rev 49827 : 8201593: Print array length in ArrayIndexOutOfBoundsException.

   1 /*
   2  * Copyright (c) 2014, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2015, 2018, 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 "gc/shared/barrierSetAssembler.hpp"
  29 #include "interpreter/bytecodeHistogram.hpp"
  30 #include "interpreter/interpreter.hpp"
  31 #include "interpreter/interpreterRuntime.hpp"
  32 #include "interpreter/interp_masm.hpp"
  33 #include "interpreter/templateInterpreterGenerator.hpp"
  34 #include "interpreter/templateTable.hpp"
  35 #include "oops/arrayOop.hpp"
  36 #include "oops/methodData.hpp"
  37 #include "oops/method.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "prims/jvmtiExport.hpp"
  40 #include "prims/jvmtiThreadState.hpp"
  41 #include "runtime/arguments.hpp"
  42 #include "runtime/deoptimization.hpp"
  43 #include "runtime/frame.inline.hpp"
  44 #include "runtime/sharedRuntime.hpp"
  45 #include "runtime/stubRoutines.hpp"
  46 #include "runtime/synchronizer.hpp"
  47 #include "runtime/timer.hpp"
  48 #include "runtime/vframeArray.hpp"
  49 #include "utilities/debug.hpp"
  50 #include "utilities/macros.hpp"
  51 
  52 #undef __
  53 #define __ _masm->
  54 
  55 // Size of interpreter code.  Increase if too small.  Interpreter will
  56 // fail with a guarantee ("not enough space for interpreter generation");
  57 // if too small.
  58 // Run with +PrintInterpreter to get the VM to print out the size.
  59 // Max size with JVMTI
  60 int TemplateInterpreter::InterpreterCodeSize = 256*K;
  61 
  62 #ifdef PRODUCT
  63 #define BLOCK_COMMENT(str) /* nothing */
  64 #else
  65 #define BLOCK_COMMENT(str) __ block_comment(str)
  66 #endif
  67 
  68 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  69 
  70 //-----------------------------------------------------------------------------
  71 
  72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  73   // Slow_signature handler that respects the PPC C calling conventions.
  74   //
  75   // We get called by the native entry code with our output register
  76   // area == 8. First we call InterpreterRuntime::get_result_handler
  77   // to copy the pointer to the signature string temporarily to the
  78   // first C-argument and to return the result_handler in
  79   // R3_RET. Since native_entry will copy the jni-pointer to the
  80   // first C-argument slot later on, it is OK to occupy this slot
  81   // temporarilly. Then we copy the argument list on the java
  82   // expression stack into native varargs format on the native stack
  83   // and load arguments into argument registers. Integer arguments in
  84   // the varargs vector will be sign-extended to 8 bytes.
  85   //
  86   // On entry:
  87   //   R3_ARG1        - intptr_t*     Address of java argument list in memory.
  88   //   R15_prev_state - BytecodeInterpreter* Address of interpreter state for
  89   //     this method
  90   //   R19_method
  91   //
  92   // On exit (just before return instruction):
  93   //   R3_RET            - contains the address of the result_handler.
  94   //   R4_ARG2           - is not updated for static methods and contains "this" otherwise.
  95   //   R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
  96   //                       ARGi contains this argument. Otherwise, ARGi is not updated.
  97   //   F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
  98 
  99   const int LogSizeOfTwoInstructions = 3;
 100 
 101   // FIXME: use Argument:: GL: Argument names different numbers!
 102   const int max_fp_register_arguments  = 13;
 103   const int max_int_register_arguments = 6;  // first 2 are reserved
 104 
 105   const Register arg_java       = R21_tmp1;
 106   const Register arg_c          = R22_tmp2;
 107   const Register signature      = R23_tmp3;  // is string
 108   const Register sig_byte       = R24_tmp4;
 109   const Register fpcnt          = R25_tmp5;
 110   const Register argcnt         = R26_tmp6;
 111   const Register intSlot        = R27_tmp7;
 112   const Register target_sp      = R28_tmp8;
 113   const FloatRegister floatSlot = F0;
 114 
 115   address entry = __ function_entry();
 116 
 117   __ save_LR_CR(R0);
 118   __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 119   // We use target_sp for storing arguments in the C frame.
 120   __ mr(target_sp, R1_SP);
 121   __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
 122 
 123   __ mr(arg_java, R3_ARG1);
 124 
 125   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
 126 
 127   // Signature is in R3_RET. Signature is callee saved.
 128   __ mr(signature, R3_RET);
 129 
 130   // Get the result handler.
 131   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
 132 
 133   {
 134     Label L;
 135     // test if static
 136     // _access_flags._flags must be at offset 0.
 137     // TODO PPC port: requires change in shared code.
 138     //assert(in_bytes(AccessFlags::flags_offset()) == 0,
 139     //       "MethodDesc._access_flags == MethodDesc._access_flags._flags");
 140     // _access_flags must be a 32 bit value.
 141     assert(sizeof(AccessFlags) == 4, "wrong size");
 142     __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
 143     // testbit with condition register.
 144     __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
 145     __ btrue(CCR0, L);
 146     // For non-static functions, pass "this" in R4_ARG2 and copy it
 147     // to 2nd C-arg slot.
 148     // We need to box the Java object here, so we use arg_java
 149     // (address of current Java stack slot) as argument and don't
 150     // dereference it as in case of ints, floats, etc.
 151     __ mr(R4_ARG2, arg_java);
 152     __ addi(arg_java, arg_java, -BytesPerWord);
 153     __ std(R4_ARG2, _abi(carg_2), target_sp);
 154     __ bind(L);
 155   }
 156 
 157   // Will be incremented directly after loop_start. argcnt=0
 158   // corresponds to 3rd C argument.
 159   __ li(argcnt, -1);
 160   // arg_c points to 3rd C argument
 161   __ addi(arg_c, target_sp, _abi(carg_3));
 162   // no floating-point args parsed so far
 163   __ li(fpcnt, 0);
 164 
 165   Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
 166   Label loop_start, loop_end;
 167   Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
 168 
 169   // signature points to '(' at entry
 170 #ifdef ASSERT
 171   __ lbz(sig_byte, 0, signature);
 172   __ cmplwi(CCR0, sig_byte, '(');
 173   __ bne(CCR0, do_dontreachhere);
 174 #endif
 175 
 176   __ bind(loop_start);
 177 
 178   __ addi(argcnt, argcnt, 1);
 179   __ lbzu(sig_byte, 1, signature);
 180 
 181   __ cmplwi(CCR0, sig_byte, ')'); // end of signature
 182   __ beq(CCR0, loop_end);
 183 
 184   __ cmplwi(CCR0, sig_byte, 'B'); // byte
 185   __ beq(CCR0, do_int);
 186 
 187   __ cmplwi(CCR0, sig_byte, 'C'); // char
 188   __ beq(CCR0, do_int);
 189 
 190   __ cmplwi(CCR0, sig_byte, 'D'); // double
 191   __ beq(CCR0, do_double);
 192 
 193   __ cmplwi(CCR0, sig_byte, 'F'); // float
 194   __ beq(CCR0, do_float);
 195 
 196   __ cmplwi(CCR0, sig_byte, 'I'); // int
 197   __ beq(CCR0, do_int);
 198 
 199   __ cmplwi(CCR0, sig_byte, 'J'); // long
 200   __ beq(CCR0, do_long);
 201 
 202   __ cmplwi(CCR0, sig_byte, 'S'); // short
 203   __ beq(CCR0, do_int);
 204 
 205   __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
 206   __ beq(CCR0, do_int);
 207 
 208   __ cmplwi(CCR0, sig_byte, 'L'); // object
 209   __ beq(CCR0, do_object);
 210 
 211   __ cmplwi(CCR0, sig_byte, '['); // array
 212   __ beq(CCR0, do_array);
 213 
 214   //  __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
 215   //  __ beq(CCR0, do_void);
 216 
 217   __ bind(do_dontreachhere);
 218 
 219   __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
 220 
 221   __ bind(do_array);
 222 
 223   {
 224     Label start_skip, end_skip;
 225 
 226     __ bind(start_skip);
 227     __ lbzu(sig_byte, 1, signature);
 228     __ cmplwi(CCR0, sig_byte, '[');
 229     __ beq(CCR0, start_skip); // skip further brackets
 230     __ cmplwi(CCR0, sig_byte, '9');
 231     __ bgt(CCR0, end_skip);   // no optional size
 232     __ cmplwi(CCR0, sig_byte, '0');
 233     __ bge(CCR0, start_skip); // skip optional size
 234     __ bind(end_skip);
 235 
 236     __ cmplwi(CCR0, sig_byte, 'L');
 237     __ beq(CCR0, do_object);  // for arrays of objects, the name of the object must be skipped
 238     __ b(do_boxed);          // otherwise, go directly to do_boxed
 239   }
 240 
 241   __ bind(do_object);
 242   {
 243     Label L;
 244     __ bind(L);
 245     __ lbzu(sig_byte, 1, signature);
 246     __ cmplwi(CCR0, sig_byte, ';');
 247     __ bne(CCR0, L);
 248    }
 249   // Need to box the Java object here, so we use arg_java (address of
 250   // current Java stack slot) as argument and don't dereference it as
 251   // in case of ints, floats, etc.
 252   Label do_null;
 253   __ bind(do_boxed);
 254   __ ld(R0,0, arg_java);
 255   __ cmpdi(CCR0, R0, 0);
 256   __ li(intSlot,0);
 257   __ beq(CCR0, do_null);
 258   __ mr(intSlot, arg_java);
 259   __ bind(do_null);
 260   __ std(intSlot, 0, arg_c);
 261   __ addi(arg_java, arg_java, -BytesPerWord);
 262   __ addi(arg_c, arg_c, BytesPerWord);
 263   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 264   __ blt(CCR0, move_intSlot_to_ARG);
 265   __ b(loop_start);
 266 
 267   __ bind(do_int);
 268   __ lwa(intSlot, 0, arg_java);
 269   __ std(intSlot, 0, arg_c);
 270   __ addi(arg_java, arg_java, -BytesPerWord);
 271   __ addi(arg_c, arg_c, BytesPerWord);
 272   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 273   __ blt(CCR0, move_intSlot_to_ARG);
 274   __ b(loop_start);
 275 
 276   __ bind(do_long);
 277   __ ld(intSlot, -BytesPerWord, arg_java);
 278   __ std(intSlot, 0, arg_c);
 279   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 280   __ addi(arg_c, arg_c, BytesPerWord);
 281   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 282   __ blt(CCR0, move_intSlot_to_ARG);
 283   __ b(loop_start);
 284 
 285   __ bind(do_float);
 286   __ lfs(floatSlot, 0, arg_java);
 287 #if defined(LINUX)
 288   // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
 289   // in the least significant word of an argument slot.
 290 #if defined(VM_LITTLE_ENDIAN)
 291   __ stfs(floatSlot, 0, arg_c);
 292 #else
 293   __ stfs(floatSlot, 4, arg_c);
 294 #endif
 295 #elif defined(AIX)
 296   // Although AIX runs on big endian CPU, float is in most significant
 297   // word of an argument slot.
 298   __ stfs(floatSlot, 0, arg_c);
 299 #else
 300 #error "unknown OS"
 301 #endif
 302   __ addi(arg_java, arg_java, -BytesPerWord);
 303   __ addi(arg_c, arg_c, BytesPerWord);
 304   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 305   __ blt(CCR0, move_floatSlot_to_FARG);
 306   __ b(loop_start);
 307 
 308   __ bind(do_double);
 309   __ lfd(floatSlot, - BytesPerWord, arg_java);
 310   __ stfd(floatSlot, 0, arg_c);
 311   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 312   __ addi(arg_c, arg_c, BytesPerWord);
 313   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 314   __ blt(CCR0, move_floatSlot_to_FARG);
 315   __ b(loop_start);
 316 
 317   __ bind(loop_end);
 318 
 319   __ pop_frame();
 320   __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 321   __ restore_LR_CR(R0);
 322 
 323   __ blr();
 324 
 325   Label move_int_arg, move_float_arg;
 326   __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 327   __ mr(R5_ARG3, intSlot);  __ b(loop_start);
 328   __ mr(R6_ARG4, intSlot);  __ b(loop_start);
 329   __ mr(R7_ARG5, intSlot);  __ b(loop_start);
 330   __ mr(R8_ARG6, intSlot);  __ b(loop_start);
 331   __ mr(R9_ARG7, intSlot);  __ b(loop_start);
 332   __ mr(R10_ARG8, intSlot); __ b(loop_start);
 333 
 334   __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 335   __ fmr(F1_ARG1, floatSlot);   __ b(loop_start);
 336   __ fmr(F2_ARG2, floatSlot);   __ b(loop_start);
 337   __ fmr(F3_ARG3, floatSlot);   __ b(loop_start);
 338   __ fmr(F4_ARG4, floatSlot);   __ b(loop_start);
 339   __ fmr(F5_ARG5, floatSlot);   __ b(loop_start);
 340   __ fmr(F6_ARG6, floatSlot);   __ b(loop_start);
 341   __ fmr(F7_ARG7, floatSlot);   __ b(loop_start);
 342   __ fmr(F8_ARG8, floatSlot);   __ b(loop_start);
 343   __ fmr(F9_ARG9, floatSlot);   __ b(loop_start);
 344   __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
 345   __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
 346   __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
 347   __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
 348 
 349   __ bind(move_intSlot_to_ARG);
 350   __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
 351   __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
 352   __ add(R11_scratch1, R0, R11_scratch1);
 353   __ mtctr(R11_scratch1/*branch_target*/);
 354   __ bctr();
 355   __ bind(move_floatSlot_to_FARG);
 356   __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
 357   __ addi(fpcnt, fpcnt, 1);
 358   __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
 359   __ add(R11_scratch1, R0, R11_scratch1);
 360   __ mtctr(R11_scratch1/*branch_target*/);
 361   __ bctr();
 362 
 363   return entry;
 364 }
 365 
 366 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 367   //
 368   // Registers alive
 369   //   R3_RET
 370   //   LR
 371   //
 372   // Registers updated
 373   //   R3_RET
 374   //
 375 
 376   Label done;
 377   address entry = __ pc();
 378 
 379   switch (type) {
 380   case T_BOOLEAN:
 381     // convert !=0 to 1
 382     __ neg(R0, R3_RET);
 383     __ orr(R0, R3_RET, R0);
 384     __ srwi(R3_RET, R0, 31);
 385     break;
 386   case T_BYTE:
 387      // sign extend 8 bits
 388      __ extsb(R3_RET, R3_RET);
 389      break;
 390   case T_CHAR:
 391      // zero extend 16 bits
 392      __ clrldi(R3_RET, R3_RET, 48);
 393      break;
 394   case T_SHORT:
 395      // sign extend 16 bits
 396      __ extsh(R3_RET, R3_RET);
 397      break;
 398   case T_INT:
 399      // sign extend 32 bits
 400      __ extsw(R3_RET, R3_RET);
 401      break;
 402   case T_LONG:
 403      break;
 404   case T_OBJECT:
 405     // JNIHandles::resolve result.
 406     __ resolve_jobject(R3_RET, R11_scratch1, R31, /* needs_frame */ true); // kills R31
 407     break;
 408   case T_FLOAT:
 409      break;
 410   case T_DOUBLE:
 411      break;
 412   case T_VOID:
 413      break;
 414   default: ShouldNotReachHere();
 415   }
 416 
 417   BIND(done);
 418   __ blr();
 419 
 420   return entry;
 421 }
 422 
 423 // Abstract method entry.
 424 //
 425 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 426   address entry = __ pc();
 427 
 428   //
 429   // Registers alive
 430   //   R16_thread     - JavaThread*
 431   //   R19_method     - callee's method (method to be invoked)
 432   //   R1_SP          - SP prepared such that caller's outgoing args are near top
 433   //   LR             - return address to caller
 434   //
 435   // Stack layout at this point:
 436   //
 437   //   0       [TOP_IJAVA_FRAME_ABI]         <-- R1_SP
 438   //           alignment (optional)
 439   //           [outgoing Java arguments]
 440   //           ...
 441   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
 442   //            ...
 443   //
 444 
 445   // Can't use call_VM here because we have not set up a new
 446   // interpreter state. Make the call to the vm and make it look like
 447   // our caller set up the JavaFrameAnchor.
 448   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
 449 
 450   // Push a new C frame and save LR.
 451   __ save_LR_CR(R0);
 452   __ push_frame_reg_args(0, R11_scratch1);
 453 
 454   // This is not a leaf but we have a JavaFrameAnchor now and we will
 455   // check (create) exceptions afterward so this is ok.
 456   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 457                   R16_thread, R19_method);
 458 
 459   // Pop the C frame and restore LR.
 460   __ pop_frame();
 461   __ restore_LR_CR(R0);
 462 
 463   // Reset JavaFrameAnchor from call_VM_leaf above.
 464   __ reset_last_Java_frame();
 465 
 466   // We don't know our caller, so jump to the general forward exception stub,
 467   // which will also pop our full frame off. Satisfy the interface of
 468   // SharedRuntime::generate_forward_exception()
 469   __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
 470   __ mtctr(R11_scratch1);
 471   __ bctr();
 472 
 473   return entry;
 474 }
 475 
 476 // Interpreter intrinsic for WeakReference.get().
 477 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
 478 //    into R8 and return quickly
 479 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
 480 //    It contains a GC barrier which puts the reference into the satb buffer
 481 //    to indicate that someone holds a strong reference to the object the
 482 //    weak ref points to!
 483 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 484   // Code: _aload_0, _getfield, _areturn
 485   // parameter size = 1
 486   //
 487   // The code that gets generated by this routine is split into 2 parts:
 488   //    1. the "intrinsified" code for G1 (or any SATB based GC),
 489   //    2. the slow path - which is an expansion of the regular method entry.
 490   //
 491   // Notes:
 492   // * In the G1 code we do not check whether we need to block for
 493   //   a safepoint. If G1 is enabled then we must execute the specialized
 494   //   code for Reference.get (except when the Reference object is null)
 495   //   so that we can log the value in the referent field with an SATB
 496   //   update buffer.
 497   //   If the code for the getfield template is modified so that the
 498   //   G1 pre-barrier code is executed when the current method is
 499   //   Reference.get() then going through the normal method entry
 500   //   will be fine.
 501   // * The G1 code can, however, check the receiver object (the instance
 502   //   of java.lang.Reference) and jump to the slow path if null. If the
 503   //   Reference object is null then we obviously cannot fetch the referent
 504   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 505   //   regular method entry code to generate the NPE.
 506   //
 507 
 508   address entry = __ pc();
 509 
 510   const int referent_offset = java_lang_ref_Reference::referent_offset;
 511   guarantee(referent_offset > 0, "referent offset not initialized");
 512 
 513   Label slow_path;
 514 
 515   // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
 516 
 517   // In the G1 code we don't check if we need to reach a safepoint. We
 518   // continue and the thread will safepoint at the next bytecode dispatch.
 519 
 520   // If the receiver is null then it is OK to jump to the slow path.
 521   __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
 522 
 523   // Check if receiver == NULL and go the slow path.
 524   __ cmpdi(CCR0, R3_RET, 0);
 525   __ beq(CCR0, slow_path);
 526 
 527   // Load the value of the referent field.
 528   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
 529   bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT,
 530                     R3_RET, referent_offset, R3_RET,
 531                     /* non-volatile temp */ R31, R11_scratch1, true);
 532 
 533   // Generate the G1 pre-barrier code to log the value of
 534   // the referent field in an SATB buffer. Note with
 535   // these parameters the pre-barrier does not generate
 536   // the load of the previous value.
 537 
 538   // Restore caller sp for c2i case.
 539 #ifdef ASSERT
 540   __ ld(R9_ARG7, 0, R1_SP);
 541   __ ld(R10_ARG8, 0, R21_sender_SP);
 542   __ cmpd(CCR0, R9_ARG7, R10_ARG8);
 543   __ asm_assert_eq("backlink", 0x544);
 544 #endif // ASSERT
 545   __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
 546 
 547   __ blr();
 548 
 549   __ bind(slow_path);
 550   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
 551   return entry;
 552 }
 553 
 554 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 555   address entry = __ pc();
 556 
 557   // Expression stack must be empty before entering the VM if an
 558   // exception happened.
 559   __ empty_expression_stack();
 560   // Throw exception.
 561   __ call_VM(noreg,
 562              CAST_FROM_FN_PTR(address,
 563                               InterpreterRuntime::throw_StackOverflowError));
 564   return entry;
 565 }
 566 
 567 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
 568   address entry = __ pc();
 569   __ empty_expression_stack();
 570   __ load_const_optimized(R4_ARG2, (address) name);
 571   // Index is in R17_tos.
 572   __ mr(R5_ARG3, R17_tos);
 573   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException));
 574   return entry;
 575 }
 576 
 577 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 578   address entry = __ pc();
 579   // Expression stack must be empty before entering the VM if an
 580   // exception happened.
 581   __ empty_expression_stack();
 582 
 583   // Load exception object.
 584   // Thread will be loaded to R3_ARG1.
 585   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
 586 #ifdef ASSERT
 587   // Above call must not return here since exception pending.
 588   __ should_not_reach_here();
 589 #endif
 590   return entry;
 591 }
 592 
 593 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 594   address entry = __ pc();
 595   //__ untested("generate_exception_handler_common");
 596   Register Rexception = R17_tos;
 597 
 598   // Expression stack must be empty before entering the VM if an exception happened.
 599   __ empty_expression_stack();
 600 
 601   __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
 602   if (pass_oop) {
 603     __ mr(R5_ARG3, Rexception);
 604     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
 605   } else {
 606     __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
 607     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
 608   }
 609 
 610   // Throw exception.
 611   __ mr(R3_ARG1, Rexception);
 612   __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
 613   __ mtctr(R11_scratch1);
 614   __ bctr();
 615 
 616   return entry;
 617 }
 618 
 619 // This entry is returned to when a call returns to the interpreter.
 620 // When we arrive here, we expect that the callee stack frame is already popped.
 621 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 622   address entry = __ pc();
 623 
 624   // Move the value out of the return register back to the TOS cache of current frame.
 625   switch (state) {
 626     case ltos:
 627     case btos:
 628     case ztos:
 629     case ctos:
 630     case stos:
 631     case atos:
 632     case itos: __ mr(R17_tos, R3_RET); break;   // RET -> TOS cache
 633     case ftos:
 634     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 635     case vtos: break;                           // Nothing to do, this was a void return.
 636     default  : ShouldNotReachHere();
 637   }
 638 
 639   __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
 640   __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
 641   __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
 642 
 643   // Compiled code destroys templateTableBase, reload.
 644   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
 645 
 646   if (state == atos) {
 647     __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
 648   }
 649 
 650   const Register cache = R11_scratch1;
 651   const Register size  = R12_scratch2;
 652   __ get_cache_and_index_at_bcp(cache, 1, index_size);
 653 
 654   // Get least significant byte of 64 bit value:
 655 #if defined(VM_LITTLE_ENDIAN)
 656   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
 657 #else
 658   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
 659 #endif
 660   __ sldi(size, size, Interpreter::logStackElementSize);
 661   __ add(R15_esp, R15_esp, size);
 662 
 663  __ check_and_handle_popframe(R11_scratch1);
 664  __ check_and_handle_earlyret(R11_scratch1);
 665 
 666   __ dispatch_next(state, step);
 667   return entry;
 668 }
 669 
 670 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 671   address entry = __ pc();
 672   // If state != vtos, we're returning from a native method, which put it's result
 673   // into the result register. So move the value out of the return register back
 674   // to the TOS cache of current frame.
 675 
 676   switch (state) {
 677     case ltos:
 678     case btos:
 679     case ztos:
 680     case ctos:
 681     case stos:
 682     case atos:
 683     case itos: __ mr(R17_tos, R3_RET); break;   // GR_RET -> TOS cache
 684     case ftos:
 685     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 686     case vtos: break;                           // Nothing to do, this was a void return.
 687     default  : ShouldNotReachHere();
 688   }
 689 
 690   // Load LcpoolCache @@@ should be already set!
 691   __ get_constant_pool_cache(R27_constPoolCache);
 692 
 693   // Handle a pending exception, fall through if none.
 694   __ check_and_forward_exception(R11_scratch1, R12_scratch2);
 695 
 696   // Start executing bytecodes.
 697   if (continuation == NULL) {
 698     __ dispatch_next(state, step);
 699   } else {
 700     __ jump_to_entry(continuation, R11_scratch1);
 701   }
 702 
 703   return entry;
 704 }
 705 
 706 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 707   address entry = __ pc();
 708 
 709   __ push(state);
 710   __ call_VM(noreg, runtime_entry);
 711   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 712 
 713   return entry;
 714 }
 715 
 716 // Helpers for commoning out cases in the various type of method entries.
 717 
 718 // Increment invocation count & check for overflow.
 719 //
 720 // Note: checking for negative value instead of overflow
 721 //       so we have a 'sticky' overflow test.
 722 //
 723 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 724   // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
 725   Register Rscratch1   = R11_scratch1;
 726   Register Rscratch2   = R12_scratch2;
 727   Register R3_counters = R3_ARG1;
 728   Label done;
 729 
 730   if (TieredCompilation) {
 731     const int increment = InvocationCounter::count_increment;
 732     Label no_mdo;
 733     if (ProfileInterpreter) {
 734       const Register Rmdo = R3_counters;
 735       // If no method data exists, go to profile_continue.
 736       __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
 737       __ cmpdi(CCR0, Rmdo, 0);
 738       __ beq(CCR0, no_mdo);
 739 
 740       // Increment invocation counter in the MDO.
 741       const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 742       __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
 743       __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
 744       __ addi(Rscratch2, Rscratch2, increment);
 745       __ stw(Rscratch2, mdo_ic_offs, Rmdo);
 746       __ and_(Rscratch1, Rscratch2, Rscratch1);
 747       __ bne(CCR0, done);
 748       __ b(*overflow);
 749     }
 750 
 751     // Increment counter in MethodCounters*.
 752     const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 753     __ bind(no_mdo);
 754     __ get_method_counters(R19_method, R3_counters, done);
 755     __ lwz(Rscratch2, mo_ic_offs, R3_counters);
 756     __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
 757     __ addi(Rscratch2, Rscratch2, increment);
 758     __ stw(Rscratch2, mo_ic_offs, R3_counters);
 759     __ and_(Rscratch1, Rscratch2, Rscratch1);
 760     __ beq(CCR0, *overflow);
 761 
 762     __ bind(done);
 763 
 764   } else {
 765 
 766     // Update standard invocation counters.
 767     Register Rsum_ivc_bec = R4_ARG2;
 768     __ get_method_counters(R19_method, R3_counters, done);
 769     __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
 770     // Increment interpreter invocation counter.
 771     if (ProfileInterpreter) {  // %%% Merge this into methodDataOop.
 772       __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
 773       __ addi(R12_scratch2, R12_scratch2, 1);
 774       __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
 775     }
 776     // Check if we must create a method data obj.
 777     if (ProfileInterpreter && profile_method != NULL) {
 778       const Register profile_limit = Rscratch1;
 779       __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters);
 780       // Test to see if we should create a method data oop.
 781       __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
 782       __ blt(CCR0, *profile_method_continue);
 783       // If no method data exists, go to profile_method.
 784       __ test_method_data_pointer(*profile_method);
 785     }
 786     // Finally check for counter overflow.
 787     if (overflow) {
 788       const Register invocation_limit = Rscratch1;
 789       __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters);
 790       __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
 791       __ bge(CCR0, *overflow);
 792     }
 793 
 794     __ bind(done);
 795   }
 796 }
 797 
 798 // Generate code to initiate compilation on invocation counter overflow.
 799 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
 800   // Generate code to initiate compilation on the counter overflow.
 801 
 802   // InterpreterRuntime::frequency_counter_overflow takes one arguments,
 803   // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
 804   // We pass zero in.
 805   // The call returns the address of the verified entry point for the method or NULL
 806   // if the compilation did not complete (either went background or bailed out).
 807   //
 808   // Unlike the C++ interpreter above: Check exceptions!
 809   // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
 810   // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
 811 
 812   __ li(R4_ARG2, 0);
 813   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
 814 
 815   // Returns verified_entry_point or NULL.
 816   // We ignore it in any case.
 817   __ b(continue_entry);
 818 }
 819 
 820 // See if we've got enough room on the stack for locals plus overhead below
 821 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 822 // without going through the signal handler, i.e., reserved and yellow zones
 823 // will not be made usable. The shadow zone must suffice to handle the
 824 // overflow.
 825 //
 826 // Kills Rmem_frame_size, Rscratch1.
 827 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
 828   Label done;
 829   assert_different_registers(Rmem_frame_size, Rscratch1);
 830 
 831   BLOCK_COMMENT("stack_overflow_check_with_compare {");
 832   __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
 833   __ ld(Rscratch1, thread_(stack_overflow_limit));
 834   __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
 835   __ bgt(CCR0/*is_stack_overflow*/, done);
 836 
 837   // The stack overflows. Load target address of the runtime stub and call it.
 838   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order");
 839   __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0);
 840   __ mtctr(Rscratch1);
 841   // Restore caller_sp.
 842 #ifdef ASSERT
 843   __ ld(Rscratch1, 0, R1_SP);
 844   __ ld(R0, 0, R21_sender_SP);
 845   __ cmpd(CCR0, R0, Rscratch1);
 846   __ asm_assert_eq("backlink", 0x547);
 847 #endif // ASSERT
 848   __ mr(R1_SP, R21_sender_SP);
 849   __ bctr();
 850 
 851   __ align(32, 12);
 852   __ bind(done);
 853   BLOCK_COMMENT("} stack_overflow_check_with_compare");
 854 }
 855 
 856 // Lock the current method, interpreter register window must be set up!
 857 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
 858   const Register Robj_to_lock = Rscratch2;
 859 
 860   {
 861     if (!flags_preloaded) {
 862       __ lwz(Rflags, method_(access_flags));
 863     }
 864 
 865 #ifdef ASSERT
 866     // Check if methods needs synchronization.
 867     {
 868       Label Lok;
 869       __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
 870       __ btrue(CCR0,Lok);
 871       __ stop("method doesn't need synchronization");
 872       __ bind(Lok);
 873     }
 874 #endif // ASSERT
 875   }
 876 
 877   // Get synchronization object to Rscratch2.
 878   {
 879     Label Lstatic;
 880     Label Ldone;
 881 
 882     __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
 883     __ btrue(CCR0, Lstatic);
 884 
 885     // Non-static case: load receiver obj from stack and we're done.
 886     __ ld(Robj_to_lock, R18_locals);
 887     __ b(Ldone);
 888 
 889     __ bind(Lstatic); // Static case: Lock the java mirror
 890     // Load mirror from interpreter frame.
 891     __ ld(Robj_to_lock, _abi(callers_sp), R1_SP);
 892     __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
 893 
 894     __ bind(Ldone);
 895     __ verify_oop(Robj_to_lock);
 896   }
 897 
 898   // Got the oop to lock => execute!
 899   __ add_monitor_to_stack(true, Rscratch1, R0);
 900 
 901   __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
 902   __ lock_object(R26_monitor, Robj_to_lock);
 903 }
 904 
 905 // Generate a fixed interpreter frame for pure interpreter
 906 // and I2N native transition frames.
 907 //
 908 // Before (stack grows downwards):
 909 //
 910 //         |  ...         |
 911 //         |------------- |
 912 //         |  java arg0   |
 913 //         |  ...         |
 914 //         |  java argn   |
 915 //         |              |   <-   R15_esp
 916 //         |              |
 917 //         |--------------|
 918 //         | abi_112      |
 919 //         |              |   <-   R1_SP
 920 //         |==============|
 921 //
 922 //
 923 // After:
 924 //
 925 //         |  ...         |
 926 //         |  java arg0   |<-   R18_locals
 927 //         |  ...         |
 928 //         |  java argn   |
 929 //         |--------------|
 930 //         |              |
 931 //         |  java locals |
 932 //         |              |
 933 //         |--------------|
 934 //         |  abi_48      |
 935 //         |==============|
 936 //         |              |
 937 //         |   istate     |
 938 //         |              |
 939 //         |--------------|
 940 //         |   monitor    |<-   R26_monitor
 941 //         |--------------|
 942 //         |              |<-   R15_esp
 943 //         | expression   |
 944 //         | stack        |
 945 //         |              |
 946 //         |--------------|
 947 //         |              |
 948 //         | abi_112      |<-   R1_SP
 949 //         |==============|
 950 //
 951 // The top most frame needs an abi space of 112 bytes. This space is needed,
 952 // since we call to c. The c function may spill their arguments to the caller
 953 // frame. When we call to java, we don't need these spill slots. In order to save
 954 // space on the stack, we resize the caller. However, java locals reside in
 955 // the caller frame and the frame has to be increased. The frame_size for the
 956 // current frame was calculated based on max_stack as size for the expression
 957 // stack. At the call, just a part of the expression stack might be used.
 958 // We don't want to waste this space and cut the frame back accordingly.
 959 // The resulting amount for resizing is calculated as follows:
 960 // resize =   (number_of_locals - number_of_arguments) * slot_size
 961 //          + (R1_SP - R15_esp) + 48
 962 //
 963 // The size for the callee frame is calculated:
 964 // framesize = 112 + max_stack + monitor + state_size
 965 //
 966 // maxstack:   Max number of slots on the expression stack, loaded from the method.
 967 // monitor:    We statically reserve room for one monitor object.
 968 // state_size: We save the current state of the interpreter to this area.
 969 //
 970 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
 971   Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
 972            top_frame_size      = R7_ARG5,
 973            Rconst_method       = R8_ARG6;
 974 
 975   assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
 976 
 977   __ ld(Rconst_method, method_(const));
 978   __ lhz(Rsize_of_parameters /* number of params */,
 979          in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
 980   if (native_call) {
 981     // If we're calling a native method, we reserve space for the worst-case signature
 982     // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
 983     // We add two slots to the parameter_count, one for the jni
 984     // environment and one for a possible native mirror.
 985     Label skip_native_calculate_max_stack;
 986     __ addi(top_frame_size, Rsize_of_parameters, 2);
 987     __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
 988     __ bge(CCR0, skip_native_calculate_max_stack);
 989     __ li(top_frame_size, Argument::n_register_parameters);
 990     __ bind(skip_native_calculate_max_stack);
 991     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 992     __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
 993     __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 994     assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
 995   } else {
 996     __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
 997     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 998     __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
 999     __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
1000     __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
1001     __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
1002     __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
1003     __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
1004   }
1005 
1006   // Compute top frame size.
1007   __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
1008 
1009   // Cut back area between esp and max_stack.
1010   __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
1011 
1012   __ round_to(top_frame_size, frame::alignment_in_bytes);
1013   __ round_to(parent_frame_resize, frame::alignment_in_bytes);
1014   // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
1015   // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
1016 
1017   if (!native_call) {
1018     // Stack overflow check.
1019     // Native calls don't need the stack size check since they have no
1020     // expression stack and the arguments are already on the stack and
1021     // we only add a handful of words to the stack.
1022     __ add(R11_scratch1, parent_frame_resize, top_frame_size);
1023     generate_stack_overflow_check(R11_scratch1, R12_scratch2);
1024   }
1025 
1026   // Set up interpreter state registers.
1027 
1028   __ add(R18_locals, R15_esp, Rsize_of_parameters);
1029   __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
1030   __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
1031 
1032   // Set method data pointer.
1033   if (ProfileInterpreter) {
1034     Label zero_continue;
1035     __ ld(R28_mdx, method_(method_data));
1036     __ cmpdi(CCR0, R28_mdx, 0);
1037     __ beq(CCR0, zero_continue);
1038     __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1039     __ bind(zero_continue);
1040   }
1041 
1042   if (native_call) {
1043     __ li(R14_bcp, 0); // Must initialize.
1044   } else {
1045     __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
1046   }
1047 
1048   // Resize parent frame.
1049   __ mflr(R12_scratch2);
1050   __ neg(parent_frame_resize, parent_frame_resize);
1051   __ resize_frame(parent_frame_resize, R11_scratch1);
1052   __ std(R12_scratch2, _abi(lr), R1_SP);
1053 
1054   // Get mirror and store it in the frame as GC root for this Method*.
1055   __ load_mirror_from_const_method(R12_scratch2, Rconst_method);
1056 
1057   __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
1058   __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
1059 
1060   // Store values.
1061   // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
1062   // in InterpreterMacroAssembler::call_from_interpreter.
1063   __ std(R19_method, _ijava_state_neg(method), R1_SP);
1064   __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP);
1065   __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1066   __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1067   __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
1068 
1069   // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1070   // be found in the frame after save_interpreter_state is done. This is always true
1071   // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1072   // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1073   // (Enhanced Stack Trace).
1074   // The signal handler does not save the interpreter state into the frame.
1075   __ li(R0, 0);
1076 #ifdef ASSERT
1077   // Fill remaining slots with constants.
1078   __ load_const_optimized(R11_scratch1, 0x5afe);
1079   __ load_const_optimized(R12_scratch2, 0xdead);
1080 #endif
1081   // We have to initialize some frame slots for native calls (accessed by GC).
1082   if (native_call) {
1083     __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
1084     __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1085     if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1086   }
1087 #ifdef ASSERT
1088   else {
1089     __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1090     __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
1091     __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
1092   }
1093   __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
1094   __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1095   __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
1096   __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
1097 #endif
1098   __ subf(R12_scratch2, top_frame_size, R1_SP);
1099   __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
1100   __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1101 
1102   // Push top frame.
1103   __ push_frame(top_frame_size, R11_scratch1);
1104 }
1105 
1106 // End of helpers
1107 
1108 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1109 
1110   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1111   bool use_instruction = false;
1112   address runtime_entry = NULL;
1113   int num_args = 1;
1114   bool double_precision = true;
1115 
1116   // PPC64 specific:
1117   switch (kind) {
1118     case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1119     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1120     case Interpreter::java_lang_math_fmaF:
1121     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1122     default: break; // Fall back to runtime call.
1123   }
1124 
1125   switch (kind) {
1126     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1127     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1128     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1129     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1130     case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt);  break;
1131     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1132     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1133     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1134     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1135     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1136     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1137     default: ShouldNotReachHere();
1138   }
1139 
1140   // Use normal entry if neither instruction nor runtime call is used.
1141   if (!use_instruction && runtime_entry == NULL) return NULL;
1142 
1143   address entry = __ pc();
1144 
1145   // Load arguments
1146   assert(num_args <= 13, "passed in registers");
1147   if (double_precision) {
1148     int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1149     for (int i = 0; i < num_args; ++i) {
1150       __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1151       offset -= 2 * Interpreter::stackElementSize;
1152     }
1153   } else {
1154     int offset = num_args * Interpreter::stackElementSize;
1155     for (int i = 0; i < num_args; ++i) {
1156       __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1157       offset -= Interpreter::stackElementSize;
1158     }
1159   }
1160 
1161   // Pop c2i arguments (if any) off when we return.
1162 #ifdef ASSERT
1163   __ ld(R9_ARG7, 0, R1_SP);
1164   __ ld(R10_ARG8, 0, R21_sender_SP);
1165   __ cmpd(CCR0, R9_ARG7, R10_ARG8);
1166   __ asm_assert_eq("backlink", 0x545);
1167 #endif // ASSERT
1168   __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1169 
1170   if (use_instruction) {
1171     switch (kind) {
1172       case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1);          break;
1173       case Interpreter::java_lang_math_abs:  __ fabs(F1_RET, F1);           break;
1174       case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1175       case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3);  break;
1176       default: ShouldNotReachHere();
1177     }
1178   } else {
1179     // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1180     //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1181     //__ call_c_and_return_to_caller(R12_scratch2);
1182 
1183     // Push a new C frame and save LR.
1184     __ save_LR_CR(R0);
1185     __ push_frame_reg_args(0, R11_scratch1);
1186 
1187     __ call_VM_leaf(runtime_entry);
1188 
1189     // Pop the C frame and restore LR.
1190     __ pop_frame();
1191     __ restore_LR_CR(R0);
1192   }
1193 
1194   __ blr();
1195 
1196   __ flush();
1197 
1198   return entry;
1199 }
1200 
1201 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1202   // Quick & dirty stack overflow checking: bang the stack & handle trap.
1203   // Note that we do the banging after the frame is setup, since the exception
1204   // handling code expects to find a valid interpreter frame on the stack.
1205   // Doing the banging earlier fails if the caller frame is not an interpreter
1206   // frame.
1207   // (Also, the exception throwing code expects to unlock any synchronized
1208   // method receiever, so do the banging after locking the receiver.)
1209 
1210   // Bang each page in the shadow zone. We can't assume it's been done for
1211   // an interpreter frame with greater than a page of locals, so each page
1212   // needs to be checked.  Only true for non-native.
1213   if (UseStackBanging) {
1214     const int page_size = os::vm_page_size();
1215     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1216     const int start_page = native_call ? n_shadow_pages : 1;
1217     BLOCK_COMMENT("bang_stack_shadow_pages:");
1218     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1219       __ bang_stack_with_offset(pages*page_size);
1220     }
1221   }
1222 }
1223 
1224 // Interpreter stub for calling a native method. (asm interpreter)
1225 // This sets up a somewhat different looking stack for calling the
1226 // native method than the typical interpreter frame setup.
1227 //
1228 // On entry:
1229 //   R19_method    - method
1230 //   R16_thread    - JavaThread*
1231 //   R15_esp       - intptr_t* sender tos
1232 //
1233 //   abstract stack (grows up)
1234 //     [  IJava (caller of JNI callee)  ]  <-- ASP
1235 //        ...
1236 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1237 
1238   address entry = __ pc();
1239 
1240   const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1241 
1242   // -----------------------------------------------------------------------------
1243   // Allocate a new frame that represents the native callee (i2n frame).
1244   // This is not a full-blown interpreter frame, but in particular, the
1245   // following registers are valid after this:
1246   // - R19_method
1247   // - R18_local (points to start of arguments to native function)
1248   //
1249   //   abstract stack (grows up)
1250   //     [  IJava (caller of JNI callee)  ]  <-- ASP
1251   //        ...
1252 
1253   const Register signature_handler_fd = R11_scratch1;
1254   const Register pending_exception    = R0;
1255   const Register result_handler_addr  = R31;
1256   const Register native_method_fd     = R11_scratch1;
1257   const Register access_flags         = R22_tmp2;
1258   const Register active_handles       = R11_scratch1; // R26_monitor saved to state.
1259   const Register sync_state           = R12_scratch2;
1260   const Register sync_state_addr      = sync_state;   // Address is dead after use.
1261   const Register suspend_flags        = R11_scratch1;
1262 
1263   //=============================================================================
1264   // Allocate new frame and initialize interpreter state.
1265 
1266   Label exception_return;
1267   Label exception_return_sync_check;
1268   Label stack_overflow_return;
1269 
1270   // Generate new interpreter state and jump to stack_overflow_return in case of
1271   // a stack overflow.
1272   //generate_compute_interpreter_state(stack_overflow_return);
1273 
1274   Register size_of_parameters = R22_tmp2;
1275 
1276   generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1277 
1278   //=============================================================================
1279   // Increment invocation counter. On overflow, entry to JNI method
1280   // will be compiled.
1281   Label invocation_counter_overflow, continue_after_compile;
1282   if (inc_counter) {
1283     if (synchronized) {
1284       // Since at this point in the method invocation the exception handler
1285       // would try to exit the monitor of synchronized methods which hasn't
1286       // been entered yet, we set the thread local variable
1287       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1288       // runtime, exception handling i.e. unlock_if_synchronized_method will
1289       // check this thread local flag.
1290       // This flag has two effects, one is to force an unwind in the topmost
1291       // interpreter frame and not perform an unlock while doing so.
1292       __ li(R0, 1);
1293       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1294     }
1295     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1296 
1297     BIND(continue_after_compile);
1298   }
1299 
1300   bang_stack_shadow_pages(true);
1301 
1302   if (inc_counter) {
1303     // Reset the _do_not_unlock_if_synchronized flag.
1304     if (synchronized) {
1305       __ li(R0, 0);
1306       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1307     }
1308   }
1309 
1310   // access_flags = method->access_flags();
1311   // Load access flags.
1312   assert(access_flags->is_nonvolatile(),
1313          "access_flags must be in a non-volatile register");
1314   // Type check.
1315   assert(4 == sizeof(AccessFlags), "unexpected field size");
1316   __ lwz(access_flags, method_(access_flags));
1317 
1318   // We don't want to reload R19_method and access_flags after calls
1319   // to some helper functions.
1320   assert(R19_method->is_nonvolatile(),
1321          "R19_method must be a non-volatile register");
1322 
1323   // Check for synchronized methods. Must happen AFTER invocation counter
1324   // check, so method is not locked if counter overflows.
1325 
1326   if (synchronized) {
1327     lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1328 
1329     // Update monitor in state.
1330     __ ld(R11_scratch1, 0, R1_SP);
1331     __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
1332   }
1333 
1334   // jvmti/jvmpi support
1335   __ notify_method_entry();
1336 
1337   //=============================================================================
1338   // Get and call the signature handler.
1339 
1340   __ ld(signature_handler_fd, method_(signature_handler));
1341   Label call_signature_handler;
1342 
1343   __ cmpdi(CCR0, signature_handler_fd, 0);
1344   __ bne(CCR0, call_signature_handler);
1345 
1346   // Method has never been called. Either generate a specialized
1347   // handler or point to the slow one.
1348   //
1349   // Pass parameter 'false' to avoid exception check in call_VM.
1350   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1351 
1352   // Check for an exception while looking up the target method. If we
1353   // incurred one, bail.
1354   __ ld(pending_exception, thread_(pending_exception));
1355   __ cmpdi(CCR0, pending_exception, 0);
1356   __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1357 
1358   // Reload signature handler, it may have been created/assigned in the meanwhile.
1359   __ ld(signature_handler_fd, method_(signature_handler));
1360   __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1361 
1362   BIND(call_signature_handler);
1363 
1364   // Before we call the signature handler we push a new frame to
1365   // protect the interpreter frame volatile registers when we return
1366   // from jni but before we can get back to Java.
1367 
1368   // First set the frame anchor while the SP/FP registers are
1369   // convenient and the slow signature handler can use this same frame
1370   // anchor.
1371 
1372   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1373   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1374 
1375   // Now the interpreter frame (and its call chain) have been
1376   // invalidated and flushed. We are now protected against eager
1377   // being enabled in native code. Even if it goes eager the
1378   // registers will be reloaded as clean and we will invalidate after
1379   // the call so no spurious flush should be possible.
1380 
1381   // Call signature handler and pass locals address.
1382   //
1383   // Our signature handlers copy required arguments to the C stack
1384   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1385   __ mr(R3_ARG1, R18_locals);
1386 #if !defined(ABI_ELFv2)
1387   __ ld(signature_handler_fd, 0, signature_handler_fd);
1388 #endif
1389 
1390   __ call_stub(signature_handler_fd);
1391 
1392   // Remove the register parameter varargs slots we allocated in
1393   // compute_interpreter_state. SP+16 ends up pointing to the ABI
1394   // outgoing argument area.
1395   //
1396   // Not needed on PPC64.
1397   //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
1398 
1399   assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1400   // Save across call to native method.
1401   __ mr(result_handler_addr, R3_RET);
1402 
1403   __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1404 
1405   // Set up fixed parameters and call the native method.
1406   // If the method is static, get mirror into R4_ARG2.
1407   {
1408     Label method_is_not_static;
1409     // Access_flags is non-volatile and still, no need to restore it.
1410 
1411     // Restore access flags.
1412     __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1413     __ bfalse(CCR0, method_is_not_static);
1414 
1415     __ ld(R11_scratch1, _abi(callers_sp), R1_SP);
1416     // Load mirror from interpreter frame.
1417     __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1418     // R4_ARG2 = &state->_oop_temp;
1419     __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1420     __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1421     BIND(method_is_not_static);
1422   }
1423 
1424   // At this point, arguments have been copied off the stack into
1425   // their JNI positions. Oops are boxed in-place on the stack, with
1426   // handles copied to arguments. The result handler address is in a
1427   // register.
1428 
1429   // Pass JNIEnv address as first parameter.
1430   __ addir(R3_ARG1, thread_(jni_environment));
1431 
1432   // Load the native_method entry before we change the thread state.
1433   __ ld(native_method_fd, method_(native_function));
1434 
1435   //=============================================================================
1436   // Transition from _thread_in_Java to _thread_in_native. As soon as
1437   // we make this change the safepoint code needs to be certain that
1438   // the last Java frame we established is good. The pc in that frame
1439   // just needs to be near here not an actual return address.
1440 
1441   // We use release_store_fence to update values like the thread state, where
1442   // we don't want the current thread to continue until all our prior memory
1443   // accesses (including the new thread state) are visible to other threads.
1444   __ li(R0, _thread_in_native);
1445   __ release();
1446 
1447   // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1448   __ stw(R0, thread_(thread_state));
1449 
1450   //=============================================================================
1451   // Call the native method. Argument registers must not have been
1452   // overwritten since "__ call_stub(signature_handler);" (except for
1453   // ARG1 and ARG2 for static methods).
1454   __ call_c(native_method_fd);
1455 
1456   __ li(R0, 0);
1457   __ ld(R11_scratch1, 0, R1_SP);
1458   __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1459   __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1460   __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1461 
1462   // Note: C++ interpreter needs the following here:
1463   // The frame_manager_lr field, which we use for setting the last
1464   // java frame, gets overwritten by the signature handler. Restore
1465   // it now.
1466   //__ get_PC_trash_LR(R11_scratch1);
1467   //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1468 
1469   // Because of GC R19_method may no longer be valid.
1470 
1471   // Block, if necessary, before resuming in _thread_in_Java state.
1472   // In order for GC to work, don't clear the last_Java_sp until after
1473   // blocking.
1474 
1475   //=============================================================================
1476   // Switch thread to "native transition" state before reading the
1477   // synchronization state. This additional state is necessary
1478   // because reading and testing the synchronization state is not
1479   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1480   // in _thread_in_native state, loads _not_synchronized and is
1481   // preempted. VM thread changes sync state to synchronizing and
1482   // suspends threads for GC. Thread A is resumed to finish this
1483   // native method, but doesn't block here since it didn't see any
1484   // synchronization in progress, and escapes.
1485 
1486   // We use release_store_fence to update values like the thread state, where
1487   // we don't want the current thread to continue until all our prior memory
1488   // accesses (including the new thread state) are visible to other threads.
1489   __ li(R0/*thread_state*/, _thread_in_native_trans);
1490   __ release();
1491   __ stw(R0/*thread_state*/, thread_(thread_state));
1492   if (UseMembar) {
1493     __ fence();
1494   }
1495   // Write serialization page so that the VM thread can do a pseudo remote
1496   // membar. We use the current thread pointer to calculate a thread
1497   // specific offset to write to within the page. This minimizes bus
1498   // traffic due to cache line collision.
1499   else {
1500     __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
1501   }
1502 
1503   // Now before we return to java we must look for a current safepoint
1504   // (a new safepoint can not start since we entered native_trans).
1505   // We must check here because a current safepoint could be modifying
1506   // the callers registers right this moment.
1507 
1508   // Acquire isn't strictly necessary here because of the fence, but
1509   // sync_state is declared to be volatile, so we do it anyway
1510   // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1511 
1512   Label do_safepoint, sync_check_done;
1513   // No synchronization in progress nor yet synchronized.
1514   __ safepoint_poll(do_safepoint, sync_state);
1515 
1516   // Not suspended.
1517   // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1518   __ lwz(suspend_flags, thread_(suspend_flags));
1519   __ cmpwi(CCR1, suspend_flags, 0);
1520   __ beq(CCR1, sync_check_done);
1521 
1522   __ bind(do_safepoint);
1523   __ isync();
1524   // Block. We do the call directly and leave the current
1525   // last_Java_frame setup undisturbed. We must save any possible
1526   // native result across the call. No oop is present.
1527 
1528   __ mr(R3_ARG1, R16_thread);
1529 #if defined(ABI_ELFv2)
1530   __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1531             relocInfo::none);
1532 #else
1533   __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
1534             relocInfo::none);
1535 #endif
1536 
1537   __ bind(sync_check_done);
1538 
1539   //=============================================================================
1540   // <<<<<< Back in Interpreter Frame >>>>>
1541 
1542   // We are in thread_in_native_trans here and back in the normal
1543   // interpreter frame. We don't have to do anything special about
1544   // safepoints and we can switch to Java mode anytime we are ready.
1545 
1546   // Note: frame::interpreter_frame_result has a dependency on how the
1547   // method result is saved across the call to post_method_exit. For
1548   // native methods it assumes that the non-FPU/non-void result is
1549   // saved in _native_lresult and a FPU result in _native_fresult. If
1550   // this changes then the interpreter_frame_result implementation
1551   // will need to be updated too.
1552 
1553   // On PPC64, we have stored the result directly after the native call.
1554 
1555   //=============================================================================
1556   // Back in Java
1557 
1558   // We use release_store_fence to update values like the thread state, where
1559   // we don't want the current thread to continue until all our prior memory
1560   // accesses (including the new thread state) are visible to other threads.
1561   __ li(R0/*thread_state*/, _thread_in_Java);
1562   __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1563   __ stw(R0/*thread_state*/, thread_(thread_state));
1564 
1565   if (CheckJNICalls) {
1566     // clear_pending_jni_exception_check
1567     __ load_const_optimized(R0, 0L);
1568     __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1569   }
1570 
1571   __ reset_last_Java_frame();
1572 
1573   // Jvmdi/jvmpi support. Whether we've got an exception pending or
1574   // not, and whether unlocking throws an exception or not, we notify
1575   // on native method exit. If we do have an exception, we'll end up
1576   // in the caller's context to handle it, so if we don't do the
1577   // notify here, we'll drop it on the floor.
1578   __ notify_method_exit(true/*native method*/,
1579                         ilgl /*illegal state (not used for native methods)*/,
1580                         InterpreterMacroAssembler::NotifyJVMTI,
1581                         false /*check_exceptions*/);
1582 
1583   //=============================================================================
1584   // Handle exceptions
1585 
1586   if (synchronized) {
1587     // Don't check for exceptions since we're still in the i2n frame. Do that
1588     // manually afterwards.
1589     __ unlock_object(R26_monitor, false); // Can also unlock methods.
1590   }
1591 
1592   // Reset active handles after returning from native.
1593   // thread->active_handles()->clear();
1594   __ ld(active_handles, thread_(active_handles));
1595   // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1596   __ li(R0, 0);
1597   __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
1598 
1599   Label exception_return_sync_check_already_unlocked;
1600   __ ld(R0/*pending_exception*/, thread_(pending_exception));
1601   __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1602   __ bne(CCR0, exception_return_sync_check_already_unlocked);
1603 
1604   //-----------------------------------------------------------------------------
1605   // No exception pending.
1606 
1607   // Move native method result back into proper registers and return.
1608   // Invoke result handler (may unbox/promote).
1609   __ ld(R11_scratch1, 0, R1_SP);
1610   __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1611   __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1612   __ call_stub(result_handler_addr);
1613 
1614   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1615 
1616   // Must use the return pc which was loaded from the caller's frame
1617   // as the VM uses return-pc-patching for deoptimization.
1618   __ mtlr(R0);
1619   __ blr();
1620 
1621   //-----------------------------------------------------------------------------
1622   // An exception is pending. We call into the runtime only if the
1623   // caller was not interpreted. If it was interpreted the
1624   // interpreter will do the correct thing. If it isn't interpreted
1625   // (call stub/compiled code) we will change our return and continue.
1626 
1627   BIND(exception_return_sync_check);
1628 
1629   if (synchronized) {
1630     // Don't check for exceptions since we're still in the i2n frame. Do that
1631     // manually afterwards.
1632     __ unlock_object(R26_monitor, false); // Can also unlock methods.
1633   }
1634   BIND(exception_return_sync_check_already_unlocked);
1635 
1636   const Register return_pc = R31;
1637 
1638   __ ld(return_pc, 0, R1_SP);
1639   __ ld(return_pc, _abi(lr), return_pc);
1640 
1641   // Get the address of the exception handler.
1642   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1643                   R16_thread,
1644                   return_pc /* return pc */);
1645   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1646 
1647   // Load the PC of the the exception handler into LR.
1648   __ mtlr(R3_RET);
1649 
1650   // Load exception into R3_ARG1 and clear pending exception in thread.
1651   __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1652   __ li(R4_ARG2, 0);
1653   __ std(R4_ARG2, thread_(pending_exception));
1654 
1655   // Load the original return pc into R4_ARG2.
1656   __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1657 
1658   // Return to exception handler.
1659   __ blr();
1660 
1661   //=============================================================================
1662   // Counter overflow.
1663 
1664   if (inc_counter) {
1665     // Handle invocation counter overflow.
1666     __ bind(invocation_counter_overflow);
1667 
1668     generate_counter_overflow(continue_after_compile);
1669   }
1670 
1671   return entry;
1672 }
1673 
1674 // Generic interpreted method entry to (asm) interpreter.
1675 //
1676 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1677   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1678   address entry = __ pc();
1679   // Generate the code to allocate the interpreter stack frame.
1680   Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1681            Rsize_of_locals     = R5_ARG3; // Written by generate_fixed_frame.
1682 
1683   // Does also a stack check to assure this frame fits on the stack.
1684   generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1685 
1686   // --------------------------------------------------------------------------
1687   // Zero out non-parameter locals.
1688   // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1689   // worth to ask the flag, just do it.
1690   Register Rslot_addr = R6_ARG4,
1691            Rnum       = R7_ARG5;
1692   Label Lno_locals, Lzero_loop;
1693 
1694   // Set up the zeroing loop.
1695   __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1696   __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1697   __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1698   __ beq(CCR0, Lno_locals);
1699   __ li(R0, 0);
1700   __ mtctr(Rnum);
1701 
1702   // The zero locals loop.
1703   __ bind(Lzero_loop);
1704   __ std(R0, 0, Rslot_addr);
1705   __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1706   __ bdnz(Lzero_loop);
1707 
1708   __ bind(Lno_locals);
1709 
1710   // --------------------------------------------------------------------------
1711   // Counter increment and overflow check.
1712   Label invocation_counter_overflow,
1713         profile_method,
1714         profile_method_continue;
1715   if (inc_counter || ProfileInterpreter) {
1716 
1717     Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1718     if (synchronized) {
1719       // Since at this point in the method invocation the exception handler
1720       // would try to exit the monitor of synchronized methods which hasn't
1721       // been entered yet, we set the thread local variable
1722       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1723       // runtime, exception handling i.e. unlock_if_synchronized_method will
1724       // check this thread local flag.
1725       // This flag has two effects, one is to force an unwind in the topmost
1726       // interpreter frame and not perform an unlock while doing so.
1727       __ li(R0, 1);
1728       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1729     }
1730 
1731     // Argument and return type profiling.
1732     __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1733 
1734     // Increment invocation counter and check for overflow.
1735     if (inc_counter) {
1736       generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1737     }
1738 
1739     __ bind(profile_method_continue);
1740   }
1741 
1742   bang_stack_shadow_pages(false);
1743 
1744   if (inc_counter || ProfileInterpreter) {
1745     // Reset the _do_not_unlock_if_synchronized flag.
1746     if (synchronized) {
1747       __ li(R0, 0);
1748       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1749     }
1750   }
1751 
1752   // --------------------------------------------------------------------------
1753   // Locking of synchronized methods. Must happen AFTER invocation_counter
1754   // check and stack overflow check, so method is not locked if overflows.
1755   if (synchronized) {
1756     lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1757   }
1758 #ifdef ASSERT
1759   else {
1760     Label Lok;
1761     __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1762     __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1763     __ asm_assert_eq("method needs synchronization", 0x8521);
1764     __ bind(Lok);
1765   }
1766 #endif // ASSERT
1767 
1768   __ verify_thread();
1769 
1770   // --------------------------------------------------------------------------
1771   // JVMTI support
1772   __ notify_method_entry();
1773 
1774   // --------------------------------------------------------------------------
1775   // Start executing instructions.
1776   __ dispatch_next(vtos);
1777 
1778   // --------------------------------------------------------------------------
1779   // Out of line counter overflow and MDO creation code.
1780   if (ProfileInterpreter) {
1781     // We have decided to profile this method in the interpreter.
1782     __ bind(profile_method);
1783     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1784     __ set_method_data_pointer_for_bcp();
1785     __ b(profile_method_continue);
1786   }
1787 
1788   if (inc_counter) {
1789     // Handle invocation counter overflow.
1790     __ bind(invocation_counter_overflow);
1791     generate_counter_overflow(profile_method_continue);
1792   }
1793   return entry;
1794 }
1795 
1796 // CRC32 Intrinsics.
1797 //
1798 // Contract on scratch and work registers.
1799 // =======================================
1800 //
1801 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1802 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1803 // You can't rely on these registers across calls.
1804 //
1805 // The generators for CRC32_update and for CRC32_updateBytes use the
1806 // scratch/work register set internally, passing the work registers
1807 // as arguments to the MacroAssembler emitters as required.
1808 //
1809 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1810 // Their contents is not constant but may change according to the requirements
1811 // of the emitted code.
1812 //
1813 // All other registers from the scratch/work register set are used "internally"
1814 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1815 // Basically, only R3_RET contains a defined value which is the function result.
1816 //
1817 /**
1818  * Method entry for static native methods:
1819  *   int java.util.zip.CRC32.update(int crc, int b)
1820  */
1821 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1822   if (UseCRC32Intrinsics) {
1823     address start = __ pc();  // Remember stub start address (is rtn value).
1824     Label slow_path;
1825 
1826     // Safepoint check
1827     const Register sync_state = R11_scratch1;
1828     __ safepoint_poll(slow_path, sync_state);
1829 
1830     // We don't generate local frame and don't align stack because
1831     // we not even call stub code (we generate the code inline)
1832     // and there is no safepoint on this path.
1833 
1834     // Load java parameters.
1835     // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1836     const Register argP    = R15_esp;
1837     const Register crc     = R3_ARG1;  // crc value
1838     const Register data    = R4_ARG2;  // address of java byte value (kernel_crc32 needs address)
1839     const Register dataLen = R5_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
1840     const Register table   = R6_ARG4;  // address of crc32 table
1841     const Register tmp     = dataLen;  // Reuse unused len register to show we don't actually need a separate tmp here.
1842 
1843     BLOCK_COMMENT("CRC32_update {");
1844 
1845     // Arguments are reversed on java expression stack
1846 #ifdef VM_LITTLE_ENDIAN
1847     __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1848                                        // Being passed as an int, the single byte is at offset +0.
1849 #else
1850     __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
1851                                        // Being passed from java as an int, the single byte is at offset +3.
1852 #endif
1853     __ lwz(crc,  2*wordSize, argP);    // Current crc state, zero extend to 64 bit to have a clean register.
1854 
1855     StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1856     __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp, true);
1857 
1858     // Restore caller sp for c2i case and return.
1859     __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1860     __ blr();
1861 
1862     // Generate a vanilla native entry as the slow path.
1863     BLOCK_COMMENT("} CRC32_update");
1864     BIND(slow_path);
1865     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1866     return start;
1867   }
1868 
1869   return NULL;
1870 }
1871 
1872 /**
1873  * Method entry for static native methods:
1874  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1875  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1876  */
1877 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1878   if (UseCRC32Intrinsics) {
1879     address start = __ pc();  // Remember stub start address (is rtn value).
1880     Label slow_path;
1881 
1882     // Safepoint check
1883     const Register sync_state = R11_scratch1;
1884     __ safepoint_poll(slow_path, sync_state);
1885 
1886     // We don't generate local frame and don't align stack because
1887     // we not even call stub code (we generate the code inline)
1888     // and there is no safepoint on this path.
1889 
1890     // Load parameters.
1891     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1892     const Register argP    = R15_esp;
1893     const Register crc     = R3_ARG1;  // crc value
1894     const Register data    = R4_ARG2;  // address of java byte array
1895     const Register dataLen = R5_ARG3;  // source data len
1896     const Register table   = R6_ARG4;  // address of crc32 table
1897 
1898     const Register t0      = R9;       // scratch registers for crc calculation
1899     const Register t1      = R10;
1900     const Register t2      = R11;
1901     const Register t3      = R12;
1902 
1903     const Register tc0     = R2;       // registers to hold pre-calculated column addresses
1904     const Register tc1     = R7;
1905     const Register tc2     = R8;
1906     const Register tc3     = table;    // table address is reconstructed at the end of kernel_crc32_* emitters
1907 
1908     const Register tmp     = t0;       // Only used very locally to calculate byte buffer address.
1909 
1910     // Arguments are reversed on java expression stack.
1911     // Calculate address of start element.
1912     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1913       BLOCK_COMMENT("CRC32_updateByteBuffer {");
1914       // crc     @ (SP + 5W) (32bit)
1915       // buf     @ (SP + 3W) (64bit ptr to long array)
1916       // off     @ (SP + 2W) (32bit)
1917       // dataLen @ (SP + 1W) (32bit)
1918       // data = buf + off
1919       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1920       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1921       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1922       __ lwz( crc,     5*wordSize, argP);  // current crc state
1923       __ add( data, data, tmp);            // Add byte buffer offset.
1924     } else {                                                         // Used for "updateBytes update".
1925       BLOCK_COMMENT("CRC32_updateBytes {");
1926       // crc     @ (SP + 4W) (32bit)
1927       // buf     @ (SP + 3W) (64bit ptr to byte array)
1928       // off     @ (SP + 2W) (32bit)
1929       // dataLen @ (SP + 1W) (32bit)
1930       // data = buf + off + base_offset
1931       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1932       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1933       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1934       __ add( data, data, tmp);            // add byte buffer offset
1935       __ lwz( crc,     4*wordSize, argP);  // current crc state
1936       __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1937     }
1938 
1939     StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
1940 
1941     // Performance measurements show the 1word and 2word variants to be almost equivalent,
1942     // with very light advantages for the 1word variant. We chose the 1word variant for
1943     // code compactness.
1944     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, true);
1945 
1946     // Restore caller sp for c2i case and return.
1947     __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
1948     __ blr();
1949 
1950     // Generate a vanilla native entry as the slow path.
1951     BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1952     BIND(slow_path);
1953     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1954     return start;
1955   }
1956 
1957   return NULL;
1958 }
1959 
1960 
1961 /**
1962  * Method entry for intrinsic-candidate (non-native) methods:
1963  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
1964  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1965  * Unlike CRC32, CRC32C does not have any methods marked as native
1966  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1967  **/
1968 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1969   if (UseCRC32CIntrinsics) {
1970     address start = __ pc();  // Remember stub start address (is rtn value).
1971 
1972     // We don't generate local frame and don't align stack because
1973     // we not even call stub code (we generate the code inline)
1974     // and there is no safepoint on this path.
1975 
1976     // Load parameters.
1977     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1978     const Register argP    = R15_esp;
1979     const Register crc     = R3_ARG1;  // crc value
1980     const Register data    = R4_ARG2;  // address of java byte array
1981     const Register dataLen = R5_ARG3;  // source data len
1982     const Register table   = R6_ARG4;  // address of crc32c table
1983 
1984     const Register t0      = R9;       // scratch registers for crc calculation
1985     const Register t1      = R10;
1986     const Register t2      = R11;
1987     const Register t3      = R12;
1988 
1989     const Register tc0     = R2;       // registers to hold pre-calculated column addresses
1990     const Register tc1     = R7;
1991     const Register tc2     = R8;
1992     const Register tc3     = table;    // table address is reconstructed at the end of kernel_crc32_* emitters
1993 
1994     const Register tmp     = t0;       // Only used very locally to calculate byte buffer address.
1995 
1996     // Arguments are reversed on java expression stack.
1997     // Calculate address of start element.
1998     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1999       BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
2000       // crc     @ (SP + 5W) (32bit)
2001       // buf     @ (SP + 3W) (64bit ptr to long array)
2002       // off     @ (SP + 2W) (32bit)
2003       // dataLen @ (SP + 1W) (32bit)
2004       // data = buf + off
2005       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
2006       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
2007       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
2008       __ lwz( crc,     5*wordSize, argP);  // current crc state
2009       __ add( data, data, tmp);            // Add byte buffer offset.
2010       __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
2011     } else {                                                         // Used for "updateBytes update".
2012       BLOCK_COMMENT("CRC32C_updateBytes {");
2013       // crc     @ (SP + 4W) (32bit)
2014       // buf     @ (SP + 3W) (64bit ptr to byte array)
2015       // off     @ (SP + 2W) (32bit)
2016       // dataLen @ (SP + 1W) (32bit)
2017       // data = buf + off + base_offset
2018       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
2019       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
2020       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
2021       __ add( data, data, tmp);            // add byte buffer offset
2022       __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
2023       __ lwz( crc,     4*wordSize, argP);  // current crc state
2024       __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
2025     }
2026 
2027     StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table);
2028 
2029     // Performance measurements show the 1word and 2word variants to be almost equivalent,
2030     // with very light advantages for the 1word variant. We chose the 1word variant for
2031     // code compactness.
2032     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, false);
2033 
2034     // Restore caller sp for c2i case and return.
2035     __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
2036     __ blr();
2037 
2038     BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
2039     return start;
2040   }
2041 
2042   return NULL;
2043 }
2044 
2045 // =============================================================================
2046 // Exceptions
2047 
2048 void TemplateInterpreterGenerator::generate_throw_exception() {
2049   Register Rexception    = R17_tos,
2050            Rcontinuation = R3_RET;
2051 
2052   // --------------------------------------------------------------------------
2053   // Entry point if an method returns with a pending exception (rethrow).
2054   Interpreter::_rethrow_exception_entry = __ pc();
2055   {
2056     __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
2057     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2058     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2059 
2060     // Compiled code destroys templateTableBase, reload.
2061     __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2062   }
2063 
2064   // Entry point if a interpreted method throws an exception (throw).
2065   Interpreter::_throw_exception_entry = __ pc();
2066   {
2067     __ mr(Rexception, R3_RET);
2068 
2069     __ verify_thread();
2070     __ verify_oop(Rexception);
2071 
2072     // Expression stack must be empty before entering the VM in case of an exception.
2073     __ empty_expression_stack();
2074     // Find exception handler address and preserve exception oop.
2075     // Call C routine to find handler and jump to it.
2076     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2077     __ mtctr(Rcontinuation);
2078     // Push exception for exception handler bytecodes.
2079     __ push_ptr(Rexception);
2080 
2081     // Jump to exception handler (may be remove activation entry!).
2082     __ bctr();
2083   }
2084 
2085   // If the exception is not handled in the current frame the frame is
2086   // removed and the exception is rethrown (i.e. exception
2087   // continuation is _rethrow_exception).
2088   //
2089   // Note: At this point the bci is still the bxi for the instruction
2090   // which caused the exception and the expression stack is
2091   // empty. Thus, for any VM calls at this point, GC will find a legal
2092   // oop map (with empty expression stack).
2093 
2094   // In current activation
2095   // tos: exception
2096   // bcp: exception bcp
2097 
2098   // --------------------------------------------------------------------------
2099   // JVMTI PopFrame support
2100 
2101   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2102   {
2103     // Set the popframe_processing bit in popframe_condition indicating that we are
2104     // currently handling popframe, so that call_VMs that may happen later do not
2105     // trigger new popframe handling cycles.
2106     __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2107     __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2108     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2109 
2110     // Empty the expression stack, as in normal exception handling.
2111     __ empty_expression_stack();
2112     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2113 
2114     // Check to see whether we are returning to a deoptimized frame.
2115     // (The PopFrame call ensures that the caller of the popped frame is
2116     // either interpreted or compiled and deoptimizes it if compiled.)
2117     // Note that we don't compare the return PC against the
2118     // deoptimization blob's unpack entry because of the presence of
2119     // adapter frames in C2.
2120     Label Lcaller_not_deoptimized;
2121     Register return_pc = R3_ARG1;
2122     __ ld(return_pc, 0, R1_SP);
2123     __ ld(return_pc, _abi(lr), return_pc);
2124     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2125     __ cmpdi(CCR0, R3_RET, 0);
2126     __ bne(CCR0, Lcaller_not_deoptimized);
2127 
2128     // The deoptimized case.
2129     // In this case, we can't call dispatch_next() after the frame is
2130     // popped, but instead must save the incoming arguments and restore
2131     // them after deoptimization has occurred.
2132     __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2133     __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2134     __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2135     __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2136     __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2137     // Save these arguments.
2138     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2139 
2140     // Inform deoptimization that it is responsible for restoring these arguments.
2141     __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2142     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2143 
2144     // Return from the current method into the deoptimization blob. Will eventually
2145     // end up in the deopt interpeter entry, deoptimization prepared everything that
2146     // we will reexecute the call that called us.
2147     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2148     __ mtlr(return_pc);
2149     __ blr();
2150 
2151     // The non-deoptimized case.
2152     __ bind(Lcaller_not_deoptimized);
2153 
2154     // Clear the popframe condition flag.
2155     __ li(R0, 0);
2156     __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2157 
2158     // Get out of the current method and re-execute the call that called us.
2159     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2160     __ restore_interpreter_state(R11_scratch1);
2161     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
2162     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
2163     if (ProfileInterpreter) {
2164       __ set_method_data_pointer_for_bcp();
2165       __ ld(R11_scratch1, 0, R1_SP);
2166       __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2167     }
2168 #if INCLUDE_JVMTI
2169     Label L_done;
2170 
2171     __ lbz(R11_scratch1, 0, R14_bcp);
2172     __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2173     __ bne(CCR0, L_done);
2174 
2175     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2176     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2177     __ ld(R4_ARG2, 0, R18_locals);
2178     __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
2179     __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
2180     __ cmpdi(CCR0, R4_ARG2, 0);
2181     __ beq(CCR0, L_done);
2182     __ std(R4_ARG2, wordSize, R15_esp);
2183     __ bind(L_done);
2184 #endif // INCLUDE_JVMTI
2185     __ dispatch_next(vtos);
2186   }
2187   // end of JVMTI PopFrame support
2188 
2189   // --------------------------------------------------------------------------
2190   // Remove activation exception entry.
2191   // This is jumped to if an interpreted method can't handle an exception itself
2192   // (we come from the throw/rethrow exception entry above). We're going to call
2193   // into the VM to find the exception handler in the caller, pop the current
2194   // frame and return the handler we calculated.
2195   Interpreter::_remove_activation_entry = __ pc();
2196   {
2197     __ pop_ptr(Rexception);
2198     __ verify_thread();
2199     __ verify_oop(Rexception);
2200     __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2201 
2202     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2203     __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2204 
2205     __ get_vm_result(Rexception);
2206 
2207     // We are done with this activation frame; find out where to go next.
2208     // The continuation point will be an exception handler, which expects
2209     // the following registers set up:
2210     //
2211     // RET:  exception oop
2212     // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2213 
2214     Register return_pc = R31; // Needs to survive the runtime call.
2215     __ ld(return_pc, 0, R1_SP);
2216     __ ld(return_pc, _abi(lr), return_pc);
2217     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2218 
2219     // Remove the current activation.
2220     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2221 
2222     __ mr(R4_ARG2, return_pc);
2223     __ mtlr(R3_RET);
2224     __ mr(R3_RET, Rexception);
2225     __ blr();
2226   }
2227 }
2228 
2229 // JVMTI ForceEarlyReturn support.
2230 // Returns "in the middle" of a method with a "fake" return value.
2231 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2232 
2233   Register Rscratch1 = R11_scratch1,
2234            Rscratch2 = R12_scratch2;
2235 
2236   address entry = __ pc();
2237   __ empty_expression_stack();
2238 
2239   __ load_earlyret_value(state, Rscratch1);
2240 
2241   __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2242   // Clear the earlyret state.
2243   __ li(R0, 0);
2244   __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2245 
2246   __ remove_activation(state, false, false);
2247   // Copied from TemplateTable::_return.
2248   // Restoration of lr done by remove_activation.
2249   switch (state) {
2250     // Narrow result if state is itos but result type is smaller.
2251     case btos:
2252     case ztos:
2253     case ctos:
2254     case stos:
2255     case itos: __ narrow(R17_tos); /* fall through */
2256     case ltos:
2257     case atos: __ mr(R3_RET, R17_tos); break;
2258     case ftos:
2259     case dtos: __ fmr(F1_RET, F15_ftos); break;
2260     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2261                // to get visible before the reference to the object gets stored anywhere.
2262                __ membar(Assembler::StoreStore); break;
2263     default  : ShouldNotReachHere();
2264   }
2265   __ blr();
2266 
2267   return entry;
2268 } // end of ForceEarlyReturn support
2269 
2270 //-----------------------------------------------------------------------------
2271 // Helper for vtos entry point generation
2272 
2273 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2274                                                          address& bep,
2275                                                          address& cep,
2276                                                          address& sep,
2277                                                          address& aep,
2278                                                          address& iep,
2279                                                          address& lep,
2280                                                          address& fep,
2281                                                          address& dep,
2282                                                          address& vep) {
2283   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2284   Label L;
2285 
2286   aep = __ pc();  __ push_ptr();  __ b(L);
2287   fep = __ pc();  __ push_f();    __ b(L);
2288   dep = __ pc();  __ push_d();    __ b(L);
2289   lep = __ pc();  __ push_l();    __ b(L);
2290   __ align(32, 12, 24); // align L
2291   bep = cep = sep =
2292   iep = __ pc();  __ push_i();
2293   vep = __ pc();
2294   __ bind(L);
2295   generate_and_dispatch(t);
2296 }
2297 
2298 //-----------------------------------------------------------------------------
2299 
2300 // Non-product code
2301 #ifndef PRODUCT
2302 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2303   //__ flush_bundle();
2304   address entry = __ pc();
2305 
2306   const char *bname = NULL;
2307   uint tsize = 0;
2308   switch(state) {
2309   case ftos:
2310     bname = "trace_code_ftos {";
2311     tsize = 2;
2312     break;
2313   case btos:
2314     bname = "trace_code_btos {";
2315     tsize = 2;
2316     break;
2317   case ztos:
2318     bname = "trace_code_ztos {";
2319     tsize = 2;
2320     break;
2321   case ctos:
2322     bname = "trace_code_ctos {";
2323     tsize = 2;
2324     break;
2325   case stos:
2326     bname = "trace_code_stos {";
2327     tsize = 2;
2328     break;
2329   case itos:
2330     bname = "trace_code_itos {";
2331     tsize = 2;
2332     break;
2333   case ltos:
2334     bname = "trace_code_ltos {";
2335     tsize = 3;
2336     break;
2337   case atos:
2338     bname = "trace_code_atos {";
2339     tsize = 2;
2340     break;
2341   case vtos:
2342     // Note: In case of vtos, the topmost of stack value could be a int or doubl
2343     // In case of a double (2 slots) we won't see the 2nd stack value.
2344     // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2345     bname = "trace_code_vtos {";
2346     tsize = 2;
2347 
2348     break;
2349   case dtos:
2350     bname = "trace_code_dtos {";
2351     tsize = 3;
2352     break;
2353   default:
2354     ShouldNotReachHere();
2355   }
2356   BLOCK_COMMENT(bname);
2357 
2358   // Support short-cut for TraceBytecodesAt.
2359   // Don't call into the VM if we don't want to trace to speed up things.
2360   Label Lskip_vm_call;
2361   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2362     int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2363     int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2364     __ ld(R11_scratch1, offs1, R11_scratch1);
2365     __ lwa(R12_scratch2, offs2, R12_scratch2);
2366     __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2367     __ blt(CCR0, Lskip_vm_call);
2368   }
2369 
2370   __ push(state);
2371   // Load 2 topmost expression stack values.
2372   __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2373   __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2374   __ mflr(R31);
2375   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2376   __ mtlr(R31);
2377   __ pop(state);
2378 
2379   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2380     __ bind(Lskip_vm_call);
2381   }
2382   __ blr();
2383   BLOCK_COMMENT("} trace_code");
2384   return entry;
2385 }
2386 
2387 void TemplateInterpreterGenerator::count_bytecode() {
2388   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2389   __ lwz(R12_scratch2, offs, R11_scratch1);
2390   __ addi(R12_scratch2, R12_scratch2, 1);
2391   __ stw(R12_scratch2, offs, R11_scratch1);
2392 }
2393 
2394 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2395   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2396   __ lwz(R12_scratch2, offs, R11_scratch1);
2397   __ addi(R12_scratch2, R12_scratch2, 1);
2398   __ stw(R12_scratch2, offs, R11_scratch1);
2399 }
2400 
2401 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2402   const Register addr = R11_scratch1,
2403                  tmp  = R12_scratch2;
2404   // Get index, shift out old bytecode, bring in new bytecode, and store it.
2405   // _index = (_index >> log2_number_of_codes) |
2406   //          (bytecode << log2_number_of_codes);
2407   int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2408   __ lwz(tmp, offs1, addr);
2409   __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2410   __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2411   __ stw(tmp, offs1, addr);
2412 
2413   // Bump bucket contents.
2414   // _counters[_index] ++;
2415   int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2416   __ sldi(tmp, tmp, LogBytesPerInt);
2417   __ add(addr, tmp, addr);
2418   __ lwz(tmp, offs2, addr);
2419   __ addi(tmp, tmp, 1);
2420   __ stw(tmp, offs2, addr);
2421 }
2422 
2423 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2424   // Call a little run-time stub to avoid blow-up for each bytecode.
2425   // The run-time runtime saves the right registers, depending on
2426   // the tosca in-state for the given template.
2427 
2428   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2429          "entry must have been generated");
2430 
2431   // Note: we destroy LR here.
2432   __ bl(Interpreter::trace_code(t->tos_in()));
2433 }
2434 
2435 void TemplateInterpreterGenerator::stop_interpreter_at() {
2436   Label L;
2437   int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2438   int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2439   __ ld(R11_scratch1, offs1, R11_scratch1);
2440   __ lwa(R12_scratch2, offs2, R12_scratch2);
2441   __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2442   __ bne(CCR0, L);
2443   __ illtrap();
2444   __ bind(L);
2445 }
2446 
2447 #endif // !PRODUCT
--- EOF ---