1 /*
   2  * Copyright (c) 2016, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2016, 2017, SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "interpreter/abstractInterpreter.hpp"
  29 #include "interpreter/bytecodeHistogram.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateInterpreterGenerator.hpp"
  34 #include "interpreter/templateTable.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 
  49 
  50 // Size of interpreter code.  Increase if too small.  Interpreter will
  51 // fail with a guarantee ("not enough space for interpreter generation");
  52 // if too small.
  53 // Run with +PrintInterpreter to get the VM to print out the size.
  54 // Max size with JVMTI
  55 int TemplateInterpreter::InterpreterCodeSize = 320*K;
  56 
  57 #undef  __
  58 #ifdef PRODUCT
  59   #define __ _masm->
  60 #else
  61   #define __ _masm->
  62 //  #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm):_masm)->
  63 #endif
  64 
  65 #define BLOCK_COMMENT(str) __ block_comment(str)
  66 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  67 
  68 #define oop_tmp_offset     _z_ijava_state_neg(oop_tmp)
  69 
  70 //-----------------------------------------------------------------------------
  71 
  72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  73   //
  74   // New slow_signature handler that respects the z/Architecture
  75   // C calling conventions.
  76   //
  77   // We get called by the native entry code with our output register
  78   // area == 8. First we call InterpreterRuntime::get_result_handler
  79   // to copy the pointer to the signature string temporarily to the
  80   // first C-argument and to return the result_handler in
  81   // Z_RET. Since native_entry will copy the jni-pointer to the
  82   // first C-argument slot later on, it's OK to occupy this slot
  83   // temporarily. Then we copy the argument list on the java
  84   // expression stack into native varargs format on the native stack
  85   // and load arguments into argument registers. Integer arguments in
  86   // the varargs vector will be sign-extended to 8 bytes.
  87   //
  88   // On entry:
  89   //   Z_ARG1  - intptr_t*       Address of java argument list in memory.
  90   //   Z_state - cppInterpreter* Address of interpreter state for
  91   //                               this method
  92   //   Z_method
  93   //
  94   // On exit (just before return instruction):
  95   //   Z_RET contains the address of the result_handler.
  96   //   Z_ARG2 is not updated for static methods and contains "this" otherwise.
  97   //   Z_ARG3-Z_ARG5 contain the first 3 arguments of types other than float and double.
  98   //   Z_FARG1-Z_FARG4 contain the first 4 arguments of type float or double.
  99 
 100   const int LogSizeOfCase = 3;
 101 
 102   const int max_fp_register_arguments   = Argument::n_float_register_parameters;
 103   const int max_int_register_arguments  = Argument::n_register_parameters - 2;  // First 2 are reserved.
 104 
 105   const Register arg_java       = Z_tmp_2;
 106   const Register arg_c          = Z_tmp_3;
 107   const Register signature      = Z_R1_scratch; // Is a string.
 108   const Register fpcnt          = Z_R0_scratch;
 109   const Register argcnt         = Z_tmp_4;
 110   const Register intSlot        = Z_tmp_1;
 111   const Register sig_end        = Z_tmp_1; // Assumed end of signature (only used in do_object).
 112   const Register target_sp      = Z_tmp_1;
 113   const FloatRegister floatSlot = Z_F1;
 114 
 115   const int d_signature         = _z_abi(gpr6); // Only spill space, register contents not affected.
 116   const int d_fpcnt             = _z_abi(gpr7); // Only spill space, register contents not affected.
 117 
 118   unsigned int entry_offset = __ offset();
 119 
 120   BLOCK_COMMENT("slow_signature_handler {");
 121 
 122   // We use target_sp for storing arguments in the C frame.
 123   __ save_return_pc();
 124   __ push_frame_abi160(4*BytesPerWord);                 // Reserve space to save the tmp_[1..4] registers.
 125   __ z_stmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // Save registers only after frame is pushed.
 126 
 127   __ z_lgr(arg_java, Z_ARG1);
 128 
 129   Register   method = Z_ARG2; // Directly load into correct argument register.
 130 
 131   __ get_method(method);
 132   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), Z_thread, method);
 133 
 134   // Move signature to callee saved register.
 135   // Don't directly write to stack. Frame is used by VM call.
 136   __ z_lgr(Z_tmp_1, Z_RET);
 137 
 138   // Reload method. Register may have been altered by VM call.
 139   __ get_method(method);
 140 
 141   // Get address of result handler.
 142   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), Z_thread, method);
 143 
 144   // Save signature address to stack.
 145   __ z_stg(Z_tmp_1, d_signature, Z_SP);
 146 
 147   // Don't overwrite return value (Z_RET, Z_ARG1) in rest of the method !
 148 
 149   {
 150     Label   isStatic;
 151 
 152     // Test if static.
 153     // We can test the bit directly.
 154     // Path is Z_method->_access_flags._flags.
 155     // We only support flag bits in the least significant byte (assert !).
 156     // Therefore add 3 to address that byte within "_flags".
 157     // Reload method. VM call above may have destroyed register contents
 158     __ get_method(method);
 159     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 160     method = noreg;  // end of life
 161     __ z_btrue(isStatic);
 162 
 163     // For non-static functions, pass "this" in Z_ARG2 and copy it to 2nd C-arg slot.
 164     // Need to box the Java object here, so we use arg_java
 165     // (address of current Java stack slot) as argument and
 166     // don't dereference it as in case of ints, floats, etc..
 167     __ z_lgr(Z_ARG2, arg_java);
 168     __ add2reg(arg_java, -BytesPerWord);
 169     __ bind(isStatic);
 170   }
 171 
 172   // argcnt == 0 corresponds to 3rd C argument.
 173   //   arg #1 (result handler) and
 174   //   arg #2 (this, for non-statics), unused else
 175   // are reserved and pre-filled above.
 176   // arg_java points to the corresponding Java argument here. It
 177   // has been decremented by one argument (this) in case of non-static.
 178   __ clear_reg(argcnt, true, false);  // Don't set CC.
 179   __ z_lg(target_sp, 0, Z_SP);
 180   __ add2reg(arg_c, _z_abi(remaining_cargs), target_sp);
 181   // No floating-point args parsed so far.
 182   __ clear_mem(Address(Z_SP, d_fpcnt), 8);
 183 
 184   NearLabel   move_intSlot_to_ARG, move_floatSlot_to_FARG;
 185   NearLabel   loop_start, loop_start_restore, loop_end;
 186   NearLabel   do_int, do_long, do_float, do_double;
 187   NearLabel   do_dontreachhere, do_object, do_array, do_boxed;
 188 
 189 #ifdef ASSERT
 190   // Signature needs to point to '(' (== 0x28) at entry.
 191   __ z_lg(signature, d_signature, Z_SP);
 192   __ z_cli(0, signature, (int) '(');
 193   __ z_brne(do_dontreachhere);
 194 #endif
 195 
 196   __ bind(loop_start_restore);
 197   __ z_lg(signature, d_signature, Z_SP);  // Restore signature ptr, destroyed by move_XX_to_ARG.
 198 
 199   BIND(loop_start);
 200   // Advance to next argument type token from the signature.
 201   __ add2reg(signature, 1);
 202 
 203   // Use CLI, works well on all CPU versions.
 204     __ z_cli(0, signature, (int) ')');
 205     __ z_bre(loop_end);                // end of signature
 206     __ z_cli(0, signature, (int) 'L');
 207     __ z_bre(do_object);               // object     #9
 208     __ z_cli(0, signature, (int) 'F');
 209     __ z_bre(do_float);                // float      #7
 210     __ z_cli(0, signature, (int) 'J');
 211     __ z_bre(do_long);                 // long       #6
 212     __ z_cli(0, signature, (int) 'B');
 213     __ z_bre(do_int);                  // byte       #1
 214     __ z_cli(0, signature, (int) 'Z');
 215     __ z_bre(do_int);                  // boolean    #2
 216     __ z_cli(0, signature, (int) 'C');
 217     __ z_bre(do_int);                  // char       #3
 218     __ z_cli(0, signature, (int) 'S');
 219     __ z_bre(do_int);                  // short      #4
 220     __ z_cli(0, signature, (int) 'I');
 221     __ z_bre(do_int);                  // int        #5
 222     __ z_cli(0, signature, (int) 'D');
 223     __ z_bre(do_double);               // double     #8
 224     __ z_cli(0, signature, (int) '[');
 225     __ z_bre(do_array);                // array      #10
 226 
 227   __ bind(do_dontreachhere);
 228 
 229   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 230 
 231   // Array argument
 232   BIND(do_array);
 233 
 234   {
 235     Label   start_skip, end_skip;
 236 
 237     __ bind(start_skip);
 238 
 239     // Advance to next type tag from signature.
 240     __ add2reg(signature, 1);
 241 
 242     // Use CLI, works well on all CPU versions.
 243     __ z_cli(0, signature, (int) '[');
 244     __ z_bre(start_skip);               // Skip further brackets.
 245 
 246     __ z_cli(0, signature, (int) '9');
 247     __ z_brh(end_skip);                 // no optional size
 248 
 249     __ z_cli(0, signature, (int) '0');
 250     __ z_brnl(start_skip);              // Skip optional size.
 251 
 252     __ bind(end_skip);
 253 
 254     __ z_cli(0, signature, (int) 'L');
 255     __ z_brne(do_boxed);                // If not array of objects: go directly to do_boxed.
 256   }
 257 
 258   //  OOP argument
 259   BIND(do_object);
 260   // Pass by an object's type name.
 261   {
 262     Label   L;
 263 
 264     __ add2reg(sig_end, 4095, signature);     // Assume object type name is shorter than 4k.
 265     __ load_const_optimized(Z_R0, (int) ';'); // Type name terminator (must be in Z_R0!).
 266     __ MacroAssembler::search_string(sig_end, signature);
 267     __ z_brl(L);
 268     __ z_illtrap();  // No semicolon found: internal error or object name too long.
 269     __ bind(L);
 270     __ z_lgr(signature, sig_end);
 271     // fallthru to do_boxed
 272   }
 273 
 274   // Need to box the Java object here, so we use arg_java
 275   // (address of current Java stack slot) as argument and
 276   // don't dereference it as in case of ints, floats, etc..
 277 
 278   // UNBOX argument
 279   // Load reference and check for NULL.
 280   Label  do_int_Entry4Boxed;
 281   __ bind(do_boxed);
 282   {
 283     __ load_and_test_long(intSlot, Address(arg_java));
 284     __ z_bre(do_int_Entry4Boxed);
 285     __ z_lgr(intSlot, arg_java);
 286     __ z_bru(do_int_Entry4Boxed);
 287   }
 288 
 289   // INT argument
 290 
 291   // (also for byte, boolean, char, short)
 292   // Use lgf for load (sign-extend) and stg for store.
 293   BIND(do_int);
 294   __ z_lgf(intSlot, 0, arg_java);
 295 
 296   __ bind(do_int_Entry4Boxed);
 297   __ add2reg(arg_java, -BytesPerWord);
 298   // If argument fits into argument register, go and handle it, otherwise continue.
 299   __ compare32_and_branch(argcnt, max_int_register_arguments,
 300                           Assembler::bcondLow, move_intSlot_to_ARG);
 301   __ z_stg(intSlot, 0, arg_c);
 302   __ add2reg(arg_c, BytesPerWord);
 303   __ z_bru(loop_start);
 304 
 305   // LONG argument
 306 
 307   BIND(do_long);
 308   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 309   __ z_lg(intSlot, BytesPerWord, arg_java);
 310   // If argument fits into argument register, go and handle it, otherwise continue.
 311   __ compare32_and_branch(argcnt, max_int_register_arguments,
 312                           Assembler::bcondLow, move_intSlot_to_ARG);
 313   __ z_stg(intSlot, 0, arg_c);
 314   __ add2reg(arg_c, BytesPerWord);
 315   __ z_bru(loop_start);
 316 
 317   // FLOAT argumen
 318 
 319   BIND(do_float);
 320   __ z_le(floatSlot, 0, arg_java);
 321   __ add2reg(arg_java, -BytesPerWord);
 322   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 323   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 324   __ z_brl(move_floatSlot_to_FARG);
 325   __ z_ste(floatSlot, 4, arg_c);
 326   __ add2reg(arg_c, BytesPerWord);
 327   __ z_bru(loop_start);
 328 
 329   // DOUBLE argument
 330 
 331   BIND(do_double);
 332   __ add2reg(arg_java, -2*BytesPerWord);  // Decrement first to have positive displacement for lg.
 333   __ z_ld(floatSlot, BytesPerWord, arg_java);
 334   assert(max_fp_register_arguments <= 255, "always true");  // safety net
 335   __ z_cli(d_fpcnt+7, Z_SP, max_fp_register_arguments);
 336   __ z_brl(move_floatSlot_to_FARG);
 337   __ z_std(floatSlot, 0, arg_c);
 338   __ add2reg(arg_c, BytesPerWord);
 339   __ z_bru(loop_start);
 340 
 341   // Method exit, all arguments proocessed.
 342   __ bind(loop_end);
 343   __ z_lmg(Z_R10, Z_R13, frame::z_abi_160_size, Z_SP); // restore registers before frame is popped.
 344   __ pop_frame();
 345   __ restore_return_pc();
 346   __ z_br(Z_R14);
 347 
 348   // Copy int arguments.
 349 
 350   Label  iarg_caselist;   // Distance between each case has to be a power of 2
 351                           // (= 1 << LogSizeOfCase).
 352   __ align(16);
 353   BIND(iarg_caselist);
 354   __ z_lgr(Z_ARG3, intSlot);    // 4 bytes
 355   __ z_bru(loop_start_restore); // 4 bytes
 356 
 357   __ z_lgr(Z_ARG4, intSlot);
 358   __ z_bru(loop_start_restore);
 359 
 360   __ z_lgr(Z_ARG5, intSlot);
 361   __ z_bru(loop_start_restore);
 362 
 363   __ align(16);
 364   __ bind(move_intSlot_to_ARG);
 365   __ z_stg(signature, d_signature, Z_SP);       // Spill since signature == Z_R1_scratch.
 366   __ z_larl(Z_R1_scratch, iarg_caselist);
 367   __ z_sllg(Z_R0_scratch, argcnt, LogSizeOfCase);
 368   __ add2reg(argcnt, 1);
 369   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 370   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 371 
 372   // Copy float arguments.
 373 
 374   Label  farg_caselist;   // Distance between each case has to be a power of 2
 375                           // (= 1 << logSizeOfCase, padded with nop.
 376   __ align(16);
 377   BIND(farg_caselist);
 378   __ z_ldr(Z_FARG1, floatSlot); // 2 bytes
 379   __ z_bru(loop_start_restore); // 4 bytes
 380   __ z_nop();                   // 2 bytes
 381 
 382   __ z_ldr(Z_FARG2, floatSlot);
 383   __ z_bru(loop_start_restore);
 384   __ z_nop();
 385 
 386   __ z_ldr(Z_FARG3, floatSlot);
 387   __ z_bru(loop_start_restore);
 388   __ z_nop();
 389 
 390   __ z_ldr(Z_FARG4, floatSlot);
 391   __ z_bru(loop_start_restore);
 392   __ z_nop();
 393 
 394   __ align(16);
 395   __ bind(move_floatSlot_to_FARG);
 396   __ z_stg(signature, d_signature, Z_SP);        // Spill since signature == Z_R1_scratch.
 397   __ z_lg(Z_R0_scratch, d_fpcnt, Z_SP);          // Need old value for indexing.
 398   __ add2mem_64(Address(Z_SP, d_fpcnt), 1, Z_R1_scratch); // Increment index.
 399   __ z_larl(Z_R1_scratch, farg_caselist);
 400   __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogSizeOfCase);
 401   __ z_agr(Z_R1_scratch, Z_R0_scratch);
 402   __ z_bcr(Assembler::bcondAlways, Z_R1_scratch);
 403 
 404   BLOCK_COMMENT("} slow_signature_handler");
 405 
 406   return __ addr_at(entry_offset);
 407 }
 408 
 409 address TemplateInterpreterGenerator::generate_result_handler_for (BasicType type) {
 410   address entry = __ pc();
 411 
 412   assert(Z_tos == Z_RET, "Result handler: must move result!");
 413   assert(Z_ftos == Z_FRET, "Result handler: must move float result!");
 414 
 415   switch (type) {
 416     case T_BOOLEAN:
 417       __ c2bool(Z_tos);
 418       break;
 419     case T_CHAR:
 420       __ and_imm(Z_tos, 0xffff);
 421       break;
 422     case T_BYTE:
 423       __ z_lbr(Z_tos, Z_tos);
 424       break;
 425     case T_SHORT:
 426       __ z_lhr(Z_tos, Z_tos);
 427       break;
 428     case T_INT:
 429     case T_LONG:
 430     case T_VOID:
 431     case T_FLOAT:
 432     case T_DOUBLE:
 433       break;
 434     case T_OBJECT:
 435       // Retrieve result from frame...
 436       __ mem2reg_opt(Z_tos, Address(Z_fp, oop_tmp_offset));
 437       // and verify it.
 438       __ verify_oop(Z_tos);
 439       break;
 440     default:
 441       ShouldNotReachHere();
 442   }
 443   __ z_br(Z_R14);      // Return from result handler.
 444   return entry;
 445 }
 446 
 447 // Abstract method entry.
 448 // Attempt to execute abstract method. Throw exception.
 449 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 450   unsigned int entry_offset = __ offset();
 451 
 452   // Caller could be the call_stub or a compiled method (x86 version is wrong!).
 453 
 454   BLOCK_COMMENT("abstract_entry {");
 455 
 456   // Implement call of InterpreterRuntime::throw_AbstractMethodError.
 457   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1);
 458   __ save_return_pc();       // Save Z_R14.
 459   __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
 460 
 461   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), Z_thread);
 462 
 463   __ pop_frame();
 464   __ restore_return_pc();    // Restore Z_R14.
 465   __ reset_last_Java_frame();
 466 
 467   // Restore caller sp for c2i case.
 468   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 469 
 470   // branch to SharedRuntime::generate_forward_exception() which handles all possible callers,
 471   // i.e. call stub, compiled method, interpreted method.
 472   __ load_absolute_address(Z_tmp_1, StubRoutines::forward_exception_entry());
 473   __ z_br(Z_tmp_1);
 474 
 475   BLOCK_COMMENT("} abstract_entry");
 476 
 477   return __ addr_at(entry_offset);
 478 }
 479 
 480 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 481 #if INCLUDE_ALL_GCS
 482   if (UseG1GC) {
 483     // Inputs:
 484     //  Z_ARG1 - receiver
 485     //
 486     // What we do:
 487     //  - Load the referent field address.
 488     //  - Load the value in the referent field.
 489     //  - Pass that value to the pre-barrier.
 490     //
 491     // In the case of G1 this will record the value of the
 492     // referent in an SATB buffer if marking is active.
 493     // This will cause concurrent marking to mark the referent
 494     // field as live.
 495 
 496     Register  scratch1 = Z_tmp_2;
 497     Register  scratch2 = Z_tmp_3;
 498     Register  pre_val  = Z_RET;   // return value
 499     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
 500     Register  Rargp    = Z_esp;
 501 
 502     Label     slow_path;
 503     address   entry = __ pc();
 504 
 505     const int referent_offset = java_lang_ref_Reference::referent_offset;
 506     guarantee(referent_offset > 0, "referent offset not initialized");
 507 
 508     BLOCK_COMMENT("Reference_get {");
 509 
 510     //  If the receiver is null then it is OK to jump to the slow path.
 511     __ load_and_test_long(pre_val, Address(Rargp, Interpreter::stackElementSize)); // Get receiver.
 512     __ z_bre(slow_path);
 513 
 514     //  Load the value of the referent field.
 515     __ load_heap_oop(pre_val, referent_offset, pre_val);
 516 
 517     // Restore caller sp for c2i case.
 518     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
 519 
 520     // Generate the G1 pre-barrier code to log the value of
 521     // the referent field in an SATB buffer.
 522     // Note:
 523     //   With these parameters the write_barrier_pre does not
 524     //   generate instructions to load the previous value.
 525     __ g1_write_barrier_pre(noreg,      // obj
 526                             noreg,      // offset
 527                             pre_val,    // pre_val
 528                             noreg,      // no new val to preserve
 529                             scratch1,   // tmp
 530                             scratch2,   // tmp
 531                             true);      // pre_val_needed
 532 
 533     __ z_br(Z_R14);
 534 
 535     // Branch to previously generated regular method entry.
 536     __ bind(slow_path);
 537 
 538     address meth_entry = Interpreter::entry_for_kind(Interpreter::zerolocals);
 539     __ jump_to_entry(meth_entry, Z_R1);
 540 
 541     BLOCK_COMMENT("} Reference_get");
 542 
 543     return entry;
 544   }
 545 #endif // INCLUDE_ALL_GCS
 546 
 547   return NULL;
 548 }
 549 
 550 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 551   address entry = __ pc();
 552 
 553   DEBUG_ONLY(__ verify_esp(Z_esp, Z_ARG5));
 554 
 555   // Restore bcp under the assumption that the current frame is still
 556   // interpreted.
 557   __ restore_bcp();
 558 
 559   // Expression stack must be empty before entering the VM if an
 560   // exception happened.
 561   __ empty_expression_stack();
 562   // Throw exception.
 563   __ call_VM(noreg,
 564              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
 565   return entry;
 566 }
 567 
 568 //
 569 // Args:
 570 //   Z_ARG3: aberrant index
 571 //
 572 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char * name) {
 573   address entry = __ pc();
 574   address excp = CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException);
 575 
 576   // Expression stack must be empty before entering the VM if an
 577   // exception happened.
 578   __ empty_expression_stack();
 579 
 580   // Setup parameters.
 581   // Leave out the name and use register for array to create more detailed exceptions.
 582   __ load_absolute_address(Z_ARG2, (address) name);
 583   __ call_VM(noreg, excp, Z_ARG2, Z_ARG3);
 584   return entry;
 585 }
 586 
 587 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 588   address entry = __ pc();
 589 
 590   // Object is at TOS.
 591   __ pop_ptr(Z_ARG2);
 592 
 593   // Expression stack must be empty before entering the VM if an
 594   // exception happened.
 595   __ empty_expression_stack();
 596 
 597   __ call_VM(Z_ARG1,
 598              CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException),
 599              Z_ARG2);
 600 
 601   DEBUG_ONLY(__ should_not_reach_here();)
 602 
 603   return entry;
 604 }
 605 
 606 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 607   assert(!pass_oop || message == NULL, "either oop or message but not both");
 608   address entry = __ pc();
 609 
 610   BLOCK_COMMENT("exception_handler_common {");
 611 
 612   // Expression stack must be empty before entering the VM if an
 613   // exception happened.
 614   __ empty_expression_stack();
 615   if (name != NULL) {
 616     __ load_absolute_address(Z_ARG2, (address)name);
 617   } else {
 618     __ clear_reg(Z_ARG2, true, false);
 619   }
 620 
 621   if (pass_oop) {
 622     __ call_VM(Z_tos,
 623                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception),
 624                Z_ARG2, Z_tos /*object (see TT::aastore())*/);
 625   } else {
 626     if (message != NULL) {
 627       __ load_absolute_address(Z_ARG3, (address)message);
 628     } else {
 629       __ clear_reg(Z_ARG3, true, false);
 630     }
 631     __ call_VM(Z_tos,
 632                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 633                Z_ARG2, Z_ARG3);
 634   }
 635   // Throw exception.
 636   __ load_absolute_address(Z_R1_scratch, Interpreter::throw_exception_entry());
 637   __ z_br(Z_R1_scratch);
 638 
 639   BLOCK_COMMENT("} exception_handler_common");
 640 
 641   return entry;
 642 }
 643 
 644 address TemplateInterpreterGenerator::generate_return_entry_for (TosState state, int step, size_t index_size) {
 645   address entry = __ pc();
 646 
 647   BLOCK_COMMENT("return_entry {");
 648 
 649   // Pop i2c extension or revert top-2-parent-resize done by interpreted callees.
 650   Register sp_before_i2c_extension = Z_bcp;
 651   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 652   __ z_lg(sp_before_i2c_extension, Address(Z_fp, _z_ijava_state_neg(top_frame_sp)));
 653   __ resize_frame_absolute(sp_before_i2c_extension, Z_locals/*tmp*/, true/*load_fp*/);
 654 
 655   // TODO(ZASM): necessary??
 656   //  // and NULL it as marker that esp is now tos until next java call
 657   //  __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 658 
 659   __ restore_bcp();
 660   __ restore_locals();
 661   __ restore_esp();
 662 
 663   if (state == atos) {
 664     __ profile_return_type(Z_tmp_1, Z_tos, Z_tmp_2);
 665   }
 666 
 667   Register cache  = Z_tmp_1;
 668   Register size   = Z_tmp_1;
 669   Register offset = Z_tmp_2;
 670   const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
 671                                     ConstantPoolCacheEntry::flags_offset());
 672   __ get_cache_and_index_at_bcp(cache, offset, 1, index_size);
 673 
 674   // #args is in rightmost byte of the _flags field.
 675   __ z_llgc(size, Address(cache, offset, flags_offset+(sizeof(size_t)-1)));
 676   __ z_sllg(size, size, Interpreter::logStackElementSize); // Each argument size in bytes.
 677   __ z_agr(Z_esp, size);                                   // Pop arguments.
 678 
 679   __ check_and_handle_popframe(Z_thread);
 680   __ check_and_handle_earlyret(Z_thread);
 681 
 682   __ dispatch_next(state, step);
 683 
 684   BLOCK_COMMENT("} return_entry");
 685 
 686   return entry;
 687 }
 688 
 689 address TemplateInterpreterGenerator::generate_deopt_entry_for (TosState state,
 690                                                                int step,
 691                                                                address continuation) {
 692   address entry = __ pc();
 693 
 694   BLOCK_COMMENT("deopt_entry {");
 695 
 696   // TODO(ZASM): necessary? NULL last_sp until next java call
 697   // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 698   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
 699   __ restore_bcp();
 700   __ restore_locals();
 701   __ restore_esp();
 702 
 703   // Handle exceptions.
 704   {
 705     Label L;
 706     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
 707     __ z_bre(L);
 708     __ call_VM(noreg,
 709                CAST_FROM_FN_PTR(address,
 710                                 InterpreterRuntime::throw_pending_exception));
 711     __ should_not_reach_here();
 712     __ bind(L);
 713   }
 714   if (continuation == NULL) {
 715     __ dispatch_next(state, step);
 716   } else {
 717     __ jump_to_entry(continuation, Z_R1_scratch);
 718   }
 719 
 720   BLOCK_COMMENT("} deopt_entry");
 721 
 722   return entry;
 723 }
 724 
 725 address TemplateInterpreterGenerator::generate_safept_entry_for (TosState state,
 726                                                                 address runtime_entry) {
 727   address entry = __ pc();
 728   __ push(state);
 729   __ call_VM(noreg, runtime_entry);
 730   __ dispatch_via(vtos, Interpreter::_normal_table.table_for (vtos));
 731   return entry;
 732 }
 733 
 734 //
 735 // Helpers for commoning out cases in the various type of method entries.
 736 //
 737 
 738 // Increment invocation count & check for overflow.
 739 //
 740 // Note: checking for negative value instead of overflow
 741 // so we have a 'sticky' overflow test.
 742 //
 743 // Z_ARG2: method (see generate_fixed_frame())
 744 //
 745 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 746   Label done;
 747   Register method = Z_ARG2; // Generate_fixed_frame() copies Z_method into Z_ARG2.
 748   Register m_counters = Z_ARG4;
 749 
 750   BLOCK_COMMENT("counter_incr {");
 751 
 752   // Note: In tiered we increment either counters in method or in MDO depending
 753   // if we are profiling or not.
 754   if (TieredCompilation) {
 755     int increment = InvocationCounter::count_increment;
 756     if (ProfileInterpreter) {
 757       NearLabel no_mdo;
 758       Register mdo = m_counters;
 759       // Are we profiling?
 760       __ load_and_test_long(mdo, method2_(method, method_data));
 761       __ branch_optimized(Assembler::bcondZero, no_mdo);
 762       // Increment counter in the MDO.
 763       const Address mdo_invocation_counter(mdo, MethodData::invocation_counter_offset() +
 764                                            InvocationCounter::counter_offset());
 765       const Address mask(mdo, MethodData::invoke_mask_offset());
 766       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 767                                  Z_R1_scratch, false, Assembler::bcondZero,
 768                                  overflow);
 769       __ z_bru(done);
 770       __ bind(no_mdo);
 771     }
 772 
 773     // Increment counter in MethodCounters.
 774     const Address invocation_counter(m_counters,
 775                                      MethodCounters::invocation_counter_offset() +
 776                                      InvocationCounter::counter_offset());
 777     // Get address of MethodCounters object.
 778     __ get_method_counters(method, m_counters, done);
 779     const Address mask(m_counters, MethodCounters::invoke_mask_offset());
 780     __ increment_mask_and_jump(invocation_counter,
 781                                increment, mask,
 782                                Z_R1_scratch, false, Assembler::bcondZero,
 783                                overflow);
 784   } else {
 785     Register counter_sum = Z_ARG3; // The result of this piece of code.
 786     Register tmp         = Z_R1_scratch;
 787 #ifdef ASSERT
 788     {
 789       NearLabel ok;
 790       __ get_method(tmp);
 791       __ compare64_and_branch(method, tmp, Assembler::bcondEqual, ok);
 792       __ z_illtrap(0x66);
 793       __ bind(ok);
 794     }
 795 #endif
 796 
 797     // Get address of MethodCounters object.
 798     __ get_method_counters(method, m_counters, done);
 799     // Update standard invocation counters.
 800     __ increment_invocation_counter(m_counters, counter_sum);
 801     if (ProfileInterpreter) {
 802       __ add2mem_32(Address(m_counters, MethodCounters::interpreter_invocation_counter_offset()), 1, tmp);
 803       if (profile_method != NULL) {
 804         const Address profile_limit(m_counters, MethodCounters::interpreter_profile_limit_offset());
 805         __ z_cl(counter_sum, profile_limit);
 806         __ branch_optimized(Assembler::bcondLow, *profile_method_continue);
 807         // If no method data exists, go to profile_method.
 808         __ test_method_data_pointer(tmp, *profile_method);
 809       }
 810     }
 811 
 812     const Address invocation_limit(m_counters, MethodCounters::interpreter_invocation_limit_offset());
 813     __ z_cl(counter_sum, invocation_limit);
 814     __ branch_optimized(Assembler::bcondNotLow, *overflow);
 815   }
 816 
 817   __ bind(done);
 818 
 819   BLOCK_COMMENT("} counter_incr");
 820 }
 821 
 822 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 823   // InterpreterRuntime::frequency_counter_overflow takes two
 824   // arguments, the first (thread) is passed by call_VM, the second
 825   // indicates if the counter overflow occurs at a backwards branch
 826   // (NULL bcp). We pass zero for it. The call returns the address
 827   // of the verified entry point for the method or NULL if the
 828   // compilation did not complete (either went background or bailed
 829   // out).
 830   __ clear_reg(Z_ARG2);
 831   __ call_VM(noreg,
 832              CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow),
 833              Z_ARG2);
 834   __ z_bru(do_continue);
 835 }
 836 
 837 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register frame_size, Register tmp1) {
 838   Register tmp2 = Z_R1_scratch;
 839   const int page_size = os::vm_page_size();
 840   NearLabel after_frame_check;
 841 
 842   BLOCK_COMMENT("counter_overflow {");
 843 
 844   assert_different_registers(frame_size, tmp1);
 845 
 846   // Stack banging is sufficient overflow check if frame_size < page_size.
 847   if (Immediate::is_uimm(page_size, 15)) {
 848     __ z_chi(frame_size, page_size);
 849     __ z_brl(after_frame_check);
 850   } else {
 851     __ load_const_optimized(tmp1, page_size);
 852     __ compareU32_and_branch(frame_size, tmp1, Assembler::bcondLow, after_frame_check);
 853   }
 854 
 855   // Get the stack base, and in debug, verify it is non-zero.
 856   __ z_lg(tmp1, thread_(stack_base));
 857 #ifdef ASSERT
 858   address reentry = NULL;
 859   NearLabel base_not_zero;
 860   __ compareU64_and_branch(tmp1, (intptr_t)0L, Assembler::bcondNotEqual, base_not_zero);
 861   reentry = __ stop_chain_static(reentry, "stack base is zero in generate_stack_overflow_check");
 862   __ bind(base_not_zero);
 863 #endif
 864 
 865   // Get the stack size, and in debug, verify it is non-zero.
 866   assert(sizeof(size_t) == sizeof(intptr_t), "wrong load size");
 867   __ z_lg(tmp2, thread_(stack_size));
 868 #ifdef ASSERT
 869   NearLabel size_not_zero;
 870   __ compareU64_and_branch(tmp2, (intptr_t)0L, Assembler::bcondNotEqual, size_not_zero);
 871   reentry = __ stop_chain_static(reentry, "stack size is zero in generate_stack_overflow_check");
 872   __ bind(size_not_zero);
 873 #endif
 874 
 875   // Compute the beginning of the protected zone minus the requested frame size.
 876   __ z_sgr(tmp1, tmp2);
 877   __ add2reg(tmp1, JavaThread::stack_guard_zone_size());
 878 
 879   // Add in the size of the frame (which is the same as subtracting it from the
 880   // SP, which would take another register.
 881   __ z_agr(tmp1, frame_size);
 882 
 883   // The frame is greater than one page in size, so check against
 884   // the bottom of the stack.
 885   __ compareU64_and_branch(Z_SP, tmp1, Assembler::bcondHigh, after_frame_check);
 886 
 887   // The stack will overflow, throw an exception.
 888 
 889   // Restore SP to sender's sp. This is necessary if the sender's frame is an
 890   // extended compiled frame (see gen_c2i_adapter()) and safer anyway in case of
 891   // JSR292 adaptations.
 892   __ resize_frame_absolute(Z_R10, tmp1, true/*load_fp*/);
 893 
 894   // Note also that the restored frame is not necessarily interpreted.
 895   // Use the shared runtime version of the StackOverflowError.
 896   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 897   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 898   __ load_absolute_address(tmp1, StubRoutines::throw_StackOverflowError_entry());
 899   __ z_br(tmp1);
 900 
 901   // If you get to here, then there is enough stack space.
 902   __ bind(after_frame_check);
 903 
 904   BLOCK_COMMENT("} counter_overflow");
 905 }
 906 
 907 // Allocate monitor and lock method (asm interpreter).
 908 //
 909 // Args:
 910 //   Z_locals: locals
 911 
 912 void TemplateInterpreterGenerator::lock_method(void) {
 913 
 914   BLOCK_COMMENT("lock_method {");
 915 
 916   // Synchronize method.
 917   const Register method = Z_tmp_2;
 918   __ get_method(method);
 919 
 920 #ifdef ASSERT
 921   address reentry = NULL;
 922   {
 923     Label L;
 924     __ testbit(method2_(method, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
 925     __ z_btrue(L);
 926     reentry = __ stop_chain_static(reentry, "method doesn't need synchronization");
 927     __ bind(L);
 928   }
 929 #endif // ASSERT
 930 
 931   // Get synchronization object.
 932   const Register object = Z_tmp_2;
 933 
 934   {
 935     Label     done;
 936     Label     static_method;
 937 
 938     __ testbit(method2_(method, access_flags), JVM_ACC_STATIC_BIT);
 939     __ z_btrue(static_method);
 940 
 941     // non-static method: Load receiver obj from stack.
 942     __ mem2reg_opt(object, Address(Z_locals, Interpreter::local_offset_in_bytes(0)));
 943     __ z_bru(done);
 944 
 945     __ bind(static_method);
 946 
 947     // Lock the java mirror.
 948     __ load_mirror(object, method);
 949 #ifdef ASSERT
 950     {
 951       NearLabel L;
 952       __ compare64_and_branch(object, (intptr_t) 0, Assembler::bcondNotEqual, L);
 953       reentry = __ stop_chain_static(reentry, "synchronization object is NULL");
 954       __ bind(L);
 955     }
 956 #endif // ASSERT
 957 
 958     __ bind(done);
 959   }
 960 
 961   __ add_monitor_to_stack(true, Z_ARG3, Z_ARG4, Z_ARG5); // Allocate monitor elem.
 962   // Store object and lock it.
 963   __ get_monitors(Z_tmp_1);
 964   __ reg2mem_opt(object, Address(Z_tmp_1, BasicObjectLock::obj_offset_in_bytes()));
 965   __ lock_object(Z_tmp_1, object);
 966 
 967   BLOCK_COMMENT("} lock_method");
 968 }
 969 
 970 // Generate a fixed interpreter frame. This is identical setup for
 971 // interpreted methods and for native methods hence the shared code.
 972 //
 973 // Registers alive
 974 //   Z_thread   - JavaThread*
 975 //   Z_SP       - old stack pointer
 976 //   Z_method   - callee's method
 977 //   Z_esp      - parameter list (slot 'above' last param)
 978 //   Z_R14      - return pc, to be stored in caller's frame
 979 //   Z_R10      - sender sp, note: Z_tmp_1 is Z_R10!
 980 //
 981 // Registers updated
 982 //   Z_SP       - new stack pointer
 983 //   Z_esp      - callee's operand stack pointer
 984 //                points to the slot above the value on top
 985 //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
 986 //   Z_bcp      - the bytecode pointer
 987 //   Z_fp       - the frame pointer, thereby killing Z_method
 988 //   Z_ARG2     - copy of Z_method
 989 //
 990 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 991 
 992   //  stack layout
 993   //
 994   //   F1 [TOP_IJAVA_FRAME_ABI]              <-- Z_SP, Z_R10 (see note below)
 995   //      [F1's operand stack (unused)]
 996   //      [F1's outgoing Java arguments]     <-- Z_esp
 997   //      [F1's operand stack (non args)]
 998   //      [monitors]      (optional)
 999   //      [IJAVA_STATE]
1000   //
1001   //   F2 [PARENT_IJAVA_FRAME_ABI]
1002   //      ...
1003   //
1004   //  0x000
1005   //
1006   // Note: Z_R10, the sender sp, will be below Z_SP if F1 was extended by a c2i adapter.
1007 
1008   //=============================================================================
1009   // Allocate space for locals other than the parameters, the
1010   // interpreter state, monitors, and the expression stack.
1011 
1012   const Register local_count     = Z_ARG5;
1013   const Register fp              = Z_tmp_2;
1014 
1015   BLOCK_COMMENT("generate_fixed_frame {");
1016 
1017   {
1018   // local registers
1019   const Register top_frame_size  = Z_ARG2;
1020   const Register sp_after_resize = Z_ARG3;
1021   const Register max_stack       = Z_ARG4;
1022 
1023   // local_count = method->constMethod->max_locals();
1024   __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1025   __ z_llgh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1026 
1027   if (native_call) {
1028     // If we're calling a native method, we replace max_stack (which is
1029     // zero) with space for the worst-case signature handler varargs
1030     // vector, which is:
1031     //   max_stack = max(Argument::n_register_parameters, parameter_count+2);
1032     //
1033     // We add two slots to the parameter_count, one for the jni
1034     // environment and one for a possible native mirror. We allocate
1035     // space for at least the number of ABI registers, even though
1036     // InterpreterRuntime::slow_signature_handler won't write more than
1037     // parameter_count+2 words when it creates the varargs vector at the
1038     // top of the stack. The generated slow signature handler will just
1039     // load trash into registers beyond the necessary number. We're
1040     // still going to cut the stack back by the ABI register parameter
1041     // count so as to get SP+16 pointing at the ABI outgoing parameter
1042     // area, so we need to allocate at least that much even though we're
1043     // going to throw it away.
1044     //
1045 
1046     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1047     __ z_llgh(max_stack,  Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1048     __ add2reg(max_stack, 2);
1049 
1050     NearLabel passing_args_on_stack;
1051 
1052     // max_stack in bytes
1053     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1054 
1055     int argument_registers_in_bytes = Argument::n_register_parameters << LogBytesPerWord;
1056     __ compare64_and_branch(max_stack, argument_registers_in_bytes, Assembler::bcondNotLow, passing_args_on_stack);
1057 
1058     __ load_const_optimized(max_stack, argument_registers_in_bytes);
1059 
1060     __ bind(passing_args_on_stack);
1061   } else {
1062     // !native_call
1063     __ z_lg(max_stack, method_(const));
1064 
1065     // Calculate number of non-parameter locals (in slots):
1066     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1067     __ z_sh(local_count, Address(Z_R1_scratch, ConstMethod::size_of_parameters_offset()));
1068 
1069     // max_stack = method->max_stack();
1070     __ z_llgh(max_stack, Address(max_stack, ConstMethod::max_stack_offset()));
1071     // max_stack in bytes
1072     __ z_sllg(max_stack, max_stack, LogBytesPerWord);
1073   }
1074 
1075   // Resize (i.e. normally shrink) the top frame F1 ...
1076   //   F1      [TOP_IJAVA_FRAME_ABI]          <-- Z_SP, Z_R10
1077   //           F1's operand stack (free)
1078   //           ...
1079   //           F1's operand stack (free)      <-- Z_esp
1080   //           F1's outgoing Java arg m
1081   //           ...
1082   //           F1's outgoing Java arg 0
1083   //           ...
1084   //
1085   //  ... into a parent frame (Z_R10 holds F1's SP before any modification, see also above)
1086   //
1087   //           +......................+
1088   //           :                      :        <-- Z_R10, saved below as F0's z_ijava_state.sender_sp
1089   //           :                      :
1090   //   F1      [PARENT_IJAVA_FRAME_ABI]        <-- Z_SP       \
1091   //           F0's non arg local                             | = delta
1092   //           ...                                            |
1093   //           F0's non arg local              <-- Z_esp      /
1094   //           F1's outgoing Java arg m
1095   //           ...
1096   //           F1's outgoing Java arg 0
1097   //           ...
1098   //
1099   // then push the new top frame F0.
1100   //
1101   //   F0      [TOP_IJAVA_FRAME_ABI]    = frame::z_top_ijava_frame_abi_size \
1102   //           [operand stack]          = max_stack                          | = top_frame_size
1103   //           [IJAVA_STATE]            = frame::z_ijava_state_size         /
1104 
1105   // sp_after_resize = Z_esp - delta
1106   //
1107   // delta = PARENT_IJAVA_FRAME_ABI + (locals_count - params_count)
1108 
1109   __ add2reg(sp_after_resize, (Interpreter::stackElementSize) - (frame::z_parent_ijava_frame_abi_size), Z_esp);
1110   __ z_sllg(Z_R0_scratch, local_count, LogBytesPerWord); // Params have already been subtracted from local_count.
1111   __ z_slgr(sp_after_resize, Z_R0_scratch);
1112 
1113   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
1114   __ add2reg(top_frame_size,
1115              frame::z_top_ijava_frame_abi_size +
1116              frame::z_ijava_state_size +
1117              frame::interpreter_frame_monitor_size() * wordSize,
1118              max_stack);
1119 
1120   if (!native_call) {
1121     // Stack overflow check.
1122     // Native calls don't need the stack size check since they have no
1123     // expression stack and the arguments are already on the stack and
1124     // we only add a handful of words to the stack.
1125     Register frame_size = max_stack; // Reuse the regiser for max_stack.
1126     __ z_lgr(frame_size, Z_SP);
1127     __ z_sgr(frame_size, sp_after_resize);
1128     __ z_agr(frame_size, top_frame_size);
1129     generate_stack_overflow_check(frame_size, fp/*tmp1*/);
1130   }
1131 
1132   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
1133   __ asm_assert_eq("killed Z_R14", 0);
1134   __ resize_frame_absolute(sp_after_resize, fp, true);
1135   __ save_return_pc(Z_R14);
1136 
1137   // ... and push the new frame F0.
1138   __ push_frame(top_frame_size, fp, true /*copy_sp*/, false);
1139   }
1140 
1141   //=============================================================================
1142   // Initialize the new frame F0: initialize interpreter state.
1143 
1144   {
1145   // locals
1146   const Register local_addr = Z_ARG4;
1147 
1148   BLOCK_COMMENT("generate_fixed_frame: initialize interpreter state {");
1149 
1150 #ifdef ASSERT
1151   // Set the magic number (using local_addr as tmp register).
1152   __ load_const_optimized(local_addr, frame::z_istate_magic_number);
1153   __ z_stg(local_addr, _z_ijava_state_neg(magic), fp);
1154 #endif
1155 
1156   // Save sender SP from F1 (i.e. before it was potentially modified by an
1157   // adapter) into F0's interpreter state. We us it as well to revert
1158   // resizing the frame above.
1159   __ z_stg(Z_R10, _z_ijava_state_neg(sender_sp), fp);
1160 
1161   // Load cp cache and save it at the and of this block.
1162   __ z_lg(Z_R1_scratch, Address(Z_method,    Method::const_offset()));
1163   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstMethod::constants_offset()));
1164   __ z_lg(Z_R1_scratch, Address(Z_R1_scratch, ConstantPool::cache_offset_in_bytes()));
1165 
1166   // z_ijava_state->method = method;
1167   __ z_stg(Z_method, _z_ijava_state_neg(method), fp);
1168 
1169   // Point locals at the first argument. Method's locals are the
1170   // parameters on top of caller's expression stack.
1171   // Tos points past last Java argument.
1172 
1173   __ z_lg(Z_locals, Address(Z_method, Method::const_offset()));
1174   __ z_llgh(Z_locals /*parameter_count words*/,
1175             Address(Z_locals, ConstMethod::size_of_parameters_offset()));
1176   __ z_sllg(Z_locals /*parameter_count bytes*/, Z_locals /*parameter_count*/, LogBytesPerWord);
1177   __ z_agr(Z_locals, Z_esp);
1178   // z_ijava_state->locals - i*BytesPerWord points to i-th Java local (i starts at 0)
1179   // z_ijava_state->locals = Z_esp + parameter_count bytes
1180   __ z_stg(Z_locals, _z_ijava_state_neg(locals), fp);
1181 
1182   // z_ijava_state->oop_temp = NULL;
1183   __ store_const(Address(fp, oop_tmp_offset), 0);
1184 
1185   // Initialize z_ijava_state->mdx.
1186   Register Rmdp = Z_bcp;
1187   // native_call: assert that mdo == NULL
1188   const bool check_for_mdo = !native_call DEBUG_ONLY(|| native_call);
1189   if (ProfileInterpreter && check_for_mdo) {
1190     Label get_continue;
1191 
1192     __ load_and_test_long(Rmdp, method_(method_data));
1193     __ z_brz(get_continue);
1194     DEBUG_ONLY(if (native_call) __ stop("native methods don't have a mdo"));
1195     __ add2reg(Rmdp, in_bytes(MethodData::data_offset()));
1196     __ bind(get_continue);
1197   }
1198   __ z_stg(Rmdp, _z_ijava_state_neg(mdx), fp);
1199 
1200   // Initialize z_ijava_state->bcp and Z_bcp.
1201   if (native_call) {
1202     __ clear_reg(Z_bcp); // Must initialize. Will get written into frame where GC reads it.
1203   } else {
1204     __ z_lg(Z_bcp, method_(const));
1205     __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset()));
1206   }
1207   __ z_stg(Z_bcp, _z_ijava_state_neg(bcp), fp);
1208 
1209   // no monitors and empty operand stack
1210   // => z_ijava_state->monitors points to the top slot in IJAVA_STATE.
1211   // => Z_ijava_state->esp points one slot above into the operand stack.
1212   // z_ijava_state->monitors = fp - frame::z_ijava_state_size - Interpreter::stackElementSize;
1213   // z_ijava_state->esp = Z_esp = z_ijava_state->monitors;
1214   __ add2reg(Z_esp, -frame::z_ijava_state_size, fp);
1215   __ z_stg(Z_esp, _z_ijava_state_neg(monitors), fp);
1216   __ add2reg(Z_esp, -Interpreter::stackElementSize);
1217   __ z_stg(Z_esp, _z_ijava_state_neg(esp), fp);
1218 
1219   // z_ijava_state->cpoolCache = Z_R1_scratch (see load above);
1220   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(cpoolCache), fp);
1221 
1222   // Get mirror and store it in the frame as GC root for this Method*.
1223   __ load_mirror(Z_R1_scratch, Z_method);
1224   __ z_stg(Z_R1_scratch, _z_ijava_state_neg(mirror), fp);
1225 
1226   BLOCK_COMMENT("} generate_fixed_frame: initialize interpreter state");
1227 
1228   //=============================================================================
1229   if (!native_call) {
1230     // Fill locals with 0x0s.
1231     NearLabel locals_zeroed;
1232     NearLabel doXC;
1233 
1234     // Local_count is already num_locals_slots - num_param_slots.
1235     __ compare64_and_branch(local_count, (intptr_t)0L, Assembler::bcondNotHigh, locals_zeroed);
1236 
1237     // Advance local_addr to point behind locals (creates positive incr. in loop).
1238     __ z_lg(Z_R1_scratch, Address(Z_method, Method::const_offset()));
1239     __ z_llgh(Z_R0_scratch, Address(Z_R1_scratch, ConstMethod::size_of_locals_offset()));
1240     __ add2reg(Z_R0_scratch, -1);
1241 
1242     __ z_lgr(local_addr/*locals*/, Z_locals);
1243     __ z_sllg(Z_R0_scratch, Z_R0_scratch, LogBytesPerWord);
1244     __ z_sllg(local_count, local_count, LogBytesPerWord); // Local_count are non param locals.
1245     __ z_sgr(local_addr, Z_R0_scratch);
1246 
1247     if (VM_Version::has_Prefetch()) {
1248       __ z_pfd(0x02, 0, Z_R0, local_addr);
1249       __ z_pfd(0x02, 256, Z_R0, local_addr);
1250     }
1251 
1252     // Can't optimise for Z10 using "compare and branch" (immediate value is too big).
1253     __ z_cghi(local_count, 256);
1254     __ z_brnh(doXC);
1255 
1256     // MVCLE: Initialize if quite a lot locals.
1257     //  __ bind(doMVCLE);
1258     __ z_lgr(Z_R0_scratch, local_addr);
1259     __ z_lgr(Z_R1_scratch, local_count);
1260     __ clear_reg(Z_ARG2);        // Src len of MVCLE is zero.
1261 
1262     __ MacroAssembler::move_long_ext(Z_R0_scratch, Z_ARG1, 0);
1263     __ z_bru(locals_zeroed);
1264 
1265     Label  XC_template;
1266     __ bind(XC_template);
1267     __ z_xc(0, 0, local_addr, 0, local_addr);
1268 
1269     __ bind(doXC);
1270     __ z_bctgr(local_count, Z_R0);                  // Get #bytes-1 for EXECUTE.
1271     if (VM_Version::has_ExecuteExtensions()) {
1272       __ z_exrl(local_count, XC_template);          // Execute XC with variable length.
1273     } else {
1274       __ z_larl(Z_R1_scratch, XC_template);
1275       __ z_ex(local_count, 0, Z_R0, Z_R1_scratch);  // Execute XC with variable length.
1276     }
1277 
1278     __ bind(locals_zeroed);
1279   }
1280 
1281   }
1282   // Finally set the frame pointer, destroying Z_method.
1283   assert(Z_fp == Z_method, "maybe set Z_fp earlier if other register than Z_method");
1284   // Oprofile analysis suggests to keep a copy in a register to be used by
1285   // generate_counter_incr().
1286   __ z_lgr(Z_ARG2, Z_method);
1287   __ z_lgr(Z_fp, fp);
1288 
1289   BLOCK_COMMENT("} generate_fixed_frame");
1290 }
1291 
1292 // Various method entries
1293 
1294 // Math function, frame manager must set up an interpreter state, etc.
1295 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1296 
1297   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1298   bool use_instruction = false;
1299   address runtime_entry = NULL;
1300   int num_args = 1;
1301   bool double_precision = true;
1302 
1303   // s390 specific:
1304   switch (kind) {
1305     case Interpreter::java_lang_math_sqrt:
1306     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1307     case Interpreter::java_lang_math_fmaF:
1308     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1309     default: break; // Fall back to runtime call.
1310   }
1311 
1312   switch (kind) {
1313     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1314     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1315     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1316     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1317     case Interpreter::java_lang_math_sqrt : /* runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); not available */ break;
1318     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1319     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1320     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1321     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1322     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1323     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1324     default: ShouldNotReachHere();
1325   }
1326 
1327   // Use normal entry if neither instruction nor runtime call is used.
1328   if (!use_instruction && runtime_entry == NULL) return NULL;
1329 
1330   address entry = __ pc();
1331 
1332   if (use_instruction) {
1333     switch (kind) {
1334       case Interpreter::java_lang_math_sqrt:
1335         // Can use memory operand directly.
1336         __ z_sqdb(Z_FRET, Interpreter::stackElementSize, Z_esp);
1337         break;
1338       case Interpreter::java_lang_math_abs:
1339         // Load operand from stack.
1340         __ mem2freg_opt(Z_FRET, Address(Z_esp, Interpreter::stackElementSize));
1341         __ z_lpdbr(Z_FRET);
1342         break;
1343       case Interpreter::java_lang_math_fmaF:
1344         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1345         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize)); // arg1
1346         __ z_maeb(Z_FRET, Z_FARG2, Address(Z_esp, 2 * Interpreter::stackElementSize));
1347         break;
1348       case Interpreter::java_lang_math_fmaD:
1349         __ mem2freg_opt(Z_FRET,  Address(Z_esp,     Interpreter::stackElementSize)); // result reg = arg3
1350         __ mem2freg_opt(Z_FARG2, Address(Z_esp, 5 * Interpreter::stackElementSize)); // arg1
1351         __ z_madb(Z_FRET, Z_FARG2, Address(Z_esp, 3 * Interpreter::stackElementSize));
1352         break;
1353       default: ShouldNotReachHere();
1354     }
1355   } else {
1356     // Load arguments
1357     assert(num_args <= 4, "passed in registers");
1358     if (double_precision) {
1359       int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1360       for (int i = 0; i < num_args; ++i) {
1361         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1362         offset -= 2 * Interpreter::stackElementSize;
1363       }
1364     } else {
1365       int offset = num_args * Interpreter::stackElementSize;
1366       for (int i = 0; i < num_args; ++i) {
1367         __ mem2freg_opt(as_FloatRegister(Z_FARG1->encoding() + 2 * i), Address(Z_esp, offset));
1368         offset -= Interpreter::stackElementSize;
1369       }
1370     }
1371     // Call runtime
1372     __ save_return_pc();       // Save Z_R14.
1373     __ push_frame_abi160(0);   // Without new frame the RT call could overwrite the saved Z_R14.
1374 
1375     __ call_VM_leaf(runtime_entry);
1376 
1377     __ pop_frame();
1378     __ restore_return_pc();    // Restore Z_R14.
1379   }
1380 
1381   // Pop c2i arguments (if any) off when we return.
1382   __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1383 
1384   __ z_br(Z_R14);
1385 
1386   return entry;
1387 }
1388 
1389 // Interpreter stub for calling a native method. (asm interpreter).
1390 // This sets up a somewhat different looking stack for calling the
1391 // native method than the typical interpreter frame setup.
1392 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1393   // Determine code generation flags.
1394   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1395 
1396   // Interpreter entry for ordinary Java methods.
1397   //
1398   // Registers alive
1399   //   Z_SP          - stack pointer
1400   //   Z_thread      - JavaThread*
1401   //   Z_method      - callee's method (method to be invoked)
1402   //   Z_esp         - operand (or expression) stack pointer of caller. one slot above last arg.
1403   //   Z_R10         - sender sp (before modifications, e.g. by c2i adapter
1404   //                   and as well by generate_fixed_frame below)
1405   //   Z_R14         - return address to caller (call_stub or c2i_adapter)
1406   //
1407   // Registers updated
1408   //   Z_SP          - stack pointer
1409   //   Z_fp          - callee's framepointer
1410   //   Z_esp         - callee's operand stack pointer
1411   //                   points to the slot above the value on top
1412   //   Z_locals      - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1413   //   Z_tos         - integer result, if any
1414   //   z_ftos        - floating point result, if any
1415   //
1416   // Stack layout at this point:
1417   //
1418   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1419   //                                                          frame was extended by c2i adapter)
1420   //           [outgoing Java arguments]     <-- Z_esp
1421   //           ...
1422   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1423   //           ...
1424   //
1425 
1426   address entry_point = __ pc();
1427 
1428   // Make sure registers are different!
1429   assert_different_registers(Z_thread, Z_method, Z_esp);
1430 
1431   BLOCK_COMMENT("native_entry {");
1432 
1433   // Make sure method is native and not abstract.
1434 #ifdef ASSERT
1435   address reentry = NULL;
1436   { Label L;
1437     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1438     __ z_btrue(L);
1439     reentry = __ stop_chain_static(reentry, "tried to execute non-native method as native");
1440     __ bind(L);
1441   }
1442   { Label L;
1443     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1444     __ z_bfalse(L);
1445     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1446     __ bind(L);
1447   }
1448 #endif // ASSERT
1449 
1450 #ifdef ASSERT
1451   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1452   __ save_return_pc(Z_R14);
1453 #endif
1454 
1455   // Generate the code to allocate the interpreter stack frame.
1456   generate_fixed_frame(true);
1457 
1458   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1459   // Since at this point in the method invocation the exception handler
1460   // would try to exit the monitor of synchronized methods which hasn't
1461   // been entered yet, we set the thread local variable
1462   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1463   // runtime, exception handling i.e. unlock_if_synchronized_method will
1464   // check this thread local flag.
1465   __ z_mvi(do_not_unlock_if_synchronized, true);
1466 
1467   // Increment invocation count and check for overflow.
1468   NearLabel invocation_counter_overflow;
1469   if (inc_counter) {
1470     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1471   }
1472 
1473   Label continue_after_compile;
1474   __ bind(continue_after_compile);
1475 
1476   bang_stack_shadow_pages(true);
1477 
1478   // Reset the _do_not_unlock_if_synchronized flag.
1479   __ z_mvi(do_not_unlock_if_synchronized, false);
1480 
1481   // Check for synchronized methods.
1482   // This mst happen AFTER invocation_counter check and stack overflow check,
1483   // so method is not locked if overflows.
1484   if (synchronized) {
1485     lock_method();
1486   } else {
1487     // No synchronization necessary.
1488 #ifdef ASSERT
1489     { Label L;
1490       __ get_method(Z_R1_scratch);
1491       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1492       __ z_bfalse(L);
1493       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1494       __ bind(L);
1495     }
1496 #endif // ASSERT
1497   }
1498 
1499   // start execution
1500 
1501   // jvmti support
1502   __ notify_method_entry();
1503 
1504   //=============================================================================
1505   // Get and call the signature handler.
1506   const Register Rmethod                 = Z_tmp_2;
1507   const Register signature_handler_entry = Z_tmp_1;
1508   const Register Rresult_handler         = Z_tmp_3;
1509   Label call_signature_handler;
1510 
1511   assert_different_registers(Z_fp, Rmethod, signature_handler_entry, Rresult_handler);
1512   assert(Rresult_handler->is_nonvolatile(), "Rresult_handler must be in a non-volatile register");
1513 
1514   // Reload method.
1515   __ get_method(Rmethod);
1516 
1517   // Check for signature handler.
1518   __ load_and_test_long(signature_handler_entry, method2_(Rmethod, signature_handler));
1519   __ z_brne(call_signature_handler);
1520 
1521   // Method has never been called. Either generate a specialized
1522   // handler or point to the slow one.
1523   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call),
1524              Rmethod);
1525 
1526   // Reload method.
1527   __ get_method(Rmethod);
1528 
1529   // Reload signature handler, it must have been created/assigned in the meantime.
1530   __ z_lg(signature_handler_entry, method2_(Rmethod, signature_handler));
1531 
1532   __ bind(call_signature_handler);
1533 
1534   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1535   __ set_top_ijava_frame_at_SP_as_last_Java_frame(Z_SP, Z_R1/*tmp*/);
1536 
1537   // Call signature handler and pass locals address in Z_ARG1.
1538   __ z_lgr(Z_ARG1, Z_locals);
1539   __ call_stub(signature_handler_entry);
1540   // Save result handler returned by signature handler.
1541   __ z_lgr(Rresult_handler, Z_RET);
1542 
1543   // Reload method (the slow signature handler may block for GC).
1544   __ get_method(Rmethod);
1545 
1546   // Pass mirror handle if static call.
1547   {
1548     Label method_is_not_static;
1549     __ testbit(method2_(Rmethod, access_flags), JVM_ACC_STATIC_BIT);
1550     __ z_bfalse(method_is_not_static);
1551     // Get mirror.
1552     __ load_mirror(Z_R1, Rmethod);
1553     // z_ijava_state.oop_temp = pool_holder->klass_part()->java_mirror();
1554     __ z_stg(Z_R1, oop_tmp_offset, Z_fp);
1555     // Pass handle to mirror as 2nd argument to JNI method.
1556     __ add2reg(Z_ARG2, oop_tmp_offset, Z_fp);
1557     __ bind(method_is_not_static);
1558   }
1559 
1560   // Pass JNIEnv address as first parameter.
1561   __ add2reg(Z_ARG1, in_bytes(JavaThread::jni_environment_offset()), Z_thread);
1562 
1563   // Note: last java frame has been set above already. The pc from there
1564   // is precise enough.
1565 
1566   // Get native function entry point before we change the thread state.
1567   __ z_lg(Z_R1/*native_method_entry*/, method2_(Rmethod, native_function));
1568 
1569   //=============================================================================
1570   // Transition from _thread_in_Java to _thread_in_native. As soon as
1571   // we make this change the safepoint code needs to be certain that
1572   // the last Java frame we established is good. The pc in that frame
1573   // just need to be near here not an actual return address.
1574 #ifdef ASSERT
1575   {
1576     NearLabel L;
1577     __ mem2reg_opt(Z_R14, Address(Z_thread, JavaThread::thread_state_offset()), false /*32 bits*/);
1578     __ compareU32_and_branch(Z_R14, _thread_in_Java, Assembler::bcondEqual, L);
1579     reentry = __ stop_chain_static(reentry, "Wrong thread state in native stub");
1580     __ bind(L);
1581   }
1582 #endif
1583 
1584   // Memory ordering: Z does not reorder store/load with subsequent load. That's strong enough.
1585   __ set_thread_state(_thread_in_native);
1586 
1587   //=============================================================================
1588   // Call the native method. Argument registers must not have been
1589   // overwritten since "__ call_stub(signature_handler);" (except for
1590   // ARG1 and ARG2 for static methods).
1591 
1592   __ call_c(Z_R1/*native_method_entry*/);
1593 
1594   // NOTE: frame::interpreter_frame_result() depends on these stores.
1595   __ z_stg(Z_RET, _z_ijava_state_neg(lresult), Z_fp);
1596   __ freg2mem_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1597   const Register Rlresult = signature_handler_entry;
1598   assert(Rlresult->is_nonvolatile(), "Rlresult must be in a non-volatile register");
1599   __ z_lgr(Rlresult, Z_RET);
1600 
1601   // Z_method may no longer be valid, because of GC.
1602 
1603   // Block, if necessary, before resuming in _thread_in_Java state.
1604   // In order for GC to work, don't clear the last_Java_sp until after
1605   // blocking.
1606 
1607   //=============================================================================
1608   // Switch thread to "native transition" state before reading the
1609   // synchronization state. This additional state is necessary
1610   // because reading and testing the synchronization state is not
1611   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1612   // in _thread_in_native state, loads _not_synchronized and is
1613   // preempted. VM thread changes sync state to synchronizing and
1614   // suspends threads for GC. Thread A is resumed to finish this
1615   // native method, but doesn't block here since it didn't see any
1616   // synchronization is progress, and escapes.
1617 
1618   __ set_thread_state(_thread_in_native_trans);
1619   if (UseMembar) {
1620     __ z_fence();
1621   } else {
1622     // Write serialization page so VM thread can do a pseudo remote
1623     // membar. We use the current thread pointer to calculate a thread
1624     // specific offset to write to within the page. This minimizes bus
1625     // traffic due to cache line collision.
1626     __ serialize_memory(Z_thread, Z_R1, Z_R0);
1627   }
1628   // Now before we return to java we must look for a current safepoint
1629   // (a new safepoint can not start since we entered native_trans).
1630   // We must check here because a current safepoint could be modifying
1631   // the callers registers right this moment.
1632 
1633   // Check for safepoint operation in progress and/or pending suspend requests.
1634   {
1635     Label Continue, do_safepoint;
1636     __ safepoint_poll(do_safepoint, Z_R1);
1637     // Check for suspend.
1638     __ load_and_test_int(Z_R0/*suspend_flags*/, thread_(suspend_flags));
1639     __ z_bre(Continue); // 0 -> no flag set -> not suspended
1640     __ bind(do_safepoint);
1641     __ z_lgr(Z_ARG1, Z_thread);
1642     __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1643     __ bind(Continue);
1644   }
1645 
1646   //=============================================================================
1647   // Back in Interpreter Frame.
1648 
1649   // We are in thread_in_native_trans here and back in the normal
1650   // interpreter frame. We don't have to do anything special about
1651   // safepoints and we can switch to Java mode anytime we are ready.
1652 
1653   // Note: frame::interpreter_frame_result has a dependency on how the
1654   // method result is saved across the call to post_method_exit. For
1655   // native methods it assumes that the non-FPU/non-void result is
1656   // saved in z_ijava_state.lresult and a FPU result in z_ijava_state.fresult. If
1657   // this changes then the interpreter_frame_result implementation
1658   // will need to be updated too.
1659 
1660   //=============================================================================
1661   // Back in Java.
1662 
1663   // Memory ordering: Z does not reorder store/load with subsequent
1664   // load. That's strong enough.
1665   __ set_thread_state(_thread_in_Java);
1666 
1667   __ reset_last_Java_frame();
1668 
1669   // We reset the JNI handle block only after unboxing the result; see below.
1670 
1671   // The method register is junk from after the thread_in_native transition
1672   // until here. Also can't call_VM until the bcp has been
1673   // restored. Need bcp for throwing exception below so get it now.
1674   __ get_method(Rmethod);
1675 
1676   // Restore Z_bcp to have legal interpreter frame,
1677   // i.e., bci == 0 <=> Z_bcp == code_base().
1678   __ z_lg(Z_bcp, Address(Rmethod, Method::const_offset())); // get constMethod
1679   __ add2reg(Z_bcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1680 
1681   if (CheckJNICalls) {
1682     // clear_pending_jni_exception_check
1683     __ clear_mem(Address(Z_thread, JavaThread::pending_jni_exception_check_fn_offset()), sizeof(oop));
1684   }
1685 
1686   // Check if the native method returns an oop, and if so, move it
1687   // from the jni handle to z_ijava_state.oop_temp. This is
1688   // necessary, because we reset the jni handle block below.
1689   // NOTE: frame::interpreter_frame_result() depends on this, too.
1690   { NearLabel no_oop_result;
1691   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1692   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_result);
1693   __ resolve_jobject(Rlresult, /* tmp1 */ Rmethod, /* tmp2 */ Z_R1);
1694   __ z_stg(Rlresult, oop_tmp_offset, Z_fp);
1695   __ bind(no_oop_result);
1696   }
1697 
1698   // Reset handle block.
1699   __ z_lg(Z_R1/*active_handles*/, thread_(active_handles));
1700   __ clear_mem(Address(Z_R1, JNIHandleBlock::top_offset_in_bytes()), 4);
1701 
1702   // Bandle exceptions (exception handling will handle unlocking!).
1703   {
1704     Label L;
1705     __ load_and_test_long(Z_R0/*pending_exception*/, thread_(pending_exception));
1706     __ z_bre(L);
1707     __ MacroAssembler::call_VM(noreg,
1708                                CAST_FROM_FN_PTR(address,
1709                                InterpreterRuntime::throw_pending_exception));
1710     __ should_not_reach_here();
1711     __ bind(L);
1712   }
1713 
1714   if (synchronized) {
1715     Register Rfirst_monitor = Z_ARG2;
1716     __ add2reg(Rfirst_monitor, -(frame::z_ijava_state_size + (int)sizeof(BasicObjectLock)), Z_fp);
1717 #ifdef ASSERT
1718     NearLabel ok;
1719     __ z_lg(Z_R1, _z_ijava_state_neg(monitors), Z_fp);
1720     __ compareU64_and_branch(Rfirst_monitor, Z_R1, Assembler::bcondEqual, ok);
1721     reentry = __ stop_chain_static(reentry, "native_entry:unlock: inconsistent z_ijava_state.monitors");
1722     __ bind(ok);
1723 #endif
1724     __ unlock_object(Rfirst_monitor);
1725   }
1726 
1727   // JVMTI support. Result has already been saved above to the frame.
1728   __ notify_method_exit(true/*native_method*/, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1729 
1730   // Move native method result back into proper registers and return.
1731   // C++ interpreter does not use result handler. So do we need to here? TODO(ZASM): check if correct.
1732   { NearLabel no_oop_or_null;
1733   __ mem2freg_opt(Z_FRET, Address(Z_fp, _z_ijava_state_neg(fresult)));
1734   __ load_and_test_long(Z_RET, Address(Z_fp, _z_ijava_state_neg(lresult)));
1735   __ z_bre(no_oop_or_null); // No unboxing if the result is NULL.
1736   __ load_absolute_address(Z_R1, AbstractInterpreter::result_handler(T_OBJECT));
1737   __ compareU64_and_branch(Z_R1, Rresult_handler, Assembler::bcondNotEqual, no_oop_or_null);
1738   __ z_lg(Z_RET, oop_tmp_offset, Z_fp);
1739   __ verify_oop(Z_RET);
1740   __ bind(no_oop_or_null);
1741   }
1742 
1743   // Pop the native method's interpreter frame.
1744   __ pop_interpreter_frame(Z_R14 /*return_pc*/, Z_ARG2/*tmp1*/, Z_ARG3/*tmp2*/);
1745 
1746   // Return to caller.
1747   __ z_br(Z_R14);
1748 
1749   if (inc_counter) {
1750     // Handle overflow of counter and compile method.
1751     __ bind(invocation_counter_overflow);
1752     generate_counter_overflow(continue_after_compile);
1753   }
1754 
1755   BLOCK_COMMENT("} native_entry");
1756 
1757   return entry_point;
1758 }
1759 
1760 //
1761 // Generic interpreted method entry to template interpreter.
1762 //
1763 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1764   address entry_point = __ pc();
1765 
1766   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1767 
1768   // Interpreter entry for ordinary Java methods.
1769   //
1770   // Registers alive
1771   //   Z_SP       - stack pointer
1772   //   Z_thread   - JavaThread*
1773   //   Z_method   - callee's method (method to be invoked)
1774   //   Z_esp      - operand (or expression) stack pointer of caller. one slot above last arg.
1775   //   Z_R10      - sender sp (before modifications, e.g. by c2i adapter
1776   //                           and as well by generate_fixed_frame below)
1777   //   Z_R14      - return address to caller (call_stub or c2i_adapter)
1778   //
1779   // Registers updated
1780   //   Z_SP       - stack pointer
1781   //   Z_fp       - callee's framepointer
1782   //   Z_esp      - callee's operand stack pointer
1783   //                points to the slot above the value on top
1784   //   Z_locals   - used to access locals: locals[i] := *(Z_locals - i*BytesPerWord)
1785   //   Z_tos      - integer result, if any
1786   //   z_ftos     - floating point result, if any
1787   //
1788   //
1789   // stack layout at this point:
1790   //
1791   //   F1      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP, Z_R10 (Z_R10 will be below Z_SP if
1792   //                                                          frame was extended by c2i adapter)
1793   //           [outgoing Java arguments]     <-- Z_esp
1794   //           ...
1795   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
1796   //           ...
1797   //
1798   // stack layout before dispatching the first bytecode:
1799   //
1800   //   F0      [TOP_IJAVA_FRAME_ABI]         <-- Z_SP
1801   //           [operand stack]               <-- Z_esp
1802   //           monitor (optional, can grow)
1803   //           [IJAVA_STATE]
1804   //   F1      [PARENT_IJAVA_FRAME_ABI]      <-- Z_fp (== *Z_SP)
1805   //           [F0's locals]                 <-- Z_locals
1806   //           [F1's operand stack]
1807   //           [F1's monitors] (optional)
1808   //           [IJAVA_STATE]
1809 
1810   // Make sure registers are different!
1811   assert_different_registers(Z_thread, Z_method, Z_esp);
1812 
1813   BLOCK_COMMENT("normal_entry {");
1814 
1815   // Make sure method is not native and not abstract.
1816   // Rethink these assertions - they can be simplified and shared.
1817 #ifdef ASSERT
1818   address reentry = NULL;
1819   { Label L;
1820     __ testbit(method_(access_flags), JVM_ACC_NATIVE_BIT);
1821     __ z_bfalse(L);
1822     reentry = __ stop_chain_static(reentry, "tried to execute native method as non-native");
1823     __ bind(L);
1824   }
1825   { Label L;
1826     __ testbit(method_(access_flags), JVM_ACC_ABSTRACT_BIT);
1827     __ z_bfalse(L);
1828     reentry = __ stop_chain_static(reentry, "tried to execute abstract method as non-abstract");
1829     __ bind(L);
1830   }
1831 #endif // ASSERT
1832 
1833 #ifdef ASSERT
1834   // Save the return PC into the callers frame for assertion in generate_fixed_frame.
1835   __ save_return_pc(Z_R14);
1836 #endif
1837 
1838   // Generate the code to allocate the interpreter stack frame.
1839   generate_fixed_frame(false);
1840 
1841   const Address do_not_unlock_if_synchronized(Z_thread, JavaThread::do_not_unlock_if_synchronized_offset());
1842   // Since at this point in the method invocation the exception handler
1843   // would try to exit the monitor of synchronized methods which hasn't
1844   // been entered yet, we set the thread local variable
1845   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1846   // runtime, exception handling i.e. unlock_if_synchronized_method will
1847   // check this thread local flag.
1848   __ z_mvi(do_not_unlock_if_synchronized, true);
1849 
1850   __ profile_parameters_type(Z_tmp_2, Z_ARG3, Z_ARG4);
1851 
1852   // Increment invocation counter and check for overflow.
1853   //
1854   // Note: checking for negative value instead of overflow so we have a 'sticky'
1855   // overflow test (may be of importance as soon as we have true MT/MP).
1856   NearLabel invocation_counter_overflow;
1857   NearLabel profile_method;
1858   NearLabel profile_method_continue;
1859   NearLabel Lcontinue;
1860   if (inc_counter) {
1861     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1862     if (ProfileInterpreter) {
1863       __ bind(profile_method_continue);
1864     }
1865   }
1866   __ bind(Lcontinue);
1867 
1868   bang_stack_shadow_pages(false);
1869 
1870   // Reset the _do_not_unlock_if_synchronized flag.
1871   __ z_mvi(do_not_unlock_if_synchronized, false);
1872 
1873   // Check for synchronized methods.
1874   // Must happen AFTER invocation_counter check and stack overflow check,
1875   // so method is not locked if overflows.
1876   if (synchronized) {
1877     // Allocate monitor and lock method.
1878     lock_method();
1879   } else {
1880 #ifdef ASSERT
1881     { Label L;
1882       __ get_method(Z_R1_scratch);
1883       __ testbit(method2_(Z_R1_scratch, access_flags), JVM_ACC_SYNCHRONIZED_BIT);
1884       __ z_bfalse(L);
1885       reentry = __ stop_chain_static(reentry, "method needs synchronization");
1886       __ bind(L);
1887     }
1888 #endif // ASSERT
1889   }
1890 
1891   // start execution
1892 
1893 #ifdef ASSERT
1894   __ verify_esp(Z_esp, Z_R1_scratch);
1895 
1896   __ verify_thread();
1897 #endif
1898 
1899   // jvmti support
1900   __ notify_method_entry();
1901 
1902   // Start executing instructions.
1903   __ dispatch_next(vtos);
1904   // Dispatch_next does not return.
1905   DEBUG_ONLY(__ should_not_reach_here());
1906 
1907   // Invocation counter overflow.
1908   if (inc_counter) {
1909     if (ProfileInterpreter) {
1910       // We have decided to profile this method in the interpreter.
1911       __ bind(profile_method);
1912 
1913       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1914       __ set_method_data_pointer_for_bcp();
1915       __ z_bru(profile_method_continue);
1916     }
1917 
1918     // Handle invocation counter overflow.
1919     __ bind(invocation_counter_overflow);
1920     generate_counter_overflow(Lcontinue);
1921   }
1922 
1923   BLOCK_COMMENT("} normal_entry");
1924 
1925   return entry_point;
1926 }
1927 
1928 
1929 /**
1930  * Method entry for static native methods:
1931  *   int java.util.zip.CRC32.update(int crc, int b)
1932  */
1933 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1934 
1935   if (UseCRC32Intrinsics) {
1936     uint64_t entry_off = __ offset();
1937     Label    slow_path;
1938 
1939     // If we need a safepoint check, generate full interpreter entry.
1940     __ safepoint_poll(slow_path, Z_R1);
1941 
1942     BLOCK_COMMENT("CRC32_update {");
1943 
1944     // We don't generate local frame and don't align stack because
1945     // we not even call stub code (we generate the code inline)
1946     // and there is no safepoint on this path.
1947 
1948     // Load java parameters.
1949     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1950     const Register argP    = Z_esp;
1951     const Register crc     = Z_ARG1;  // crc value
1952     const Register data    = Z_ARG2;  // address of java byte value (kernel_crc32 needs address)
1953     const Register dataLen = Z_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1954     const Register table   = Z_ARG4;  // address of crc32 table
1955 
1956     // Arguments are reversed on java expression stack.
1957     __ z_la(data, 3+1*wordSize, argP);  // byte value (stack address).
1958                                         // Being passed as an int, the single byte is at offset +3.
1959     __ z_llgf(crc, 2 * wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register.
1960 
1961     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
1962     __ kernel_crc32_singleByte(crc, data, dataLen, table, Z_R1, true);
1963 
1964     // Restore caller sp for c2i case.
1965     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
1966 
1967     __ z_br(Z_R14);
1968 
1969     BLOCK_COMMENT("} CRC32_update");
1970 
1971     // Use a previously generated vanilla native entry as the slow path.
1972     BIND(slow_path);
1973     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
1974     return __ addr_at(entry_off);
1975   }
1976 
1977   return NULL;
1978 }
1979 
1980 
1981 /**
1982  * Method entry for static native methods:
1983  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1984  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1985  */
1986 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1987 
1988   if (UseCRC32Intrinsics) {
1989     uint64_t entry_off = __ offset();
1990     Label    slow_path;
1991 
1992     // If we need a safepoint check, generate full interpreter entry.
1993     __ safepoint_poll(slow_path, Z_R1);
1994 
1995     // We don't generate local frame and don't align stack because
1996     // we call stub code and there is no safepoint on this path.
1997 
1998     // Load parameters.
1999     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
2000     const Register argP    = Z_esp;
2001     const Register crc     = Z_ARG1;  // crc value
2002     const Register data    = Z_ARG2;  // address of java byte array
2003     const Register dataLen = Z_ARG3;  // source data len
2004     const Register table   = Z_ARG4;  // address of crc32 table
2005     const Register t0      = Z_R10;   // work reg for kernel* emitters
2006     const Register t1      = Z_R11;   // work reg for kernel* emitters
2007     const Register t2      = Z_R12;   // work reg for kernel* emitters
2008     const Register t3      = Z_R13;   // work reg for kernel* emitters
2009 
2010     // Arguments are reversed on java expression stack.
2011     // Calculate address of start element.
2012     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
2013       // crc     @ (SP + 5W) (32bit)
2014       // buf     @ (SP + 3W) (64bit ptr to long array)
2015       // off     @ (SP + 2W) (32bit)
2016       // dataLen @ (SP + 1W) (32bit)
2017       // data = buf + off
2018       BLOCK_COMMENT("CRC32_updateByteBuffer {");
2019       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
2020       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2021       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2022       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2023     } else {                                                         // Used for "updateBytes update".
2024       // crc     @ (SP + 4W) (32bit)
2025       // buf     @ (SP + 3W) (64bit ptr to byte array)
2026       // off     @ (SP + 2W) (32bit)
2027       // dataLen @ (SP + 1W) (32bit)
2028       // data = buf + off + base_offset
2029       BLOCK_COMMENT("CRC32_updateBytes {");
2030       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
2031       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2032       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2033       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process
2034       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2035     }
2036 
2037     StubRoutines::zarch::generate_load_crc_table_addr(_masm, table);
2038 
2039     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2040     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
2041     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, true);
2042     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
2043 
2044     // Restore caller sp for c2i case.
2045     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2046 
2047     __ z_br(Z_R14);
2048 
2049     BLOCK_COMMENT("} CRC32_update{Bytes|ByteBuffer}");
2050 
2051     // Use a previously generated vanilla native entry as the slow path.
2052     BIND(slow_path);
2053     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), Z_R1);
2054     return __ addr_at(entry_off);
2055   }
2056 
2057   return NULL;
2058 }
2059 
2060 
2061 /**
2062  * Method entry for intrinsic-candidate (non-native) methods:
2063  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
2064  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
2065  * Unlike CRC32, CRC32C does not have any methods marked as native
2066  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
2067  */
2068 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
2069 
2070   if (UseCRC32CIntrinsics) {
2071     uint64_t entry_off = __ offset();
2072 
2073     // We don't generate local frame and don't align stack because
2074     // we call stub code and there is no safepoint on this path.
2075 
2076     // Load parameters.
2077     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
2078     const Register argP    = Z_esp;
2079     const Register crc     = Z_ARG1;  // crc value
2080     const Register data    = Z_ARG2;  // address of java byte array
2081     const Register dataLen = Z_ARG3;  // source data len
2082     const Register table   = Z_ARG4;  // address of crc32 table
2083     const Register t0      = Z_R10;   // work reg for kernel* emitters
2084     const Register t1      = Z_R11;   // work reg for kernel* emitters
2085     const Register t2      = Z_R12;   // work reg for kernel* emitters
2086     const Register t3      = Z_R13;   // work reg for kernel* emitters
2087 
2088     // Arguments are reversed on java expression stack.
2089     // Calculate address of start element.
2090     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateByteBuffer direct".
2091       // crc     @ (SP + 5W) (32bit)
2092       // buf     @ (SP + 3W) (64bit ptr to long array)
2093       // off     @ (SP + 2W) (32bit)
2094       // dataLen @ (SP + 1W) (32bit)
2095       // data = buf + off
2096       BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
2097       __ z_llgf(crc,    5*wordSize, argP);  // current crc state
2098       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2099       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2100       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
2101       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
2102     } else {                                                                // Used for "updateBytes update".
2103       // crc     @ (SP + 4W) (32bit)
2104       // buf     @ (SP + 3W) (64bit ptr to byte array)
2105       // off     @ (SP + 2W) (32bit)
2106       // dataLen @ (SP + 1W) (32bit)
2107       // data = buf + off + base_offset
2108       BLOCK_COMMENT("CRC32C_updateBytes {");
2109       __ z_llgf(crc,    4*wordSize, argP);  // current crc state
2110       __ z_lg(data,     3*wordSize, argP);  // start of byte buffer
2111       __ z_agf(data,    2*wordSize, argP);  // Add byte buffer offset.
2112       __ z_lgf(dataLen, 1*wordSize, argP);  // #bytes to process, calculated as
2113       __ z_sgf(dataLen, Address(argP, 2*wordSize));  // (end_index - offset)
2114       __ z_aghi(data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2115     }
2116 
2117     StubRoutines::zarch::generate_load_crc32c_table_addr(_masm, table);
2118 
2119     __ resize_frame(-(6*8), Z_R0, true); // Resize frame to provide add'l space to spill 5 registers.
2120     __ z_stmg(t0, t3, 1*8, Z_SP);        // Spill regs 10..13 to make them available as work registers.
2121     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, false);
2122     __ z_lmg(t0, t3, 1*8, Z_SP);         // Spill regs 10..13 back from stack.
2123 
2124     // Restore caller sp for c2i case.
2125     __ resize_frame_absolute(Z_R10, Z_R0, true); // Cut the stack back to where the caller started.
2126 
2127     __ z_br(Z_R14);
2128 
2129     BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
2130     return __ addr_at(entry_off);
2131   }
2132 
2133   return NULL;
2134 }
2135 
2136 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
2137   // Quick & dirty stack overflow checking: bang the stack & handle trap.
2138   // Note that we do the banging after the frame is setup, since the exception
2139   // handling code expects to find a valid interpreter frame on the stack.
2140   // Doing the banging earlier fails if the caller frame is not an interpreter
2141   // frame.
2142   // (Also, the exception throwing code expects to unlock any synchronized
2143   // method receiver, so do the banging after locking the receiver.)
2144 
2145   // Bang each page in the shadow zone. We can't assume it's been done for
2146   // an interpreter frame with greater than a page of locals, so each page
2147   // needs to be checked. Only true for non-native. For native, we only bang the last page.
2148   if (UseStackBanging) {
2149     const int page_size      = os::vm_page_size();
2150     const int n_shadow_pages = (int)(JavaThread::stack_shadow_zone_size()/page_size);
2151     const int start_page_num = native_call ? n_shadow_pages : 1;
2152     for (int pages = start_page_num; pages <= n_shadow_pages; pages++) {
2153       __ bang_stack_with_offset(pages*page_size);
2154     }
2155   }
2156 }
2157 
2158 //-----------------------------------------------------------------------------
2159 // Exceptions
2160 
2161 void TemplateInterpreterGenerator::generate_throw_exception() {
2162 
2163   BLOCK_COMMENT("throw_exception {");
2164 
2165   // Entry point in previous activation (i.e., if the caller was interpreted).
2166   Interpreter::_rethrow_exception_entry = __ pc();
2167   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Frame accessors use Z_fp.
2168   // Z_ARG1 (==Z_tos): exception
2169   // Z_ARG2          : Return address/pc that threw exception.
2170   __ restore_bcp();    // R13 points to call/send.
2171   __ restore_locals();
2172 
2173   // Fallthrough, no need to restore Z_esp.
2174 
2175   // Entry point for exceptions thrown within interpreter code.
2176   Interpreter::_throw_exception_entry = __ pc();
2177   // Expression stack is undefined here.
2178   // Z_ARG1 (==Z_tos): exception
2179   // Z_bcp: exception bcp
2180   __ verify_oop(Z_ARG1);
2181   __ z_lgr(Z_ARG2, Z_ARG1);
2182 
2183   // Expression stack must be empty before entering the VM in case of
2184   // an exception.
2185   __ empty_expression_stack();
2186   // Find exception handler address and preserve exception oop.
2187   const Register Rpreserved_exc_oop = Z_tmp_1;
2188   __ call_VM(Rpreserved_exc_oop,
2189              CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception),
2190              Z_ARG2);
2191   // Z_RET: exception handler entry point
2192   // Z_bcp: bcp for exception handler
2193   __ push_ptr(Rpreserved_exc_oop); // Push exception which is now the only value on the stack.
2194   __ z_br(Z_RET); // Jump to exception handler (may be _remove_activation_entry!).
2195 
2196   // If the exception is not handled in the current frame the frame is
2197   // removed and the exception is rethrown (i.e. exception
2198   // continuation is _rethrow_exception).
2199   //
2200   // Note: At this point the bci is still the bci for the instruction
2201   // which caused the exception and the expression stack is
2202   // empty. Thus, for any VM calls at this point, GC will find a legal
2203   // oop map (with empty expression stack).
2204 
2205   //
2206   // JVMTI PopFrame support
2207   //
2208 
2209   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2210   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2211   __ empty_expression_stack();
2212   // Set the popframe_processing bit in pending_popframe_condition
2213   // indicating that we are currently handling popframe, so that
2214   // call_VMs that may happen later do not trigger new popframe
2215   // handling cycles.
2216   __ load_sized_value(Z_tmp_1, Address(Z_thread, JavaThread::popframe_condition_offset()), 4, false /*signed*/);
2217   __ z_oill(Z_tmp_1, JavaThread::popframe_processing_bit);
2218   __ z_sty(Z_tmp_1, thread_(popframe_condition));
2219 
2220   {
2221     // Check to see whether we are returning to a deoptimized frame.
2222     // (The PopFrame call ensures that the caller of the popped frame is
2223     // either interpreted or compiled and deoptimizes it if compiled.)
2224     // In this case, we can't call dispatch_next() after the frame is
2225     // popped, but instead must save the incoming arguments and restore
2226     // them after deoptimization has occurred.
2227     //
2228     // Note that we don't compare the return PC against the
2229     // deoptimization blob's unpack entry because of the presence of
2230     // adapter frames in C2.
2231     NearLabel caller_not_deoptimized;
2232     __ z_lg(Z_ARG1, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2233     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), Z_ARG1);
2234     __ compareU64_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, caller_not_deoptimized);
2235 
2236     // Compute size of arguments for saving when returning to
2237     // deoptimized caller.
2238     __ get_method(Z_ARG2);
2239     __ z_lg(Z_ARG2, Address(Z_ARG2, Method::const_offset()));
2240     __ z_llgh(Z_ARG2, Address(Z_ARG2, ConstMethod::size_of_parameters_offset()));
2241     __ z_sllg(Z_ARG2, Z_ARG2, Interpreter::logStackElementSize); // slots 2 bytes
2242     __ restore_locals();
2243     // Compute address of args to be saved.
2244     __ z_lgr(Z_ARG3, Z_locals);
2245     __ z_slgr(Z_ARG3, Z_ARG2);
2246     __ add2reg(Z_ARG3, wordSize);
2247     // Save these arguments.
2248     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args),
2249                     Z_thread, Z_ARG2, Z_ARG3);
2250 
2251     __ remove_activation(vtos, Z_R14,
2252                          /* throw_monitor_exception */ false,
2253                          /* install_monitor_exception */ false,
2254                          /* notify_jvmdi */ false);
2255 
2256     // Inform deoptimization that it is responsible for restoring
2257     // these arguments.
2258     __ store_const(thread_(popframe_condition),
2259                    JavaThread::popframe_force_deopt_reexecution_bit,
2260                    Z_tmp_1, false);
2261 
2262     // Continue in deoptimization handler.
2263     __ z_br(Z_R14);
2264 
2265     __ bind(caller_not_deoptimized);
2266   }
2267 
2268   // Clear the popframe condition flag.
2269   __ clear_mem(thread_(popframe_condition), sizeof(int));
2270 
2271   __ remove_activation(vtos,
2272                        noreg,  // Retaddr is not used.
2273                        false,  // throw_monitor_exception
2274                        false,  // install_monitor_exception
2275                        false); // notify_jvmdi
2276   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2277   __ restore_bcp();
2278   __ restore_locals();
2279   __ restore_esp();
2280   // The method data pointer was incremented already during
2281   // call profiling. We have to restore the mdp for the current bcp.
2282   if (ProfileInterpreter) {
2283     __ set_method_data_pointer_for_bcp();
2284   }
2285 #if INCLUDE_JVMTI
2286   {
2287     Label L_done;
2288 
2289     __ z_cli(0, Z_bcp, Bytecodes::_invokestatic);
2290     __ z_brc(Assembler::bcondNotEqual, L_done);
2291 
2292     // The member name argument must be restored if _invokestatic is
2293     // re-executed after a PopFrame call.  Detect such a case in the
2294     // InterpreterRuntime function and return the member name
2295     // argument, or NULL.
2296     __ z_lg(Z_ARG2, Address(Z_locals));
2297     __ get_method(Z_ARG3);
2298     __ call_VM(Z_tmp_1,
2299                CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null),
2300                Z_ARG2, Z_ARG3, Z_bcp);
2301 
2302     __ z_ltgr(Z_tmp_1, Z_tmp_1);
2303     __ z_brc(Assembler::bcondEqual, L_done);
2304 
2305     __ z_stg(Z_tmp_1, Address(Z_esp, wordSize));
2306     __ bind(L_done);
2307   }
2308 #endif // INCLUDE_JVMTI
2309   __ dispatch_next(vtos);
2310   // End of PopFrame support.
2311   Interpreter::_remove_activation_entry = __ pc();
2312 
2313   // In between activations - previous activation type unknown yet
2314   // compute continuation point - the continuation point expects the
2315   // following registers set up:
2316   //
2317   // Z_ARG1 (==Z_tos): exception
2318   // Z_ARG2          : return address/pc that threw exception
2319 
2320   Register return_pc = Z_tmp_1;
2321   Register handler   = Z_tmp_2;
2322    assert(return_pc->is_nonvolatile(), "use non-volatile reg. to preserve exception pc");
2323    assert(handler->is_nonvolatile(),   "use non-volatile reg. to handler pc");
2324   __ asm_assert_ijava_state_magic(return_pc/*tmp*/); // The top frame should be an interpreter frame.
2325   __ z_lg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_fp);
2326 
2327   // Moved removing the activation after VM call, because the new top
2328   // frame does not necessarily have the z_abi_160 required for a VM
2329   // call (e.g. if it is compiled).
2330 
2331   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2332                                          SharedRuntime::exception_handler_for_return_address),
2333                         Z_thread, return_pc);
2334   __ z_lgr(handler, Z_RET); // Save exception handler.
2335 
2336   // Preserve exception over this code sequence.
2337   __ pop_ptr(Z_ARG1);
2338   __ set_vm_result(Z_ARG1);
2339   // Remove the activation (without doing throws on illegalMonitorExceptions).
2340   __ remove_activation(vtos, noreg/*ret.pc already loaded*/, false/*throw exc*/, true/*install exc*/, false/*notify jvmti*/);
2341   __ z_lg(Z_fp, _z_abi(callers_sp), Z_SP); // Restore frame pointer.
2342 
2343   __ get_vm_result(Z_ARG1);     // Restore exception.
2344   __ verify_oop(Z_ARG1);
2345   __ z_lgr(Z_ARG2, return_pc);  // Restore return address.
2346 
2347 #ifdef ASSERT
2348   // The return_pc in the new top frame is dead... at least that's my
2349   // current understanding. To assert this I overwrite it.
2350   // Note: for compiled frames the handler is the deopt blob
2351   // which writes Z_ARG2 into the return_pc slot.
2352   __ load_const_optimized(return_pc, 0xb00b1);
2353   __ z_stg(return_pc, _z_parent_ijava_frame_abi(return_pc), Z_SP);
2354 #endif
2355 
2356   // Z_ARG1 (==Z_tos): exception
2357   // Z_ARG2          : return address/pc that threw exception
2358 
2359   // Note that an "issuing PC" is actually the next PC after the call.
2360   __ z_br(handler);         // Jump to exception handler of caller.
2361 
2362   BLOCK_COMMENT("} throw_exception");
2363 }
2364 
2365 //
2366 // JVMTI ForceEarlyReturn support
2367 //
2368 address TemplateInterpreterGenerator::generate_earlyret_entry_for (TosState state) {
2369   address entry = __ pc();
2370 
2371   BLOCK_COMMENT("earlyret_entry {");
2372 
2373   __ z_lg(Z_fp, _z_parent_ijava_frame_abi(callers_sp), Z_SP);
2374   __ restore_bcp();
2375   __ restore_locals();
2376   __ restore_esp();
2377   __ empty_expression_stack();
2378   __ load_earlyret_value(state);
2379 
2380   Register RjvmtiState = Z_tmp_1;
2381   __ z_lg(RjvmtiState, thread_(jvmti_thread_state));
2382   __ store_const(Address(RjvmtiState, JvmtiThreadState::earlyret_state_offset()),
2383                  JvmtiThreadState::earlyret_inactive, 4, 4, Z_R0_scratch);
2384 
2385   if (state == itos) {
2386     // Narrow result if state is itos but result type is smaller.
2387     // Need to narrow in the return bytecode rather than in generate_return_entry
2388     // since compiled code callers expect the result to already be narrowed.
2389     __ narrow(Z_tos, Z_tmp_1); /* fall through */
2390   }
2391   __ remove_activation(state,
2392                        Z_tmp_1, // retaddr
2393                        false,   // throw_monitor_exception
2394                        false,   // install_monitor_exception
2395                        true);   // notify_jvmdi
2396   __ z_br(Z_tmp_1);
2397 
2398   BLOCK_COMMENT("} earlyret_entry");
2399 
2400   return entry;
2401 }
2402 
2403 //-----------------------------------------------------------------------------
2404 // Helper for vtos entry point generation.
2405 
2406 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2407                                                          address& bep,
2408                                                          address& cep,
2409                                                          address& sep,
2410                                                          address& aep,
2411                                                          address& iep,
2412                                                          address& lep,
2413                                                          address& fep,
2414                                                          address& dep,
2415                                                          address& vep) {
2416   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2417   Label L;
2418   aep = __ pc(); __ push_ptr(); __ z_bru(L);
2419   fep = __ pc(); __ push_f();   __ z_bru(L);
2420   dep = __ pc(); __ push_d();   __ z_bru(L);
2421   lep = __ pc(); __ push_l();   __ z_bru(L);
2422   bep = cep = sep =
2423   iep = __ pc(); __ push_i();
2424   vep = __ pc();
2425   __ bind(L);
2426   generate_and_dispatch(t);
2427 }
2428 
2429 //-----------------------------------------------------------------------------
2430 
2431 #ifndef PRODUCT
2432 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2433   address entry = __ pc();
2434   NearLabel counter_below_trace_threshold;
2435 
2436   if (TraceBytecodesAt > 0) {
2437     // Skip runtime call, if the trace threshold is not yet reached.
2438     __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2439     __ load_absolute_address(Z_tmp_2, (address)&TraceBytecodesAt);
2440     __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2441     __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2442     __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, counter_below_trace_threshold);
2443   }
2444 
2445   int offset2 = state == ltos || state == dtos ? 2 : 1;
2446 
2447   __ push(state);
2448   // Preserved return pointer is in Z_R14.
2449   // InterpreterRuntime::trace_bytecode() preserved and returns the value passed as second argument.
2450   __ z_lgr(Z_ARG2, Z_R14);
2451   __ z_lg(Z_ARG3, Address(Z_esp, Interpreter::expr_offset_in_bytes(0)));
2452   if (WizardMode) {
2453     __ z_lgr(Z_ARG4, Z_esp); // Trace Z_esp in WizardMode.
2454   } else {
2455     __ z_lg(Z_ARG4, Address(Z_esp, Interpreter::expr_offset_in_bytes(offset2)));
2456   }
2457   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), Z_ARG2, Z_ARG3, Z_ARG4);
2458   __ z_lgr(Z_R14, Z_RET); // Estore return address (see above).
2459   __ pop(state);
2460 
2461   __ bind(counter_below_trace_threshold);
2462   __ z_br(Z_R14); // return
2463 
2464   return entry;
2465 }
2466 
2467 // Make feasible for old CPUs.
2468 void TemplateInterpreterGenerator::count_bytecode() {
2469   __ load_absolute_address(Z_R1_scratch, (address) &BytecodeCounter::_counter_value);
2470   __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2471 }
2472 
2473 void TemplateInterpreterGenerator::histogram_bytecode(Template * t) {
2474   __ load_absolute_address(Z_R1_scratch, (address)&BytecodeHistogram::_counters[ t->bytecode() ]);
2475   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2476 }
2477 
2478 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template * t) {
2479   Address  index_addr(Z_tmp_1, (intptr_t) 0);
2480   Register index = Z_tmp_2;
2481 
2482   // Load previous index.
2483   __ load_absolute_address(Z_tmp_1, (address) &BytecodePairHistogram::_index);
2484   __ mem2reg_opt(index, index_addr, false);
2485 
2486   // Mask with current bytecode and store as new previous index.
2487   __ z_srl(index, BytecodePairHistogram::log2_number_of_codes);
2488   __ load_const_optimized(Z_R0_scratch,
2489                           (int)t->bytecode() << BytecodePairHistogram::log2_number_of_codes);
2490   __ z_or(index, Z_R0_scratch);
2491   __ reg2mem_opt(index, index_addr, false);
2492 
2493   // Load counter array's address.
2494   __ z_lgfr(index, index);   // Sign extend for addressing.
2495   __ z_sllg(index, index, LogBytesPerInt);  // index2bytes
2496   __ load_absolute_address(Z_R1_scratch,
2497                            (address) &BytecodePairHistogram::_counters);
2498   // Add index and increment counter.
2499   __ z_agr(Z_R1_scratch, index);
2500   __ add2mem_32(Address(Z_R1_scratch), 1, Z_tmp_1);
2501 }
2502 
2503 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2504   // Call a little run-time stub to avoid blow-up for each bytecode.
2505   // The run-time runtime saves the right registers, depending on
2506   // the tosca in-state for the given template.
2507   address entry = Interpreter::trace_code(t->tos_in());
2508   guarantee(entry != NULL, "entry must have been generated");
2509   __ call_stub(entry);
2510 }
2511 
2512 void TemplateInterpreterGenerator::stop_interpreter_at() {
2513   NearLabel L;
2514 
2515   __ load_absolute_address(Z_tmp_1, (address)&BytecodeCounter::_counter_value);
2516   __ load_absolute_address(Z_tmp_2, (address)&StopInterpreterAt);
2517   __ load_sized_value(Z_tmp_1, Address(Z_tmp_1), 4, false /*signed*/);
2518   __ load_sized_value(Z_tmp_2, Address(Z_tmp_2), 8, false /*signed*/);
2519   __ compareU64_and_branch(Z_tmp_1, Z_tmp_2, Assembler::bcondLow, L);
2520   assert(Z_tmp_1->is_nonvolatile(), "must be nonvolatile to preserve Z_tos");
2521   assert(Z_F8->is_nonvolatile(), "must be nonvolatile to preserve Z_ftos");
2522   __ z_lgr(Z_tmp_1, Z_tos);      // Save tos.
2523   __ z_lgr(Z_tmp_2, Z_bytecode); // Save Z_bytecode.
2524   __ z_ldr(Z_F8, Z_ftos);        // Save ftos.
2525   // Use -XX:StopInterpreterAt=<num> to set the limit
2526   // and break at breakpoint().
2527   __ call_VM(noreg, CAST_FROM_FN_PTR(address, breakpoint), false);
2528   __ z_lgr(Z_tos, Z_tmp_1);      // Restore tos.
2529   __ z_lgr(Z_bytecode, Z_tmp_2); // Save Z_bytecode.
2530   __ z_ldr(Z_ftos, Z_F8);        // Restore ftos.
2531   __ bind(L);
2532 }
2533 
2534 #endif // !PRODUCT