1 /*
   2  * Copyright (c) 2014, 2025, 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 16 bit value.
 148     assert(sizeof(AccessFlags) == 2, "wrong size");
 149     __ lhz(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 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
 700   if (!Continuations::enabled()) return nullptr;
 701   address start = __ pc();
 702 
 703   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
 704   __ restore_interpreter_state(R11_scratch1, false, true /*restore_top_frame_sp*/);
 705   __ blr();
 706 
 707   return start;
 708 }
 709 
 710 // Helpers for commoning out cases in the various type of method entries.
 711 
 712 // Increment invocation count & check for overflow.
 713 //
 714 // Note: checking for negative value instead of overflow
 715 //       so we have a 'sticky' overflow test.
 716 //
 717 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 718   // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
 719   Register Rscratch1   = R11_scratch1;
 720   Register Rscratch2   = R12_scratch2;
 721   Register R3_counters = R3_ARG1;
 722   Label done;
 723 
 724   const int increment = InvocationCounter::count_increment;
 725   Label no_mdo;
 726   if (ProfileInterpreter) {
 727     const Register Rmdo = R3_counters;
 728     __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
 729     __ cmpdi(CCR0, Rmdo, 0);
 730     __ beq(CCR0, no_mdo);
 731 
 732     // Increment invocation counter in the MDO.
 733     const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 734     __ lwz(Rscratch2, mdo_ic_offs, Rmdo);
 735     __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo);
 736     __ addi(Rscratch2, Rscratch2, increment);
 737     __ stw(Rscratch2, mdo_ic_offs, Rmdo);
 738     __ and_(Rscratch1, Rscratch2, Rscratch1);
 739     __ bne(CCR0, done);
 740     __ b(*overflow);
 741   }
 742 
 743   // Increment counter in MethodCounters*.
 744   const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
 745   __ bind(no_mdo);
 746   __ get_method_counters(R19_method, R3_counters, done);
 747   __ lwz(Rscratch2, mo_ic_offs, R3_counters);
 748   __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters);
 749   __ addi(Rscratch2, Rscratch2, increment);
 750   __ stw(Rscratch2, mo_ic_offs, R3_counters);
 751   __ and_(Rscratch1, Rscratch2, Rscratch1);
 752   __ beq(CCR0, *overflow);
 753 
 754   __ bind(done);
 755 }
 756 
 757 // Generate code to initiate compilation on invocation counter overflow.
 758 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
 759   // Generate code to initiate compilation on the counter overflow.
 760 
 761   // InterpreterRuntime::frequency_counter_overflow takes one arguments,
 762   // which indicates if the counter overflow occurs at a backwards branch (null bcp)
 763   // We pass zero in.
 764   // The call returns the address of the verified entry point for the method or null
 765   // if the compilation did not complete (either went background or bailed out).
 766   //
 767   // Unlike the C++ interpreter above: Check exceptions!
 768   // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
 769   // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
 770 
 771   __ li(R4_ARG2, 0);
 772   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
 773 
 774   // Returns verified_entry_point or null.
 775   // We ignore it in any case.
 776   __ b(continue_entry);
 777 }
 778 
 779 // See if we've got enough room on the stack for locals plus overhead below
 780 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 781 // without going through the signal handler, i.e., reserved and yellow zones
 782 // will not be made usable. The shadow zone must suffice to handle the
 783 // overflow.
 784 //
 785 // Kills Rmem_frame_size, Rscratch1.
 786 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
 787   Label done;
 788   assert_different_registers(Rmem_frame_size, Rscratch1);
 789 
 790   BLOCK_COMMENT("stack_overflow_check_with_compare {");
 791   __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size);
 792   __ ld(Rscratch1, thread_(stack_overflow_limit));
 793   __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1);
 794   __ bgt(CCR0/*is_stack_overflow*/, done);
 795 
 796   // The stack overflows. Load target address of the runtime stub and call it.
 797   assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "generated in wrong order");
 798   __ load_const_optimized(Rscratch1, (SharedRuntime::throw_StackOverflowError_entry()), R0);
 799   __ mtctr(Rscratch1);
 800   // Restore caller_sp (c2i adapter may exist, but no shrinking of interpreted caller frame).
 801 #ifdef ASSERT
 802   Label frame_not_shrunk;
 803   __ cmpld(CCR0, R1_SP, R21_sender_SP);
 804   __ ble(CCR0, frame_not_shrunk);
 805   __ stop("frame shrunk");
 806   __ bind(frame_not_shrunk);
 807   __ ld(Rscratch1, 0, R1_SP);
 808   __ ld(R0, 0, R21_sender_SP);
 809   __ cmpd(CCR0, R0, Rscratch1);
 810   __ asm_assert_eq("backlink");
 811 #endif // ASSERT
 812   __ mr(R1_SP, R21_sender_SP);
 813   __ bctr();
 814 
 815   __ align(32, 12);
 816   __ bind(done);
 817   BLOCK_COMMENT("} stack_overflow_check_with_compare");
 818 }
 819 
 820 // Lock the current method, interpreter register window must be set up!
 821 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
 822   const Register Robj_to_lock = Rscratch2;
 823 
 824   {
 825     if (!flags_preloaded) {
 826       __ lhz(Rflags, method_(access_flags));
 827     }
 828 
 829 #ifdef ASSERT
 830     // Check if methods needs synchronization.
 831     {
 832       Label Lok;
 833       __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
 834       __ btrue(CCR0,Lok);
 835       __ stop("method doesn't need synchronization");
 836       __ bind(Lok);
 837     }
 838 #endif // ASSERT
 839   }
 840 
 841   // Get synchronization object to Rscratch2.
 842   {
 843     Label Lstatic;
 844     Label Ldone;
 845 
 846     __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
 847     __ btrue(CCR0, Lstatic);
 848 
 849     // Non-static case: load receiver obj from stack and we're done.
 850     __ ld(Robj_to_lock, R18_locals);
 851     __ b(Ldone);
 852 
 853     __ bind(Lstatic); // Static case: Lock the java mirror
 854     // Load mirror from interpreter frame.
 855     __ ld(Robj_to_lock, _abi0(callers_sp), R1_SP);
 856     __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock);
 857 
 858     __ bind(Ldone);
 859     __ verify_oop(Robj_to_lock);
 860   }
 861 
 862   // Got the oop to lock => execute!
 863   __ add_monitor_to_stack(true, Rscratch1, R0);
 864 
 865   __ std(Robj_to_lock, in_bytes(BasicObjectLock::obj_offset()), R26_monitor);
 866   __ lock_object(R26_monitor, Robj_to_lock);
 867 }
 868 
 869 // Generate a fixed interpreter frame for pure interpreter
 870 // and I2N native transition frames.
 871 //
 872 // Before (stack grows downwards):
 873 //
 874 //         |  ...         |
 875 //         |------------- |
 876 //         |  java arg0   |
 877 //         |  ...         |
 878 //         |  java argn   |
 879 //         |              |   <-   R15_esp
 880 //         |              |
 881 //         |--------------|
 882 //         | abi_112      |
 883 //         |              |   <-   R1_SP
 884 //         |==============|
 885 //
 886 //
 887 // After:
 888 //
 889 //         |  ...         |
 890 //         |  java arg0   |<-   R18_locals
 891 //         |  ...         |
 892 //         |  java argn   |
 893 //         |--------------|
 894 //         |              |
 895 //         |  java locals |
 896 //         |              |
 897 //         |--------------|
 898 //         |  abi_48      |
 899 //         |==============|
 900 //         |              |
 901 //         |   istate     |
 902 //         |              |
 903 //         |--------------|
 904 //         |   monitor    |<-   R26_monitor
 905 //         |--------------|
 906 //         |              |<-   R15_esp
 907 //         | expression   |
 908 //         | stack        |
 909 //         |              |
 910 //         |--------------|
 911 //         |              |
 912 //         | abi_112      |<-   R1_SP
 913 //         |==============|
 914 //
 915 // The top most frame needs an abi space of 112 bytes. This space is needed,
 916 // since we call to c. The c function may spill their arguments to the caller
 917 // frame. When we call to java, we don't need these spill slots. In order to save
 918 // space on the stack, we resize the caller. However, java locals reside in
 919 // the caller frame and the frame has to be increased. The frame_size for the
 920 // current frame was calculated based on max_stack as size for the expression
 921 // stack. At the call, just a part of the expression stack might be used.
 922 // We don't want to waste this space and cut the frame back accordingly.
 923 // The resulting amount for resizing is calculated as follows:
 924 // resize =   (number_of_locals - number_of_arguments) * slot_size
 925 //          + (R1_SP - R15_esp) + 48
 926 //
 927 // The size for the callee frame is calculated:
 928 // framesize = 112 + max_stack + monitor + state_size
 929 //
 930 // maxstack:   Max number of slots on the expression stack, loaded from the method.
 931 // monitor:    We statically reserve room for one monitor object.
 932 // state_size: We save the current state of the interpreter to this area.
 933 //
 934 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
 935   Register Rparent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
 936            Rtop_frame_size      = R7_ARG5,
 937            Rconst_method        = R8_ARG6,
 938            Rconst_pool          = R9_ARG7,
 939            Rmirror              = R10_ARG8;
 940 
 941   assert_different_registers(Rsize_of_parameters, Rsize_of_locals, Rparent_frame_resize, Rtop_frame_size,
 942                              Rconst_method, Rconst_pool);
 943 
 944   __ ld(Rconst_method, method_(const));
 945   __ lhz(Rsize_of_parameters /* number of params */,
 946          in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
 947   if (native_call) {
 948     // If we're calling a native method, we reserve space for the worst-case signature
 949     // handler varargs vector, which is max(Argument::n_int_register_parameters_c, parameter_count+2).
 950     // We add two slots to the parameter_count, one for the jni
 951     // environment and one for a possible native mirror.
 952     Label skip_native_calculate_max_stack;
 953     __ addi(Rtop_frame_size, Rsize_of_parameters, 2);
 954     __ cmpwi(CCR0, Rtop_frame_size, Argument::n_int_register_parameters_c);
 955     __ bge(CCR0, skip_native_calculate_max_stack);
 956     __ li(Rtop_frame_size, Argument::n_int_register_parameters_c);
 957     __ bind(skip_native_calculate_max_stack);
 958     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 959     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 960     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 961     assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
 962   } else {
 963     __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
 964     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
 965     __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
 966     __ lhz(Rtop_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
 967     __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
 968     __ sub(Rparent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
 969     __ sldi(Rtop_frame_size, Rtop_frame_size, Interpreter::logStackElementSize);
 970     __ add(Rparent_frame_resize, Rparent_frame_resize, R11_scratch1);
 971   }
 972 
 973   // Compute top frame size.
 974   __ addi(Rtop_frame_size, Rtop_frame_size, frame::top_ijava_frame_abi_size + frame::ijava_state_size);
 975 
 976   // Cut back area between esp and max_stack.
 977   __ addi(Rparent_frame_resize, Rparent_frame_resize, frame::parent_ijava_frame_abi_size - Interpreter::stackElementSize);
 978 
 979   __ round_to(Rtop_frame_size, frame::alignment_in_bytes);
 980   __ round_to(Rparent_frame_resize, frame::alignment_in_bytes);
 981   // Rparent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
 982   // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
 983 
 984   if (!native_call) {
 985     // Stack overflow check.
 986     // Native calls don't need the stack size check since they have no
 987     // expression stack and the arguments are already on the stack and
 988     // we only add a handful of words to the stack.
 989     __ add(R11_scratch1, Rparent_frame_resize, Rtop_frame_size);
 990     generate_stack_overflow_check(R11_scratch1, R12_scratch2);
 991   }
 992 
 993   // Set up interpreter state registers.
 994 
 995   __ add(R18_locals, R15_esp, Rsize_of_parameters);
 996   __ ld(Rconst_pool, in_bytes(ConstMethod::constants_offset()), Rconst_method);
 997   __ ld(R27_constPoolCache, ConstantPool::cache_offset(), Rconst_pool);
 998 
 999   // Set method data pointer.
1000   if (ProfileInterpreter) {
1001     Label zero_continue;
1002     __ ld(R28_mdx, method_(method_data));
1003     __ cmpdi(CCR0, R28_mdx, 0);
1004     __ beq(CCR0, zero_continue);
1005     __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
1006     __ bind(zero_continue);
1007   }
1008 
1009   if (native_call) {
1010     __ li(R14_bcp, 0); // Must initialize.
1011   } else {
1012     __ addi(R14_bcp, Rconst_method, in_bytes(ConstMethod::codes_offset()));
1013   }
1014 
1015   // Resize parent frame.
1016   __ mflr(R12_scratch2);
1017   __ neg(Rparent_frame_resize, Rparent_frame_resize);
1018   __ resize_frame(Rparent_frame_resize, R11_scratch1);
1019   __ std(R12_scratch2, _abi0(lr), R1_SP);
1020 
1021   // Get mirror and store it in the frame as GC root for this Method*.
1022   __ ld(Rmirror, ConstantPool::pool_holder_offset(), Rconst_pool);
1023   __ ld(Rmirror, in_bytes(Klass::java_mirror_offset()), Rmirror);
1024   __ resolve_oop_handle(Rmirror, R11_scratch1, R12_scratch2, MacroAssembler::PRESERVATION_FRAME_LR_GP_REGS);
1025 
1026   __ addi(R26_monitor, R1_SP, -frame::ijava_state_size);
1027   __ addi(R15_esp, R26_monitor, -Interpreter::stackElementSize);
1028 
1029   // Store values.
1030   __ std(R19_method, _ijava_state_neg(method), R1_SP);
1031   __ std(Rmirror, _ijava_state_neg(mirror), R1_SP);
1032   __ sub(R12_scratch2, R18_locals, R1_SP);
1033   __ srdi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1034   // Store relativized R18_locals, see frame::interpreter_frame_locals().
1035   __ std(R12_scratch2, _ijava_state_neg(locals), R1_SP);
1036   __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
1037 
1038   // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
1039   // be found in the frame after save_interpreter_state is done. This is always true
1040   // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
1041   // because e.g. frame::interpreter_frame_bcp() will not access the correct value
1042   // (Enhanced Stack Trace).
1043   // The signal handler does not save the interpreter state into the frame.
1044 
1045   // We have to initialize some of these frame slots for native calls (accessed by GC).
1046   // Also initialize them for non-native calls for better tool support (even though
1047   // you may not get the most recent version as described above).
1048   __ li(R0, 0);
1049   __ li(R12_scratch2, -(frame::ijava_state_size / wordSize));
1050   __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
1051   __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
1052   if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
1053   __ sub(R12_scratch2, R15_esp, R1_SP);
1054   __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1055   __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
1056   __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); // only used for native_call
1057 
1058   // Store sender's SP and this frame's top SP.
1059   __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
1060   __ neg(R12_scratch2, Rtop_frame_size);
1061   __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1062   // Store relativized top_frame_sp
1063   __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
1064 
1065   // Push top frame.
1066   __ push_frame(Rtop_frame_size, R11_scratch1);
1067 }
1068 
1069 // End of helpers
1070 
1071 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1072 
1073   // Decide what to do: Use same platform specific instructions and runtime calls as compilers.
1074   bool use_instruction = false;
1075   address runtime_entry = nullptr;
1076   int num_args = 1;
1077   bool double_precision = true;
1078 
1079   // PPC64 specific:
1080   switch (kind) {
1081     case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break;
1082     case Interpreter::java_lang_math_abs:  use_instruction = true; break;
1083     case Interpreter::java_lang_math_fmaF:
1084     case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break;
1085     default: break; // Fall back to runtime call.
1086   }
1087 
1088   switch (kind) {
1089     case Interpreter::java_lang_math_sin  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);   break;
1090     case Interpreter::java_lang_math_cos  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);   break;
1091     case Interpreter::java_lang_math_tan  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);   break;
1092     case Interpreter::java_lang_math_tanh : /* run interpreted */ break;
1093     case Interpreter::java_lang_math_abs  : /* run interpreted */ break;
1094     case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt);  break;
1095     case Interpreter::java_lang_math_log  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);   break;
1096     case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break;
1097     case Interpreter::java_lang_math_pow  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break;
1098     case Interpreter::java_lang_math_exp  : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);   break;
1099     case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break;
1100     case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break;
1101     default: ShouldNotReachHere();
1102   }
1103 
1104   // Use normal entry if neither instruction nor runtime call is used.
1105   if (!use_instruction && runtime_entry == nullptr) return nullptr;
1106 
1107   address entry = __ pc();
1108 
1109   // Load arguments
1110   assert(num_args <= 13, "passed in registers");
1111   if (double_precision) {
1112     int offset = (2 * num_args - 1) * Interpreter::stackElementSize;
1113     for (int i = 0; i < num_args; ++i) {
1114       __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1115       offset -= 2 * Interpreter::stackElementSize;
1116     }
1117   } else {
1118     int offset = num_args * Interpreter::stackElementSize;
1119     for (int i = 0; i < num_args; ++i) {
1120       __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp);
1121       offset -= Interpreter::stackElementSize;
1122     }
1123   }
1124 
1125   if (use_instruction) {
1126     switch (kind) {
1127       case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1);          break;
1128       case Interpreter::java_lang_math_abs:  __ fabs(F1_RET, F1);           break;
1129       case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break;
1130       case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3);  break;
1131       default: ShouldNotReachHere();
1132     }
1133   } else {
1134     // Comment: Can use tail call if the unextended frame is always C ABI compliant:
1135     //__ load_const_optimized(R12_scratch2, runtime_entry, R0);
1136     //__ call_c_and_return_to_caller(R12_scratch2);
1137 
1138     // Push a new C frame and save LR.
1139     __ save_LR(R0);
1140     __ push_frame_reg_args(0, R11_scratch1);
1141 
1142     __ call_VM_leaf(runtime_entry);
1143 
1144     // Pop the C frame and restore LR.
1145     __ pop_frame();
1146     __ restore_LR(R0);
1147   }
1148 
1149   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1150   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1151   __ blr();
1152 
1153   __ flush();
1154 
1155   return entry;
1156 }
1157 
1158 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
1159   if (!VM_Version::supports_float16()) return nullptr;
1160 
1161   address entry = __ pc();
1162 
1163   __ lfs(F1, Interpreter::stackElementSize, R15_esp);
1164   __ f2hf(R3_RET, F1, F0);
1165 
1166   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1167   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1168   __ blr();
1169 
1170   __ flush();
1171 
1172   return entry;
1173 }
1174 
1175 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
1176   if (!VM_Version::supports_float16()) return nullptr;
1177 
1178   address entry = __ pc();
1179 
1180   // Note: Could also use:
1181   //__ li(R3, Interpreter::stackElementSize);
1182   //__ lfiwax(F1_RET, R15_esp, R3); // short stored as 32 bit integer
1183   //__ xscvhpdp(F1_RET->to_vsr(), F1_RET->to_vsr());
1184   __ lwa(R3, Interpreter::stackElementSize, R15_esp);
1185   __ hf2f(F1_RET, R3);
1186 
1187   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1188   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1189   __ blr();
1190 
1191   __ flush();
1192 
1193   return entry;
1194 }
1195 
1196 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1197   // Quick & dirty stack overflow checking: bang the stack & handle trap.
1198   // Note that we do the banging after the frame is setup, since the exception
1199   // handling code expects to find a valid interpreter frame on the stack.
1200   // Doing the banging earlier fails if the caller frame is not an interpreter
1201   // frame.
1202   // (Also, the exception throwing code expects to unlock any synchronized
1203   // method receiever, so do the banging after locking the receiver.)
1204 
1205   // Bang each page in the shadow zone. We can't assume it's been done for
1206   // an interpreter frame with greater than a page of locals, so each page
1207   // needs to be checked.  Only true for non-native.
1208   const size_t page_size = os::vm_page_size();
1209   const int n_shadow_pages = StackOverflow::stack_shadow_zone_size() / page_size;
1210   const int start_page = native_call ? n_shadow_pages : 1;
1211   BLOCK_COMMENT("bang_stack_shadow_pages:");
1212   for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1213     __ bang_stack_with_offset(pages*page_size);
1214   }
1215 }
1216 
1217 // Interpreter stub for calling a native method. (asm interpreter)
1218 // This sets up a somewhat different looking stack for calling the
1219 // native method than the typical interpreter frame setup.
1220 //
1221 // On entry:
1222 //   R19_method    - method
1223 //   R16_thread    - JavaThread*
1224 //   R15_esp       - intptr_t* sender tos
1225 //
1226 //   abstract stack (grows up)
1227 //     [  IJava (caller of JNI callee)  ]  <-- ASP
1228 //        ...
1229 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
1230 
1231   address entry = __ pc();
1232 
1233   const bool inc_counter = UseCompiler || CountCompiledCalls;
1234 
1235   // -----------------------------------------------------------------------------
1236   // Allocate a new frame that represents the native callee (i2n frame).
1237   // This is not a full-blown interpreter frame, but in particular, the
1238   // following registers are valid after this:
1239   // - R19_method
1240   // - R18_local (points to start of arguments to native function)
1241   //
1242   //   abstract stack (grows up)
1243   //     [  IJava (caller of JNI callee)  ]  <-- ASP
1244   //        ...
1245 
1246   const Register signature_handler_fd = R11_scratch1;
1247   const Register pending_exception    = R0;
1248   const Register result_handler_addr  = R31;
1249   const Register native_method_fd     = R12_scratch2; // preferred in MacroAssembler::branch_to
1250   const Register access_flags         = R22_tmp2;
1251   const Register active_handles       = R11_scratch1; // R26_monitor saved to state.
1252   const Register sync_state           = R12_scratch2;
1253   const Register sync_state_addr      = sync_state;   // Address is dead after use.
1254   const Register suspend_flags        = R11_scratch1;
1255 
1256   //=============================================================================
1257   // Allocate new frame and initialize interpreter state.
1258 
1259   Label exception_return;
1260   Label exception_return_sync_check;
1261   Label stack_overflow_return;
1262 
1263   Register size_of_parameters = R22_tmp2;
1264 
1265   generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
1266 
1267   //=============================================================================
1268   // Increment invocation counter. On overflow, entry to JNI method
1269   // will be compiled.
1270   Label invocation_counter_overflow, continue_after_compile;
1271   if (inc_counter) {
1272     if (synchronized) {
1273       // Since at this point in the method invocation the exception handler
1274       // would try to exit the monitor of synchronized methods which hasn't
1275       // been entered yet, we set the thread local variable
1276       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1277       // runtime, exception handling i.e. unlock_if_synchronized_method will
1278       // check this thread local flag.
1279       // This flag has two effects, one is to force an unwind in the topmost
1280       // interpreter frame and not perform an unlock while doing so.
1281       __ li(R0, 1);
1282       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1283     }
1284     generate_counter_incr(&invocation_counter_overflow);
1285 
1286     BIND(continue_after_compile);
1287   }
1288 
1289   bang_stack_shadow_pages(true);
1290 
1291   if (inc_counter) {
1292     // Reset the _do_not_unlock_if_synchronized flag.
1293     if (synchronized) {
1294       __ li(R0, 0);
1295       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1296     }
1297   }
1298 
1299   // access_flags = method->access_flags();
1300   // Load access flags.
1301   assert(__ nonvolatile_accross_vthread_preemtion(access_flags),
1302          "access_flags not preserved");
1303   // Type check.
1304   assert(2 == sizeof(AccessFlags), "unexpected field size");
1305   __ lhz(access_flags, method_(access_flags));
1306 
1307   // We don't want to reload R19_method and access_flags after calls
1308   // to some helper functions.
1309   assert(R19_method->is_nonvolatile(),
1310          "R19_method must be a non-volatile register");
1311 
1312   // Check for synchronized methods. Must happen AFTER invocation counter
1313   // check, so method is not locked if counter overflows.
1314 
1315   if (synchronized) {
1316     lock_method(access_flags, R11_scratch1, R12_scratch2, true);
1317 
1318     // Update monitor in state.
1319     __ ld(R11_scratch1, 0, R1_SP);
1320     __ sub(R12_scratch2, R26_monitor, R11_scratch1);
1321     __ sradi(R12_scratch2, R12_scratch2, Interpreter::logStackElementSize);
1322     __ std(R12_scratch2, _ijava_state_neg(monitors), R11_scratch1);
1323   }
1324 
1325   // jvmti/jvmpi support
1326   __ notify_method_entry();
1327 
1328   //=============================================================================
1329   // Get and call the signature handler.
1330 
1331   __ ld(signature_handler_fd, method_(signature_handler));
1332   Label call_signature_handler;
1333 
1334   __ cmpdi(CCR0, signature_handler_fd, 0);
1335   __ bne(CCR0, call_signature_handler);
1336 
1337   // Method has never been called. Either generate a specialized
1338   // handler or point to the slow one.
1339   //
1340   // Pass parameter 'false' to avoid exception check in call_VM.
1341   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
1342 
1343   // Check for an exception while looking up the target method. If we
1344   // incurred one, bail.
1345   __ ld(pending_exception, thread_(pending_exception));
1346   __ cmpdi(CCR0, pending_exception, 0);
1347   __ bne(CCR0, exception_return_sync_check); // Has pending exception.
1348 
1349   // Reload signature handler, it may have been created/assigned in the meanwhile.
1350   __ ld(signature_handler_fd, method_(signature_handler));
1351   __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
1352 
1353   BIND(call_signature_handler);
1354 
1355   // Before we call the signature handler we push a new frame to
1356   // protect the interpreter frame volatile registers when we return
1357   // from jni but before we can get back to Java.
1358 
1359   // First set the frame anchor while the SP/FP registers are
1360   // convenient and the slow signature handler can use this same frame
1361   // anchor.
1362 
1363   bool support_vthread_preemption = Continuations::enabled() && LockingMode != LM_LEGACY;
1364 
1365   // We have a TOP_IJAVA_FRAME here, which belongs to us.
1366   Label last_java_pc;
1367   Label *resume_pc = support_vthread_preemption ? &last_java_pc : nullptr;
1368   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R3_ARG1/*tmp*/, resume_pc);
1369 
1370   // Now the interpreter frame (and its call chain) have been
1371   // invalidated and flushed. We are now protected against eager
1372   // being enabled in native code. Even if it goes eager the
1373   // registers will be reloaded as clean and we will invalidate after
1374   // the call so no spurious flush should be possible.
1375 
1376   // Call signature handler and pass locals address.
1377   //
1378   // Our signature handlers copy required arguments to the C stack
1379   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
1380   __ mr(R3_ARG1, R18_locals);
1381 #if !defined(ABI_ELFv2)
1382   __ ld(signature_handler_fd, 0, signature_handler_fd);
1383 #endif
1384 
1385   __ call_stub(signature_handler_fd);
1386 
1387   assert(__ nonvolatile_accross_vthread_preemtion(result_handler_addr),
1388          "result_handler_addr not preserved");
1389   // Save across call to native method.
1390   __ mr(result_handler_addr, R3_RET);
1391   __ ld(R11_scratch1, _abi0(callers_sp), R1_SP); // load FP
1392 
1393   __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
1394 
1395   // Set up fixed parameters and call the native method.
1396   // If the method is static, get mirror into R4_ARG2.
1397   {
1398     Label method_is_not_static;
1399     // Access_flags is non-volatile and still, no need to restore it.
1400 
1401     // Restore access flags.
1402     __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
1403     __ bfalse(CCR0, method_is_not_static);
1404 
1405     // Load mirror from interpreter frame (FP in R11_scratch1)
1406     __ ld(R21_tmp1, _ijava_state_neg(mirror), R11_scratch1);
1407     // R4_ARG2 = &state->_oop_temp;
1408     __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
1409     __ std(R21_tmp1/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
1410     BIND(method_is_not_static);
1411   }
1412 
1413   // At this point, arguments have been copied off the stack into
1414   // their JNI positions. Oops are boxed in-place on the stack, with
1415   // handles copied to arguments. The result handler address is in a
1416   // register.
1417 
1418   // Pass JNIEnv address as first parameter.
1419   __ addir(R3_ARG1, thread_(jni_environment));
1420 
1421   // Load the native_method entry before we change the thread state.
1422   __ ld(native_method_fd, method_(native_function));
1423 
1424   //=============================================================================
1425   // Transition from _thread_in_Java to _thread_in_native. As soon as
1426   // we make this change the safepoint code needs to be certain that
1427   // the last Java frame we established is good. The pc in that frame
1428   // just needs to be near here not an actual return address.
1429 
1430   // We use release_store_fence to update values like the thread state, where
1431   // we don't want the current thread to continue until all our prior memory
1432   // accesses (including the new thread state) are visible to other threads.
1433   __ li(R0, _thread_in_native);
1434   __ release();
1435 
1436   // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
1437   __ stw(R0, thread_(thread_state));
1438 
1439   //=============================================================================
1440   // Call the native method. Argument registers must not have been
1441   // overwritten since "__ call_stub(signature_handler);" (except for
1442   // ARG1 and ARG2 for static methods).
1443 
1444   if (support_vthread_preemption) {
1445     // result_handler_addr is a nonvolatile register. Its value will be preserved across
1446     // the native call but only if the call isn't preempted. To preserve its value even
1447     // in the case of preemption we save it in the lresult slot. It is restored at
1448     // resume_pc if, and only if the call was preempted. This works because only
1449     // j.l.Object::wait calls are preempted which don't return a result.
1450     __ std(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1451   }
1452   __ push_cont_fastpath();
1453   __ call_c(native_method_fd);
1454   __ pop_cont_fastpath();
1455 
1456   __ li(R0, 0);
1457   __ ld(R11_scratch1, 0, R1_SP);
1458   __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1459   __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1460   __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
1461 
1462   // Note: C++ interpreter needs the following here:
1463   // The frame_manager_lr field, which we use for setting the last
1464   // java frame, gets overwritten by the signature handler. Restore
1465   // it now.
1466   //__ get_PC_trash_LR(R11_scratch1);
1467   //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
1468 
1469   // Because of GC R19_method may no longer be valid.
1470 
1471   // Block, if necessary, before resuming in _thread_in_Java state.
1472   // In order for GC to work, don't clear the last_Java_sp until after
1473   // blocking.
1474 
1475   //=============================================================================
1476   // Switch thread to "native transition" state before reading the
1477   // synchronization state. This additional state is necessary
1478   // because reading and testing the synchronization state is not
1479   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
1480   // in _thread_in_native state, loads _not_synchronized and is
1481   // preempted. VM thread changes sync state to synchronizing and
1482   // suspends threads for GC. Thread A is resumed to finish this
1483   // native method, but doesn't block here since it didn't see any
1484   // synchronization in progress, and escapes.
1485 
1486   // We use release_store_fence to update values like the thread state, where
1487   // we don't want the current thread to continue until all our prior memory
1488   // accesses (including the new thread state) are visible to other threads.
1489   __ li(R0/*thread_state*/, _thread_in_native_trans);
1490   __ release();
1491   __ stw(R0/*thread_state*/, thread_(thread_state));
1492   if (!UseSystemMemoryBarrier) {
1493     __ fence();
1494   }
1495 
1496   // Now before we return to java we must look for a current safepoint
1497   // (a new safepoint can not start since we entered native_trans).
1498   // We must check here because a current safepoint could be modifying
1499   // the callers registers right this moment.
1500 
1501   // Acquire isn't strictly necessary here because of the fence, but
1502   // sync_state is declared to be volatile, so we do it anyway
1503   // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
1504 
1505   Label do_safepoint, sync_check_done;
1506   // No synchronization in progress nor yet synchronized.
1507   __ safepoint_poll(do_safepoint, sync_state, true /* at_return */, false /* in_nmethod */);
1508 
1509   // Not suspended.
1510   // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
1511   __ lwz(suspend_flags, thread_(suspend_flags));
1512   __ cmpwi(CCR1, suspend_flags, 0);
1513   __ beq(CCR1, sync_check_done);
1514 
1515   __ bind(do_safepoint);
1516   __ isync();
1517   // Block. We do the call directly and leave the current
1518   // last_Java_frame setup undisturbed. We must save any possible
1519   // native result across the call. No oop is present.
1520 
1521   __ mr(R3_ARG1, R16_thread);
1522   __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1523 
1524   __ bind(sync_check_done);
1525 
1526   //=============================================================================
1527   // <<<<<< Back in Interpreter Frame >>>>>
1528 
1529   // We are in thread_in_native_trans here and back in the normal
1530   // interpreter frame. We don't have to do anything special about
1531   // safepoints and we can switch to Java mode anytime we are ready.
1532 
1533   // Note: frame::interpreter_frame_result has a dependency on how the
1534   // method result is saved across the call to post_method_exit. For
1535   // native methods it assumes that the non-FPU/non-void result is
1536   // saved in _native_lresult and a FPU result in _native_fresult. If
1537   // this changes then the interpreter_frame_result implementation
1538   // will need to be updated too.
1539 
1540   // On PPC64, we have stored the result directly after the native call.
1541 
1542   //=============================================================================
1543   // Back in Java
1544 
1545   // We use release_store_fence to update values like the thread state, where
1546   // we don't want the current thread to continue until all our prior memory
1547   // accesses (including the new thread state) are visible to other threads.
1548   __ li(R0/*thread_state*/, _thread_in_Java);
1549   __ lwsync(); // Acquire safepoint and suspend state, release thread state.
1550   __ stw(R0/*thread_state*/, thread_(thread_state));
1551 
1552   if (support_vthread_preemption) {
1553     // Check preemption for Object.wait()
1554     Label not_preempted;
1555     __ ld(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1556     __ cmpdi(CCR0, R0, 0);
1557     __ beq(CCR0, not_preempted);
1558     __ mtlr(R0);
1559     __ li(R0, 0);
1560     __ std(R0, in_bytes(JavaThread::preempt_alternate_return_offset()), R16_thread);
1561     __ blr();
1562 
1563     // Execution will be resumed here when the vthread becomes runnable again.
1564     __ bind(*resume_pc);
1565     __ restore_after_resume(R11_scratch1 /* fp */);
1566     // We saved the result handler before the call
1567     __ ld(result_handler_addr, _ijava_state_neg(lresult), R11_scratch1);
1568 #ifdef ASSERT
1569     // Clobber result slots. Only native methods returning void can be preemted currently.
1570     __ load_const(R3_RET, UCONST64(0xbad01001));
1571     __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1572     __ std(R3_RET, _ijava_state_neg(fresult), R11_scratch1);
1573     // reset_last_Java_frame() below asserts that a last java sp is set
1574     __ asm_assert_mem8_is_zero(in_bytes(JavaThread::last_Java_sp_offset()),
1575         R16_thread, FILE_AND_LINE ": Last java sp should not be set when resuming");
1576     __ std(R3_RET, in_bytes(JavaThread::last_Java_sp_offset()), R16_thread);
1577 #endif
1578     __ bind(not_preempted);
1579   }
1580 
1581   if (CheckJNICalls) {
1582     // clear_pending_jni_exception_check
1583     __ load_const_optimized(R0, 0L);
1584     __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread);
1585   }
1586 
1587   __ reset_last_Java_frame();
1588 
1589   // Jvmdi/jvmpi support. Whether we've got an exception pending or
1590   // not, and whether unlocking throws an exception or not, we notify
1591   // on native method exit. If we do have an exception, we'll end up
1592   // in the caller's context to handle it, so if we don't do the
1593   // notify here, we'll drop it on the floor.
1594   __ notify_method_exit(true/*native method*/,
1595                         ilgl /*illegal state (not used for native methods)*/,
1596                         InterpreterMacroAssembler::NotifyJVMTI,
1597                         false /*check_exceptions*/);
1598 
1599   //=============================================================================
1600   // Handle exceptions
1601 
1602   if (synchronized) {
1603     __ unlock_object(R26_monitor); // Can also unlock methods.
1604   }
1605 
1606   // Reset active handles after returning from native.
1607   // thread->active_handles()->clear();
1608   __ ld(active_handles, thread_(active_handles));
1609   // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
1610   __ li(R0, 0);
1611   __ stw(R0, in_bytes(JNIHandleBlock::top_offset()), active_handles);
1612 
1613   Label exception_return_sync_check_already_unlocked;
1614   __ ld(R0/*pending_exception*/, thread_(pending_exception));
1615   __ cmpdi(CCR0, R0/*pending_exception*/, 0);
1616   __ bne(CCR0, exception_return_sync_check_already_unlocked);
1617 
1618   //-----------------------------------------------------------------------------
1619   // No exception pending.
1620 
1621   // Move native method result back into proper registers and return.
1622   // Invoke result handler (may unbox/promote).
1623   __ ld(R11_scratch1, 0, R1_SP);
1624   __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
1625   __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
1626   __ call_stub(result_handler_addr);
1627 
1628   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
1629 
1630   // Must use the return pc which was loaded from the caller's frame
1631   // as the VM uses return-pc-patching for deoptimization.
1632   __ mtlr(R0);
1633   __ blr();
1634 
1635   //-----------------------------------------------------------------------------
1636   // An exception is pending. We call into the runtime only if the
1637   // caller was not interpreted. If it was interpreted the
1638   // interpreter will do the correct thing. If it isn't interpreted
1639   // (call stub/compiled code) we will change our return and continue.
1640 
1641   BIND(exception_return_sync_check);
1642 
1643   if (synchronized) {
1644     __ unlock_object(R26_monitor); // Can also unlock methods.
1645   }
1646   BIND(exception_return_sync_check_already_unlocked);
1647 
1648   const Register return_pc = R31;
1649 
1650   __ ld(return_pc, 0, R1_SP);
1651   __ ld(return_pc, _abi0(lr), return_pc);
1652 
1653   // Get the address of the exception handler.
1654   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1655                   R16_thread,
1656                   return_pc /* return pc */);
1657   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
1658 
1659   // Load the PC of the exception handler into LR.
1660   __ mtlr(R3_RET);
1661 
1662   // Load exception into R3_ARG1 and clear pending exception in thread.
1663   __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
1664   __ li(R4_ARG2, 0);
1665   __ std(R4_ARG2, thread_(pending_exception));
1666 
1667   // Load the original return pc into R4_ARG2.
1668   __ mr(R4_ARG2/*issuing_pc*/, return_pc);
1669 
1670   // Return to exception handler.
1671   __ blr();
1672 
1673   //=============================================================================
1674   // Counter overflow.
1675 
1676   if (inc_counter) {
1677     // Handle invocation counter overflow.
1678     __ bind(invocation_counter_overflow);
1679 
1680     generate_counter_overflow(continue_after_compile);
1681   }
1682 
1683   return entry;
1684 }
1685 
1686 // Generic interpreted method entry to (asm) interpreter.
1687 //
1688 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1689   bool inc_counter = UseCompiler || CountCompiledCalls;
1690   address entry = __ pc();
1691   // Generate the code to allocate the interpreter stack frame.
1692   Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
1693            Rsize_of_locals     = R5_ARG3; // Written by generate_fixed_frame.
1694 
1695   // Does also a stack check to assure this frame fits on the stack.
1696   generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
1697 
1698   // --------------------------------------------------------------------------
1699   // Zero out non-parameter locals.
1700   // Note: *Always* zero out non-parameter locals as Sparc does. It's not
1701   // worth to ask the flag, just do it.
1702   Register Rslot_addr = R6_ARG4,
1703            Rnum       = R7_ARG5;
1704   Label Lno_locals, Lzero_loop;
1705 
1706   // Set up the zeroing loop.
1707   __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
1708   __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
1709   __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
1710   __ beq(CCR0, Lno_locals);
1711   __ li(R0, 0);
1712   __ mtctr(Rnum);
1713 
1714   // The zero locals loop.
1715   __ bind(Lzero_loop);
1716   __ std(R0, 0, Rslot_addr);
1717   __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
1718   __ bdnz(Lzero_loop);
1719 
1720   __ bind(Lno_locals);
1721 
1722   // --------------------------------------------------------------------------
1723   // Counter increment and overflow check.
1724   Label invocation_counter_overflow;
1725   Label continue_after_compile;
1726   if (inc_counter || ProfileInterpreter) {
1727 
1728     Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
1729     if (synchronized) {
1730       // Since at this point in the method invocation the exception handler
1731       // would try to exit the monitor of synchronized methods which hasn't
1732       // been entered yet, we set the thread local variable
1733       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1734       // runtime, exception handling i.e. unlock_if_synchronized_method will
1735       // check this thread local flag.
1736       // This flag has two effects, one is to force an unwind in the topmost
1737       // interpreter frame and not perform an unlock while doing so.
1738       __ li(R0, 1);
1739       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1740     }
1741 
1742     // Argument and return type profiling.
1743     __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
1744 
1745     // Increment invocation counter and check for overflow.
1746     if (inc_counter) {
1747       generate_counter_incr(&invocation_counter_overflow);
1748     }
1749 
1750     __ bind(continue_after_compile);
1751   }
1752 
1753   bang_stack_shadow_pages(false);
1754 
1755   if (inc_counter || ProfileInterpreter) {
1756     // Reset the _do_not_unlock_if_synchronized flag.
1757     if (synchronized) {
1758       __ li(R0, 0);
1759       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
1760     }
1761   }
1762 
1763   // --------------------------------------------------------------------------
1764   // Locking of synchronized methods. Must happen AFTER invocation_counter
1765   // check and stack overflow check, so method is not locked if overflows.
1766   if (synchronized) {
1767     lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
1768   }
1769 #ifdef ASSERT
1770   else {
1771     Label Lok;
1772     __ lhz(R0, in_bytes(Method::access_flags_offset()), R19_method);
1773     __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
1774     __ asm_assert_eq("method needs synchronization");
1775     __ bind(Lok);
1776   }
1777 #endif // ASSERT
1778 
1779   // --------------------------------------------------------------------------
1780   // JVMTI support
1781   __ notify_method_entry();
1782 
1783   // --------------------------------------------------------------------------
1784   // Start executing instructions.
1785   __ dispatch_next(vtos);
1786 
1787   // --------------------------------------------------------------------------
1788   if (inc_counter) {
1789     // Handle invocation counter overflow.
1790     __ bind(invocation_counter_overflow);
1791     generate_counter_overflow(continue_after_compile);
1792   }
1793   return entry;
1794 }
1795 
1796 // CRC32 Intrinsics.
1797 //
1798 // Contract on scratch and work registers.
1799 // =======================================
1800 //
1801 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
1802 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
1803 // You can't rely on these registers across calls.
1804 //
1805 // The generators for CRC32_update and for CRC32_updateBytes use the
1806 // scratch/work register set internally, passing the work registers
1807 // as arguments to the MacroAssembler emitters as required.
1808 //
1809 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
1810 // Their contents is not constant but may change according to the requirements
1811 // of the emitted code.
1812 //
1813 // All other registers from the scratch/work register set are used "internally"
1814 // and contain garbage (i.e. unpredictable values) once blr() is reached.
1815 // Basically, only R3_RET contains a defined value which is the function result.
1816 //
1817 /**
1818  * Method entry for static native methods:
1819  *   int java.util.zip.CRC32.update(int crc, int b)
1820  */
1821 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1822   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1823   address start = __ pc();  // Remember stub start address (is rtn value).
1824   Label slow_path;
1825 
1826   // Safepoint check
1827   const Register sync_state = R11_scratch1;
1828   __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1829 
1830   // We don't generate local frame and don't align stack because
1831   // we not even call stub code (we generate the code inline)
1832   // and there is no safepoint on this path.
1833 
1834   // Load java parameters.
1835   // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
1836   const Register argP    = R15_esp;
1837   const Register crc     = R3_ARG1;  // crc value
1838   const Register data    = R4_ARG2;
1839   const Register table   = R5_ARG3;  // address of crc32 table
1840 
1841   BLOCK_COMMENT("CRC32_update {");
1842 
1843   // Arguments are reversed on java expression stack
1844 #ifdef VM_LITTLE_ENDIAN
1845   int data_offs = 0+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1846                                      // Being passed as an int, the single byte is at offset +0.
1847 #else
1848   int data_offs = 3+1*wordSize;      // (stack) address of byte value. Emitter expects address, not value.
1849                                      // Being passed from java as an int, the single byte is at offset +3.
1850 #endif
1851   __ lwz(crc, 2*wordSize, argP);     // Current crc state, zero extend to 64 bit to have a clean register.
1852   __ lbz(data, data_offs, argP);     // Byte from buffer, zero-extended.
1853   __ load_const_optimized(table, StubRoutines::crc_table_addr(), R0);
1854   __ kernel_crc32_singleByteReg(crc, data, table, true);
1855 
1856   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1857   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1858   __ blr();
1859 
1860   // Generate a vanilla native entry as the slow path.
1861   BLOCK_COMMENT("} CRC32_update");
1862   BIND(slow_path);
1863   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1864   return start;
1865 }
1866 
1867 /**
1868  * Method entry for static native methods:
1869  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
1870  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
1871  */
1872 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1873   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1874   address start = __ pc();  // Remember stub start address (is rtn value).
1875   Label slow_path;
1876 
1877   // Safepoint check
1878   const Register sync_state = R11_scratch1;
1879   __ safepoint_poll(slow_path, sync_state, false /* at_return */, false /* in_nmethod */);
1880 
1881   // We don't generate local frame and don't align stack because
1882   // we not even call stub code (we generate the code inline)
1883   // and there is no safepoint on this path.
1884 
1885   // Load parameters.
1886   // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1887   const Register argP    = R15_esp;
1888   const Register crc     = R3_ARG1;  // crc value
1889   const Register data    = R4_ARG2;  // address of java byte array
1890   const Register dataLen = R5_ARG3;  // source data len
1891   const Register tmp     = R11_scratch1;
1892 
1893   // Arguments are reversed on java expression stack.
1894   // Calculate address of start element.
1895   if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
1896     BLOCK_COMMENT("CRC32_updateByteBuffer {");
1897     // crc     @ (SP + 5W) (32bit)
1898     // buf     @ (SP + 3W) (64bit ptr to long array)
1899     // off     @ (SP + 2W) (32bit)
1900     // dataLen @ (SP + 1W) (32bit)
1901     // data = buf + off
1902     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1903     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1904     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1905     __ lwz( crc,     5*wordSize, argP);  // current crc state
1906     __ add( data, data, tmp);            // Add byte buffer offset.
1907   } else {                                                         // Used for "updateBytes update".
1908     BLOCK_COMMENT("CRC32_updateBytes {");
1909     // crc     @ (SP + 4W) (32bit)
1910     // buf     @ (SP + 3W) (64bit ptr to byte array)
1911     // off     @ (SP + 2W) (32bit)
1912     // dataLen @ (SP + 1W) (32bit)
1913     // data = buf + off + base_offset
1914     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1915     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1916     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1917     __ add( data, data, tmp);            // add byte buffer offset
1918     __ lwz( crc,     4*wordSize, argP);  // current crc state
1919     __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1920   }
1921 
1922   __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, false);
1923 
1924   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1925   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1926   __ blr();
1927 
1928   // Generate a vanilla native entry as the slow path.
1929   BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
1930   BIND(slow_path);
1931   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
1932   return start;
1933 }
1934 
1935 
1936 /**
1937  * Method entry for intrinsic-candidate (non-native) methods:
1938  *   int java.util.zip.CRC32C.updateBytes(           int crc, byte[] b,  int off, int end)
1939  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end)
1940  * Unlike CRC32, CRC32C does not have any methods marked as native
1941  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1942  **/
1943 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1944   assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1945   address start = __ pc();  // Remember stub start address (is rtn value).
1946 
1947   // We don't generate local frame and don't align stack because
1948   // we not even call stub code (we generate the code inline)
1949   // and there is no safepoint on this path.
1950 
1951   // Load parameters.
1952   // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
1953   const Register argP    = R15_esp;
1954   const Register crc     = R3_ARG1;  // crc value
1955   const Register data    = R4_ARG2;  // address of java byte array
1956   const Register dataLen = R5_ARG3;  // source data len
1957   const Register tmp     = R11_scratch1;
1958 
1959   // Arguments are reversed on java expression stack.
1960   // Calculate address of start element.
1961   if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer".
1962     BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {");
1963     // crc     @ (SP + 5W) (32bit)
1964     // buf     @ (SP + 3W) (64bit ptr to long array)
1965     // off     @ (SP + 2W) (32bit)
1966     // dataLen @ (SP + 1W) (32bit)
1967     // data = buf + off
1968     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1969     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1970     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1971     __ lwz( crc,     5*wordSize, argP);  // current crc state
1972     __ add( data, data, tmp);            // Add byte buffer offset.
1973     __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1974   } else {                                                         // Used for "updateBytes update".
1975     BLOCK_COMMENT("CRC32C_updateBytes {");
1976     // crc     @ (SP + 4W) (32bit)
1977     // buf     @ (SP + 3W) (64bit ptr to byte array)
1978     // off     @ (SP + 2W) (32bit)
1979     // dataLen @ (SP + 1W) (32bit)
1980     // data = buf + off + base_offset
1981     __ ld(  data,    3*wordSize, argP);  // start of byte buffer
1982     __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
1983     __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
1984     __ add( data, data, tmp);            // add byte buffer offset
1985     __ sub( dataLen, dataLen, tmp);      // (end_index - offset)
1986     __ lwz( crc,     4*wordSize, argP);  // current crc state
1987     __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1988   }
1989 
1990   __ crc32(crc, data, dataLen, R2, R6, R7, R8, R9, R10, R11, R12, true);
1991 
1992   // Restore caller sp for c2i case (from compiled) and for resized sender frame (from interpreted).
1993   __ resize_frame_absolute(R21_sender_SP, R11_scratch1, R0);
1994   __ blr();
1995 
1996   BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}");
1997   return start;
1998 }
1999 
2000 // Not supported
2001 address TemplateInterpreterGenerator::generate_currentThread() { return nullptr; }
2002 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
2003 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
2004 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
2005 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
2006 
2007 // =============================================================================
2008 // Exceptions
2009 
2010 void TemplateInterpreterGenerator::generate_throw_exception() {
2011   Register Rexception    = R17_tos,
2012            Rcontinuation = R3_RET;
2013 
2014   // --------------------------------------------------------------------------
2015   // Entry point if an method returns with a pending exception (rethrow).
2016   Interpreter::_rethrow_exception_entry = __ pc();
2017   {
2018     __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2019 
2020     // Compiled code destroys templateTableBase, reload.
2021     __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
2022   }
2023 
2024   // Entry point if a interpreted method throws an exception (throw).
2025   Interpreter::_throw_exception_entry = __ pc();
2026   {
2027     __ mr(Rexception, R3_RET);
2028 
2029     __ verify_oop(Rexception);
2030 
2031     // Expression stack must be empty before entering the VM in case of an exception.
2032     __ empty_expression_stack();
2033     // Find exception handler address and preserve exception oop.
2034     // Call C routine to find handler and jump to it.
2035     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
2036     __ mtctr(Rcontinuation);
2037     // Push exception for exception handler bytecodes.
2038     __ push_ptr(Rexception);
2039 
2040     // Jump to exception handler (may be remove activation entry!).
2041     __ bctr();
2042   }
2043 
2044   // If the exception is not handled in the current frame the frame is
2045   // removed and the exception is rethrown (i.e. exception
2046   // continuation is _rethrow_exception).
2047   //
2048   // Note: At this point the bci is still the bxi for the instruction
2049   // which caused the exception and the expression stack is
2050   // empty. Thus, for any VM calls at this point, GC will find a legal
2051   // oop map (with empty expression stack).
2052 
2053   // In current activation
2054   // tos: exception
2055   // bcp: exception bcp
2056 
2057   // --------------------------------------------------------------------------
2058   // JVMTI PopFrame support
2059 
2060   Interpreter::_remove_activation_preserving_args_entry = __ pc();
2061   {
2062     // Set the popframe_processing bit in popframe_condition indicating that we are
2063     // currently handling popframe, so that call_VMs that may happen later do not
2064     // trigger new popframe handling cycles.
2065     __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2066     __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
2067     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2068 
2069     // Empty the expression stack, as in normal exception handling.
2070     __ empty_expression_stack();
2071     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
2072 
2073     // Check to see whether we are returning to a deoptimized frame.
2074     // (The PopFrame call ensures that the caller of the popped frame is
2075     // either interpreted or compiled and deoptimizes it if compiled.)
2076     // Note that we don't compare the return PC against the
2077     // deoptimization blob's unpack entry because of the presence of
2078     // adapter frames in C2.
2079     Label Lcaller_not_deoptimized;
2080     Register return_pc = R3_ARG1;
2081     __ ld(return_pc, 0, R1_SP);
2082     __ ld(return_pc, _abi0(lr), return_pc);
2083     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
2084     __ cmpdi(CCR0, R3_RET, 0);
2085     __ bne(CCR0, Lcaller_not_deoptimized);
2086 
2087     // The deoptimized case.
2088     // In this case, we can't call dispatch_next() after the frame is
2089     // popped, but instead must save the incoming arguments and restore
2090     // them after deoptimization has occurred.
2091     __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
2092     __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
2093     __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
2094     __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
2095     __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
2096     // Save these arguments.
2097     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
2098 
2099     // Inform deoptimization that it is responsible for restoring these arguments.
2100     __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
2101     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2102 
2103     // Return from the current method into the deoptimization blob. Will eventually
2104     // end up in the deopt interpreter entry, deoptimization prepared everything that
2105     // we will reexecute the call that called us.
2106     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
2107     __ mtlr(return_pc);
2108     __ pop_cont_fastpath();
2109     __ blr();
2110 
2111     // The non-deoptimized case.
2112     __ bind(Lcaller_not_deoptimized);
2113 
2114     // Clear the popframe condition flag.
2115     __ li(R0, 0);
2116     __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
2117 
2118     // Get out of the current method and re-execute the call that called us.
2119     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2120     __ pop_cont_fastpath();
2121     __ restore_interpreter_state(R11_scratch1, false /*bcp_and_mdx_only*/, true /*restore_top_frame_sp*/);
2122     if (ProfileInterpreter) {
2123       __ set_method_data_pointer_for_bcp();
2124       __ ld(R11_scratch1, 0, R1_SP);
2125       __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
2126     }
2127 #if INCLUDE_JVMTI
2128     Label L_done;
2129 
2130     __ lbz(R11_scratch1, 0, R14_bcp);
2131     __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
2132     __ bne(CCR0, L_done);
2133 
2134     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2135     // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
2136     __ ld(R4_ARG2, 0, R18_locals);
2137     __ call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp);
2138 
2139     __ cmpdi(CCR0, R4_ARG2, 0);
2140     __ beq(CCR0, L_done);
2141     __ std(R4_ARG2, wordSize, R15_esp);
2142     __ bind(L_done);
2143 #endif // INCLUDE_JVMTI
2144     __ dispatch_next(vtos);
2145   }
2146   // end of JVMTI PopFrame support
2147 
2148   // --------------------------------------------------------------------------
2149   // Remove activation exception entry.
2150   // This is jumped to if an interpreted method can't handle an exception itself
2151   // (we come from the throw/rethrow exception entry above). We're going to call
2152   // into the VM to find the exception handler in the caller, pop the current
2153   // frame and return the handler we calculated.
2154   Interpreter::_remove_activation_entry = __ pc();
2155   {
2156     __ pop_ptr(Rexception);
2157     __ verify_oop(Rexception);
2158     __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
2159 
2160     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
2161     __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
2162 
2163     __ get_vm_result(Rexception);
2164 
2165     // We are done with this activation frame; find out where to go next.
2166     // The continuation point will be an exception handler, which expects
2167     // the following registers set up:
2168     //
2169     // RET:  exception oop
2170     // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
2171 
2172     Register return_pc = R31; // Needs to survive the runtime call.
2173     __ ld(return_pc, 0, R1_SP);
2174     __ ld(return_pc, _abi0(lr), return_pc);
2175     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
2176 
2177     // Remove the current activation.
2178     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
2179     __ pop_cont_fastpath();
2180 
2181     __ mr(R4_ARG2, return_pc);
2182     __ mtlr(R3_RET);
2183     __ mr(R3_RET, Rexception);
2184     __ blr();
2185   }
2186 }
2187 
2188 // JVMTI ForceEarlyReturn support.
2189 // Returns "in the middle" of a method with a "fake" return value.
2190 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2191 
2192   Register Rscratch1 = R11_scratch1,
2193            Rscratch2 = R12_scratch2;
2194 
2195   address entry = __ pc();
2196   __ empty_expression_stack();
2197 
2198   __ load_earlyret_value(state, Rscratch1);
2199 
2200   __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
2201   // Clear the earlyret state.
2202   __ li(R0, 0);
2203   __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
2204 
2205   __ remove_activation(state, false, false);
2206   // Copied from TemplateTable::_return.
2207   // Restoration of lr done by remove_activation.
2208   switch (state) {
2209     // Narrow result if state is itos but result type is smaller.
2210     case btos:
2211     case ztos:
2212     case ctos:
2213     case stos:
2214     case itos: __ narrow(R17_tos); /* fall through */
2215     case ltos:
2216     case atos: __ mr(R3_RET, R17_tos); break;
2217     case ftos:
2218     case dtos: __ fmr(F1_RET, F15_ftos); break;
2219     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
2220                // to get visible before the reference to the object gets stored anywhere.
2221                __ membar(Assembler::StoreStore); break;
2222     default  : ShouldNotReachHere();
2223   }
2224   __ blr();
2225 
2226   return entry;
2227 } // end of ForceEarlyReturn support
2228 
2229 //-----------------------------------------------------------------------------
2230 // Helper for vtos entry point generation
2231 
2232 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2233                                                          address& bep,
2234                                                          address& cep,
2235                                                          address& sep,
2236                                                          address& aep,
2237                                                          address& iep,
2238                                                          address& lep,
2239                                                          address& fep,
2240                                                          address& dep,
2241                                                          address& vep) {
2242   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2243   Label L;
2244 
2245   aep = __ pc();  __ push_ptr();  __ b(L);
2246   fep = __ pc();  __ push_f();    __ b(L);
2247   dep = __ pc();  __ push_d();    __ b(L);
2248   lep = __ pc();  __ push_l();    __ b(L);
2249   __ align(32, 12, 24); // align L
2250   bep = cep = sep =
2251   iep = __ pc();  __ push_i();
2252   vep = __ pc();
2253   __ bind(L);
2254   generate_and_dispatch(t);
2255 }
2256 
2257 //-----------------------------------------------------------------------------
2258 
2259 void TemplateInterpreterGenerator::count_bytecode() {
2260   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
2261   __ lwz(R12_scratch2, offs, R11_scratch1);
2262   __ addi(R12_scratch2, R12_scratch2, 1);
2263   __ stw(R12_scratch2, offs, R11_scratch1);
2264 }
2265 
2266 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2267   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
2268   __ lwz(R12_scratch2, offs, R11_scratch1);
2269   __ addi(R12_scratch2, R12_scratch2, 1);
2270   __ stw(R12_scratch2, offs, R11_scratch1);
2271 }
2272 
2273 // Non-product code
2274 #ifndef PRODUCT
2275 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2276   //__ flush_bundle();
2277   address entry = __ pc();
2278 
2279   const char *bname = nullptr;
2280   uint tsize = 0;
2281   switch(state) {
2282   case ftos:
2283     bname = "trace_code_ftos {";
2284     tsize = 2;
2285     break;
2286   case btos:
2287     bname = "trace_code_btos {";
2288     tsize = 2;
2289     break;
2290   case ztos:
2291     bname = "trace_code_ztos {";
2292     tsize = 2;
2293     break;
2294   case ctos:
2295     bname = "trace_code_ctos {";
2296     tsize = 2;
2297     break;
2298   case stos:
2299     bname = "trace_code_stos {";
2300     tsize = 2;
2301     break;
2302   case itos:
2303     bname = "trace_code_itos {";
2304     tsize = 2;
2305     break;
2306   case ltos:
2307     bname = "trace_code_ltos {";
2308     tsize = 3;
2309     break;
2310   case atos:
2311     bname = "trace_code_atos {";
2312     tsize = 2;
2313     break;
2314   case vtos:
2315     // Note: In case of vtos, the topmost of stack value could be a int or doubl
2316     // In case of a double (2 slots) we won't see the 2nd stack value.
2317     // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
2318     bname = "trace_code_vtos {";
2319     tsize = 2;
2320 
2321     break;
2322   case dtos:
2323     bname = "trace_code_dtos {";
2324     tsize = 3;
2325     break;
2326   default:
2327     ShouldNotReachHere();
2328   }
2329   BLOCK_COMMENT(bname);
2330 
2331   // Support short-cut for TraceBytecodesAt.
2332   // Don't call into the VM if we don't want to trace to speed up things.
2333   Label Lskip_vm_call;
2334   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2335     int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
2336     int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2337     __ ld(R11_scratch1, offs1, R11_scratch1);
2338     __ lwa(R12_scratch2, offs2, R12_scratch2);
2339     __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2340     __ blt(CCR0, Lskip_vm_call);
2341   }
2342 
2343   __ push(state);
2344   // Load 2 topmost expression stack values.
2345   __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
2346   __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
2347   __ mflr(R31);
2348   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
2349   __ mtlr(R31);
2350   __ pop(state);
2351 
2352   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
2353     __ bind(Lskip_vm_call);
2354   }
2355   __ blr();
2356   BLOCK_COMMENT("} trace_code");
2357   return entry;
2358 }
2359 
2360 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2361   const Register addr = R11_scratch1,
2362                  tmp  = R12_scratch2;
2363   // Get index, shift out old bytecode, bring in new bytecode, and store it.
2364   // _index = (_index >> log2_number_of_codes) |
2365   //          (bytecode << log2_number_of_codes);
2366   int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
2367   __ lwz(tmp, offs1, addr);
2368   __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
2369   __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2370   __ stw(tmp, offs1, addr);
2371 
2372   // Bump bucket contents.
2373   // _counters[_index] ++;
2374   int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
2375   __ sldi(tmp, tmp, LogBytesPerInt);
2376   __ add(addr, tmp, addr);
2377   __ lwz(tmp, offs2, addr);
2378   __ addi(tmp, tmp, 1);
2379   __ stw(tmp, offs2, addr);
2380 }
2381 
2382 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2383   // Call a little run-time stub to avoid blow-up for each bytecode.
2384   // The run-time runtime saves the right registers, depending on
2385   // the tosca in-state for the given template.
2386 
2387   assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2388          "entry must have been generated");
2389 
2390   // Note: we destroy LR here.
2391   __ bl(Interpreter::trace_code(t->tos_in()));
2392 }
2393 
2394 void TemplateInterpreterGenerator::stop_interpreter_at() {
2395   Label L;
2396   int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
2397   int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
2398   __ ld(R11_scratch1, offs1, R11_scratch1);
2399   __ lwa(R12_scratch2, offs2, R12_scratch2);
2400   __ cmpd(CCR0, R12_scratch2, R11_scratch1);
2401   __ bne(CCR0, L);
2402   __ illtrap();
2403   __ bind(L);
2404 }
2405 
2406 #endif // !PRODUCT