1 /*
   2  * Copyright (c) 2014, 2021, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2015, 2021 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "classfile/javaClasses.hpp"
  29 #include "gc/shared/barrierSetAssembler.hpp"
  30 #include "interpreter/bytecodeHistogram.hpp"
  31 #include "interpreter/interpreter.hpp"
  32 #include "interpreter/interpreterRuntime.hpp"
  33 #include "interpreter/interp_masm.hpp"
  34 #include "interpreter/templateInterpreterGenerator.hpp"
  35 #include "interpreter/templateTable.hpp"
  36 #include "oops/arrayOop.hpp"
  37 #include "oops/methodData.hpp"
  38 #include "oops/method.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "prims/jvmtiExport.hpp"
  41 #include "prims/jvmtiThreadState.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/deoptimization.hpp"
  44 #include "runtime/frame.inline.hpp"
  45 #include "runtime/jniHandles.hpp"
  46 #include "runtime/sharedRuntime.hpp"
  47 #include "runtime/stubRoutines.hpp"
  48 #include "runtime/synchronizer.hpp"
  49 #include "runtime/timer.hpp"
  50 #include "runtime/vframeArray.hpp"
  51 #include "runtime/vm_version.hpp"
  52 #include "utilities/debug.hpp"
  53 #include "utilities/macros.hpp"
  54 
  55 #undef __
  56 #define __ _masm->
  57 
  58 // Size of interpreter code.  Increase if too small.  Interpreter will
  59 // fail with a guarantee ("not enough space for interpreter generation");
  60 // if too small.
  61 // Run with +PrintInterpreter to get the VM to print out the size.
  62 // Max size with JVMTI
  63 int TemplateInterpreter::InterpreterCodeSize = 256*K;
  64 
  65 #ifdef PRODUCT
  66 #define BLOCK_COMMENT(str) /* nothing */
  67 #else
  68 #define BLOCK_COMMENT(str) __ block_comment(str)
  69 #endif
  70 
  71 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  72 
  73 //-----------------------------------------------------------------------------
  74 
  75 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  76   // Slow_signature handler that respects the PPC C calling conventions.
  77   //
  78   // We get called by the native entry code with our output register
  79   // area == 8. First we call InterpreterRuntime::get_result_handler
  80   // to copy the pointer to the signature string temporarily to the
  81   // first C-argument and to return the result_handler in
  82   // R3_RET. Since native_entry will copy the jni-pointer to the
  83   // first C-argument slot later on, it is OK to occupy this slot
  84   // temporarilly. Then we copy the argument list on the java
  85   // expression stack into native varargs format on the native stack
  86   // and load arguments into argument registers. Integer arguments in
  87   // the varargs vector will be sign-extended to 8 bytes.
  88   //
  89   // On entry:
  90   //   R3_ARG1        - intptr_t*     Address of java argument list in memory.
  91   //   R15_prev_state - BytecodeInterpreter* Address of interpreter state for
  92   //     this method
  93   //   R19_method
  94   //
  95   // On exit (just before return instruction):
  96   //   R3_RET            - contains the address of the result_handler.
  97   //   R4_ARG2           - is not updated for static methods and contains "this" otherwise.
  98   //   R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
  99   //                       ARGi contains this argument. Otherwise, ARGi is not updated.
 100   //   F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
 101 
 102   const int LogSizeOfTwoInstructions = 3;
 103 
 104   // FIXME: use Argument:: GL: Argument names different numbers!
 105   const int max_fp_register_arguments  = 13;
 106   const int max_int_register_arguments = 6;  // first 2 are reserved
 107 
 108   const Register arg_java       = R21_tmp1;
 109   const Register arg_c          = R22_tmp2;
 110   const Register signature      = R23_tmp3;  // is string
 111   const Register sig_byte       = R24_tmp4;
 112   const Register fpcnt          = R25_tmp5;
 113   const Register argcnt         = R26_tmp6;
 114   const Register intSlot        = R27_tmp7;
 115   const Register target_sp      = R28_tmp8;
 116   const FloatRegister floatSlot = F0;
 117 
 118   address entry = __ function_entry();
 119 
 120   __ save_LR_CR(R0);
 121   __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 122   // We use target_sp for storing arguments in the C frame.
 123   __ mr(target_sp, R1_SP);
 124   __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
 125 
 126   __ mr(arg_java, R3_ARG1);
 127 
 128   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
 129 
 130   // Signature is in R3_RET. Signature is callee saved.
 131   __ mr(signature, R3_RET);
 132 
 133   // Get the result handler.
 134   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
 135 
 136   {
 137     Label L;
 138     // test if static
 139     // _access_flags._flags must be at offset 0.
 140     // TODO PPC port: requires change in shared code.
 141     //assert(in_bytes(AccessFlags::flags_offset()) == 0,
 142     //       "MethodDesc._access_flags == MethodDesc._access_flags._flags");
 143     // _access_flags must be a 32 bit value.
 144     assert(sizeof(AccessFlags) == 4, "wrong size");
 145     __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
 146     // testbit with condition register.
 147     __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
 148     __ btrue(CCR0, L);
 149     // For non-static functions, pass "this" in R4_ARG2 and copy it
 150     // to 2nd C-arg slot.
 151     // We need to box the Java object here, so we use arg_java
 152     // (address of current Java stack slot) as argument and don't
 153     // dereference it as in case of ints, floats, etc.
 154     __ mr(R4_ARG2, arg_java);
 155     __ addi(arg_java, arg_java, -BytesPerWord);
 156     __ std(R4_ARG2, _abi0(carg_2), target_sp);
 157     __ bind(L);
 158   }
 159 
 160   // Will be incremented directly after loop_start. argcnt=0
 161   // corresponds to 3rd C argument.
 162   __ li(argcnt, -1);
 163   // arg_c points to 3rd C argument
 164   __ addi(arg_c, target_sp, _abi0(carg_3));
 165   // no floating-point args parsed so far
 166   __ li(fpcnt, 0);
 167 
 168   Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
 169   Label loop_start, loop_end;
 170   Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
 171 
 172   // signature points to '(' at entry
 173 #ifdef ASSERT
 174   __ lbz(sig_byte, 0, signature);
 175   __ cmplwi(CCR0, sig_byte, '(');
 176   __ bne(CCR0, do_dontreachhere);
 177 #endif
 178 
 179   __ bind(loop_start);
 180 
 181   __ addi(argcnt, argcnt, 1);
 182   __ lbzu(sig_byte, 1, signature);
 183 
 184   __ cmplwi(CCR0, sig_byte, ')'); // end of signature
 185   __ beq(CCR0, loop_end);
 186 
 187   __ cmplwi(CCR0, sig_byte, 'B'); // byte
 188   __ beq(CCR0, do_int);
 189 
 190   __ cmplwi(CCR0, sig_byte, 'C'); // char
 191   __ beq(CCR0, do_int);
 192 
 193   __ cmplwi(CCR0, sig_byte, 'D'); // double
 194   __ beq(CCR0, do_double);
 195 
 196   __ cmplwi(CCR0, sig_byte, 'F'); // float
 197   __ beq(CCR0, do_float);
 198 
 199   __ cmplwi(CCR0, sig_byte, 'I'); // int
 200   __ beq(CCR0, do_int);
 201 
 202   __ cmplwi(CCR0, sig_byte, 'J'); // long
 203   __ beq(CCR0, do_long);
 204 
 205   __ cmplwi(CCR0, sig_byte, 'S'); // short
 206   __ beq(CCR0, do_int);
 207 
 208   __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
 209   __ beq(CCR0, do_int);
 210 
 211   __ cmplwi(CCR0, sig_byte, 'L'); // object
 212   __ beq(CCR0, do_object);
 213 
 214   __ cmplwi(CCR0, sig_byte, '['); // array
 215   __ beq(CCR0, do_array);
 216 
 217   //  __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
 218   //  __ beq(CCR0, do_void);
 219 
 220   __ bind(do_dontreachhere);
 221 
 222   __ unimplemented("ShouldNotReachHere in slow_signature_handler");
 223 
 224   __ bind(do_array);
 225 
 226   {
 227     Label start_skip, end_skip;
 228 
 229     __ bind(start_skip);
 230     __ lbzu(sig_byte, 1, signature);
 231     __ cmplwi(CCR0, sig_byte, '[');
 232     __ beq(CCR0, start_skip); // skip further brackets
 233     __ cmplwi(CCR0, sig_byte, '9');
 234     __ bgt(CCR0, end_skip);   // no optional size
 235     __ cmplwi(CCR0, sig_byte, '0');
 236     __ bge(CCR0, start_skip); // skip optional size
 237     __ bind(end_skip);
 238 
 239     __ cmplwi(CCR0, sig_byte, 'L');
 240     __ beq(CCR0, do_object);  // for arrays of objects, the name of the object must be skipped
 241     __ b(do_boxed);          // otherwise, go directly to do_boxed
 242   }
 243 
 244   __ bind(do_object);
 245   {
 246     Label L;
 247     __ bind(L);
 248     __ lbzu(sig_byte, 1, signature);
 249     __ cmplwi(CCR0, sig_byte, ';');
 250     __ bne(CCR0, L);
 251    }
 252   // Need to box the Java object here, so we use arg_java (address of
 253   // current Java stack slot) as argument and don't dereference it as
 254   // in case of ints, floats, etc.
 255   Label do_null;
 256   __ bind(do_boxed);
 257   __ ld(R0,0, arg_java);
 258   __ cmpdi(CCR0, R0, 0);
 259   __ li(intSlot,0);
 260   __ beq(CCR0, do_null);
 261   __ mr(intSlot, arg_java);
 262   __ bind(do_null);
 263   __ std(intSlot, 0, arg_c);
 264   __ addi(arg_java, arg_java, -BytesPerWord);
 265   __ addi(arg_c, arg_c, BytesPerWord);
 266   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 267   __ blt(CCR0, move_intSlot_to_ARG);
 268   __ b(loop_start);
 269 
 270   __ bind(do_int);
 271   __ lwa(intSlot, 0, arg_java);
 272   __ std(intSlot, 0, arg_c);
 273   __ addi(arg_java, arg_java, -BytesPerWord);
 274   __ addi(arg_c, arg_c, BytesPerWord);
 275   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 276   __ blt(CCR0, move_intSlot_to_ARG);
 277   __ b(loop_start);
 278 
 279   __ bind(do_long);
 280   __ ld(intSlot, -BytesPerWord, arg_java);
 281   __ std(intSlot, 0, arg_c);
 282   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 283   __ addi(arg_c, arg_c, BytesPerWord);
 284   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 285   __ blt(CCR0, move_intSlot_to_ARG);
 286   __ b(loop_start);
 287 
 288   __ bind(do_float);
 289   __ lfs(floatSlot, 0, arg_java);
 290 #if defined(LINUX)
 291   // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
 292   // in the least significant word of an argument slot.
 293 #if defined(VM_LITTLE_ENDIAN)
 294   __ stfs(floatSlot, 0, arg_c);
 295 #else
 296   __ stfs(floatSlot, 4, arg_c);
 297 #endif
 298 #elif defined(AIX)
 299   // Although AIX runs on big endian CPU, float is in most significant
 300   // word of an argument slot.
 301   __ stfs(floatSlot, 0, arg_c);
 302 #else
 303 #error "unknown OS"
 304 #endif
 305   __ addi(arg_java, arg_java, -BytesPerWord);
 306   __ addi(arg_c, arg_c, BytesPerWord);
 307   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 308   __ blt(CCR0, move_floatSlot_to_FARG);
 309   __ b(loop_start);
 310 
 311   __ bind(do_double);
 312   __ lfd(floatSlot, - BytesPerWord, arg_java);
 313   __ stfd(floatSlot, 0, arg_c);
 314   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 315   __ addi(arg_c, arg_c, BytesPerWord);
 316   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 317   __ blt(CCR0, move_floatSlot_to_FARG);
 318   __ b(loop_start);
 319 
 320   __ bind(loop_end);
 321 
 322   __ pop_frame();
 323   __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 324   __ restore_LR_CR(R0);
 325 
 326   __ blr();
 327 
 328   Label move_int_arg, move_float_arg;
 329   __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 330   __ mr(R5_ARG3, intSlot);  __ b(loop_start);
 331   __ mr(R6_ARG4, intSlot);  __ b(loop_start);
 332   __ mr(R7_ARG5, intSlot);  __ b(loop_start);
 333   __ mr(R8_ARG6, intSlot);  __ b(loop_start);
 334   __ mr(R9_ARG7, intSlot);  __ b(loop_start);
 335   __ mr(R10_ARG8, intSlot); __ b(loop_start);
 336 
 337   __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 338   __ fmr(F1_ARG1, floatSlot);   __ b(loop_start);
 339   __ fmr(F2_ARG2, floatSlot);   __ b(loop_start);
 340   __ fmr(F3_ARG3, floatSlot);   __ b(loop_start);
 341   __ fmr(F4_ARG4, floatSlot);   __ b(loop_start);
 342   __ fmr(F5_ARG5, floatSlot);   __ b(loop_start);
 343   __ fmr(F6_ARG6, floatSlot);   __ b(loop_start);
 344   __ fmr(F7_ARG7, floatSlot);   __ b(loop_start);
 345   __ fmr(F8_ARG8, floatSlot);   __ b(loop_start);
 346   __ fmr(F9_ARG9, floatSlot);   __ b(loop_start);
 347   __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
 348   __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
 349   __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
 350   __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
 351 
 352   __ bind(move_intSlot_to_ARG);
 353   __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
 354   __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
 355   __ add(R11_scratch1, R0, R11_scratch1);
 356   __ mtctr(R11_scratch1/*branch_target*/);
 357   __ bctr();
 358   __ bind(move_floatSlot_to_FARG);
 359   __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
 360   __ addi(fpcnt, fpcnt, 1);
 361   __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
 362   __ add(R11_scratch1, R0, R11_scratch1);
 363   __ mtctr(R11_scratch1/*branch_target*/);
 364   __ bctr();
 365 
 366   return entry;
 367 }
 368 
 369 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 370   //
 371   // Registers alive
 372   //   R3_RET
 373   //   LR
 374   //
 375   // Registers updated
 376   //   R3_RET
 377   //
 378 
 379   Label done;
 380   address entry = __ pc();
 381 
 382   switch (type) {
 383   case T_BOOLEAN:
 384     // convert !=0 to 1
 385     __ neg(R0, R3_RET);
 386     __ orr(R0, R3_RET, R0);
 387     __ srwi(R3_RET, R0, 31);
 388     break;
 389   case T_BYTE:
 390      // sign extend 8 bits
 391      __ extsb(R3_RET, R3_RET);
 392      break;
 393   case T_CHAR:
 394      // zero extend 16 bits
 395      __ clrldi(R3_RET, R3_RET, 48);
 396      break;
 397   case T_SHORT:
 398      // sign extend 16 bits
 399      __ extsh(R3_RET, R3_RET);
 400      break;
 401   case T_INT:
 402      // sign extend 32 bits
 403      __ extsw(R3_RET, R3_RET);
 404      break;
 405   case T_LONG:
 406      break;
 407   case T_OBJECT:
 408     // JNIHandles::resolve result.
 409     __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31
 410     break;
 411   case T_FLOAT:
 412      break;
 413   case T_DOUBLE:
 414      break;
 415   case T_VOID:
 416      break;
 417   default: ShouldNotReachHere();
 418   }
 419 
 420   BIND(done);
 421   __ blr();
 422 
 423   return entry;
 424 }
 425 
 426 // Abstract method entry.
 427 //
 428 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 429   address entry = __ pc();
 430 
 431   //
 432   // Registers alive
 433   //   R16_thread     - JavaThread*
 434   //   R19_method     - callee's method (method to be invoked)
 435   //   R1_SP          - SP prepared such that caller's outgoing args are near top
 436   //   LR             - return address to caller
 437   //
 438   // Stack layout at this point:
 439   //
 440   //   0       [TOP_IJAVA_FRAME_ABI]         <-- R1_SP
 441   //           alignment (optional)
 442   //           [outgoing Java arguments]
 443   //           ...
 444   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
 445   //            ...
 446   //
 447 
 448   // Can't use call_VM here because we have not set up a new
 449   // interpreter state. Make the call to the vm and make it look like
 450   // our caller set up the JavaFrameAnchor.
 451   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
 452 
 453   // Push a new C frame and save LR.
 454   __ save_LR_CR(R0);
 455   __ push_frame_reg_args(0, R11_scratch1);
 456 
 457   // This is not a leaf but we have a JavaFrameAnchor now and we will
 458   // check (create) exceptions afterward so this is ok.
 459   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 460                   R16_thread, R19_method);
 461 
 462   // Pop the C frame and restore LR.
 463   __ pop_frame();
 464   __ restore_LR_CR(R0);
 465 
 466   // Reset JavaFrameAnchor from call_VM_leaf above.
 467   __ reset_last_Java_frame();
 468 
 469   // We don't know our caller, so jump to the general forward exception stub,
 470   // which will also pop our full frame off. Satisfy the interface of
 471   // SharedRuntime::generate_forward_exception()
 472   __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
 473   __ mtctr(R11_scratch1);
 474   __ bctr();
 475 
 476   return entry;
 477 }
 478 
 479 // Interpreter intrinsic for WeakReference.get().
 480 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
 481 //    into R8 and return quickly
 482 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
 483 //    It contains a GC barrier which puts the reference into the satb buffer
 484 //    to indicate that someone holds a strong reference to the object the
 485 //    weak ref points to!
 486 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 487   // Code: _aload_0, _getfield, _areturn
 488   // parameter size = 1
 489   //
 490   // The code that gets generated by this routine is split into 2 parts:
 491   //    1. the "intrinsified" code for G1 (or any SATB based GC),
 492   //    2. the slow path - which is an expansion of the regular method entry.
 493   //
 494   // Notes:
 495   // * In the G1 code we do not check whether we need to block for
 496   //   a safepoint. If G1 is enabled then we must execute the specialized
 497   //   code for Reference.get (except when the Reference object is null)
 498   //   so that we can log the value in the referent field with an SATB
 499   //   update buffer.
 500   //   If the code for the getfield template is modified so that the
 501   //   G1 pre-barrier code is executed when the current method is
 502   //   Reference.get() then going through the normal method entry
 503   //   will be fine.
 504   // * The G1 code can, however, check the receiver object (the instance
 505   //   of java.lang.Reference) and jump to the slow path if null. If the
 506   //   Reference object is null then we obviously cannot fetch the referent
 507   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 508   //   regular method entry code to generate the NPE.
 509   //
 510 
 511   address entry = __ pc();
 512 
 513   const int referent_offset = java_lang_ref_Reference::referent_offset();
 514 
 515   Label slow_path;
 516 
 517   // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
 518 
 519   // In the G1 code we don't check if we need to reach a safepoint. We
 520   // continue and the thread will safepoint at the next bytecode dispatch.
 521 
 522   // If the receiver is null then it is OK to jump to the slow path.
 523   __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
 524 
 525   // Check if receiver == NULL and go the slow path.
 526   __ cmpdi(CCR0, R3_RET, 0);
 527   __ beq(CCR0, slow_path);
 528 
 529   __ load_heap_oop(R3_RET, referent_offset, R3_RET,
 530                    /* non-volatile temp */ R31, R11_scratch1,
 531                    MacroAssembler::PRESERVATION_FRAME_LR,
 532                    ON_WEAK_OOP_REF);
 533 
 534   // Generate the G1 pre-barrier code to log the value of
 535   // the referent field in an SATB buffer. Note with
 536   // these parameters the pre-barrier does not generate
 537   // the load of the previous value.
 538 
 539   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
 540   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
 541 
 542   __ blr();
 543 
 544   __ bind(slow_path);
 545   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
 546   return entry;
 547 }
 548 
 549 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 550   address entry = __ pc();
 551 
 552   // Expression stack must be empty before entering the VM if an
 553   // exception happened.
 554   __ empty_expression_stack();
 555   // Throw exception.
 556   __ call_VM(noreg,
 557              CAST_FROM_FN_PTR(address,
 558                               InterpreterRuntime::throw_StackOverflowError));
 559   return entry;
 560 }
 561 
 562 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 563   address entry = __ pc();
 564   __ empty_expression_stack();
 565   // R4_ARG2 already contains the array.
 566   // Index is in R17_tos.
 567   __ mr(R5_ARG3, R17_tos);
 568   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
 569   return entry;
 570 }
 571 
 572 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 573   address entry = __ pc();
 574   // Expression stack must be empty before entering the VM if an
 575   // exception happened.
 576   __ empty_expression_stack();
 577 
 578   // Load exception object.
 579   // Thread will be loaded to R3_ARG1.
 580   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
 581 #ifdef ASSERT
 582   // Above call must not return here since exception pending.
 583   __ should_not_reach_here();
 584 #endif
 585   return entry;
 586 }
 587 
 588 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 589   address entry = __ pc();
 590   //__ untested("generate_exception_handler_common");
 591   Register Rexception = R17_tos;
 592 
 593   // Expression stack must be empty before entering the VM if an exception happened.
 594   __ empty_expression_stack();
 595 
 596   __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
 597   if (pass_oop) {
 598     __ mr(R5_ARG3, Rexception);
 599     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
 600   } else {
 601     __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
 602     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
 603   }
 604 
 605   // Throw exception.
 606   __ mr(R3_ARG1, Rexception);
 607   __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
 608   __ mtctr(R11_scratch1);
 609   __ bctr();
 610 
 611   return entry;
 612 }
 613 
 614 // This entry is returned to when a call returns to the interpreter.
 615 // When we arrive here, we expect that the callee stack frame is already popped.
 616 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 617   address entry = __ pc();
 618 
 619   // Move the value out of the return register back to the TOS cache of current frame.
 620   switch (state) {
 621     case ltos:
 622     case btos:
 623     case ztos:
 624     case ctos:
 625     case stos:
 626     case atos:
 627     case itos: __ mr(R17_tos, R3_RET); break;   // RET -> TOS cache
 628     case ftos:
 629     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 630     case vtos: break;                           // Nothing to do, this was a void return.
 631     default  : ShouldNotReachHere();
 632   }
 633 
 634   __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
 635   __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
 636   __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
 637 
 638   // Compiled code destroys templateTableBase, reload.
 639   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
 640 
 641   if (state == atos) {
 642     __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
 643   }
 644 
 645   const Register cache = R11_scratch1;
 646   const Register size  = R12_scratch2;
 647   __ get_cache_and_index_at_bcp(cache, 1, index_size);
 648 
 649   // Get least significant byte of 64 bit value:
 650 #if defined(VM_LITTLE_ENDIAN)
 651   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
 652 #else
 653   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
 654 #endif
 655   __ sldi(size, size, Interpreter::logStackElementSize);
 656   __ add(R15_esp, R15_esp, size);
 657 
 658  __ check_and_handle_popframe(R11_scratch1);
 659  __ check_and_handle_earlyret(R11_scratch1);
 660 
 661   __ dispatch_next(state, step);
 662   return entry;
 663 }
 664 
 665 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 666   address entry = __ pc();
 667   // If state != vtos, we're returning from a native method, which put it's result
 668   // into the result register. So move the value out of the return register back
 669   // to the TOS cache of current frame.
 670 
 671   switch (state) {
 672     case ltos:
 673     case btos:
 674     case ztos:
 675     case ctos:
 676     case stos:
 677     case atos:
 678     case itos: __ mr(R17_tos, R3_RET); break;   // GR_RET -> TOS cache
 679     case ftos:
 680     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 681     case vtos: break;                           // Nothing to do, this was a void return.
 682     default  : ShouldNotReachHere();
 683   }
 684 
 685   // Load LcpoolCache @@@ should be already set!
 686   __ get_constant_pool_cache(R27_constPoolCache);
 687 
 688   // Handle a pending exception, fall through if none.
 689   __ check_and_forward_exception(R11_scratch1, R12_scratch2);
 690 
 691   // Start executing bytecodes.
 692   if (continuation == NULL) {
 693     __ dispatch_next(state, step);
 694   } else {
 695     __ jump_to_entry(continuation, R11_scratch1);
 696   }
 697 
 698   return entry;
 699 }
 700 
 701 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 702   address entry = __ pc();
 703 
 704   __ push(state);
 705   __ call_VM(noreg, runtime_entry);
 706   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 707 
 708   return entry;
 709 }
 710 
 711 // Helpers for commoning out cases in the various type of method entries.
 712 
 713 // Increment invocation count & check for overflow.
 714 //
 715 // Note: checking for negative value instead of overflow
 716 //       so we have a 'sticky' overflow test.
 717 //
 718 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 719   // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
 720   Register Rscratch1   = R11_scratch1;
 721   Register Rscratch2   = R12_scratch2;
 722   Register R3_counters = R3_ARG1;
 723   Label done;
 724 
 725   const int increment = InvocationCounter::count_increment;
 726   Label no_mdo;
 727   if (ProfileInterpreter) {
 728     const Register Rmdo = R3_counters;
 729     __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
 730     __ cmpdi(CCR0, Rmdo, 0);
 731     __ beq(CCR0, no_mdo);
 732 
 733     // Increment invocation counter in the MDO.
 734     const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 735     __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
 736     __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
 737     __ addi(Rscratch2, Rscratch2, increment);
 738     __ stw(Rscratch2, mdo_ic_offs, Rmdo);
 739     __ and_(Rscratch1, Rscratch2, Rscratch1);
 740     __ bne(CCR0, done);
 741     __ b(*overflow);
 742   }
 743 
 744   // Increment counter in MethodCounters*.
 745   const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 746   __ bind(no_mdo);
 747   __ get_method_counters(R19_method, R3_counters, done);
 748   __ lwz(Rscratch2, mo_ic_offs, R3_counters);
 749   __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
 750   __ addi(Rscratch2, Rscratch2, increment);
 751   __ stw(Rscratch2, mo_ic_offs, R3_counters);
 752   __ and_(Rscratch1, Rscratch2, Rscratch1);
 753   __ beq(CCR0, *overflow);
 754 
 755   __ bind(done);
 756 }
 757 
 758 // Generate code to initiate compilation on invocation counter overflow.
 759 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
 760   // Generate code to initiate compilation on the counter overflow.
 761 
 762   // InterpreterRuntime::frequency_counter_overflow takes one arguments,
 763   // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
 764   // We pass zero in.
 765   // The call returns the address of the verified entry point for the method or NULL
 766   // if the compilation did not complete (either went background or bailed out).
 767   //
 768   // Unlike the C++ interpreter above: Check exceptions!
 769   // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
 770   // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
 771 
 772   __ li(R4_ARG2, 0);
 773   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
 774 
 775   // Returns verified_entry_point or NULL.
 776   // We ignore it in any case.
 777   __ b(continue_entry);
 778 }
 779 
 780 // See if we've got enough room on the stack for locals plus overhead below
 781 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 782 // without going through the signal handler, i.e., reserved and yellow zones
 783 // will not be made usable. The shadow zone must suffice to handle the
 784 // overflow.
 785 //
 786 // Kills Rmem_frame_size, Rscratch1.
 787 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
 788   Label done;
 789   assert_different_registers(Rmem_frame_size, Rscratch1);
 790 
 791   BLOCK_COMMENT("stack_overflow_check_with_compare {");
 792   __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
 793   __ ld(Rscratch1, thread_(stack_overflow_limit));
 794   __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
 795   __ bgt(CCR0/*is_stack_overflow*/, done);
 796 
 797   // The stack overflows. Load target address of the runtime stub and call it.
 798   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
 799   __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
 800   __ mtctr(Rscratch1);
 801   // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
 802 #ifdef ASSERT
 803   Label frame_not_shrunk;
 804   __ cmpld(CCR0, R1_SP, R21_sender_SP);
 805   __ ble(CCR0, frame_not_shrunk);
 806   __ stop("frame shrunk");
 807   __ bind(frame_not_shrunk);
 808   __ ld(Rscratch1, 0, R1_SP);
 809   __ ld(R0, 0, R21_sender_SP);
 810   __ cmpd(CCR0, R0, Rscratch1);
 811   __ asm_assert_eq("backlink");
 812 #endif // ASSERT
 813   __ mr(R1_SP, R21_sender_SP);
 814   __ bctr();
 815 
 816   __ align(32, 12);
 817   __ bind(done);
 818   BLOCK_COMMENT("} stack_overflow_check_with_compare");
 819 }
 820 
 821 // Lock the current method, interpreter register window must be set up!
 822 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
 823   const Register Robj_to_lock = Rscratch2;
 824 
 825   {
 826     if (!flags_preloaded) {
 827       __ lwz(Rflags, method_(access_flags));
 828     }
 829 
 830 #ifdef ASSERT
 831     // Check if methods needs synchronization.
 832     {
 833       Label Lok;
 834       __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
 835       __ btrue(CCR0,Lok);
 836       __ stop("method doesn't need synchronization");
 837       __ bind(Lok);
 838     }
 839 #endif // ASSERT
 840   }
 841 
 842   // Get synchronization object to Rscratch2.
 843   {
 844     Label Lstatic;
 845     Label Ldone;
 846 
 847     __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
 848     __ btrue(CCR0, Lstatic);
 849 
 850     // Non-static case: load receiver obj from stack and we're done.
 851     __ ld(Robj_to_lock, R18_locals);
 852     __ b(Ldone);
 853 
 854     __ bind(Lstatic); // Static case: Lock the java mirror
 855     // Load mirror from interpreter frame.
 856     __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
 857     __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
 858 
 859     __ bind(Ldone);
 860     __ verify_oop(Robj_to_lock);
 861   }
 862 
 863   // Got the oop to lock => execute!
 864   __ add_monitor_to_stack(true, Rscratch1, R0);
 865 
 866   __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
 867   __ lock_object(R26_monitor, Robj_to_lock);
 868 }
 869 
 870 // Generate a fixed interpreter frame for pure interpreter
 871 // and I2N native transition frames.
 872 //
 873 // Before (stack grows downwards):
 874 //
 875 //         |  ...         |
 876 //         |------------- |
 877 //         |  java arg0   |
 878 //         |  ...         |
 879 //         |  java argn   |
 880 //         |              |   <-   R15_esp
 881 //         |              |
 882 //         |--------------|
 883 //         | abi_112      |
 884 //         |              |   <-   R1_SP
 885 //         |==============|
 886 //
 887 //
 888 // After:
 889 //
 890 //         |  ...         |
 891 //         |  java arg0   |<-   R18_locals
 892 //         |  ...         |
 893 //         |  java argn   |
 894 //         |--------------|
 895 //         |              |
 896 //         |  java locals |
 897 //         |              |
 898 //         |--------------|
 899 //         |  abi_48      |
 900 //         |==============|
 901 //         |              |
 902 //         |   istate     |
 903 //         |              |
 904 //         |--------------|
 905 //         |   monitor    |<-   R26_monitor
 906 //         |--------------|
 907 //         |              |<-   R15_esp
 908 //         | expression   |
 909 //         | stack        |
 910 //         |              |
 911 //         |--------------|
 912 //         |              |
 913 //         | abi_112      |<-   R1_SP
 914 //         |==============|
 915 //
 916 // The top most frame needs an abi space of 112 bytes. This space is needed,
 917 // since we call to c. The c function may spill their arguments to the caller
 918 // frame. When we call to java, we don't need these spill slots. In order to save
 919 // space on the stack, we resize the caller. However, java locals reside in
 920 // the caller frame and the frame has to be increased. The frame_size for the
 921 // current frame was calculated based on max_stack as size for the expression
 922 // stack. At the call, just a part of the expression stack might be used.
 923 // We don't want to waste this space and cut the frame back accordingly.
 924 // The resulting amount for resizing is calculated as follows:
 925 // resize =   (number_of_locals - number_of_arguments) * slot_size
 926 //          + (R1_SP - R15_esp) + 48
 927 //
 928 // The size for the callee frame is calculated:
 929 // framesize = 112 + max_stack + monitor + state_size
 930 //
 931 // maxstack:   Max number of slots on the expression stack, loaded from the method.
 932 // monitor:    We statically reserve room for one monitor object.
 933 // state_size: We save the current state of the interpreter to this area.
 934 //
 935 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
 936   Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
 937            Rtop_frame_size      = R7_ARG5,
 938            Rconst_method        = R8_ARG6,
 939            Rconst_pool          = R9_ARG7,
 940            Rmirror              = R10_ARG8;
 941 
 942   assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
 943                              Rconst_method, Rconst_pool);
 944 
 945   __ ld(Rconst_method, method_(const));
 946   __ lhz(Rsize_of_parameters /* number of params */,
 947          in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
 948   if (native_call) {
 949     // If we're calling a native method, we reserve space for the worst-case signature
 950     // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
 951     // We add two slots to the parameter_count, one for the jni
 952     // environment and one for a possible native mirror.
 953     Label skip_native_calculate_max_stack;
 954     __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
 955     __ cmpwi(CCR0, Rtop_frame_size, Argument::n_register_parameters);
 956     __ bge(CCR0, skip_native_calculate_max_stack);
 957     __ li(Rtop_frame_size, Argument::n_register_parameters);
 958     __ bind(skip_native_calculate_max_stack);
 959     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 960     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 961     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 962     assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
 963   } else {
 964     __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
 965     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 966     __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
 967     __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
 968     __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
 969     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 970     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 971     __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
 972   }
 973 
 974   // Compute top frame size.
 975   __ addi(Rtop_frame_size, Rtop_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
 976 
 977   // Cut back area between esp and max_stack.
 978   __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
 979 
 980   __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
 981   __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
 982   // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
 983   // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
 984 
 985   if (!native_call) {
 986     // Stack overflow check.
 987     // Native calls don't need the stack size check since they have no
 988     // expression stack and the arguments are already on the stack and
 989     // we only add a handful of words to the stack.
 990     __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
 991     generate_stack_overflow_check(R11_scratch1, R12_scratch2);
 992   }
 993 
 994   // Set up interpreter state registers.
 995 
 996   __ add(R18_locals, R15_esp, Rsize_of_parameters);
 997   __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
 998   __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), Rconst_pool);
 999 
1000   // Set method data pointer.
1001   if (ProfileInterpreter) {
1002     Label zero_continue;
1003     __ ld(R28_mdx, method_(method_data));
1004     __ cmpdi(CCR0, R28_mdx, 0);
1005     __ beq(CCR0, zero_continue);
1006     __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1007     __ bind(zero_continue);
1008   }
1009 
1010   if (native_call) {
1011     __ li(R14_bcp, 0); // Must initialize.
1012   } else {
1013     __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1014   }
1015 
1016   // Resize parent frame.
1017   __ mflr(R12_scratch2);
1018   __ neg(Rparent_frame_resize, Rparent_frame_resize);
1019   __ resize_frame(Rparent_frame_resize, R11_scratch1);
1020   __ std(R12_scratch2, _abi0(lr), R1_SP);
1021 
1022   // Get mirror and store it in the frame as GC root for this Method*.
1023   __ ld(Rmirror, ConstantPool::pool_holder_offset_in_bytes(), Rconst_pool);
1024   __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1025   __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1026 
1027   __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1028   __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1029 
1030   // Store values.
1031   __ std(R19_method, _ijava_state_neg(method), R1_SP);
1032   __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1033   __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1034   __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1035 
1036   // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1037   // be found in the frame after save_interpreter_state is done. This is always true
1038   // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1039   // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1040   // (Enhanced Stack Trace).
1041   // The signal handler does not save the interpreter state into the frame.
1042 
1043   // We have to initialize some of these frame slots for native calls (accessed by GC).
1044   // Also initialize them for non-native calls for better tool support (even though
1045   // you may not get the most recent version as described above).
1046   __ li(R0, 0);
1047   __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1048   __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1049   if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1050   __ std(R15_esp, _ijava_state_neg(esp), R1_SP);
1051   __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1052 
1053   // Store sender's SP and this frame's top SP.
1054   __ subf(R12_scratch2, Rtop_frame_size, R1_SP);
1055   __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1056   __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1057 
1058   // Push top frame.
1059   __ push_frame(Rtop_frame_size, R11_scratch1);
1060 }
1061 
1062 // End of helpers
1063 
1064 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1065 
1066   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1067   bool use_instruction = false;
1068   address runtime_entry = NULL;
1069   int num_args = 1;
1070   bool double_precision = true;
1071 
1072   // PPC64 specific:
1073   switch (kind) {
1074     case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1075     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1076     case Interpreter::java_lang_math_fmaF:
1077     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1078     default: break; // Fall back to runtime call.
1079   }
1080 
1081   switch (kind) {
1082     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1083     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1084     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1085     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1086     case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt);  break;
1087     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1088     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1089     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1090     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1091     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1092     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1093     default: ShouldNotReachHere();
1094   }
1095 
1096   // Use normal entry if neither instruction nor runtime call is used.
1097   if (!use_instruction && runtime_entry == NULL) return NULL;
1098 
1099   address entry = __ pc();
1100 
1101   // Load arguments
1102   assert(num_args <= 13, "passed in registers");
1103   if (double_precision) {
1104     int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1105     for (int i = 0; i < num_args; ++i) {
1106       __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1107       offset -= 2 * Interpreter::stackElementSize;
1108     }
1109   } else {
1110     int offset = num_args * Interpreter::stackElementSize;
1111     for (int i = 0; i < num_args; ++i) {
1112       __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1113       offset -= Interpreter::stackElementSize;
1114     }
1115   }
1116 
1117   if (use_instruction) {
1118     switch (kind) {
1119       case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1);          break;
1120       case Interpreter::java_lang_math_abs:  __ fabs(F1_RET, F1);           break;
1121       case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1122       case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3);  break;
1123       default: ShouldNotReachHere();
1124     }
1125   } else {
1126     // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1127     //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1128     //__ call_c_and_return_to_caller(R12_scratch2);
1129 
1130     // Push a new C frame and save LR.
1131     __ save_LR_CR(R0);
1132     __ push_frame_reg_args(0, R11_scratch1);
1133 
1134     __ call_VM_leaf(runtime_entry);
1135 
1136     // Pop the C frame and restore LR.
1137     __ pop_frame();
1138     __ restore_LR_CR(R0);
1139   }
1140 
1141   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1142   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1143   __ blr();
1144 
1145   __ flush();
1146 
1147   return entry;
1148 }
1149 
1150 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1151   // Quick & dirty stack overflow checking: bang the stack & handle trap.
1152   // Note that we do the banging after the frame is setup, since the exception
1153   // handling code expects to find a valid interpreter frame on the stack.
1154   // Doing the banging earlier fails if the caller frame is not an interpreter
1155   // frame.
1156   // (Also, the exception throwing code expects to unlock any synchronized
1157   // method receiever, so do the banging after locking the receiver.)
1158 
1159   // Bang each page in the shadow zone. We can't assume it's been done for
1160   // an interpreter frame with greater than a page of locals, so each page
1161   // needs to be checked.  Only true for non-native.
1162   const int page_size = os::vm_page_size();
1163   const int n_shadow_pages = ((int)StackOverflow::stack_shadow_zone_size()) / page_size;
1164   const int start_page = native_call ? n_shadow_pages : 1;
1165   BLOCK_COMMENT("bang_stack_shadow_pages:");
1166   for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1167     __ bang_stack_with_offset(pages*page_size);
1168   }
1169 }
1170 
1171 // Interpreter stub for calling a native method. (asm interpreter)
1172 // This sets up a somewhat different looking stack for calling the
1173 // native method than the typical interpreter frame setup.
1174 //
1175 // On entry:
1176 //   R19_method    - method
1177 //   R16_thread    - JavaThread*
1178 //   R15_esp       - intptr_t* sender tos
1179 //
1180 //   abstract stack (grows up)
1181 //     [  IJava (caller of JNI callee)  ]  <-- ASP
1182 //        ...
1183 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1184 
1185   address entry = __ pc();
1186 
1187   const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1188 
1189   // -----------------------------------------------------------------------------
1190   // Allocate a new frame that represents the native callee (i2n frame).
1191   // This is not a full-blown interpreter frame, but in particular, the
1192   // following registers are valid after this:
1193   // - R19_method
1194   // - R18_local (points to start of arguments to native function)
1195   //
1196   //   abstract stack (grows up)
1197   //     [  IJava (caller of JNI callee)  ]  <-- ASP
1198   //        ...
1199 
1200   const Register signature_handler_fd = R11_scratch1;
1201   const Register pending_exception    = R0;
1202   const Register result_handler_addr  = R31;
1203   const Register native_method_fd     = R11_scratch1;
1204   const Register access_flags         = R22_tmp2;
1205   const Register active_handles       = R11_scratch1; // R26_monitor saved to state.
1206   const Register sync_state           = R12_scratch2;
1207   const Register sync_state_addr      = sync_state;   // Address is dead after use.
1208   const Register suspend_flags        = R11_scratch1;
1209 
1210   //=============================================================================
1211   // Allocate new frame and initialize interpreter state.
1212 
1213   Label exception_return;
1214   Label exception_return_sync_check;
1215   Label stack_overflow_return;
1216 
1217   // Generate new interpreter state and jump to stack_overflow_return in case of
1218   // a stack overflow.
1219   //generate_compute_interpreter_state(stack_overflow_return);
1220 
1221   Register size_of_parameters = R22_tmp2;
1222 
1223   generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1224 
1225   //=============================================================================
1226   // Increment invocation counter. On overflow, entry to JNI method
1227   // will be compiled.
1228   Label invocation_counter_overflow, continue_after_compile;
1229   if (inc_counter) {
1230     if (synchronized) {
1231       // Since at this point in the method invocation the exception handler
1232       // would try to exit the monitor of synchronized methods which hasn't
1233       // been entered yet, we set the thread local variable
1234       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1235       // runtime, exception handling i.e. unlock_if_synchronized_method will
1236       // check this thread local flag.
1237       // This flag has two effects, one is to force an unwind in the topmost
1238       // interpreter frame and not perform an unlock while doing so.
1239       __ li(R0, 1);
1240       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1241     }
1242     generate_counter_incr(&invocation_counter_overflow);
1243 
1244     BIND(continue_after_compile);
1245   }
1246 
1247   bang_stack_shadow_pages(true);
1248 
1249   if (inc_counter) {
1250     // Reset the _do_not_unlock_if_synchronized flag.
1251     if (synchronized) {
1252       __ li(R0, 0);
1253       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1254     }
1255   }
1256 
1257   // access_flags = method->access_flags();
1258   // Load access flags.
1259   assert(access_flags->is_nonvolatile(),
1260          "access_flags must be in a non-volatile register");
1261   // Type check.
1262   assert(4 == sizeof(AccessFlags), "unexpected field size");
1263   __ lwz(access_flags, method_(access_flags));
1264 
1265   // We don't want to reload R19_method and access_flags after calls
1266   // to some helper functions.
1267   assert(R19_method->is_nonvolatile(),
1268          "R19_method must be a non-volatile register");
1269 
1270   // Check for synchronized methods. Must happen AFTER invocation counter
1271   // check, so method is not locked if counter overflows.
1272 
1273   if (synchronized) {
1274     lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1275 
1276     // Update monitor in state.
1277     __ ld(R11_scratch1, 0, R1_SP);
1278     __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1279   }
1280 
1281   // jvmti/jvmpi support
1282   __ notify_method_entry();
1283 
1284   //=============================================================================
1285   // Get and call the signature handler.
1286 
1287   __ ld(signature_handler_fd, method_(signature_handler));
1288   Label call_signature_handler;
1289 
1290   __ cmpdi(CCR0, signature_handler_fd, 0);
1291   __ bne(CCR0, call_signature_handler);
1292 
1293   // Method has never been called. Either generate a specialized
1294   // handler or point to the slow one.
1295   //
1296   // Pass parameter 'false' to avoid exception check in call_VM.
1297   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1298 
1299   // Check for an exception while looking up the target method. If we
1300   // incurred one, bail.
1301   __ ld(pending_exception, thread_(pending_exception));
1302   __ cmpdi(CCR0, pending_exception, 0);
1303   __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1304 
1305   // Reload signature handler, it may have been created/assigned in the meanwhile.
1306   __ ld(signature_handler_fd, method_(signature_handler));
1307   __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1308 
1309   BIND(call_signature_handler);
1310 
1311   // Before we call the signature handler we push a new frame to
1312   // protect the interpreter frame volatile registers when we return
1313   // from jni but before we can get back to Java.
1314 
1315   // First set the frame anchor while the SP/FP registers are
1316   // convenient and the slow signature handler can use this same frame
1317   // anchor.
1318 
1319   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1320   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1321 
1322   // Now the interpreter frame (and its call chain) have been
1323   // invalidated and flushed. We are now protected against eager
1324   // being enabled in native code. Even if it goes eager the
1325   // registers will be reloaded as clean and we will invalidate after
1326   // the call so no spurious flush should be possible.
1327 
1328   // Call signature handler and pass locals address.
1329   //
1330   // Our signature handlers copy required arguments to the C stack
1331   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1332   __ mr(R3_ARG1, R18_locals);
1333 #if !defined(ABI_ELFv2)
1334   __ ld(signature_handler_fd, 0, signature_handler_fd);
1335 #endif
1336 
1337   __ call_stub(signature_handler_fd);
1338 
1339   // Remove the register parameter varargs slots we allocated in
1340   // compute_interpreter_state. SP+16 ends up pointing to the ABI
1341   // outgoing argument area.
1342   //
1343   // Not needed on PPC64.
1344   //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1345 
1346   assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1347   // Save across call to native method.
1348   __ mr(result_handler_addr, R3_RET);
1349 
1350   __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1351 
1352   // Set up fixed parameters and call the native method.
1353   // If the method is static, get mirror into R4_ARG2.
1354   {
1355     Label method_is_not_static;
1356     // Access_flags is non-volatile and still, no need to restore it.
1357 
1358     // Restore access flags.
1359     __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1360     __ bfalse(CCR0, method_is_not_static);
1361 
1362     __ ld(R11_scratch1, _abi0(callers_sp), R1_SP);
1363     // Load mirror from interpreter frame.
1364     __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1365     // R4_ARG2 = &state->_oop_temp;
1366     __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1367     __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1368     BIND(method_is_not_static);
1369   }
1370 
1371   // At this point, arguments have been copied off the stack into
1372   // their JNI positions. Oops are boxed in-place on the stack, with
1373   // handles copied to arguments. The result handler address is in a
1374   // register.
1375 
1376   // Pass JNIEnv address as first parameter.
1377   __ addir(R3_ARG1, thread_(jni_environment));
1378 
1379   // Load the native_method entry before we change the thread state.
1380   __ ld(native_method_fd, method_(native_function));
1381 
1382   //=============================================================================
1383   // Transition from _thread_in_Java to _thread_in_native. As soon as
1384   // we make this change the safepoint code needs to be certain that
1385   // the last Java frame we established is good. The pc in that frame
1386   // just needs to be near here not an actual return address.
1387 
1388   // We use release_store_fence to update values like the thread state, where
1389   // we don't want the current thread to continue until all our prior memory
1390   // accesses (including the new thread state) are visible to other threads.
1391   __ li(R0, _thread_in_native);
1392   __ release();
1393 
1394   // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1395   __ stw(R0, thread_(thread_state));
1396 
1397   //=============================================================================
1398   // Call the native method. Argument registers must not have been
1399   // overwritten since "__ call_stub(signature_handler);" (except for
1400   // ARG1 and ARG2 for static methods).
1401   __ call_c(native_method_fd);
1402 
1403   __ li(R0, 0);
1404   __ ld(R11_scratch1, 0, R1_SP);
1405   __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1406   __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1407   __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1408 
1409   // Note: C++ interpreter needs the following here:
1410   // The frame_manager_lr field, which we use for setting the last
1411   // java frame, gets overwritten by the signature handler. Restore
1412   // it now.
1413   //__ get_PC_trash_LR(R11_scratch1);
1414   //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1415 
1416   // Because of GC R19_method may no longer be valid.
1417 
1418   // Block, if necessary, before resuming in _thread_in_Java state.
1419   // In order for GC to work, don't clear the last_Java_sp until after
1420   // blocking.
1421 
1422   //=============================================================================
1423   // Switch thread to "native transition" state before reading the
1424   // synchronization state. This additional state is necessary
1425   // because reading and testing the synchronization state is not
1426   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1427   // in _thread_in_native state, loads _not_synchronized and is
1428   // preempted. VM thread changes sync state to synchronizing and
1429   // suspends threads for GC. Thread A is resumed to finish this
1430   // native method, but doesn't block here since it didn't see any
1431   // synchronization in progress, and escapes.
1432 
1433   // We use release_store_fence to update values like the thread state, where
1434   // we don't want the current thread to continue until all our prior memory
1435   // accesses (including the new thread state) are visible to other threads.
1436   __ li(R0/*thread_state*/, _thread_in_native_trans);
1437   __ release();
1438   __ stw(R0/*thread_state*/, thread_(thread_state));
1439   __ fence();
1440 
1441   // Now before we return to java we must look for a current safepoint
1442   // (a new safepoint can not start since we entered native_trans).
1443   // We must check here because a current safepoint could be modifying
1444   // the callers registers right this moment.
1445 
1446   // Acquire isn't strictly necessary here because of the fence, but
1447   // sync_state is declared to be volatile, so we do it anyway
1448   // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1449 
1450   Label do_safepoint, sync_check_done;
1451   // No synchronization in progress nor yet synchronized.
1452   __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1453 
1454   // Not suspended.
1455   // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1456   __ lwz(suspend_flags, thread_(suspend_flags));
1457   __ cmpwi(CCR1, suspend_flags, 0);
1458   __ beq(CCR1, sync_check_done);
1459 
1460   __ bind(do_safepoint);
1461   __ isync();
1462   // Block. We do the call directly and leave the current
1463   // last_Java_frame setup undisturbed. We must save any possible
1464   // native result across the call. No oop is present.
1465 
1466   __ mr(R3_ARG1, R16_thread);
1467 #if defined(ABI_ELFv2)
1468   __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1469             relocInfo::none);
1470 #else
1471   __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1472             relocInfo::none);
1473 #endif
1474 
1475   __ bind(sync_check_done);
1476 
1477   //=============================================================================
1478   // <<<<<< Back in Interpreter Frame >>>>>
1479 
1480   // We are in thread_in_native_trans here and back in the normal
1481   // interpreter frame. We don't have to do anything special about
1482   // safepoints and we can switch to Java mode anytime we are ready.
1483 
1484   // Note: frame::interpreter_frame_result has a dependency on how the
1485   // method result is saved across the call to post_method_exit. For
1486   // native methods it assumes that the non-FPU/non-void result is
1487   // saved in _native_lresult and a FPU result in _native_fresult. If
1488   // this changes then the interpreter_frame_result implementation
1489   // will need to be updated too.
1490 
1491   // On PPC64, we have stored the result directly after the native call.
1492 
1493   //=============================================================================
1494   // Back in Java
1495 
1496   // We use release_store_fence to update values like the thread state, where
1497   // we don't want the current thread to continue until all our prior memory
1498   // accesses (including the new thread state) are visible to other threads.
1499   __ li(R0/*thread_state*/, _thread_in_Java);
1500   __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1501   __ stw(R0/*thread_state*/, thread_(thread_state));
1502 
1503   if (CheckJNICalls) {
1504     // clear_pending_jni_exception_check
1505     __ load_const_optimized(R0, 0L);
1506     __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1507   }
1508 
1509   __ reset_last_Java_frame();
1510 
1511   // Jvmdi/jvmpi support. Whether we've got an exception pending or
1512   // not, and whether unlocking throws an exception or not, we notify
1513   // on native method exit. If we do have an exception, we'll end up
1514   // in the caller's context to handle it, so if we don't do the
1515   // notify here, we'll drop it on the floor.
1516   __ notify_method_exit(true/*native method*/,
1517                         ilgl /*illegal state (not used for native methods)*/,
1518                         InterpreterMacroAssembler::NotifyJVMTI,
1519                         false /*check_exceptions*/);
1520 
1521   //=============================================================================
1522   // Handle exceptions
1523 
1524   if (synchronized) {
1525     __ unlock_object(R26_monitor); // Can also unlock methods.
1526   }
1527 
1528   // Reset active handles after returning from native.
1529   // thread->active_handles()->clear();
1530   __ ld(active_handles, thread_(active_handles));
1531   // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1532   __ li(R0, 0);
1533   __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1534 
1535   Label exception_return_sync_check_already_unlocked;
1536   __ ld(R0/*pending_exception*/, thread_(pending_exception));
1537   __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1538   __ bne(CCR0, exception_return_sync_check_already_unlocked);
1539 
1540   //-----------------------------------------------------------------------------
1541   // No exception pending.
1542 
1543   // Move native method result back into proper registers and return.
1544   // Invoke result handler (may unbox/promote).
1545   __ ld(R11_scratch1, 0, R1_SP);
1546   __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1547   __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1548   __ call_stub(result_handler_addr);
1549 
1550   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1551 
1552   // Must use the return pc which was loaded from the caller's frame
1553   // as the VM uses return-pc-patching for deoptimization.
1554   __ mtlr(R0);
1555   __ blr();
1556 
1557   //-----------------------------------------------------------------------------
1558   // An exception is pending. We call into the runtime only if the
1559   // caller was not interpreted. If it was interpreted the
1560   // interpreter will do the correct thing. If it isn't interpreted
1561   // (call stub/compiled code) we will change our return and continue.
1562 
1563   BIND(exception_return_sync_check);
1564 
1565   if (synchronized) {
1566     __ unlock_object(R26_monitor); // Can also unlock methods.
1567   }
1568   BIND(exception_return_sync_check_already_unlocked);
1569 
1570   const Register return_pc = R31;
1571 
1572   __ ld(return_pc, 0, R1_SP);
1573   __ ld(return_pc, _abi0(lr), return_pc);
1574 
1575   // Get the address of the exception handler.
1576   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1577                   R16_thread,
1578                   return_pc /* return pc */);
1579   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1580 
1581   // Load the PC of the the exception handler into LR.
1582   __ mtlr(R3_RET);
1583 
1584   // Load exception into R3_ARG1 and clear pending exception in thread.
1585   __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1586   __ li(R4_ARG2, 0);
1587   __ std(R4_ARG2, thread_(pending_exception));
1588 
1589   // Load the original return pc into R4_ARG2.
1590   __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1591 
1592   // Return to exception handler.
1593   __ blr();
1594 
1595   //=============================================================================
1596   // Counter overflow.
1597 
1598   if (inc_counter) {
1599     // Handle invocation counter overflow.
1600     __ bind(invocation_counter_overflow);
1601 
1602     generate_counter_overflow(continue_after_compile);
1603   }
1604 
1605   return entry;
1606 }
1607 
1608 // Generic interpreted method entry to (asm) interpreter.
1609 //
1610 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1611   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1612   address entry = __ pc();
1613   // Generate the code to allocate the interpreter stack frame.
1614   Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1615            Rsize_of_locals     = R5_ARG3; // Written by generate_fixed_frame.
1616 
1617   // Does also a stack check to assure this frame fits on the stack.
1618   generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1619 
1620   // --------------------------------------------------------------------------
1621   // Zero out non-parameter locals.
1622   // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1623   // worth to ask the flag, just do it.
1624   Register Rslot_addr = R6_ARG4,
1625            Rnum       = R7_ARG5;
1626   Label Lno_locals, Lzero_loop;
1627 
1628   // Set up the zeroing loop.
1629   __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1630   __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1631   __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1632   __ beq(CCR0, Lno_locals);
1633   __ li(R0, 0);
1634   __ mtctr(Rnum);
1635 
1636   // The zero locals loop.
1637   __ bind(Lzero_loop);
1638   __ std(R0, 0, Rslot_addr);
1639   __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1640   __ bdnz(Lzero_loop);
1641 
1642   __ bind(Lno_locals);
1643 
1644   // --------------------------------------------------------------------------
1645   // Counter increment and overflow check.
1646   Label invocation_counter_overflow;
1647   Label continue_after_compile;
1648   if (inc_counter || ProfileInterpreter) {
1649 
1650     Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1651     if (synchronized) {
1652       // Since at this point in the method invocation the exception handler
1653       // would try to exit the monitor of synchronized methods which hasn't
1654       // been entered yet, we set the thread local variable
1655       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1656       // runtime, exception handling i.e. unlock_if_synchronized_method will
1657       // check this thread local flag.
1658       // This flag has two effects, one is to force an unwind in the topmost
1659       // interpreter frame and not perform an unlock while doing so.
1660       __ li(R0, 1);
1661       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1662     }
1663 
1664     // Argument and return type profiling.
1665     __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1666 
1667     // Increment invocation counter and check for overflow.
1668     if (inc_counter) {
1669       generate_counter_incr(&invocation_counter_overflow);
1670     }
1671 
1672     __ bind(continue_after_compile);
1673   }
1674 
1675   bang_stack_shadow_pages(false);
1676 
1677   if (inc_counter || ProfileInterpreter) {
1678     // Reset the _do_not_unlock_if_synchronized flag.
1679     if (synchronized) {
1680       __ li(R0, 0);
1681       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1682     }
1683   }
1684 
1685   // --------------------------------------------------------------------------
1686   // Locking of synchronized methods. Must happen AFTER invocation_counter
1687   // check and stack overflow check, so method is not locked if overflows.
1688   if (synchronized) {
1689     lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1690   }
1691 #ifdef ASSERT
1692   else {
1693     Label Lok;
1694     __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1695     __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1696     __ asm_assert_eq("method needs synchronization");
1697     __ bind(Lok);
1698   }
1699 #endif // ASSERT
1700 
1701   __ verify_thread();
1702 
1703   // --------------------------------------------------------------------------
1704   // JVMTI support
1705   __ notify_method_entry();
1706 
1707   // --------------------------------------------------------------------------
1708   // Start executing instructions.
1709   __ dispatch_next(vtos);
1710 
1711   // --------------------------------------------------------------------------
1712   if (inc_counter) {
1713     // Handle invocation counter overflow.
1714     __ bind(invocation_counter_overflow);
1715     generate_counter_overflow(continue_after_compile);
1716   }
1717   return entry;
1718 }
1719 
1720 // CRC32 Intrinsics.
1721 //
1722 // Contract on scratch and work registers.
1723 // =======================================
1724 //
1725 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1726 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1727 // You can't rely on these registers across calls.
1728 //
1729 // The generators for CRC32_update and for CRC32_updateBytes use the
1730 // scratch/work register set internally, passing the work registers
1731 // as arguments to the MacroAssembler emitters as required.
1732 //
1733 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1734 // Their contents is not constant but may change according to the requirements
1735 // of the emitted code.
1736 //
1737 // All other registers from the scratch/work register set are used "internally"
1738 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1739 // Basically, only R3_RET contains a defined value which is the function result.
1740 //
1741 /**
1742  * Method entry for static native methods:
1743  *   int java.util.zip.CRC32.update(int crc, int b)
1744  */
1745 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1746   if (UseCRC32Intrinsics) {
1747     address start = __ pc();  // Remember stub start address (is rtn value).
1748     Label slow_path;
1749 
1750     // Safepoint check
1751     const Register sync_state = R11_scratch1;
1752     __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1753 
1754     // We don't generate local frame and don't align stack because
1755     // we not even call stub code (we generate the code inline)
1756     // and there is no safepoint on this path.
1757 
1758     // Load java parameters.
1759     // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1760     const Register argP    = R15_esp;
1761     const Register crc     = R3_ARG1;  // crc value
1762     const Register data    = R4_ARG2;
1763     const Register table   = R5_ARG3;  // address of crc32 table
1764 
1765     BLOCK_COMMENT("CRC32_update {");
1766 
1767     // Arguments are reversed on java expression stack
1768 #ifdef VM_LITTLE_ENDIAN
1769     int data_offs = 0+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1770                                        // Being passed as an int, the single byte is at offset +0.
1771 #else
1772     int data_offs = 3+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1773                                        // Being passed from java as an int, the single byte is at offset +3.
1774 #endif
1775     __ lwz(crc, 2*wordSize, argP);     // Current crc state, zero extend to 64 bit to have a clean register.
1776     __ lbz(data, data_offs, argP);     // Byte from buffer, zero-extended.
1777     __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1778     __ kernel_crc32_singleByteReg(crc, data, table, true);
1779 
1780     // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1781     __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1782     __ blr();
1783 
1784     // Generate a vanilla native entry as the slow path.
1785     BLOCK_COMMENT("} CRC32_update");
1786     BIND(slow_path);
1787     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1788     return start;
1789   }
1790 
1791   return NULL;
1792 }
1793 
1794 /**
1795  * Method entry for static native methods:
1796  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1797  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1798  */
1799 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1800   if (UseCRC32Intrinsics) {
1801     address start = __ pc();  // Remember stub start address (is rtn value).
1802     Label slow_path;
1803 
1804     // Safepoint check
1805     const Register sync_state = R11_scratch1;
1806     __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1807 
1808     // We don't generate local frame and don't align stack because
1809     // we not even call stub code (we generate the code inline)
1810     // and there is no safepoint on this path.
1811 
1812     // Load parameters.
1813     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1814     const Register argP    = R15_esp;
1815     const Register crc     = R3_ARG1;  // crc value
1816     const Register data    = R4_ARG2;  // address of java byte array
1817     const Register dataLen = R5_ARG3;  // source data len
1818     const Register tmp     = R11_scratch1;
1819 
1820     // Arguments are reversed on java expression stack.
1821     // Calculate address of start element.
1822     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1823       BLOCK_COMMENT("CRC32_updateByteBuffer {");
1824       // crc     @ (SP + 5W) (32bit)
1825       // buf     @ (SP + 3W) (64bit ptr to long array)
1826       // off     @ (SP + 2W) (32bit)
1827       // dataLen @ (SP + 1W) (32bit)
1828       // data = buf + off
1829       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1830       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1831       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1832       __ lwz( crc,     5*wordSize, argP);  // current crc state
1833       __ add( data, data, tmp);            // Add byte buffer offset.
1834     } else {                                                         // Used for "updateBytes update".
1835       BLOCK_COMMENT("CRC32_updateBytes {");
1836       // crc     @ (SP + 4W) (32bit)
1837       // buf     @ (SP + 3W) (64bit ptr to byte array)
1838       // off     @ (SP + 2W) (32bit)
1839       // dataLen @ (SP + 1W) (32bit)
1840       // data = buf + off + base_offset
1841       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1842       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1843       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1844       __ add( data, data, tmp);            // add byte buffer offset
1845       __ lwz( crc,     4*wordSize, argP);  // current crc state
1846       __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1847     }
1848 
1849     __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1850 
1851     // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1852     __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1853     __ blr();
1854 
1855     // Generate a vanilla native entry as the slow path.
1856     BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1857     BIND(slow_path);
1858     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1859     return start;
1860   }
1861 
1862   return NULL;
1863 }
1864 
1865 
1866 /**
1867  * Method entry for intrinsic-candidate (non-native) methods:
1868  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
1869  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1870  * Unlike CRC32, CRC32C does not have any methods marked as native
1871  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1872  **/
1873 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1874   if (UseCRC32CIntrinsics) {
1875     address start = __ pc();  // Remember stub start address (is rtn value).
1876 
1877     // We don't generate local frame and don't align stack because
1878     // we not even call stub code (we generate the code inline)
1879     // and there is no safepoint on this path.
1880 
1881     // Load parameters.
1882     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1883     const Register argP    = R15_esp;
1884     const Register crc     = R3_ARG1;  // crc value
1885     const Register data    = R4_ARG2;  // address of java byte array
1886     const Register dataLen = R5_ARG3;  // source data len
1887     const Register tmp     = R11_scratch1;
1888 
1889     // Arguments are reversed on java expression stack.
1890     // Calculate address of start element.
1891     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1892       BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1893       // crc     @ (SP + 5W) (32bit)
1894       // buf     @ (SP + 3W) (64bit ptr to long array)
1895       // off     @ (SP + 2W) (32bit)
1896       // dataLen @ (SP + 1W) (32bit)
1897       // data = buf + off
1898       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1899       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1900       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1901       __ lwz( crc,     5*wordSize, argP);  // current crc state
1902       __ add( data, data, tmp);            // Add byte buffer offset.
1903       __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1904     } else {                                                         // Used for "updateBytes update".
1905       BLOCK_COMMENT("CRC32C_updateBytes {");
1906       // crc     @ (SP + 4W) (32bit)
1907       // buf     @ (SP + 3W) (64bit ptr to byte array)
1908       // off     @ (SP + 2W) (32bit)
1909       // dataLen @ (SP + 1W) (32bit)
1910       // data = buf + off + base_offset
1911       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1912       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1913       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1914       __ add( data, data, tmp);            // add byte buffer offset
1915       __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1916       __ lwz( crc,     4*wordSize, argP);  // current crc state
1917       __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1918     }
1919 
1920     __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1921 
1922     // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1923     __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1924     __ blr();
1925 
1926     BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1927     return start;
1928   }
1929 
1930   return NULL;
1931 }
1932 
1933 // =============================================================================
1934 // Exceptions
1935 
1936 void TemplateInterpreterGenerator::generate_throw_exception() {
1937   Register Rexception    = R17_tos,
1938            Rcontinuation = R3_RET;
1939 
1940   // --------------------------------------------------------------------------
1941   // Entry point if an method returns with a pending exception (rethrow).
1942   Interpreter::_rethrow_exception_entry = __ pc();
1943   {
1944     __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
1945     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
1946     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
1947 
1948     // Compiled code destroys templateTableBase, reload.
1949     __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1950   }
1951 
1952   // Entry point if a interpreted method throws an exception (throw).
1953   Interpreter::_throw_exception_entry = __ pc();
1954   {
1955     __ mr(Rexception, R3_RET);
1956 
1957     __ verify_thread();
1958     __ verify_oop(Rexception);
1959 
1960     // Expression stack must be empty before entering the VM in case of an exception.
1961     __ empty_expression_stack();
1962     // Find exception handler address and preserve exception oop.
1963     // Call C routine to find handler and jump to it.
1964     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1965     __ mtctr(Rcontinuation);
1966     // Push exception for exception handler bytecodes.
1967     __ push_ptr(Rexception);
1968 
1969     // Jump to exception handler (may be remove activation entry!).
1970     __ bctr();
1971   }
1972 
1973   // If the exception is not handled in the current frame the frame is
1974   // removed and the exception is rethrown (i.e. exception
1975   // continuation is _rethrow_exception).
1976   //
1977   // Note: At this point the bci is still the bxi for the instruction
1978   // which caused the exception and the expression stack is
1979   // empty. Thus, for any VM calls at this point, GC will find a legal
1980   // oop map (with empty expression stack).
1981 
1982   // In current activation
1983   // tos: exception
1984   // bcp: exception bcp
1985 
1986   // --------------------------------------------------------------------------
1987   // JVMTI PopFrame support
1988 
1989   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1990   {
1991     // Set the popframe_processing bit in popframe_condition indicating that we are
1992     // currently handling popframe, so that call_VMs that may happen later do not
1993     // trigger new popframe handling cycles.
1994     __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1995     __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1996     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1997 
1998     // Empty the expression stack, as in normal exception handling.
1999     __ empty_expression_stack();
2000     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2001 
2002     // Check to see whether we are returning to a deoptimized frame.
2003     // (The PopFrame call ensures that the caller of the popped frame is
2004     // either interpreted or compiled and deoptimizes it if compiled.)
2005     // Note that we don't compare the return PC against the
2006     // deoptimization blob's unpack entry because of the presence of
2007     // adapter frames in C2.
2008     Label Lcaller_not_deoptimized;
2009     Register return_pc = R3_ARG1;
2010     __ ld(return_pc, 0, R1_SP);
2011     __ ld(return_pc, _abi0(lr), return_pc);
2012     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2013     __ cmpdi(CCR0, R3_RET, 0);
2014     __ bne(CCR0, Lcaller_not_deoptimized);
2015 
2016     // The deoptimized case.
2017     // In this case, we can't call dispatch_next() after the frame is
2018     // popped, but instead must save the incoming arguments and restore
2019     // them after deoptimization has occurred.
2020     __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2021     __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2022     __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2023     __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2024     __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2025     // Save these arguments.
2026     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2027 
2028     // Inform deoptimization that it is responsible for restoring these arguments.
2029     __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2030     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2031 
2032     // Return from the current method into the deoptimization blob. Will eventually
2033     // end up in the deopt interpeter entry, deoptimization prepared everything that
2034     // we will reexecute the call that called us.
2035     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2036     __ mtlr(return_pc);
2037     __ blr();
2038 
2039     // The non-deoptimized case.
2040     __ bind(Lcaller_not_deoptimized);
2041 
2042     // Clear the popframe condition flag.
2043     __ li(R0, 0);
2044     __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2045 
2046     // Get out of the current method and re-execute the call that called us.
2047     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2048     __ restore_interpreter_state(R11_scratch1);
2049     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2050     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2051     if (ProfileInterpreter) {
2052       __ set_method_data_pointer_for_bcp();
2053       __ ld(R11_scratch1, 0, R1_SP);
2054       __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2055     }
2056 #if INCLUDE_JVMTI
2057     Label L_done;
2058 
2059     __ lbz(R11_scratch1, 0, R14_bcp);
2060     __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2061     __ bne(CCR0, L_done);
2062 
2063     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2064     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2065     __ ld(R4_ARG2, 0, R18_locals);
2066     __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2067 
2068     __ cmpdi(CCR0, R4_ARG2, 0);
2069     __ beq(CCR0, L_done);
2070     __ std(R4_ARG2, wordSize, R15_esp);
2071     __ bind(L_done);
2072 #endif // INCLUDE_JVMTI
2073     __ dispatch_next(vtos);
2074   }
2075   // end of JVMTI PopFrame support
2076 
2077   // --------------------------------------------------------------------------
2078   // Remove activation exception entry.
2079   // This is jumped to if an interpreted method can't handle an exception itself
2080   // (we come from the throw/rethrow exception entry above). We're going to call
2081   // into the VM to find the exception handler in the caller, pop the current
2082   // frame and return the handler we calculated.
2083   Interpreter::_remove_activation_entry = __ pc();
2084   {
2085     __ pop_ptr(Rexception);
2086     __ verify_thread();
2087     __ verify_oop(Rexception);
2088     __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2089 
2090     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2091     __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2092 
2093     __ get_vm_result(Rexception);
2094 
2095     // We are done with this activation frame; find out where to go next.
2096     // The continuation point will be an exception handler, which expects
2097     // the following registers set up:
2098     //
2099     // RET:  exception oop
2100     // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2101 
2102     Register return_pc = R31; // Needs to survive the runtime call.
2103     __ ld(return_pc, 0, R1_SP);
2104     __ ld(return_pc, _abi0(lr), return_pc);
2105     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2106 
2107     // Remove the current activation.
2108     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2109 
2110     __ mr(R4_ARG2, return_pc);
2111     __ mtlr(R3_RET);
2112     __ mr(R3_RET, Rexception);
2113     __ blr();
2114   }
2115 }
2116 
2117 // JVMTI ForceEarlyReturn support.
2118 // Returns "in the middle" of a method with a "fake" return value.
2119 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2120 
2121   Register Rscratch1 = R11_scratch1,
2122            Rscratch2 = R12_scratch2;
2123 
2124   address entry = __ pc();
2125   __ empty_expression_stack();
2126 
2127   __ load_earlyret_value(state, Rscratch1);
2128 
2129   __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2130   // Clear the earlyret state.
2131   __ li(R0, 0);
2132   __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2133 
2134   __ remove_activation(state, false, false);
2135   // Copied from TemplateTable::_return.
2136   // Restoration of lr done by remove_activation.
2137   switch (state) {
2138     // Narrow result if state is itos but result type is smaller.
2139     case btos:
2140     case ztos:
2141     case ctos:
2142     case stos:
2143     case itos: __ narrow(R17_tos); /* fall through */
2144     case ltos:
2145     case atos: __ mr(R3_RET, R17_tos); break;
2146     case ftos:
2147     case dtos: __ fmr(F1_RET, F15_ftos); break;
2148     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2149                // to get visible before the reference to the object gets stored anywhere.
2150                __ membar(Assembler::StoreStore); break;
2151     default  : ShouldNotReachHere();
2152   }
2153   __ blr();
2154 
2155   return entry;
2156 } // end of ForceEarlyReturn support
2157 
2158 //-----------------------------------------------------------------------------
2159 // Helper for vtos entry point generation
2160 
2161 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2162                                                          address& bep,
2163                                                          address& cep,
2164                                                          address& sep,
2165                                                          address& aep,
2166                                                          address& iep,
2167                                                          address& lep,
2168                                                          address& fep,
2169                                                          address& dep,
2170                                                          address& vep) {
2171   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2172   Label L;
2173 
2174   aep = __ pc();  __ push_ptr();  __ b(L);
2175   fep = __ pc();  __ push_f();    __ b(L);
2176   dep = __ pc();  __ push_d();    __ b(L);
2177   lep = __ pc();  __ push_l();    __ b(L);
2178   __ align(32, 12, 24); // align L
2179   bep = cep = sep =
2180   iep = __ pc();  __ push_i();
2181   vep = __ pc();
2182   __ bind(L);
2183   generate_and_dispatch(t);
2184 }
2185 
2186 //-----------------------------------------------------------------------------
2187 
2188 // Non-product code
2189 #ifndef PRODUCT
2190 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2191   //__ flush_bundle();
2192   address entry = __ pc();
2193 
2194   const char *bname = NULL;
2195   uint tsize = 0;
2196   switch(state) {
2197   case ftos:
2198     bname = "trace_code_ftos {";
2199     tsize = 2;
2200     break;
2201   case btos:
2202     bname = "trace_code_btos {";
2203     tsize = 2;
2204     break;
2205   case ztos:
2206     bname = "trace_code_ztos {";
2207     tsize = 2;
2208     break;
2209   case ctos:
2210     bname = "trace_code_ctos {";
2211     tsize = 2;
2212     break;
2213   case stos:
2214     bname = "trace_code_stos {";
2215     tsize = 2;
2216     break;
2217   case itos:
2218     bname = "trace_code_itos {";
2219     tsize = 2;
2220     break;
2221   case ltos:
2222     bname = "trace_code_ltos {";
2223     tsize = 3;
2224     break;
2225   case atos:
2226     bname = "trace_code_atos {";
2227     tsize = 2;
2228     break;
2229   case vtos:
2230     // Note: In case of vtos, the topmost of stack value could be a int or doubl
2231     // In case of a double (2 slots) we won't see the 2nd stack value.
2232     // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2233     bname = "trace_code_vtos {";
2234     tsize = 2;
2235 
2236     break;
2237   case dtos:
2238     bname = "trace_code_dtos {";
2239     tsize = 3;
2240     break;
2241   default:
2242     ShouldNotReachHere();
2243   }
2244   BLOCK_COMMENT(bname);
2245 
2246   // Support short-cut for TraceBytecodesAt.
2247   // Don't call into the VM if we don't want to trace to speed up things.
2248   Label Lskip_vm_call;
2249   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2250     int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2251     int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2252     __ ld(R11_scratch1, offs1, R11_scratch1);
2253     __ lwa(R12_scratch2, offs2, R12_scratch2);
2254     __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2255     __ blt(CCR0, Lskip_vm_call);
2256   }
2257 
2258   __ push(state);
2259   // Load 2 topmost expression stack values.
2260   __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2261   __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2262   __ mflr(R31);
2263   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2264   __ mtlr(R31);
2265   __ pop(state);
2266 
2267   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2268     __ bind(Lskip_vm_call);
2269   }
2270   __ blr();
2271   BLOCK_COMMENT("} trace_code");
2272   return entry;
2273 }
2274 
2275 void TemplateInterpreterGenerator::count_bytecode() {
2276   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2277   __ lwz(R12_scratch2, offs, R11_scratch1);
2278   __ addi(R12_scratch2, R12_scratch2, 1);
2279   __ stw(R12_scratch2, offs, R11_scratch1);
2280 }
2281 
2282 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2283   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2284   __ lwz(R12_scratch2, offs, R11_scratch1);
2285   __ addi(R12_scratch2, R12_scratch2, 1);
2286   __ stw(R12_scratch2, offs, R11_scratch1);
2287 }
2288 
2289 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2290   const Register addr = R11_scratch1,
2291                  tmp  = R12_scratch2;
2292   // Get index, shift out old bytecode, bring in new bytecode, and store it.
2293   // _index = (_index >> log2_number_of_codes) |
2294   //          (bytecode << log2_number_of_codes);
2295   int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2296   __ lwz(tmp, offs1, addr);
2297   __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2298   __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2299   __ stw(tmp, offs1, addr);
2300 
2301   // Bump bucket contents.
2302   // _counters[_index] ++;
2303   int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2304   __ sldi(tmp, tmp, LogBytesPerInt);
2305   __ add(addr, tmp, addr);
2306   __ lwz(tmp, offs2, addr);
2307   __ addi(tmp, tmp, 1);
2308   __ stw(tmp, offs2, addr);
2309 }
2310 
2311 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2312   // Call a little run-time stub to avoid blow-up for each bytecode.
2313   // The run-time runtime saves the right registers, depending on
2314   // the tosca in-state for the given template.
2315 
2316   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2317          "entry must have been generated");
2318 
2319   // Note: we destroy LR here.
2320   __ bl(Interpreter::trace_code(t->tos_in()));
2321 }
2322 
2323 void TemplateInterpreterGenerator::stop_interpreter_at() {
2324   Label L;
2325   int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2326   int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2327   __ ld(R11_scratch1, offs1, R11_scratch1);
2328   __ lwa(R12_scratch2, offs2, R12_scratch2);
2329   __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2330   __ bne(CCR0, L);
2331   __ illtrap();
2332   __ bind(L);
2333 }
2334 
2335 #endif // !PRODUCT