1 /*
   2  * Copyright (c) 2014, 2024, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2015, 2024 SAP SE. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/macroAssembler.inline.hpp"
  28 #include "classfile/javaClasses.hpp"
  29 #include "compiler/disassembler.hpp"
  30 #include "gc/shared/barrierSetAssembler.hpp"
  31 #include "interpreter/bytecodeHistogram.hpp"
  32 #include "interpreter/interpreter.hpp"
  33 #include "interpreter/interpreterRuntime.hpp"
  34 #include "interpreter/interp_masm.hpp"
  35 #include "interpreter/templateInterpreterGenerator.hpp"
  36 #include "interpreter/templateTable.hpp"
  37 #include "oops/arrayOop.hpp"
  38 #include "oops/method.hpp"
  39 #include "oops/methodCounters.hpp"
  40 #include "oops/methodData.hpp"
  41 #include "oops/oop.inline.hpp"
  42 #include "oops/resolvedIndyEntry.hpp"
  43 #include "oops/resolvedMethodEntry.hpp"
  44 #include "prims/jvmtiExport.hpp"
  45 #include "prims/jvmtiThreadState.hpp"
  46 #include "runtime/arguments.hpp"
  47 #include "runtime/deoptimization.hpp"
  48 #include "runtime/frame.inline.hpp"
  49 #include "runtime/jniHandles.hpp"
  50 #include "runtime/sharedRuntime.hpp"
  51 #include "runtime/stubRoutines.hpp"
  52 #include "runtime/synchronizer.hpp"
  53 #include "runtime/timer.hpp"
  54 #include "runtime/vframeArray.hpp"
  55 #include "runtime/vm_version.hpp"
  56 #include "utilities/debug.hpp"
  57 #include "utilities/macros.hpp"
  58 
  59 #undef __
  60 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
  61 
  62 // Size of interpreter code.  Increase if too small.  Interpreter will
  63 // fail with a guarantee ("not enough space for interpreter generation");
  64 // if too small.
  65 // Run with +PrintInterpreter to get the VM to print out the size.
  66 // Max size with JVMTI
  67 int TemplateInterpreter::InterpreterCodeSize = 256*K;
  68 
  69 #ifdef PRODUCT
  70 #define BLOCK_COMMENT(str) /* nothing */
  71 #else
  72 #define BLOCK_COMMENT(str) __ block_comment(str)
  73 #endif
  74 
  75 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
  76 
  77 //-----------------------------------------------------------------------------
  78 
  79 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  80   // Slow_signature handler that respects the PPC C calling conventions.
  81   //
  82   // We get called by the native entry code with our output register
  83   // area == 8. First we call InterpreterRuntime::get_result_handler
  84   // to copy the pointer to the signature string temporarily to the
  85   // first C-argument and to return the result_handler in
  86   // R3_RET. Since native_entry will copy the jni-pointer to the
  87   // first C-argument slot later on, it is OK to occupy this slot
  88   // temporarily. Then we copy the argument list on the java
  89   // expression stack into native varargs format on the native stack
  90   // and load arguments into argument registers. Integer arguments in
  91   // the varargs vector will be sign-extended to 8 bytes.
  92   //
  93   // On entry:
  94   //   R3_ARG1        - intptr_t*     Address of java argument list in memory.
  95   //   R15_prev_state - BytecodeInterpreter* Address of interpreter state for
  96   //     this method
  97   //   R19_method
  98   //
  99   // On exit (just before return instruction):
 100   //   R3_RET            - contains the address of the result_handler.
 101   //   R4_ARG2           - is not updated for static methods and contains "this" otherwise.
 102   //   R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
 103   //                       ARGi contains this argument. Otherwise, ARGi is not updated.
 104   //   F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
 105 
 106   const int LogSizeOfTwoInstructions = 3;
 107 
 108   // FIXME: use Argument:: GL: Argument names different numbers!
 109   const int max_fp_register_arguments  = 13;
 110   const int max_int_register_arguments = 6;  // first 2 are reserved
 111 
 112   const Register arg_java       = R21_tmp1;
 113   const Register arg_c          = R22_tmp2;
 114   const Register signature      = R23_tmp3;  // is string
 115   const Register sig_byte       = R24_tmp4;
 116   const Register fpcnt          = R25_tmp5;
 117   const Register argcnt         = R26_tmp6;
 118   const Register intSlot        = R27_tmp7;
 119   const Register target_sp      = R28_tmp8;
 120   const FloatRegister floatSlot = F0;
 121 
 122   address entry = __ function_entry();
 123 
 124   __ save_LR(R0);
 125   __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 126   // We use target_sp for storing arguments in the C frame.
 127   __ mr(target_sp, R1_SP);
 128   __ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
 129 
 130   __ mr(arg_java, R3_ARG1);
 131 
 132   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
 133 
 134   // Signature is in R3_RET. Signature is callee saved.
 135   __ mr(signature, R3_RET);
 136 
 137   // Get the result handler.
 138   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
 139 
 140   {
 141     Label L;
 142     // test if static
 143     // _access_flags._flags must be at offset 0.
 144     // TODO PPC port: requires change in shared code.
 145     //assert(in_bytes(AccessFlags::flags_offset()) == 0,
 146     //       "MethodDesc._access_flags == MethodDesc._access_flags._flags");
 147     // _access_flags must be a 32 bit value.
 148     assert(sizeof(AccessFlags) == 4, "wrong size");
 149     __ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
 150     // testbit with condition register.
 151     __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
 152     __ btrue(CCR0, L);
 153     // For non-static functions, pass "this" in R4_ARG2 and copy it
 154     // to 2nd C-arg slot.
 155     // We need to box the Java object here, so we use arg_java
 156     // (address of current Java stack slot) as argument and don't
 157     // dereference it as in case of ints, floats, etc.
 158     __ mr(R4_ARG2, arg_java);
 159     __ addi(arg_java, arg_java, -BytesPerWord);
 160     __ std(R4_ARG2, _abi0(carg_2), target_sp);
 161     __ bind(L);
 162   }
 163 
 164   // Will be incremented directly after loop_start. argcnt=0
 165   // corresponds to 3rd C argument.
 166   __ li(argcnt, -1);
 167   // arg_c points to 3rd C argument
 168   __ addi(arg_c, target_sp, _abi0(carg_3));
 169   // no floating-point args parsed so far
 170   __ li(fpcnt, 0);
 171 
 172   Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
 173   Label loop_start, loop_end;
 174   Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
 175 
 176   // signature points to '(' at entry
 177 #ifdef ASSERT
 178   __ lbz(sig_byte, 0, signature);
 179   __ cmplwi(CCR0, sig_byte, '(');
 180   __ bne(CCR0, do_dontreachhere);
 181 #endif
 182 
 183   __ bind(loop_start);
 184 
 185   __ addi(argcnt, argcnt, 1);
 186   __ lbzu(sig_byte, 1, signature);
 187 
 188   __ cmplwi(CCR0, sig_byte, ')'); // end of signature
 189   __ beq(CCR0, loop_end);
 190 
 191   __ cmplwi(CCR0, sig_byte, 'B'); // byte
 192   __ beq(CCR0, do_int);
 193 
 194   __ cmplwi(CCR0, sig_byte, 'C'); // char
 195   __ beq(CCR0, do_int);
 196 
 197   __ cmplwi(CCR0, sig_byte, 'D'); // double
 198   __ beq(CCR0, do_double);
 199 
 200   __ cmplwi(CCR0, sig_byte, 'F'); // float
 201   __ beq(CCR0, do_float);
 202 
 203   __ cmplwi(CCR0, sig_byte, 'I'); // int
 204   __ beq(CCR0, do_int);
 205 
 206   __ cmplwi(CCR0, sig_byte, 'J'); // long
 207   __ beq(CCR0, do_long);
 208 
 209   __ cmplwi(CCR0, sig_byte, 'S'); // short
 210   __ beq(CCR0, do_int);
 211 
 212   __ cmplwi(CCR0, sig_byte, 'Z'); // boolean
 213   __ beq(CCR0, do_int);
 214 
 215   __ cmplwi(CCR0, sig_byte, 'L'); // object
 216   __ beq(CCR0, do_object);
 217 
 218   __ cmplwi(CCR0, sig_byte, '['); // array
 219   __ beq(CCR0, do_array);
 220 
 221   //  __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
 222   //  __ beq(CCR0, do_void);
 223 
 224   __ bind(do_dontreachhere);
 225 
 226   __ unimplemented("ShouldNotReachHere in slow_signature_handler");
 227 
 228   __ bind(do_array);
 229 
 230   {
 231     Label start_skip, end_skip;
 232 
 233     __ bind(start_skip);
 234     __ lbzu(sig_byte, 1, signature);
 235     __ cmplwi(CCR0, sig_byte, '[');
 236     __ beq(CCR0, start_skip); // skip further brackets
 237     __ cmplwi(CCR0, sig_byte, '9');
 238     __ bgt(CCR0, end_skip);   // no optional size
 239     __ cmplwi(CCR0, sig_byte, '0');
 240     __ bge(CCR0, start_skip); // skip optional size
 241     __ bind(end_skip);
 242 
 243     __ cmplwi(CCR0, sig_byte, 'L');
 244     __ beq(CCR0, do_object);  // for arrays of objects, the name of the object must be skipped
 245     __ b(do_boxed);          // otherwise, go directly to do_boxed
 246   }
 247 
 248   __ bind(do_object);
 249   {
 250     Label L;
 251     __ bind(L);
 252     __ lbzu(sig_byte, 1, signature);
 253     __ cmplwi(CCR0, sig_byte, ';');
 254     __ bne(CCR0, L);
 255    }
 256   // Need to box the Java object here, so we use arg_java (address of
 257   // current Java stack slot) as argument and don't dereference it as
 258   // in case of ints, floats, etc.
 259   Label do_null;
 260   __ bind(do_boxed);
 261   __ ld(R0,0, arg_java);
 262   __ cmpdi(CCR0, R0, 0);
 263   __ li(intSlot,0);
 264   __ beq(CCR0, do_null);
 265   __ mr(intSlot, arg_java);
 266   __ bind(do_null);
 267   __ std(intSlot, 0, arg_c);
 268   __ addi(arg_java, arg_java, -BytesPerWord);
 269   __ addi(arg_c, arg_c, BytesPerWord);
 270   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 271   __ blt(CCR0, move_intSlot_to_ARG);
 272   __ b(loop_start);
 273 
 274   __ bind(do_int);
 275   __ lwa(intSlot, 0, arg_java);
 276   __ std(intSlot, 0, arg_c);
 277   __ addi(arg_java, arg_java, -BytesPerWord);
 278   __ addi(arg_c, arg_c, BytesPerWord);
 279   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 280   __ blt(CCR0, move_intSlot_to_ARG);
 281   __ b(loop_start);
 282 
 283   __ bind(do_long);
 284   __ ld(intSlot, -BytesPerWord, arg_java);
 285   __ std(intSlot, 0, arg_c);
 286   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 287   __ addi(arg_c, arg_c, BytesPerWord);
 288   __ cmplwi(CCR0, argcnt, max_int_register_arguments);
 289   __ blt(CCR0, move_intSlot_to_ARG);
 290   __ b(loop_start);
 291 
 292   __ bind(do_float);
 293   __ lfs(floatSlot, 0, arg_java);
 294   __ stfs(floatSlot, Argument::float_on_stack_offset_in_bytes_c, arg_c);
 295   __ addi(arg_java, arg_java, -BytesPerWord);
 296   __ addi(arg_c, arg_c, BytesPerWord);
 297   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 298   __ blt(CCR0, move_floatSlot_to_FARG);
 299   __ b(loop_start);
 300 
 301   __ bind(do_double);
 302   __ lfd(floatSlot, - BytesPerWord, arg_java);
 303   __ stfd(floatSlot, 0, arg_c);
 304   __ addi(arg_java, arg_java, - 2 * BytesPerWord);
 305   __ addi(arg_c, arg_c, BytesPerWord);
 306   __ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
 307   __ blt(CCR0, move_floatSlot_to_FARG);
 308   __ b(loop_start);
 309 
 310   __ bind(loop_end);
 311 
 312   __ pop_frame();
 313   __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
 314   __ restore_LR(R0);
 315 
 316   __ blr();
 317 
 318   Label move_int_arg, move_float_arg;
 319   __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 320   __ mr(R5_ARG3, intSlot);  __ b(loop_start);
 321   __ mr(R6_ARG4, intSlot);  __ b(loop_start);
 322   __ mr(R7_ARG5, intSlot);  __ b(loop_start);
 323   __ mr(R8_ARG6, intSlot);  __ b(loop_start);
 324   __ mr(R9_ARG7, intSlot);  __ b(loop_start);
 325   __ mr(R10_ARG8, intSlot); __ b(loop_start);
 326 
 327   __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
 328   __ fmr(F1_ARG1, floatSlot);   __ b(loop_start);
 329   __ fmr(F2_ARG2, floatSlot);   __ b(loop_start);
 330   __ fmr(F3_ARG3, floatSlot);   __ b(loop_start);
 331   __ fmr(F4_ARG4, floatSlot);   __ b(loop_start);
 332   __ fmr(F5_ARG5, floatSlot);   __ b(loop_start);
 333   __ fmr(F6_ARG6, floatSlot);   __ b(loop_start);
 334   __ fmr(F7_ARG7, floatSlot);   __ b(loop_start);
 335   __ fmr(F8_ARG8, floatSlot);   __ b(loop_start);
 336   __ fmr(F9_ARG9, floatSlot);   __ b(loop_start);
 337   __ fmr(F10_ARG10, floatSlot); __ b(loop_start);
 338   __ fmr(F11_ARG11, floatSlot); __ b(loop_start);
 339   __ fmr(F12_ARG12, floatSlot); __ b(loop_start);
 340   __ fmr(F13_ARG13, floatSlot); __ b(loop_start);
 341 
 342   __ bind(move_intSlot_to_ARG);
 343   __ sldi(R0, argcnt, LogSizeOfTwoInstructions);
 344   __ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
 345   __ add(R11_scratch1, R0, R11_scratch1);
 346   __ mtctr(R11_scratch1/*branch_target*/);
 347   __ bctr();
 348   __ bind(move_floatSlot_to_FARG);
 349   __ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
 350   __ addi(fpcnt, fpcnt, 1);
 351   __ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
 352   __ add(R11_scratch1, R0, R11_scratch1);
 353   __ mtctr(R11_scratch1/*branch_target*/);
 354   __ bctr();
 355 
 356   return entry;
 357 }
 358 
 359 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 360   //
 361   // Registers alive
 362   //   R3_RET
 363   //   LR
 364   //
 365   // Registers updated
 366   //   R3_RET
 367   //
 368 
 369   Label done;
 370   address entry = __ pc();
 371 
 372   switch (type) {
 373   case T_BOOLEAN:
 374     // convert !=0 to 1
 375     __ normalize_bool(R3_RET);
 376     break;
 377   case T_BYTE:
 378      // sign extend 8 bits
 379      __ extsb(R3_RET, R3_RET);
 380      break;
 381   case T_CHAR:
 382      // zero extend 16 bits
 383      __ clrldi(R3_RET, R3_RET, 48);
 384      break;
 385   case T_SHORT:
 386      // sign extend 16 bits
 387      __ extsh(R3_RET, R3_RET);
 388      break;
 389   case T_INT:
 390      // sign extend 32 bits
 391      __ extsw(R3_RET, R3_RET);
 392      break;
 393   case T_LONG:
 394      break;
 395   case T_OBJECT:
 396     // JNIHandles::resolve result.
 397     __ resolve_jobject(R3_RET, R11_scratch1, R31, MacroAssembler::PRESERVATION_FRAME_LR); // kills R31
 398     break;
 399   case T_FLOAT:
 400      break;
 401   case T_DOUBLE:
 402      break;
 403   case T_VOID:
 404      break;
 405   default: ShouldNotReachHere();
 406   }
 407 
 408   BIND(done);
 409   __ blr();
 410 
 411   return entry;
 412 }
 413 
 414 // Abstract method entry.
 415 //
 416 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 417   address entry = __ pc();
 418 
 419   //
 420   // Registers alive
 421   //   R16_thread     - JavaThread*
 422   //   R19_method     - callee's method (method to be invoked)
 423   //   R1_SP          - SP prepared such that caller's outgoing args are near top
 424   //   LR             - return address to caller
 425   //
 426   // Stack layout at this point:
 427   //
 428   //   0       [TOP_IJAVA_FRAME_ABI]         <-- R1_SP
 429   //           alignment (optional)
 430   //           [outgoing Java arguments]
 431   //           ...
 432   //   PARENT  [PARENT_IJAVA_FRAME_ABI]
 433   //            ...
 434   //
 435 
 436   // Can't use call_VM here because we have not set up a new
 437   // interpreter state. Make the call to the vm and make it look like
 438   // our caller set up the JavaFrameAnchor.
 439   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
 440 
 441   // Push a new C frame and save LR.
 442   __ save_LR(R0);
 443   __ push_frame_reg_args(0, R11_scratch1);
 444 
 445   // This is not a leaf but we have a JavaFrameAnchor now and we will
 446   // check (create) exceptions afterward so this is ok.
 447   __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 448                   R16_thread, R19_method);
 449 
 450   // Pop the C frame and restore LR.
 451   __ pop_frame();
 452   __ restore_LR(R0);
 453 
 454   // Reset JavaFrameAnchor from call_VM_leaf above.
 455   __ reset_last_Java_frame();
 456 
 457   // We don't know our caller, so jump to the general forward exception stub,
 458   // which will also pop our full frame off. Satisfy the interface of
 459   // SharedRuntime::generate_forward_exception()
 460   __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0);
 461   __ mtctr(R11_scratch1);
 462   __ bctr();
 463 
 464   return entry;
 465 }
 466 
 467 // Interpreter intrinsic for WeakReference.get().
 468 // 1. Don't push a full blown frame and go on dispatching, but fetch the value
 469 //    into R8 and return quickly
 470 // 2. If G1 is active we *must* execute this intrinsic for corrrectness:
 471 //    It contains a GC barrier which puts the reference into the satb buffer
 472 //    to indicate that someone holds a strong reference to the object the
 473 //    weak ref points to!
 474 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 475   // Code: _aload_0, _getfield, _areturn
 476   // parameter size = 1
 477   //
 478   // The code that gets generated by this routine is split into 2 parts:
 479   //    1. the "intrinsified" code for G1 (or any SATB based GC),
 480   //    2. the slow path - which is an expansion of the regular method entry.
 481   //
 482   // Notes:
 483   // * In the G1 code we do not check whether we need to block for
 484   //   a safepoint. If G1 is enabled then we must execute the specialized
 485   //   code for Reference.get (except when the Reference object is null)
 486   //   so that we can log the value in the referent field with an SATB
 487   //   update buffer.
 488   //   If the code for the getfield template is modified so that the
 489   //   G1 pre-barrier code is executed when the current method is
 490   //   Reference.get() then going through the normal method entry
 491   //   will be fine.
 492   // * The G1 code can, however, check the receiver object (the instance
 493   //   of java.lang.Reference) and jump to the slow path if null. If the
 494   //   Reference object is null then we obviously cannot fetch the referent
 495   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 496   //   regular method entry code to generate the NPE.
 497   //
 498 
 499   address entry = __ pc();
 500 
 501   const int referent_offset = java_lang_ref_Reference::referent_offset();
 502 
 503   Label slow_path;
 504 
 505   // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
 506 
 507   // In the G1 code we don't check if we need to reach a safepoint. We
 508   // continue and the thread will safepoint at the next bytecode dispatch.
 509 
 510   // If the receiver is null then it is OK to jump to the slow path.
 511   __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver
 512 
 513   // Check if receiver == nullptr and go the slow path.
 514   __ cmpdi(CCR0, R3_RET, 0);
 515   __ beq(CCR0, slow_path);
 516 
 517   __ load_heap_oop(R3_RET, referent_offset, R3_RET,
 518                    /* non-volatile temp */ R31, R11_scratch1,
 519                    MacroAssembler::PRESERVATION_FRAME_LR,
 520                    ON_WEAK_OOP_REF);
 521 
 522   // Generate the G1 pre-barrier code to log the value of
 523   // the referent field in an SATB buffer. Note with
 524   // these parameters the pre-barrier does not generate
 525   // the load of the previous value.
 526 
 527   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
 528   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
 529 
 530   __ blr();
 531 
 532   __ bind(slow_path);
 533   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
 534   return entry;
 535 }
 536 
 537 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 538   address entry = __ pc();
 539 
 540   // Expression stack must be empty before entering the VM if an
 541   // exception happened.
 542   __ empty_expression_stack();
 543   // Throw exception.
 544   __ call_VM(noreg,
 545              CAST_FROM_FN_PTR(address,
 546                               InterpreterRuntime::throw_StackOverflowError));
 547   return entry;
 548 }
 549 
 550 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 551   address entry = __ pc();
 552   __ empty_expression_stack();
 553   // R4_ARG2 already contains the array.
 554   // Index is in R17_tos.
 555   __ mr(R5_ARG3, R17_tos);
 556   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R4_ARG2, R5_ARG3);
 557   return entry;
 558 }
 559 
 560 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 561   address entry = __ pc();
 562   // Expression stack must be empty before entering the VM if an
 563   // exception happened.
 564   __ empty_expression_stack();
 565 
 566   // Load exception object.
 567   // Thread will be loaded to R3_ARG1.
 568   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
 569 #ifdef ASSERT
 570   // Above call must not return here since exception pending.
 571   __ should_not_reach_here();
 572 #endif
 573   return entry;
 574 }
 575 
 576 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 577   address entry = __ pc();
 578   //__ untested("generate_exception_handler_common");
 579   Register Rexception = R17_tos;
 580 
 581   // Expression stack must be empty before entering the VM if an exception happened.
 582   __ empty_expression_stack();
 583 
 584   __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
 585   if (pass_oop) {
 586     __ mr(R5_ARG3, Rexception);
 587     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception));
 588   } else {
 589     __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
 590     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception));
 591   }
 592 
 593   // Throw exception.
 594   __ mr(R3_ARG1, Rexception);
 595   __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
 596   __ mtctr(R11_scratch1);
 597   __ bctr();
 598 
 599   return entry;
 600 }
 601 
 602 // This entry is returned to when a call returns to the interpreter.
 603 // When we arrive here, we expect that the callee stack frame is already popped.
 604 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 605   address entry = __ pc();
 606 
 607   // Move the value out of the return register back to the TOS cache of current frame.
 608   switch (state) {
 609     case ltos:
 610     case btos:
 611     case ztos:
 612     case ctos:
 613     case stos:
 614     case atos:
 615     case itos: __ mr(R17_tos, R3_RET); break;   // RET -> TOS cache
 616     case ftos:
 617     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 618     case vtos: break;                           // Nothing to do, this was a void return.
 619     default  : ShouldNotReachHere();
 620   }
 621 
 622   __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
 623 
 624   // Compiled code destroys templateTableBase, reload.
 625   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
 626 
 627   if (state == atos) {
 628     __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
 629   }
 630 
 631   const Register cache = R11_scratch1;
 632   const Register size  = R12_scratch2;
 633   if (index_size == sizeof(u4)) {
 634     __ load_resolved_indy_entry(cache, size /* tmp */);
 635     __ lhz(size, in_bytes(ResolvedIndyEntry::num_parameters_offset()), cache);
 636   } else {
 637     assert(index_size == sizeof(u2), "Can only be u2");
 638     __ load_method_entry(cache, size /* tmp */);
 639     __ lhz(size, in_bytes(ResolvedMethodEntry::num_parameters_offset()), cache);
 640   }
 641   __ sldi(size, size, Interpreter::logStackElementSize);
 642   __ add(R15_esp, R15_esp, size);
 643 
 644  __ check_and_handle_popframe(R11_scratch1);
 645  __ check_and_handle_earlyret(R11_scratch1);
 646 
 647   __ dispatch_next(state, step);
 648   return entry;
 649 }
 650 
 651 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step, address continuation) {
 652   address entry = __ pc();
 653   // If state != vtos, we're returning from a native method, which put it's result
 654   // into the result register. So move the value out of the return register back
 655   // to the TOS cache of current frame.
 656 
 657   switch (state) {
 658     case ltos:
 659     case btos:
 660     case ztos:
 661     case ctos:
 662     case stos:
 663     case atos:
 664     case itos: __ mr(R17_tos, R3_RET); break;   // GR_RET -> TOS cache
 665     case ftos:
 666     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
 667     case vtos: break;                           // Nothing to do, this was a void return.
 668     default  : ShouldNotReachHere();
 669   }
 670 
 671   // Load LcpoolCache @@@ should be already set!
 672   __ get_constant_pool_cache(R27_constPoolCache);
 673 
 674   // Handle a pending exception, fall through if none.
 675   __ check_and_forward_exception(R11_scratch1, R12_scratch2);
 676 
 677   // Start executing bytecodes.
 678   if (continuation == nullptr) {
 679     __ dispatch_next(state, step);
 680   } else {
 681     __ jump_to_entry(continuation, R11_scratch1);
 682   }
 683 
 684   return entry;
 685 }
 686 
 687 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 688   address entry = __ pc();
 689 
 690   __ push(state);
 691   __ push_cont_fastpath();
 692   __ call_VM(noreg, runtime_entry);
 693   __ pop_cont_fastpath();
 694   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 695 
 696   return entry;
 697 }
 698 
 699 // Helpers for commoning out cases in the various type of method entries.
 700 
 701 // Increment invocation count & check for overflow.
 702 //
 703 // Note: checking for negative value instead of overflow
 704 //       so we have a 'sticky' overflow test.
 705 //
 706 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 707   // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
 708   Register Rscratch1   = R11_scratch1;
 709   Register Rscratch2   = R12_scratch2;
 710   Register R3_counters = R3_ARG1;
 711   Label done;
 712 
 713   const int increment = InvocationCounter::count_increment;
 714   Label no_mdo;
 715   if (ProfileInterpreter) {
 716     const Register Rmdo = R3_counters;
 717     __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
 718     __ cmpdi(CCR0, Rmdo, 0);
 719     __ beq(CCR0, no_mdo);
 720 
 721     // Increment invocation counter in the MDO.
 722     const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 723     __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
 724     __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
 725     __ addi(Rscratch2, Rscratch2, increment);
 726     __ stw(Rscratch2, mdo_ic_offs, Rmdo);
 727     __ and_(Rscratch1, Rscratch2, Rscratch1);
 728     __ bne(CCR0, done);
 729     __ b(*overflow);
 730   }
 731 
 732   // Increment counter in MethodCounters*.
 733   const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 734   __ bind(no_mdo);
 735   __ get_method_counters(R19_method, R3_counters, done);
 736   __ lwz(Rscratch2, mo_ic_offs, R3_counters);
 737   __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
 738   __ addi(Rscratch2, Rscratch2, increment);
 739   __ stw(Rscratch2, mo_ic_offs, R3_counters);
 740   __ and_(Rscratch1, Rscratch2, Rscratch1);
 741   __ beq(CCR0, *overflow);
 742 
 743   __ bind(done);
 744 }
 745 
 746 // Generate code to initiate compilation on invocation counter overflow.
 747 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
 748   // Generate code to initiate compilation on the counter overflow.
 749 
 750   // InterpreterRuntime::frequency_counter_overflow takes one arguments,
 751   // which indicates if the counter overflow occurs at a backwards branch (null bcp)
 752   // We pass zero in.
 753   // The call returns the address of the verified entry point for the method or null
 754   // if the compilation did not complete (either went background or bailed out).
 755   //
 756   // Unlike the C++ interpreter above: Check exceptions!
 757   // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
 758   // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
 759 
 760   __ li(R4_ARG2, 0);
 761   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
 762 
 763   // Returns verified_entry_point or null.
 764   // We ignore it in any case.
 765   __ b(continue_entry);
 766 }
 767 
 768 // See if we've got enough room on the stack for locals plus overhead below
 769 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 770 // without going through the signal handler, i.e., reserved and yellow zones
 771 // will not be made usable. The shadow zone must suffice to handle the
 772 // overflow.
 773 //
 774 // Kills Rmem_frame_size, Rscratch1.
 775 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
 776   Label done;
 777   assert_different_registers(Rmem_frame_size, Rscratch1);
 778 
 779   BLOCK_COMMENT("stack_overflow_check_with_compare {");
 780   __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
 781   __ ld(Rscratch1, thread_(stack_overflow_limit));
 782   __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
 783   __ bgt(CCR0/*is_stack_overflow*/, done);
 784 
 785   // The stack overflows. Load target address of the runtime stub and call it.
 786   assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "generated in wrong order");
 787   __ load_const_optimized(Rscratch1, (SharedRuntime::throw_StackOverflowError_entry()), R0);
 788   __ mtctr(Rscratch1);
 789   // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
 790 #ifdef ASSERT
 791   Label frame_not_shrunk;
 792   __ cmpld(CCR0, R1_SP, R21_sender_SP);
 793   __ ble(CCR0, frame_not_shrunk);
 794   __ stop("frame shrunk");
 795   __ bind(frame_not_shrunk);
 796   __ ld(Rscratch1, 0, R1_SP);
 797   __ ld(R0, 0, R21_sender_SP);
 798   __ cmpd(CCR0, R0, Rscratch1);
 799   __ asm_assert_eq("backlink");
 800 #endif // ASSERT
 801   __ mr(R1_SP, R21_sender_SP);
 802   __ bctr();
 803 
 804   __ align(32, 12);
 805   __ bind(done);
 806   BLOCK_COMMENT("} stack_overflow_check_with_compare");
 807 }
 808 
 809 // Lock the current method, interpreter register window must be set up!
 810 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
 811   const Register Robj_to_lock = Rscratch2;
 812 
 813   {
 814     if (!flags_preloaded) {
 815       __ lwz(Rflags, method_(access_flags));
 816     }
 817 
 818 #ifdef ASSERT
 819     // Check if methods needs synchronization.
 820     {
 821       Label Lok;
 822       __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
 823       __ btrue(CCR0,Lok);
 824       __ stop("method doesn't need synchronization");
 825       __ bind(Lok);
 826     }
 827 #endif // ASSERT
 828   }
 829 
 830   // Get synchronization object to Rscratch2.
 831   {
 832     Label Lstatic;
 833     Label Ldone;
 834 
 835     __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
 836     __ btrue(CCR0, Lstatic);
 837 
 838     // Non-static case: load receiver obj from stack and we're done.
 839     __ ld(Robj_to_lock, R18_locals);
 840     __ b(Ldone);
 841 
 842     __ bind(Lstatic); // Static case: Lock the java mirror
 843     // Load mirror from interpreter frame.
 844     __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
 845     __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
 846 
 847     __ bind(Ldone);
 848     __ verify_oop(Robj_to_lock);
 849   }
 850 
 851   // Got the oop to lock => execute!
 852   __ add_monitor_to_stack(true, Rscratch1, R0);
 853 
 854   __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor);
 855   __ lock_object(R26_monitor, Robj_to_lock);
 856 }
 857 
 858 // Generate a fixed interpreter frame for pure interpreter
 859 // and I2N native transition frames.
 860 //
 861 // Before (stack grows downwards):
 862 //
 863 //         |  ...         |
 864 //         |------------- |
 865 //         |  java arg0   |
 866 //         |  ...         |
 867 //         |  java argn   |
 868 //         |              |   <-   R15_esp
 869 //         |              |
 870 //         |--------------|
 871 //         | abi_112      |
 872 //         |              |   <-   R1_SP
 873 //         |==============|
 874 //
 875 //
 876 // After:
 877 //
 878 //         |  ...         |
 879 //         |  java arg0   |<-   R18_locals
 880 //         |  ...         |
 881 //         |  java argn   |
 882 //         |--------------|
 883 //         |              |
 884 //         |  java locals |
 885 //         |              |
 886 //         |--------------|
 887 //         |  abi_48      |
 888 //         |==============|
 889 //         |              |
 890 //         |   istate     |
 891 //         |              |
 892 //         |--------------|
 893 //         |   monitor    |<-   R26_monitor
 894 //         |--------------|
 895 //         |              |<-   R15_esp
 896 //         | expression   |
 897 //         | stack        |
 898 //         |              |
 899 //         |--------------|
 900 //         |              |
 901 //         | abi_112      |<-   R1_SP
 902 //         |==============|
 903 //
 904 // The top most frame needs an abi space of 112 bytes. This space is needed,
 905 // since we call to c. The c function may spill their arguments to the caller
 906 // frame. When we call to java, we don't need these spill slots. In order to save
 907 // space on the stack, we resize the caller. However, java locals reside in
 908 // the caller frame and the frame has to be increased. The frame_size for the
 909 // current frame was calculated based on max_stack as size for the expression
 910 // stack. At the call, just a part of the expression stack might be used.
 911 // We don't want to waste this space and cut the frame back accordingly.
 912 // The resulting amount for resizing is calculated as follows:
 913 // resize =   (number_of_locals - number_of_arguments) * slot_size
 914 //          + (R1_SP - R15_esp) + 48
 915 //
 916 // The size for the callee frame is calculated:
 917 // framesize = 112 + max_stack + monitor + state_size
 918 //
 919 // maxstack:   Max number of slots on the expression stack, loaded from the method.
 920 // monitor:    We statically reserve room for one monitor object.
 921 // state_size: We save the current state of the interpreter to this area.
 922 //
 923 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
 924   Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
 925            Rtop_frame_size      = R7_ARG5,
 926            Rconst_method        = R8_ARG6,
 927            Rconst_pool          = R9_ARG7,
 928            Rmirror              = R10_ARG8;
 929 
 930   assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
 931                              Rconst_method, Rconst_pool);
 932 
 933   __ ld(Rconst_method, method_(const));
 934   __ lhz(Rsize_of_parameters /* number of params */,
 935          in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
 936   if (native_call) {
 937     // If we're calling a native method, we reserve space for the worst-case signature
 938     // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2).
 939     // We add two slots to the parameter_count, one for the jni
 940     // environment and one for a possible native mirror.
 941     Label skip_native_calculate_max_stack;
 942     __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
 943     __ cmpwi(CCR0, Rtop_frame_size, Argument::n_int_register_parameters_c);
 944     __ bge(CCR0, skip_native_calculate_max_stack);
 945     __ li(Rtop_frame_size, Argument::n_int_register_parameters_c);
 946     __ bind(skip_native_calculate_max_stack);
 947     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 948     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 949     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 950     assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
 951   } else {
 952     __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
 953     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 954     __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
 955     __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
 956     __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
 957     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 958     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 959     __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
 960   }
 961 
 962   // Compute top frame size.
 963   __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
 964 
 965   // Cut back area between esp and max_stack.
 966   __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize);
 967 
 968   __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
 969   __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
 970   // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
 971   // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
 972 
 973   if (!native_call) {
 974     // Stack overflow check.
 975     // Native calls don't need the stack size check since they have no
 976     // expression stack and the arguments are already on the stack and
 977     // we only add a handful of words to the stack.
 978     __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
 979     generate_stack_overflow_check(R11_scratch1, R12_scratch2);
 980   }
 981 
 982   // Set up interpreter state registers.
 983 
 984   __ add(R18_locals, R15_esp, Rsize_of_parameters);
 985   __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
 986   __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool);
 987 
 988   // Set method data pointer.
 989   if (ProfileInterpreter) {
 990     Label zero_continue;
 991     __ ld(R28_mdx, method_(method_data));
 992     __ cmpdi(CCR0, R28_mdx, 0);
 993     __ beq(CCR0, zero_continue);
 994     __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
 995     __ bind(zero_continue);
 996   }
 997 
 998   if (native_call) {
 999     __ li(R14_bcp, 0); // Must initialize.
1000   } else {
1001     __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset()));
1002   }
1003 
1004   // Resize parent frame.
1005   __ mflr(R12_scratch2);
1006   __ neg(Rparent_frame_resize, Rparent_frame_resize);
1007   __ resize_frame(Rparent_frame_resize, R11_scratch1);
1008   __ std(R12_scratch2, _abi0(lr), R1_SP);
1009 
1010   // Get mirror and store it in the frame as GC root for this Method*.
1011   __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool);
1012   __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1013   __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1014 
1015   __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1016   __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1017 
1018   // Store values.
1019   __ std(R19_method, _ijava_state_neg(method), R1_SP);
1020   __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1021   __ sub(R12_scratch2, R18_locals, R1_SP);
1022   __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1023   // Store relativized R18_locals, see frame::interpreter_frame_locals().
1024   __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP);
1025   __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1026 
1027   // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1028   // be found in the frame after save_interpreter_state is done. This is always true
1029   // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1030   // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1031   // (Enhanced Stack Trace).
1032   // The signal handler does not save the interpreter state into the frame.
1033 
1034   // We have to initialize some of these frame slots for native calls (accessed by GC).
1035   // Also initialize them for non-native calls for better tool support (even though
1036   // you may not get the most recent version as described above).
1037   __ li(R0, 0);
1038   __ li(R12_scratch2, -(frame::ijava_state_size / wordSize));
1039   __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1040   __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1041   if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1042   __ sub(R12_scratch2, R15_esp, R1_SP);
1043   __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1044   __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1045   __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1046 
1047   // Store sender's SP and this frame's top SP.
1048   __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1049   __ neg(R12_scratch2, Rtop_frame_size);
1050   __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1051   // Store relativized top_frame_sp
1052   __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1053 
1054   // Push top frame.
1055   __ push_frame(Rtop_frame_size, R11_scratch1);
1056 }
1057 
1058 // End of helpers
1059 
1060 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1061 
1062   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1063   bool use_instruction = false;
1064   address runtime_entry = nullptr;
1065   int num_args = 1;
1066   bool double_precision = true;
1067 
1068   // PPC64 specific:
1069   switch (kind) {
1070     case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1071     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1072     case Interpreter::java_lang_math_fmaF:
1073     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1074     default: break; // Fall back to runtime call.
1075   }
1076 
1077   switch (kind) {
1078     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1079     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1080     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1081     case Interpreter::java_lang_math_tanh : /* run interpreted */ break;
1082     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1083     case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt);  break;
1084     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1085     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1086     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1087     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1088     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1089     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1090     default: ShouldNotReachHere();
1091   }
1092 
1093   // Use normal entry if neither instruction nor runtime call is used.
1094   if (!use_instruction && runtime_entry == nullptr) return nullptr;
1095 
1096   address entry = __ pc();
1097 
1098   // Load arguments
1099   assert(num_args <= 13, "passed in registers");
1100   if (double_precision) {
1101     int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1102     for (int i = 0; i < num_args; ++i) {
1103       __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1104       offset -= 2 * Interpreter::stackElementSize;
1105     }
1106   } else {
1107     int offset = num_args * Interpreter::stackElementSize;
1108     for (int i = 0; i < num_args; ++i) {
1109       __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1110       offset -= Interpreter::stackElementSize;
1111     }
1112   }
1113 
1114   if (use_instruction) {
1115     switch (kind) {
1116       case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1);          break;
1117       case Interpreter::java_lang_math_abs:  __ fabs(F1_RET, F1);           break;
1118       case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1119       case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3);  break;
1120       default: ShouldNotReachHere();
1121     }
1122   } else {
1123     // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1124     //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1125     //__ call_c_and_return_to_caller(R12_scratch2);
1126 
1127     // Push a new C frame and save LR.
1128     __ save_LR(R0);
1129     __ push_frame_reg_args(0, R11_scratch1);
1130 
1131     __ call_VM_leaf(runtime_entry);
1132 
1133     // Pop the C frame and restore LR.
1134     __ pop_frame();
1135     __ restore_LR(R0);
1136   }
1137 
1138   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1139   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1140   __ blr();
1141 
1142   __ flush();
1143 
1144   return entry;
1145 }
1146 
1147 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1148   // Quick & dirty stack overflow checking: bang the stack & handle trap.
1149   // Note that we do the banging after the frame is setup, since the exception
1150   // handling code expects to find a valid interpreter frame on the stack.
1151   // Doing the banging earlier fails if the caller frame is not an interpreter
1152   // frame.
1153   // (Also, the exception throwing code expects to unlock any synchronized
1154   // method receiever, so do the banging after locking the receiver.)
1155 
1156   // Bang each page in the shadow zone. We can't assume it's been done for
1157   // an interpreter frame with greater than a page of locals, so each page
1158   // needs to be checked.  Only true for non-native.
1159   const size_t page_size = os::vm_page_size();
1160   const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size;
1161   const int start_page = native_call ? n_shadow_pages : 1;
1162   BLOCK_COMMENT("bang_stack_shadow_pages:");
1163   for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1164     __ bang_stack_with_offset(pages*page_size);
1165   }
1166 }
1167 
1168 // Interpreter stub for calling a native method. (asm interpreter)
1169 // This sets up a somewhat different looking stack for calling the
1170 // native method than the typical interpreter frame setup.
1171 //
1172 // On entry:
1173 //   R19_method    - method
1174 //   R16_thread    - JavaThread*
1175 //   R15_esp       - intptr_t* sender tos
1176 //
1177 //   abstract stack (grows up)
1178 //     [  IJava (caller of JNI callee)  ]  <-- ASP
1179 //        ...
1180 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1181 
1182   address entry = __ pc();
1183 
1184   const bool inc_counter = UseCompiler || CountCompiledCalls;
1185 
1186   // -----------------------------------------------------------------------------
1187   // Allocate a new frame that represents the native callee (i2n frame).
1188   // This is not a full-blown interpreter frame, but in particular, the
1189   // following registers are valid after this:
1190   // - R19_method
1191   // - R18_local (points to start of arguments to native function)
1192   //
1193   //   abstract stack (grows up)
1194   //     [  IJava (caller of JNI callee)  ]  <-- ASP
1195   //        ...
1196 
1197   const Register signature_handler_fd = R11_scratch1;
1198   const Register pending_exception    = R0;
1199   const Register result_handler_addr  = R31;
1200   const Register native_method_fd     = R11_scratch1;
1201   const Register access_flags         = R22_tmp2;
1202   const Register active_handles       = R11_scratch1; // R26_monitor saved to state.
1203   const Register sync_state           = R12_scratch2;
1204   const Register sync_state_addr      = sync_state;   // Address is dead after use.
1205   const Register suspend_flags        = R11_scratch1;
1206 
1207   //=============================================================================
1208   // Allocate new frame and initialize interpreter state.
1209 
1210   Label exception_return;
1211   Label exception_return_sync_check;
1212   Label stack_overflow_return;
1213 
1214   // Generate new interpreter state and jump to stack_overflow_return in case of
1215   // a stack overflow.
1216   //generate_compute_interpreter_state(stack_overflow_return);
1217 
1218   Register size_of_parameters = R22_tmp2;
1219 
1220   generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1221 
1222   //=============================================================================
1223   // Increment invocation counter. On overflow, entry to JNI method
1224   // will be compiled.
1225   Label invocation_counter_overflow, continue_after_compile;
1226   if (inc_counter) {
1227     if (synchronized) {
1228       // Since at this point in the method invocation the exception handler
1229       // would try to exit the monitor of synchronized methods which hasn't
1230       // been entered yet, we set the thread local variable
1231       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1232       // runtime, exception handling i.e. unlock_if_synchronized_method will
1233       // check this thread local flag.
1234       // This flag has two effects, one is to force an unwind in the topmost
1235       // interpreter frame and not perform an unlock while doing so.
1236       __ li(R0, 1);
1237       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1238     }
1239     generate_counter_incr(&invocation_counter_overflow);
1240 
1241     BIND(continue_after_compile);
1242   }
1243 
1244   bang_stack_shadow_pages(true);
1245 
1246   if (inc_counter) {
1247     // Reset the _do_not_unlock_if_synchronized flag.
1248     if (synchronized) {
1249       __ li(R0, 0);
1250       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1251     }
1252   }
1253 
1254   // access_flags = method->access_flags();
1255   // Load access flags.
1256   assert(access_flags->is_nonvolatile(),
1257          "access_flags must be in a non-volatile register");
1258   // Type check.
1259   assert(4 == sizeof(AccessFlags), "unexpected field size");
1260   __ lwz(access_flags, method_(access_flags));
1261 
1262   // We don't want to reload R19_method and access_flags after calls
1263   // to some helper functions.
1264   assert(R19_method->is_nonvolatile(),
1265          "R19_method must be a non-volatile register");
1266 
1267   // Check for synchronized methods. Must happen AFTER invocation counter
1268   // check, so method is not locked if counter overflows.
1269 
1270   if (synchronized) {
1271     lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1272 
1273     // Update monitor in state.
1274     __ ld(R11_scratch1, 0, R1_SP);
1275     __ sub(R12_scratch2, R26_monitor, R11_scratch1);
1276     __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1277     __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1);
1278   }
1279 
1280   // jvmti/jvmpi support
1281   __ notify_method_entry();
1282 
1283   //=============================================================================
1284   // Get and call the signature handler.
1285 
1286   __ ld(signature_handler_fd, method_(signature_handler));
1287   Label call_signature_handler;
1288 
1289   __ cmpdi(CCR0, signature_handler_fd, 0);
1290   __ bne(CCR0, call_signature_handler);
1291 
1292   // Method has never been called. Either generate a specialized
1293   // handler or point to the slow one.
1294   //
1295   // Pass parameter 'false' to avoid exception check in call_VM.
1296   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1297 
1298   // Check for an exception while looking up the target method. If we
1299   // incurred one, bail.
1300   __ ld(pending_exception, thread_(pending_exception));
1301   __ cmpdi(CCR0, pending_exception, 0);
1302   __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1303 
1304   // Reload signature handler, it may have been created/assigned in the meanwhile.
1305   __ ld(signature_handler_fd, method_(signature_handler));
1306   __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1307 
1308   BIND(call_signature_handler);
1309 
1310   // Before we call the signature handler we push a new frame to
1311   // protect the interpreter frame volatile registers when we return
1312   // from jni but before we can get back to Java.
1313 
1314   // First set the frame anchor while the SP/FP registers are
1315   // convenient and the slow signature handler can use this same frame
1316   // anchor.
1317 
1318   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1319   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
1320 
1321   // Now the interpreter frame (and its call chain) have been
1322   // invalidated and flushed. We are now protected against eager
1323   // being enabled in native code. Even if it goes eager the
1324   // registers will be reloaded as clean and we will invalidate after
1325   // the call so no spurious flush should be possible.
1326 
1327   // Call signature handler and pass locals address.
1328   //
1329   // Our signature handlers copy required arguments to the C stack
1330   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1331   __ mr(R3_ARG1, R18_locals);
1332 #if !defined(ABI_ELFv2)
1333   __ ld(signature_handler_fd, 0, signature_handler_fd);
1334 #endif
1335 
1336   __ call_stub(signature_handler_fd);
1337 
1338   // Remove the register parameter varargs slots we allocated in
1339   // compute_interpreter_state. SP+16 ends up pointing to the ABI
1340   // outgoing argument area.
1341   //
1342   // Not needed on PPC64.
1343   //__ add(SP, SP, Argument::n_int_register_parameters_c*BytesPerWord);
1344 
1345   assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
1346   // Save across call to native method.
1347   __ mr(result_handler_addr, R3_RET);
1348 
1349   __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1350 
1351   // Set up fixed parameters and call the native method.
1352   // If the method is static, get mirror into R4_ARG2.
1353   {
1354     Label method_is_not_static;
1355     // Access_flags is non-volatile and still, no need to restore it.
1356 
1357     // Restore access flags.
1358     __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1359     __ bfalse(CCR0, method_is_not_static);
1360 
1361     __ ld(R11_scratch1, _abi0(callers_sp), R1_SP);
1362     // Load mirror from interpreter frame.
1363     __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1);
1364     // R4_ARG2 = &state->_oop_temp;
1365     __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1366     __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1367     BIND(method_is_not_static);
1368   }
1369 
1370   // At this point, arguments have been copied off the stack into
1371   // their JNI positions. Oops are boxed in-place on the stack, with
1372   // handles copied to arguments. The result handler address is in a
1373   // register.
1374 
1375   // Pass JNIEnv address as first parameter.
1376   __ addir(R3_ARG1, thread_(jni_environment));
1377 
1378   // Load the native_method entry before we change the thread state.
1379   __ ld(native_method_fd, method_(native_function));
1380 
1381   //=============================================================================
1382   // Transition from _thread_in_Java to _thread_in_native. As soon as
1383   // we make this change the safepoint code needs to be certain that
1384   // the last Java frame we established is good. The pc in that frame
1385   // just needs to be near here not an actual return address.
1386 
1387   // We use release_store_fence to update values like the thread state, where
1388   // we don't want the current thread to continue until all our prior memory
1389   // accesses (including the new thread state) are visible to other threads.
1390   __ li(R0, _thread_in_native);
1391   __ release();
1392 
1393   // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1394   __ stw(R0, thread_(thread_state));
1395 
1396   //=============================================================================
1397   // Call the native method. Argument registers must not have been
1398   // overwritten since "__ call_stub(signature_handler);" (except for
1399   // ARG1 and ARG2 for static methods).
1400   __ call_c(native_method_fd);
1401 
1402   __ li(R0, 0);
1403   __ ld(R11_scratch1, 0, R1_SP);
1404   __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1405   __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1406   __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1407 
1408   // Note: C++ interpreter needs the following here:
1409   // The frame_manager_lr field, which we use for setting the last
1410   // java frame, gets overwritten by the signature handler. Restore
1411   // it now.
1412   //__ get_PC_trash_LR(R11_scratch1);
1413   //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1414 
1415   // Because of GC R19_method may no longer be valid.
1416 
1417   // Block, if necessary, before resuming in _thread_in_Java state.
1418   // In order for GC to work, don't clear the last_Java_sp until after
1419   // blocking.
1420 
1421   //=============================================================================
1422   // Switch thread to "native transition" state before reading the
1423   // synchronization state. This additional state is necessary
1424   // because reading and testing the synchronization state is not
1425   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1426   // in _thread_in_native state, loads _not_synchronized and is
1427   // preempted. VM thread changes sync state to synchronizing and
1428   // suspends threads for GC. Thread A is resumed to finish this
1429   // native method, but doesn't block here since it didn't see any
1430   // synchronization in progress, and escapes.
1431 
1432   // We use release_store_fence to update values like the thread state, where
1433   // we don't want the current thread to continue until all our prior memory
1434   // accesses (including the new thread state) are visible to other threads.
1435   __ li(R0/*thread_state*/, _thread_in_native_trans);
1436   __ release();
1437   __ stw(R0/*thread_state*/, thread_(thread_state));
1438   if (!UseSystemMemoryBarrier) {
1439     __ fence();
1440   }
1441 
1442   // Now before we return to java we must look for a current safepoint
1443   // (a new safepoint can not start since we entered native_trans).
1444   // We must check here because a current safepoint could be modifying
1445   // the callers registers right this moment.
1446 
1447   // Acquire isn't strictly necessary here because of the fence, but
1448   // sync_state is declared to be volatile, so we do it anyway
1449   // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1450 
1451   Label do_safepoint, sync_check_done;
1452   // No synchronization in progress nor yet synchronized.
1453   __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1454 
1455   // Not suspended.
1456   // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1457   __ lwz(suspend_flags, thread_(suspend_flags));
1458   __ cmpwi(CCR1, suspend_flags, 0);
1459   __ beq(CCR1, sync_check_done);
1460 
1461   __ bind(do_safepoint);
1462   __ isync();
1463   // Block. We do the call directly and leave the current
1464   // last_Java_frame setup undisturbed. We must save any possible
1465   // native result across the call. No oop is present.
1466 
1467   __ mr(R3_ARG1, R16_thread);
1468   __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1469 
1470   __ bind(sync_check_done);
1471 
1472   //=============================================================================
1473   // <<<<<< Back in Interpreter Frame >>>>>
1474 
1475   // We are in thread_in_native_trans here and back in the normal
1476   // interpreter frame. We don't have to do anything special about
1477   // safepoints and we can switch to Java mode anytime we are ready.
1478 
1479   // Note: frame::interpreter_frame_result has a dependency on how the
1480   // method result is saved across the call to post_method_exit. For
1481   // native methods it assumes that the non-FPU/non-void result is
1482   // saved in _native_lresult and a FPU result in _native_fresult. If
1483   // this changes then the interpreter_frame_result implementation
1484   // will need to be updated too.
1485 
1486   // On PPC64, we have stored the result directly after the native call.
1487 
1488   //=============================================================================
1489   // Back in Java
1490 
1491   // We use release_store_fence to update values like the thread state, where
1492   // we don't want the current thread to continue until all our prior memory
1493   // accesses (including the new thread state) are visible to other threads.
1494   __ li(R0/*thread_state*/, _thread_in_Java);
1495   __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1496   __ stw(R0/*thread_state*/, thread_(thread_state));
1497 
1498   if (CheckJNICalls) {
1499     // clear_pending_jni_exception_check
1500     __ load_const_optimized(R0, 0L);
1501     __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1502   }
1503 
1504   __ reset_last_Java_frame();
1505 
1506   // Jvmdi/jvmpi support. Whether we've got an exception pending or
1507   // not, and whether unlocking throws an exception or not, we notify
1508   // on native method exit. If we do have an exception, we'll end up
1509   // in the caller's context to handle it, so if we don't do the
1510   // notify here, we'll drop it on the floor.
1511   __ notify_method_exit(true/*native method*/,
1512                         ilgl /*illegal state (not used for native methods)*/,
1513                         InterpreterMacroAssembler::NotifyJVMTI,
1514                         false /*check_exceptions*/);
1515 
1516   //=============================================================================
1517   // Handle exceptions
1518 
1519   if (synchronized) {
1520     __ unlock_object(R26_monitor); // Can also unlock methods.
1521   }
1522 
1523   // Reset active handles after returning from native.
1524   // thread->active_handles()->clear();
1525   __ ld(active_handles, thread_(active_handles));
1526   // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1527   __ li(R0, 0);
1528   __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles);
1529 
1530   Label exception_return_sync_check_already_unlocked;
1531   __ ld(R0/*pending_exception*/, thread_(pending_exception));
1532   __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1533   __ bne(CCR0, exception_return_sync_check_already_unlocked);
1534 
1535   //-----------------------------------------------------------------------------
1536   // No exception pending.
1537 
1538   // Move native method result back into proper registers and return.
1539   // Invoke result handler (may unbox/promote).
1540   __ ld(R11_scratch1, 0, R1_SP);
1541   __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1542   __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1543   __ call_stub(result_handler_addr);
1544 
1545   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1546 
1547   // Must use the return pc which was loaded from the caller's frame
1548   // as the VM uses return-pc-patching for deoptimization.
1549   __ mtlr(R0);
1550   __ blr();
1551 
1552   //-----------------------------------------------------------------------------
1553   // An exception is pending. We call into the runtime only if the
1554   // caller was not interpreted. If it was interpreted the
1555   // interpreter will do the correct thing. If it isn't interpreted
1556   // (call stub/compiled code) we will change our return and continue.
1557 
1558   BIND(exception_return_sync_check);
1559 
1560   if (synchronized) {
1561     __ unlock_object(R26_monitor); // Can also unlock methods.
1562   }
1563   BIND(exception_return_sync_check_already_unlocked);
1564 
1565   const Register return_pc = R31;
1566 
1567   __ ld(return_pc, 0, R1_SP);
1568   __ ld(return_pc, _abi0(lr), return_pc);
1569 
1570   // Get the address of the exception handler.
1571   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1572                   R16_thread,
1573                   return_pc /* return pc */);
1574   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1575 
1576   // Load the PC of the exception handler into LR.
1577   __ mtlr(R3_RET);
1578 
1579   // Load exception into R3_ARG1 and clear pending exception in thread.
1580   __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1581   __ li(R4_ARG2, 0);
1582   __ std(R4_ARG2, thread_(pending_exception));
1583 
1584   // Load the original return pc into R4_ARG2.
1585   __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1586 
1587   // Return to exception handler.
1588   __ blr();
1589 
1590   //=============================================================================
1591   // Counter overflow.
1592 
1593   if (inc_counter) {
1594     // Handle invocation counter overflow.
1595     __ bind(invocation_counter_overflow);
1596 
1597     generate_counter_overflow(continue_after_compile);
1598   }
1599 
1600   return entry;
1601 }
1602 
1603 // Generic interpreted method entry to (asm) interpreter.
1604 //
1605 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1606   bool inc_counter = UseCompiler || CountCompiledCalls;
1607   address entry = __ pc();
1608   // Generate the code to allocate the interpreter stack frame.
1609   Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1610            Rsize_of_locals     = R5_ARG3; // Written by generate_fixed_frame.
1611 
1612   // Does also a stack check to assure this frame fits on the stack.
1613   generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1614 
1615   // --------------------------------------------------------------------------
1616   // Zero out non-parameter locals.
1617   // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1618   // worth to ask the flag, just do it.
1619   Register Rslot_addr = R6_ARG4,
1620            Rnum       = R7_ARG5;
1621   Label Lno_locals, Lzero_loop;
1622 
1623   // Set up the zeroing loop.
1624   __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1625   __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1626   __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1627   __ beq(CCR0, Lno_locals);
1628   __ li(R0, 0);
1629   __ mtctr(Rnum);
1630 
1631   // The zero locals loop.
1632   __ bind(Lzero_loop);
1633   __ std(R0, 0, Rslot_addr);
1634   __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1635   __ bdnz(Lzero_loop);
1636 
1637   __ bind(Lno_locals);
1638 
1639   // --------------------------------------------------------------------------
1640   // Counter increment and overflow check.
1641   Label invocation_counter_overflow;
1642   Label continue_after_compile;
1643   if (inc_counter || ProfileInterpreter) {
1644 
1645     Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1646     if (synchronized) {
1647       // Since at this point in the method invocation the exception handler
1648       // would try to exit the monitor of synchronized methods which hasn't
1649       // been entered yet, we set the thread local variable
1650       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1651       // runtime, exception handling i.e. unlock_if_synchronized_method will
1652       // check this thread local flag.
1653       // This flag has two effects, one is to force an unwind in the topmost
1654       // interpreter frame and not perform an unlock while doing so.
1655       __ li(R0, 1);
1656       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1657     }
1658 
1659     // Argument and return type profiling.
1660     __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1661 
1662     // Increment invocation counter and check for overflow.
1663     if (inc_counter) {
1664       generate_counter_incr(&invocation_counter_overflow);
1665     }
1666 
1667     __ bind(continue_after_compile);
1668   }
1669 
1670   bang_stack_shadow_pages(false);
1671 
1672   if (inc_counter || ProfileInterpreter) {
1673     // Reset the _do_not_unlock_if_synchronized flag.
1674     if (synchronized) {
1675       __ li(R0, 0);
1676       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1677     }
1678   }
1679 
1680   // --------------------------------------------------------------------------
1681   // Locking of synchronized methods. Must happen AFTER invocation_counter
1682   // check and stack overflow check, so method is not locked if overflows.
1683   if (synchronized) {
1684     lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1685   }
1686 #ifdef ASSERT
1687   else {
1688     Label Lok;
1689     __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1690     __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1691     __ asm_assert_eq("method needs synchronization");
1692     __ bind(Lok);
1693   }
1694 #endif // ASSERT
1695 
1696   // --------------------------------------------------------------------------
1697   // JVMTI support
1698   __ notify_method_entry();
1699 
1700   // --------------------------------------------------------------------------
1701   // Start executing instructions.
1702   __ dispatch_next(vtos);
1703 
1704   // --------------------------------------------------------------------------
1705   if (inc_counter) {
1706     // Handle invocation counter overflow.
1707     __ bind(invocation_counter_overflow);
1708     generate_counter_overflow(continue_after_compile);
1709   }
1710   return entry;
1711 }
1712 
1713 // CRC32 Intrinsics.
1714 //
1715 // Contract on scratch and work registers.
1716 // =======================================
1717 //
1718 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1719 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1720 // You can't rely on these registers across calls.
1721 //
1722 // The generators for CRC32_update and for CRC32_updateBytes use the
1723 // scratch/work register set internally, passing the work registers
1724 // as arguments to the MacroAssembler emitters as required.
1725 //
1726 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1727 // Their contents is not constant but may change according to the requirements
1728 // of the emitted code.
1729 //
1730 // All other registers from the scratch/work register set are used "internally"
1731 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1732 // Basically, only R3_RET contains a defined value which is the function result.
1733 //
1734 /**
1735  * Method entry for static native methods:
1736  *   int java.util.zip.CRC32.update(int crc, int b)
1737  */
1738 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1739   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1740   address start = __ pc();  // Remember stub start address (is rtn value).
1741   Label slow_path;
1742 
1743   // Safepoint check
1744   const Register sync_state = R11_scratch1;
1745   __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1746 
1747   // We don't generate local frame and don't align stack because
1748   // we not even call stub code (we generate the code inline)
1749   // and there is no safepoint on this path.
1750 
1751   // Load java parameters.
1752   // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1753   const Register argP    = R15_esp;
1754   const Register crc     = R3_ARG1;  // crc value
1755   const Register data    = R4_ARG2;
1756   const Register table   = R5_ARG3;  // address of crc32 table
1757 
1758   BLOCK_COMMENT("CRC32_update {");
1759 
1760   // Arguments are reversed on java expression stack
1761 #ifdef VM_LITTLE_ENDIAN
1762   int data_offs = 0+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1763                                      // Being passed as an int, the single byte is at offset +0.
1764 #else
1765   int data_offs = 3+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1766                                      // Being passed from java as an int, the single byte is at offset +3.
1767 #endif
1768   __ lwz(crc, 2*wordSize, argP);     // Current crc state, zero extend to 64 bit to have a clean register.
1769   __ lbz(data, data_offs, argP);     // Byte from buffer, zero-extended.
1770   __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1771   __ kernel_crc32_singleByteReg(crc, data, table, true);
1772 
1773   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1774   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1775   __ blr();
1776 
1777   // Generate a vanilla native entry as the slow path.
1778   BLOCK_COMMENT("} CRC32_update");
1779   BIND(slow_path);
1780   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1781   return start;
1782 }
1783 
1784 /**
1785  * Method entry for static native methods:
1786  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1787  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1788  */
1789 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1790   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1791   address start = __ pc();  // Remember stub start address (is rtn value).
1792   Label slow_path;
1793 
1794   // Safepoint check
1795   const Register sync_state = R11_scratch1;
1796   __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1797 
1798   // We don't generate local frame and don't align stack because
1799   // we not even call stub code (we generate the code inline)
1800   // and there is no safepoint on this path.
1801 
1802   // Load parameters.
1803   // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1804   const Register argP    = R15_esp;
1805   const Register crc     = R3_ARG1;  // crc value
1806   const Register data    = R4_ARG2;  // address of java byte array
1807   const Register dataLen = R5_ARG3;  // source data len
1808   const Register tmp     = R11_scratch1;
1809 
1810   // Arguments are reversed on java expression stack.
1811   // Calculate address of start element.
1812   if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1813     BLOCK_COMMENT("CRC32_updateByteBuffer {");
1814     // crc     @ (SP + 5W) (32bit)
1815     // buf     @ (SP + 3W) (64bit ptr to long array)
1816     // off     @ (SP + 2W) (32bit)
1817     // dataLen @ (SP + 1W) (32bit)
1818     // data = buf + off
1819     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1820     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1821     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1822     __ lwz( crc,     5*wordSize, argP);  // current crc state
1823     __ add( data, data, tmp);            // Add byte buffer offset.
1824   } else {                                                         // Used for "updateBytes update".
1825     BLOCK_COMMENT("CRC32_updateBytes {");
1826     // crc     @ (SP + 4W) (32bit)
1827     // buf     @ (SP + 3W) (64bit ptr to byte array)
1828     // off     @ (SP + 2W) (32bit)
1829     // dataLen @ (SP + 1W) (32bit)
1830     // data = buf + off + base_offset
1831     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1832     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1833     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1834     __ add( data, data, tmp);            // add byte buffer offset
1835     __ lwz( crc,     4*wordSize, argP);  // current crc state
1836     __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1837   }
1838 
1839   __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1840 
1841   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1842   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1843   __ blr();
1844 
1845   // Generate a vanilla native entry as the slow path.
1846   BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1847   BIND(slow_path);
1848   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1849   return start;
1850 }
1851 
1852 
1853 /**
1854  * Method entry for intrinsic-candidate (non-native) methods:
1855  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
1856  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1857  * Unlike CRC32, CRC32C does not have any methods marked as native
1858  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1859  **/
1860 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1861   assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1862   address start = __ pc();  // Remember stub start address (is rtn value).
1863 
1864   // We don't generate local frame and don't align stack because
1865   // we not even call stub code (we generate the code inline)
1866   // and there is no safepoint on this path.
1867 
1868   // Load parameters.
1869   // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1870   const Register argP    = R15_esp;
1871   const Register crc     = R3_ARG1;  // crc value
1872   const Register data    = R4_ARG2;  // address of java byte array
1873   const Register dataLen = R5_ARG3;  // source data len
1874   const Register tmp     = R11_scratch1;
1875 
1876   // Arguments are reversed on java expression stack.
1877   // Calculate address of start element.
1878   if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1879     BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1880     // crc     @ (SP + 5W) (32bit)
1881     // buf     @ (SP + 3W) (64bit ptr to long array)
1882     // off     @ (SP + 2W) (32bit)
1883     // dataLen @ (SP + 1W) (32bit)
1884     // data = buf + off
1885     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1886     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1887     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1888     __ lwz( crc,     5*wordSize, argP);  // current crc state
1889     __ add( data, data, tmp);            // Add byte buffer offset.
1890     __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1891   } else {                                                         // Used for "updateBytes update".
1892     BLOCK_COMMENT("CRC32C_updateBytes {");
1893     // crc     @ (SP + 4W) (32bit)
1894     // buf     @ (SP + 3W) (64bit ptr to byte array)
1895     // off     @ (SP + 2W) (32bit)
1896     // dataLen @ (SP + 1W) (32bit)
1897     // data = buf + off + base_offset
1898     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1899     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1900     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1901     __ add( data, data, tmp);            // add byte buffer offset
1902     __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1903     __ lwz( crc,     4*wordSize, argP);  // current crc state
1904     __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1905   }
1906 
1907   __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1908 
1909   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1910   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1911   __ blr();
1912 
1913   BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1914   return start;
1915 }
1916 
1917 // Not supported
1918 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; }
1919 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1920 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1921 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1922 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1923 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() { return nullptr; }
1924 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() { return nullptr; }
1925 
1926 // =============================================================================
1927 // Exceptions
1928 
1929 void TemplateInterpreterGenerator::generate_throw_exception() {
1930   Register Rexception    = R17_tos,
1931            Rcontinuation = R3_RET;
1932 
1933   // --------------------------------------------------------------------------
1934   // Entry point if an method returns with a pending exception (rethrow).
1935   Interpreter::_rethrow_exception_entry = __ pc();
1936   {
1937     __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
1938 
1939     // Compiled code destroys templateTableBase, reload.
1940     __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
1941   }
1942 
1943   // Entry point if a interpreted method throws an exception (throw).
1944   Interpreter::_throw_exception_entry = __ pc();
1945   {
1946     __ mr(Rexception, R3_RET);
1947 
1948     __ verify_oop(Rexception);
1949 
1950     // Expression stack must be empty before entering the VM in case of an exception.
1951     __ empty_expression_stack();
1952     // Find exception handler address and preserve exception oop.
1953     // Call C routine to find handler and jump to it.
1954     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
1955     __ mtctr(Rcontinuation);
1956     // Push exception for exception handler bytecodes.
1957     __ push_ptr(Rexception);
1958 
1959     // Jump to exception handler (may be remove activation entry!).
1960     __ bctr();
1961   }
1962 
1963   // If the exception is not handled in the current frame the frame is
1964   // removed and the exception is rethrown (i.e. exception
1965   // continuation is _rethrow_exception).
1966   //
1967   // Note: At this point the bci is still the bxi for the instruction
1968   // which caused the exception and the expression stack is
1969   // empty. Thus, for any VM calls at this point, GC will find a legal
1970   // oop map (with empty expression stack).
1971 
1972   // In current activation
1973   // tos: exception
1974   // bcp: exception bcp
1975 
1976   // --------------------------------------------------------------------------
1977   // JVMTI PopFrame support
1978 
1979   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1980   {
1981     // Set the popframe_processing bit in popframe_condition indicating that we are
1982     // currently handling popframe, so that call_VMs that may happen later do not
1983     // trigger new popframe handling cycles.
1984     __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1985     __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
1986     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
1987 
1988     // Empty the expression stack, as in normal exception handling.
1989     __ empty_expression_stack();
1990     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1991 
1992     // Check to see whether we are returning to a deoptimized frame.
1993     // (The PopFrame call ensures that the caller of the popped frame is
1994     // either interpreted or compiled and deoptimizes it if compiled.)
1995     // Note that we don't compare the return PC against the
1996     // deoptimization blob's unpack entry because of the presence of
1997     // adapter frames in C2.
1998     Label Lcaller_not_deoptimized;
1999     Register return_pc = R3_ARG1;
2000     __ ld(return_pc, 0, R1_SP);
2001     __ ld(return_pc, _abi0(lr), return_pc);
2002     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2003     __ cmpdi(CCR0, R3_RET, 0);
2004     __ bne(CCR0, Lcaller_not_deoptimized);
2005 
2006     // The deoptimized case.
2007     // In this case, we can't call dispatch_next() after the frame is
2008     // popped, but instead must save the incoming arguments and restore
2009     // them after deoptimization has occurred.
2010     __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2011     __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2012     __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2013     __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2014     __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2015     // Save these arguments.
2016     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2017 
2018     // Inform deoptimization that it is responsible for restoring these arguments.
2019     __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2020     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2021 
2022     // Return from the current method into the deoptimization blob. Will eventually
2023     // end up in the deopt interpreter entry, deoptimization prepared everything that
2024     // we will reexecute the call that called us.
2025     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2026     __ mtlr(return_pc);
2027     __ pop_cont_fastpath();
2028     __ blr();
2029 
2030     // The non-deoptimized case.
2031     __ bind(Lcaller_not_deoptimized);
2032 
2033     // Clear the popframe condition flag.
2034     __ li(R0, 0);
2035     __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2036 
2037     // Get out of the current method and re-execute the call that called us.
2038     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2039     __ pop_cont_fastpath();
2040     __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2041     if (ProfileInterpreter) {
2042       __ set_method_data_pointer_for_bcp();
2043       __ ld(R11_scratch1, 0, R1_SP);
2044       __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2045     }
2046 #if INCLUDE_JVMTI
2047     Label L_done;
2048 
2049     __ lbz(R11_scratch1, 0, R14_bcp);
2050     __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2051     __ bne(CCR0, L_done);
2052 
2053     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2054     // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
2055     __ ld(R4_ARG2, 0, R18_locals);
2056     __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2057 
2058     __ cmpdi(CCR0, R4_ARG2, 0);
2059     __ beq(CCR0, L_done);
2060     __ std(R4_ARG2, wordSize, R15_esp);
2061     __ bind(L_done);
2062 #endif // INCLUDE_JVMTI
2063     __ dispatch_next(vtos);
2064   }
2065   // end of JVMTI PopFrame support
2066 
2067   // --------------------------------------------------------------------------
2068   // Remove activation exception entry.
2069   // This is jumped to if an interpreted method can't handle an exception itself
2070   // (we come from the throw/rethrow exception entry above). We're going to call
2071   // into the VM to find the exception handler in the caller, pop the current
2072   // frame and return the handler we calculated.
2073   Interpreter::_remove_activation_entry = __ pc();
2074   {
2075     __ pop_ptr(Rexception);
2076     __ verify_oop(Rexception);
2077     __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2078 
2079     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2080     __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2081 
2082     __ get_vm_result(Rexception);
2083 
2084     // We are done with this activation frame; find out where to go next.
2085     // The continuation point will be an exception handler, which expects
2086     // the following registers set up:
2087     //
2088     // RET:  exception oop
2089     // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2090 
2091     Register return_pc = R31; // Needs to survive the runtime call.
2092     __ ld(return_pc, 0, R1_SP);
2093     __ ld(return_pc, _abi0(lr), return_pc);
2094     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2095 
2096     // Remove the current activation.
2097     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2098     __ pop_cont_fastpath();
2099 
2100     __ mr(R4_ARG2, return_pc);
2101     __ mtlr(R3_RET);
2102     __ mr(R3_RET, Rexception);
2103     __ blr();
2104   }
2105 }
2106 
2107 // JVMTI ForceEarlyReturn support.
2108 // Returns "in the middle" of a method with a "fake" return value.
2109 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2110 
2111   Register Rscratch1 = R11_scratch1,
2112            Rscratch2 = R12_scratch2;
2113 
2114   address entry = __ pc();
2115   __ empty_expression_stack();
2116 
2117   __ load_earlyret_value(state, Rscratch1);
2118 
2119   __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2120   // Clear the earlyret state.
2121   __ li(R0, 0);
2122   __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2123 
2124   __ remove_activation(state, false, false);
2125   // Copied from TemplateTable::_return.
2126   // Restoration of lr done by remove_activation.
2127   switch (state) {
2128     // Narrow result if state is itos but result type is smaller.
2129     case btos:
2130     case ztos:
2131     case ctos:
2132     case stos:
2133     case itos: __ narrow(R17_tos); /* fall through */
2134     case ltos:
2135     case atos: __ mr(R3_RET, R17_tos); break;
2136     case ftos:
2137     case dtos: __ fmr(F1_RET, F15_ftos); break;
2138     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2139                // to get visible before the reference to the object gets stored anywhere.
2140                __ membar(Assembler::StoreStore); break;
2141     default  : ShouldNotReachHere();
2142   }
2143   __ blr();
2144 
2145   return entry;
2146 } // end of ForceEarlyReturn support
2147 
2148 //-----------------------------------------------------------------------------
2149 // Helper for vtos entry point generation
2150 
2151 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2152                                                          address& bep,
2153                                                          address& cep,
2154                                                          address& sep,
2155                                                          address& aep,
2156                                                          address& iep,
2157                                                          address& lep,
2158                                                          address& fep,
2159                                                          address& dep,
2160                                                          address& vep) {
2161   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2162   Label L;
2163 
2164   aep = __ pc();  __ push_ptr();  __ b(L);
2165   fep = __ pc();  __ push_f();    __ b(L);
2166   dep = __ pc();  __ push_d();    __ b(L);
2167   lep = __ pc();  __ push_l();    __ b(L);
2168   __ align(32, 12, 24); // align L
2169   bep = cep = sep =
2170   iep = __ pc();  __ push_i();
2171   vep = __ pc();
2172   __ bind(L);
2173   generate_and_dispatch(t);
2174 }
2175 
2176 //-----------------------------------------------------------------------------
2177 
2178 // Non-product code
2179 #ifndef PRODUCT
2180 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2181   //__ flush_bundle();
2182   address entry = __ pc();
2183 
2184   const char *bname = nullptr;
2185   uint tsize = 0;
2186   switch(state) {
2187   case ftos:
2188     bname = "trace_code_ftos {";
2189     tsize = 2;
2190     break;
2191   case btos:
2192     bname = "trace_code_btos {";
2193     tsize = 2;
2194     break;
2195   case ztos:
2196     bname = "trace_code_ztos {";
2197     tsize = 2;
2198     break;
2199   case ctos:
2200     bname = "trace_code_ctos {";
2201     tsize = 2;
2202     break;
2203   case stos:
2204     bname = "trace_code_stos {";
2205     tsize = 2;
2206     break;
2207   case itos:
2208     bname = "trace_code_itos {";
2209     tsize = 2;
2210     break;
2211   case ltos:
2212     bname = "trace_code_ltos {";
2213     tsize = 3;
2214     break;
2215   case atos:
2216     bname = "trace_code_atos {";
2217     tsize = 2;
2218     break;
2219   case vtos:
2220     // Note: In case of vtos, the topmost of stack value could be a int or doubl
2221     // In case of a double (2 slots) we won't see the 2nd stack value.
2222     // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2223     bname = "trace_code_vtos {";
2224     tsize = 2;
2225 
2226     break;
2227   case dtos:
2228     bname = "trace_code_dtos {";
2229     tsize = 3;
2230     break;
2231   default:
2232     ShouldNotReachHere();
2233   }
2234   BLOCK_COMMENT(bname);
2235 
2236   // Support short-cut for TraceBytecodesAt.
2237   // Don't call into the VM if we don't want to trace to speed up things.
2238   Label Lskip_vm_call;
2239   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2240     int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2241     int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2242     __ ld(R11_scratch1, offs1, R11_scratch1);
2243     __ lwa(R12_scratch2, offs2, R12_scratch2);
2244     __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2245     __ blt(CCR0, Lskip_vm_call);
2246   }
2247 
2248   __ push(state);
2249   // Load 2 topmost expression stack values.
2250   __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2251   __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2252   __ mflr(R31);
2253   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2254   __ mtlr(R31);
2255   __ pop(state);
2256 
2257   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2258     __ bind(Lskip_vm_call);
2259   }
2260   __ blr();
2261   BLOCK_COMMENT("} trace_code");
2262   return entry;
2263 }
2264 
2265 void TemplateInterpreterGenerator::count_bytecode() {
2266   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2267   __ lwz(R12_scratch2, offs, R11_scratch1);
2268   __ addi(R12_scratch2, R12_scratch2, 1);
2269   __ stw(R12_scratch2, offs, R11_scratch1);
2270 }
2271 
2272 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2273   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2274   __ lwz(R12_scratch2, offs, R11_scratch1);
2275   __ addi(R12_scratch2, R12_scratch2, 1);
2276   __ stw(R12_scratch2, offs, R11_scratch1);
2277 }
2278 
2279 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2280   const Register addr = R11_scratch1,
2281                  tmp  = R12_scratch2;
2282   // Get index, shift out old bytecode, bring in new bytecode, and store it.
2283   // _index = (_index >> log2_number_of_codes) |
2284   //          (bytecode << log2_number_of_codes);
2285   int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2286   __ lwz(tmp, offs1, addr);
2287   __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2288   __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2289   __ stw(tmp, offs1, addr);
2290 
2291   // Bump bucket contents.
2292   // _counters[_index] ++;
2293   int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2294   __ sldi(tmp, tmp, LogBytesPerInt);
2295   __ add(addr, tmp, addr);
2296   __ lwz(tmp, offs2, addr);
2297   __ addi(tmp, tmp, 1);
2298   __ stw(tmp, offs2, addr);
2299 }
2300 
2301 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2302   // Call a little run-time stub to avoid blow-up for each bytecode.
2303   // The run-time runtime saves the right registers, depending on
2304   // the tosca in-state for the given template.
2305 
2306   assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2307          "entry must have been generated");
2308 
2309   // Note: we destroy LR here.
2310   __ bl(Interpreter::trace_code(t->tos_in()));
2311 }
2312 
2313 void TemplateInterpreterGenerator::stop_interpreter_at() {
2314   Label L;
2315   int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2316   int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2317   __ ld(R11_scratch1, offs1, R11_scratch1);
2318   __ lwa(R12_scratch2, offs2, R12_scratch2);
2319   __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2320   __ bne(CCR0, L);
2321   __ illtrap();
2322   __ bind(L);
2323 }
2324 
2325 #endif // !PRODUCT