseMembar) {
1623     __ z_fence();
1624   } else {
1625     // Write serialization page so VM thread can do a pseudo remote
1626     // membar. We use the current thread pointer to calculate a thread
1627     // specific offset to write to within the page. This minimizes bus
1628     // traffic due to cache line collision.
1629     __ serialize_memory(Z_thread, Z_R1, Z_R0);
1630   }
1631   // Now before we return to java we must look for a current safepoint
1632   // (a new safepoint can not start since we entered native_trans).
1633   // We must check here because a current safepoint could be modifying
1634   // the callers registers right this moment.
1635 
1636   // Check for safepoint operation in progress and/or pending suspend requests.
1637   {
1638     Label Continue, do_safepoint;
1639     __ generate_safepoint_check(do_safepoint, Z_R1, true);
1640     // Check for suspend.
1641     __ load_and_test_int(Z_R0/*suspend_flags*/, thread_(suspend_flags));
1642     __ z_bre(Continue); // 0 -> no flag set -> not suspended
1643     __ bind(do_safepoint);
1644     __ z_lgr(Z_ARG1, Z_thread);
1645     __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1646     __ bind(Continue);
1647   }
1648 
1649   //=============================================================================
1650   // Back in Interpreter Frame.
1651 
1652   // We are in thread_in_native_trans here and back in the normal
1653   // interpreter frame. We don't have to do anything special about
1654   // safepoints and we can switch to Java mode anytime we are ready.
1655 
1656   // Note: frame::interpreter_frame_result has a dependency on how the
1657   // method result is saved across the call to post_method_exit. For
1658   // native methods it assumes that the non-FPU/non-void result is
1659   // saved in z_ijava_state.lresult and a FPU result in z_ijava_state.fresult. If
1660   // this changes then the interpreter_frame_result implementation
1661   // will need to be updated too.
1662 
1663   //=============================================================================
1664   // Back in Java.
1665 
1666   // Memory ordering: Z does not reorder store/load with subsequent
1667   // load. That's strong enough.
1668   __ set_thread_state(_thread_in_Java);
1669 
1670   __ reset_last_Java_frame();
1671 
1672   // We reset the JNI handle block only after unboxing the result; see below.
1673 
1674   // The method register is junk from after the thread_in_native transition
1675   // until here. Also can't call_VM until the bcp has been
1676   // restored. Need bcp for throwing exception below so get it now.
1677   __ get_method(Rmethod);
1678 
1679   // Restore Z_bcp to have legal interpreter frame,
1680   // i.e., bci == 0 <=> Z_bcp == code_base().
1681   __ z_lg(Z_bcp, Address(Rmethod, Method::const_offset())); // get constMethod
1682   __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1683 
1684   if (CheckJNICalls) {
1685     // clear_pending_jni_exception_check
1686     __ clear_mem(Address(Z_thread, JavaThread::pending_jni_exception_check_fn_offset()), sizeof(oop));
1687   }
1688 
1689   // Check if the native method returns an oop, and if so, move it
1690   // from the jni handle to z_ijava_state.oop_temp. This is
1691   // necessary, because we reset the jni handle block below.
1692   // NOTE: frame::interpreter_frame_result() depends on this, too.
1693   { NearLabel no_oop_result, store_oop_result;
1694   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1695   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_result);
1696   __ compareU64_and_branch(Rlresult, (intptr_t)0L, Assembler::bcondEqual, store_oop_result);
1697   __ z_lg(Rlresult, 0, Rlresult);  // unbox
1698   __ bind(store_oop_result);
1699   __ z_stg(Rlresult, oop_tmp_offset, Z_fp);
1700   __ verify_oop(Rlresult);
1701   __ bind(no_oop_result);
1702   }
1703 
1704   // Reset handle block.
1705   __ z_lg(Z_R1/*active_handles*/, thread_(active_handles));
1706   __ clear_mem(Address(Z_R1, JNIHandleBlock::top_offset_in_bytes()), 4);
1707 
1708   // Bandle exceptions (exception handling will handle unlocking!).
1709   {
1710     Label L;
1711     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
1712     __ z_bre(L);
1713     __ MacroAssembler::call_VM(noreg,
1714                                CAST_FROM_FN_PTR(address,
1715                                InterpreterRuntime::throw_pending_exception));
1716     __ should_not_reach_here();
1717     __ bind(L);
1718   }
1719 
1720   if (synchronized) {
1721     Register Rfirst_monitor = Z_ARG2;
1722     __ add2reg(Rfirst_monitor, -(frame::z_ijava_state_size + (int)sizeof(BasicObjectLock)), Z_fp);
1723 #ifdef ASSERT
1724     NearLabel ok;
1725     __ z_lg(Z_R1, _z_ijava_state_neg(monitors), Z_fp);
1726     __ compareU64_and_branch(Rfirst_monitor, Z_R1, Assembler::bcondEqual, ok);
1727     reentry = __ stop_chain_static(reentry, "native_entry:unlock: inconsistent z_ijava_state.monitors");
1728     __ bind(ok);
1729 #endif
1730     __ unlock_object(Rfirst_monitor);
1731   }
1732 
1733   // JVMTI support. Result has already been saved above to the frame.
1734   __ notify_method_exit(true/*native_method*/, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1735 
1736   // Move native method result back into proper registers and return.
1737   // C++ interpreter does not use result handler. So do we need to here? TODO(ZASM): check if correct.
1738   { NearLabel no_oop_or_null;
1739   __ mem2freg_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1740   __ load_and_test_long(Z_RET, Address(Z_fp, _z_ijava_state_neg(lresult)));
1741   __ z_bre(no_oop_or_null); // No unboxing if the result is NULL.
1742   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1743   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_or_null);
1744   __ z_lg(Z_RET, oop_tmp_offset, Z_fp);
1745   __ verify_oop(Z_RET);
1746   __ bind(no_oop_or_null);
1747   }
1748 
1749   // Pop the native method's interpreter frame.
1750   __ pop_interpreter_frame(Z_R14 /*return_pc*/, Z_ARG2/*tmp1*/, Z_ARG3/*tmp2*/);
1751 
1752   // Return to caller.
1753   __ z_br(Z_R14);
1754 
1755   if (inc_counter) {
1756     // Handle overflow of counter and compile method.
1757     __ bind(invocation_counter_overflow);
1758     generate_counter_overflow(continue_after_compile);
1759   }
1760 
1761   BLOCK_COMMENT("} native_entry");
1762 
1763   return entry_point;
1764 }
1765 
1766 //
1767 // Generic interpreted method entry to template interpreter.
1768 //
1769 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1770   address entry_point = __ pc();
1771 
1772   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1773 
1774   // Interpreter entry for ordinary Java methods.
1775   //
1776   // Registers alive
1777   //   Z_SP       - stack pointer
1778   //   Z_thread   - JavaThread*
1779   //   Z_method   - callee's method (method to be invoked)
1780   //   Z_esp      - operand (or expression) stack pointer of caller. one slot above last arg.
1781   //   Z_R10      - sender sp (before modifications, e.g. by c2i adapter
1782   //                           and as well by generate_fixed_frame below)
1783   //   Z_R14      - return address to caller (call_stub or c2i_adapter)
1784   //
1785   // Registers updated
1786   //   Z_SP       - stack pointer
1787   //   Z_fp       - callee's framepointer
1788   //   Z_esp      - callee's operand stack pointer
1789   //                points to the slot above the value on top
1790   //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1791   //   Z_tos      - integer result, if any
1792   //   z_ftos     - floating point result, if any
1793   //
1794   //
1795   // stack layout at this point:
1796   //
1797   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1798   //                                                          frame was extended by c2i adapter)
1799   //           [outgoing Java arguments]     <-- Z_esp
1800   //           ...
1801   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1802   //           ...
1803   //
1804   // stack layout before dispatching the first bytecode:
1805   //
1806   //   F0      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP
1807   //           [operand stack]               <-- Z_esp
1808   //           monitor (optional, can grow)
1809   //           [IJAVA_STATE]
1810   //   F1      [PARENT_IJAVA_FRAME_ABI]      <-- Z_fp (== *Z_SP)
1811   //           [F0's locals]                 <-- Z_locals
1812   //           [F1's operand stack]
1813   //           [F1's monitors] (optional)
1814   //           [IJAVA_STATE]
1815 
1816   // Make sure registers are different!
1817   assert_different_registers(Z_thread, Z_method, Z_esp);
1818 
1819   BLOCK_COMMENT("normal_entry {");
1820 
1821   // Make sure method is not native and not abstract.
1822   // Rethink these assertions - they can be simplified and shared.
1823 #ifdef ASSERT
1824   address reentry = NULL;
1825   { Label L;
1826     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1827     __ z_bfalse(L);
1828     reentry = __ stop_chain_static(reentry, "tried to execute native method as non-native");
1829     __ bind(L);
1830   }
1831   { Label L;
1832     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1833     __ z_bfalse(L);
1834     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1835     __ bind(L);
1836   }
1837 #endif // ASSERT
1838 
1839 #ifdef ASSERT
1840   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1841   __ save_return_pc(Z_R14);
1842 #endif
1843 
1844   // Generate the code to allocate the interpreter stack frame.
1845   generate_fixed_frame(false);
1846 
1847   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1848   // Since at this point in the method invocation the exception handler
1849   // would try to exit the monitor of synchronized methods which hasn't
1850   // been entered yet, we set the thread local variable
1851   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1852   // runtime, exception handling i.e. unlock_if_synchronized_method will
1853   // check this thread local flag.
1854   __ z_mvi(do_not_unlock_if_synchronized, true);
1855 
1856   __ profile_parameters_type(Z_tmp_2, Z_ARG3, Z_ARG4);
1857 
1858   // Increment invocation counter and check for overflow.
1859   //
1860   // Note: checking for negative value instead of overflow so we have a 'sticky'
1861   // overflow test (may be of importance as soon as we have true MT/MP).
1862   NearLabel invocation_counter_overflow;
1863   NearLabel profile_method;
1864   NearLabel profile_method_continue;
1865   NearLabel Lcontinue;
1866   if (inc_counter) {
1867     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1868     if (ProfileInterpreter) {
1869       __ bind(profile_method_continue);
1870     }
1871   }
1872   __ bind(Lcontinue);
1873 
1874   bang_stack_shadow_pages(false);
1875 
1876   // Reset the _do_not_unlock_if_synchronized flag.
1877   __ z_mvi(do_not_unlock_if_synchronized, false);
1878 
1879   // Check for synchronized methods.
1880   // Must happen AFTER invocation_counter check and stack overflow check,
1881   // so method is not locked if overflows.
1882   if (synchronized) {
1883     // Allocate monitor and lock method.
1884     lock_method();
1885   } else {
1886 #ifdef ASSERT
1887     { Label L;
1888       __ get_method(Z_R1_scratch);
1889       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1890       __ z_bfalse(L);
1891       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1892       __ bind(L);
1893     }
1894 #endif // ASSERT
1895   }
1896 
1897   // start execution
1898 
1899 #ifdef ASSERT
1900   __ verify_esp(Z_esp, Z_R1_scratch);
1901 
1902   __ verify_thread();
1903 #endif
1904 
1905   // jvmti support
1906   __ notify_method_entry();
1907 
1908   // Start executing instructions.
1909   __ dispatch_next(vtos);
1910   // Dispatch_next does not return.
1911   DEBUG_ONLY(__ should_not_reach_here());
1912 
1913   // Invocation counter overflow.
1914   if (inc_counter) {
1915     if (ProfileInterpreter) {
1916       // We have decided to profile this method in the interpreter.
1917       __ bind(profile_method);
1918 
1919       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1920       __ set_method_data_pointer_for_bcp();
1921       __ z_bru(profile_method_continue);
1922     }
1923 
1924     // Handle invocation counter overflow.
1925     __ bind(invocation_counter_overflow);
1926     generate_counter_overflow(Lcontinue);
1927   }
1928 
1929   BLOCK_COMMENT("} normal_entry");
1930 
1931   return entry_point;
1932 }
1933 
1934 // Method entry for static native methods:
1935 //   int java.util.zip.CRC32.update(int crc, int b)
1936 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1937 
1938   if (UseCRC32Intrinsics) {
1939     uint64_t entry_off = __ offset();
1940     Label    slow_path;
1941 
1942     // If we need a safepoint check, generate full interpreter entry.
1943     __ generate_safepoint_check(slow_path, Z_R1, false);
1944 
1945     BLOCK_COMMENT("CRC32_update {");
1946 
1947     // We don't generate local frame and don't align stack because
1948     // we not even call stub code (we generate the code inline)
1949     // and there is no safepoint on this path.
1950 
1951     // Load java parameters.
1952     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1953     const Register argP    = Z_esp;
1954     const Register crc     = Z_ARG1;  // crc value
1955     const Register data    = Z_ARG2;  // address of java byte value (kernel_crc32 needs address)
1956     const Register dataLen = Z_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1957     const Register table   = Z_ARG4;  // address of crc32 table
1958 
1959     // Arguments are reversed on java expression stack.
1960     __ z_la(data, 3+1*wordSize, argP);  // byte value (stack address).
1961                                         // Being passed as an int, the single byte is at offset +3.
1962     __ z_llgf(crc, 2 * wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1963 
1964     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1965     __ kernel_crc32_singleByte(crc, data, dataLen, table, Z_R1);
1966 
1967     // Restore caller sp for c2i case.
1968     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1969 
1970     __ z_br(Z_R14);
1971 
1972     BLOCK_COMMENT("} CRC32_update");
1973 
1974     // Use a previously generated vanilla native entry as the slow path.
1975     BIND(slow_path);
1976     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1977     return __ addr_at(entry_off);
1978   }
1979 
1980   return NULL;
1981 }
1982 
1983 
1984 // Method entry for static native methods:
1985 //   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1986 //   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
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     __ generate_safepoint_check(slow_path, Z_R1, false);
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);
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 // Not supported
2062 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
2063   return NULL;
2064 }
2065 
2066 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
2067   // Quick & dirty stack overflow checking: bang the stack & handle trap.
2068   // Note that we do the banging after the frame is setup, since the exception
2069   // handling code expects to find a valid interpreter frame on the stack.
2070   // Doing the banging earlier fails if the caller frame is not an interpreter
2071   // frame.
2072   // (Also, the exception throwing code expects to unlock any synchronized
2073   // method receiver, so do the banging after locking the receiver.)
2074 
2075   // Bang each page in the shadow zone. We can't assume it's been done for
2076   // an interpreter frame with greater than a page of locals, so each page
2077   // needs to be checked. Only true for non-native. For native, we only bang the last page.
2078   if (UseStackBanging) {
2079     const int page_size      = os::vm_page_size();
2080     const int n_shadow_pages = (int)(JavaThread::stack_shadow_zone_size()/page_size);
2081     const int start_page_num = native_call ? n_shadow_pages : 1;
2082     for (int pages = start_page_num; pages <= n_shadow_pages; pages++) {
2083       __ bang_stack_with_offset(pages*page_size);
2084     }
2085   }
2086 }
2087 
2088 //-----------------------------------------------------------------------------
2089 // Exceptions
2090 
2091 void TemplateInterpreterGenerator::generate_throw_exception() {
2092 
2093   BLOCK_COMMENT("throw_exception {");
2094 
2095   // Entry point in previous activation (i.e., if the caller was interpreted).
2096   Interpreter::_rethrow_exception_entry = __ pc();
2097   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Frame accessors use Z_fp.
2098   // Z_ARG1 (==Z_tos): exception
2099   // Z_ARG2          : Return address/pc that threw exception.
2100   __ restore_bcp();    // R13 points to call/send.
2101   __ restore_locals();
2102 
2103   // Fallthrough, no need to restore Z_esp.
2104 
2105   // Entry point for exceptions thrown within interpreter code.
2106   Interpreter::_throw_exception_entry = __ pc();
2107   // Expression stack is undefined here.
2108   // Z_ARG1 (==Z_tos): exception
2109   // Z_bcp: exception bcp
2110   __ verify_oop(Z_ARG1);
2111   __ z_lgr(Z_ARG2, Z_ARG1);
2112 
2113   // Expression stack must be empty before entering the VM in case of
2114   // an exception.
2115   __ empty_expression_stack();
2116   // Find exception handler address and preserve exception oop.
2117   const Register Rpreserved_exc_oop = Z_tmp_1;
2118   __ call_VM(Rpreserved_exc_oop,
2119              CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception),
2120              Z_ARG2);
2121   // Z_RET: exception handler entry point
2122   // Z_bcp: bcp for exception handler
2123   __ push_ptr(Rpreserved_exc_oop); // Push exception which is now the only value on the stack.
2124   __ z_br(Z_RET); // Jump to exception handler (may be _remove_activation_entry!).
2125 
2126   // If the exception is not handled in the current frame the frame is
2127   // removed and the exception is rethrown (i.e. exception
2128   // continuation is _rethrow_exception).
2129   //
2130   // Note: At this point the bci is still the bci for the instruction
2131   // which caused the exception and the expression stack is
2132   // empty. Thus, for any VM calls at this point, GC will find a legal
2133   // oop map (with empty expression stack).
2134 
2135   //
2136   // JVMTI PopFrame support
2137   //
2138 
2139   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2140   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2141   __ empty_expression_stack();
2142   // Set the popframe_processing bit in pending_popframe_condition
2143   // indicating that we are currently handling popframe, so that
2144   // call_VMs that may happen later do not trigger new popframe
2145   // handling cycles.
2146   __ load_sized_value(Z_tmp_1, Address(Z_thread, JavaThread::popframe_condition_offset()), 4, false /*signed*/);
2147   __ z_oill(Z_tmp_1, JavaThread::popframe_processing_bit);
2148   __ z_sty(Z_tmp_1, thread_(popframe_condition));
2149 
2150   {
2151     // Check to see whether we are returning to a deoptimized frame.
2152     // (The PopFrame call ensures that the caller of the popped frame is
2153     // either interpreted or compiled and deoptimizes it if compiled.)
2154     // In this case, we can't call dispatch_next() after the frame is
2155     // popped, but instead must save the incoming arguments and restore
2156     // them after deoptimization has occurred.
2157     //
2158     // Note that we don't compare the return PC against the
2159     // deoptimization blob's unpack entry because of the presence of
2160     // adapter frames in C2.
2161     NearLabel caller_not_deoptimized;
2162     __ z_lg(Z_ARG1, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2163     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), Z_ARG1);
2164     __ compareU64_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, caller_not_deoptimized);
2165 
2166     // Compute size of arguments for saving when returning to
2167     // deoptimized caller.
2168     __ get_method(Z_ARG2);
2169     __ z_lg(Z_ARG2, Address(Z_ARG2, Method::const_offset()));
2170     __ z_llgh(Z_ARG2, Address(Z_ARG2, ConstMethod::size_of_parameters_offset()));
2171     __ z_sllg(Z_ARG2, Z_ARG2, Interpreter::logStackElementSize); // slots 2 bytes
2172     __ restore_locals();
2173     // Compute address of args to be saved.
2174     __ z_lgr(Z_ARG3, Z_locals);
2175     __ z_slgr(Z_ARG3, Z_ARG2);
2176     __ add2reg(Z_ARG3, wordSize);
2177     // Save these arguments.
2178     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args),
2179                     Z_thread, Z_ARG2, Z_ARG3);
2180 
2181     __ remove_activation(vtos, Z_R14,
2182                          /* throw_monitor_exception */ false,
2183                          /* install_monitor_exception */ false,
2184                          /* notify_jvmdi */ false);
2185 
2186     // Inform deoptimization that it is responsible for restoring
2187     // these arguments.
2188     __ store_const(thread_(popframe_condition),
2189                    JavaThread::popframe_force_deopt_reexecution_bit,
2190                    Z_tmp_1, false);
2191 
2192     // Continue in deoptimization handler.
2193     __ z_br(Z_R14);
2194 
2195     __ bind(caller_not_deoptimized);
2196   }
2197 
2198   // Clear the popframe condition flag.
2199   __ clear_mem(thread_(popframe_condition), sizeof(int));
2200 
2201   __ remove_activation(vtos,
2202                        noreg,  // Retaddr is not used.
2203                        false,  // throw_monitor_exception
2204                        false,  // install_monitor_exception
2205                        false); // notify_jvmdi
2206   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2207   __ restore_bcp();
2208   __ restore_locals();
2209   __ restore_esp();
2210   // The method data pointer was incremented already during
2211   // call profiling. We have to restore the mdp for the current bcp.
2212   if (ProfileInterpreter) {
2213     __ set_method_data_pointer_for_bcp();
2214   }
2215 #if INCLUDE_JVMTI
2216   {
2217     Label L_done;
2218 
2219     __ z_cli(0, Z_bcp, Bytecodes::_invokestatic);
2220     __ z_brc(Assembler::bcondNotEqual, L_done);
2221 
2222     // The member name argument must be restored if _invokestatic is
2223     // re-executed after a PopFrame call.  Detect such a case in the
2224     // InterpreterRuntime function and return the member name
2225     // argument, or NULL.
2226     __ z_lg(Z_ARG2, Address(Z_locals));
2227     __ get_method(Z_ARG3);
2228     __ call_VM(Z_tmp_1,
2229                CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
2230                Z_ARG2, Z_ARG3, Z_bcp);
2231 
2232     __ z_ltgr(Z_tmp_1, Z_tmp_1);
2233     __ z_brc(Assembler::bcondEqual, L_done);
2234 
2235     __ z_stg(Z_tmp_1, Address(Z_esp, wordSize));
2236     __ bind(L_done);
2237   }
2238 #endif // INCLUDE_JVMTI
2239   __ dispatch_next(vtos);
2240   // End of PopFrame support.
2241   Interpreter::_remove_activation_entry = __ pc();
2242 
2243   // In between activations - previous activation type unknown yet
2244   // compute continuation point - the continuation point expects the
2245   // following registers set up:
2246   //
2247   // Z_ARG1 (==Z_tos): exception
2248   // Z_ARG2          : return address/pc that threw exception
2249 
2250   Register return_pc = Z_tmp_1;
2251   Register handler   = Z_tmp_2;
2252    assert(return_pc->is_nonvolatile(), "use non-volatile reg. to preserve exception pc");
2253    assert(handler->is_nonvolatile(),   "use non-volatile reg. to handler pc");
2254   __ asm_assert_ijava_state_magic(return_pc/*tmp*/); // The top frame should be an interpreter frame.
2255   __ z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2256 
2257   // Moved removing the activation after VM call, because the new top
2258   // frame does not necessarily have the z_abi_160 required for a VM
2259   // call (e.g. if it is compiled).
2260 
2261   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2262                                          SharedRuntime::exception_handler_for_return_address),
2263                         Z_thread, return_pc);
2264   __ z_lgr(handler, Z_RET); // Save exception handler.
2265 
2266   // Preserve exception over this code sequence.
2267   __ pop_ptr(Z_ARG1);
2268   __ set_vm_result(Z_ARG1);
2269   // Remove the activation (without doing throws on illegalMonitorExceptions).
2270   __ remove_activation(vtos, noreg/*ret.pc already loaded*/, false/*throw exc*/, true/*install exc*/, false/*notify jvmti*/);
2271   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2272 
2273   __ get_vm_result(Z_ARG1);     // Restore exception.
2274   __ verify_oop(Z_ARG1);
2275   __ z_lgr(Z_ARG2, return_pc);  // Restore return address.
2276 
2277 #ifdef ASSERT
2278   // The return_pc in the new top frame is dead... at least that's my
2279   // current understanding. To assert this I overwrite it.
2280   // Note: for compiled frames the handler is the deopt blob
2281   // which writes Z_ARG2 into the return_pc slot.
2282   __ load_const_optimized(return_pc, 0xb00b1);
2283   __ z_stg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2284 #endif
2285 
2286   // Z_ARG1 (==Z_tos): exception
2287   // Z_ARG2          : return address/pc that threw exception
2288 
2289   // Note that an "issuing PC" is actually the next PC after the call.
2290   __ z_br(handler);         // Jump to exception handler of caller.
2291 
2292   BLOCK_COMMENT("} throw_exception");
2293 }
2294 
2295 //
2296 // JVMTI ForceEarlyReturn support
2297 //
2298 address TemplateInterpreterGenerator::generate_earlyret_entry_for (TosState state) {
2299   address entry = __ pc();
2300 
2301   BLOCK_COMMENT("earlyret_entry {");
2302 
2303   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2304   __ restore_bcp();
2305   __ restore_locals();
2306   __ restore_esp();
2307   __ empty_expression_stack();
2308   __ load_earlyret_value(state);
2309 
2310   Register RjvmtiState = Z_tmp_1;
2311   __ z_lg(RjvmtiState, thread_(jvmti_thread_state));
2312   __ store_const(Address(RjvmtiState, JvmtiThreadState::earlyret_state_offset()),
2313                  JvmtiThreadState::earlyret_inactive, 4, 4, Z_R0_scratch);
2314 
2315   __ remove_activation(state,
2316                        Z_tmp_1, // retaddr
2317                        false,   // throw_monitor_exception
2318                        false,   // install_monitor_exception
2319                        true);   // notify_jvmdi
2320   __ z_br(Z_tmp_1);
2321 
2322   BLOCK_COMMENT("} earlyret_entry");
2323 
2324   return entry;
2325 }
2326 
2327 //-----------------------------------------------------------------------------
2328 // Helper for vtos entry point generation.
2329 
2330 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2331                                                          address& bep,
2332                                                          address& cep,
2333                                                          address& sep,
2334                                                          address& aep,
2335                                                          address& iep,
2336                                                          address& lep,
2337                                                          address& fep,
2338                                                          address& dep,
2339                                                          address& vep) {
2340   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2341   Label L;
2342   aep = __ pc(); __ push_ptr(); __ z_bru(L);
2343   fep = __ pc(); __ push_f();   __ z_bru(L);
2344   dep = __ pc(); __ push_d();   __ z_bru(L);
2345   lep = __ pc(); __ push_l();   __ z_bru(L);
2346   bep = cep = sep =
2347   iep = __ pc(); __ push_i();
2348   vep = __ pc();
2349   __ bind(L);
2350   generate_and_dispatch(t);
2351 }
2352 
2353 //-----------------------------------------------------------------------------
2354 
2355 #ifndef PRODUCT
2356 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2357   address entry = __ pc();
2358   NearLabel counter_below_trace_threshold;
2359 
2360   if (TraceBytecodesAt > 0) {
2361     // Skip runtime call, if the trace threshold is not yet reached.
2362     __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2363     __ load_absolute_address(Z_tmp_2, (address)&TraceBytecodesAt);
2364     __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2365     __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2366     __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, counter_below_trace_threshold);
2367   }
2368 
2369   int offset2 = state == ltos || state == dtos ? 2 : 1;
2370 
2371   __ push(state);
2372   // Preserved return pointer is in Z_R14.
2373   // InterpreterRuntime::trace_bytecode() preserved and returns the value passed as second argument.
2374   __ z_lgr(Z_ARG2, Z_R14);
2375   __ z_lg(Z_ARG3, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)));
2376   if (WizardMode) {
2377     __ z_lgr(Z_ARG4, Z_esp); // Trace Z_esp in WizardMode.
2378   } else {
2379     __ z_lg(Z_ARG4, Address(Z_esp, Interpreter::expr_offset_in_bytes(offset2)));
2380   }
2381   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), Z_ARG2, Z_ARG3, Z_ARG4);
2382   __ z_lgr(Z_R14, Z_RET); // Estore return address (see above).
2383   __ pop(state);
2384 
2385   __ bind(counter_below_trace_threshold);
2386   __ z_br(Z_R14); // return
2387 
2388   return entry;
2389 }
2390 
2391 // Make feasible for old CPUs.
2392 void TemplateInterpreterGenerator::count_bytecode() {
2393   __ load_absolute_address(Z_R1_scratch, (address) &BytecodeCounter::_counter_value);
2394   __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2395 }
2396 
2397 void TemplateInterpreterGenerator::histogram_bytecode(Template * t) {
2398   __ load_absolute_address(Z_R1_scratch, (address)&BytecodeHistogram::_counters[ t->bytecode() ]);
2399   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2400 }
2401 
2402 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template * t) {
2403   Address  index_addr(Z_tmp_1, (intptr_t) 0);
2404   Register index = Z_tmp_2;
2405 
2406   // Load previous index.
2407   __ load_absolute_address(Z_tmp_1, (address) &BytecodePairHistogram::_index);
2408   __ mem2reg_opt(index, index_addr, false);
2409 
2410   // Mask with current bytecode and store as new previous index.
2411   __ z_srl(index, BytecodePairHistogram::log2_number_of_codes);
2412   __ load_const_optimized(Z_R0_scratch,
2413                           (int)t->bytecode() << BytecodePairHistogram::log2_number_of_codes);
2414   __ z_or(index, Z_R0_scratch);
2415   __ reg2mem_opt(index, index_addr, false);
2416 
2417   // Load counter array's address.
2418   __ z_lgfr(index, index);   // Sign extend for addressing.
2419   __ z_sllg(index, index, LogBytesPerInt);  // index2bytes
2420   __ load_absolute_address(Z_R1_scratch,
2421                            (address) &BytecodePairHistogram::_counters);
2422   // Add index and increment counter.
2423   __ z_agr(Z_R1_scratch, index);
2424   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2425 }
2426 
2427 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2428   // Call a little run-time stub to avoid blow-up for each bytecode.
2429   // The run-time runtime saves the right registers, depending on
2430   // the tosca in-state for the given template.
2431   address entry = Interpreter::trace_code(t->tos_in());
2432   guarantee(entry != NULL, "entry must have been generated");
2433   __ call_stub(entry);
2434 }
2435 
2436 void TemplateInterpreterGenerator::stop_interpreter_at() {
2437   NearLabel L;
2438 
2439   __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2440   __ load_absolute_address(Z_tmp_2, (address)&StopInterpreterAt);
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, L);
2444   assert(Z_tmp_1->is_nonvolatile(), "must be nonvolatile to preserve Z_tos");
2445   assert(Z_F8->is_nonvolatile(), "must be nonvolatile to preserve Z_ftos");
2446   __ z_lgr(Z_tmp_1, Z_tos);      // Save tos.
2447   __ z_lgr(Z_tmp_2, Z_bytecode); // Save Z_bytecode.
2448   __ z_ldr(Z_F8, Z_ftos);        // Save ftos.
2449   // Use -XX:StopInterpreterAt=<num> to set the limit
2450   // and break at breakpoint().
2451   __ call_VM(noreg, CAST_FROM_FN_PTR(address, breakpoint), false);
2452   __ z_lgr(Z_tos, Z_tmp_1);      // Restore tos.
2453   __ z_lgr(Z_bytecode, Z_tmp_2); // Save Z_bytecode.
2454   __ z_ldr(Z_ftos, Z_F8);        // Restore ftos.
2455   __ bind(L);
2456 }
2457 
2458 #endif // !PRODUCT