1 /*
   2  * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2020, Red Hat Inc. 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/compiler_globals.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 "interpreter/bytecodeTracer.hpp"
  38 #include "memory/resourceArea.hpp"
  39 #include "oops/arrayOop.hpp"
  40 #include "oops/method.hpp"
  41 #include "oops/methodCounters.hpp"
  42 #include "oops/methodData.hpp"
  43 #include "oops/oop.inline.hpp"
  44 #include "oops/inlineKlass.hpp"
  45 #include "oops/resolvedIndyEntry.hpp"
  46 #include "oops/resolvedMethodEntry.hpp"
  47 #include "prims/jvmtiExport.hpp"
  48 #include "prims/jvmtiThreadState.hpp"
  49 #include "runtime/arguments.hpp"
  50 #include "runtime/deoptimization.hpp"
  51 #include "runtime/frame.inline.hpp"
  52 #include "runtime/globals.hpp"
  53 #include "runtime/jniHandles.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/stubRoutines.hpp"
  56 #include "runtime/synchronizer.hpp"
  57 #include "runtime/timer.hpp"
  58 #include "runtime/vframeArray.hpp"
  59 #include "utilities/checkedCast.hpp"
  60 #include "utilities/debug.hpp"
  61 #include "utilities/powerOfTwo.hpp"
  62 #include <sys/types.h>
  63 
  64 // Size of interpreter code.  Increase if too small.  Interpreter will
  65 // fail with a guarantee ("not enough space for interpreter generation");
  66 // if too small.
  67 // Run with +PrintInterpreter to get the VM to print out the size.
  68 // Max size with JVMTI
  69 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
  70 
  71 #define __ _masm->
  72 
  73 //-----------------------------------------------------------------------------
  74 
  75 extern "C" void entry(CodeBuffer*);
  76 
  77 //-----------------------------------------------------------------------------
  78 
  79 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  80   address entry = __ pc();
  81 
  82   __ andr(esp, esp, -16);
  83   __ mov(c_rarg3, esp);
  84   // rmethod
  85   // rlocals
  86   // c_rarg3: first stack arg - wordSize
  87 
  88   // adjust sp
  89   __ sub(sp, c_rarg3, 18 * wordSize);
  90   __ str(lr, Address(__ pre(sp, -2 * wordSize)));
  91   __ call_VM(noreg,
  92              CAST_FROM_FN_PTR(address,
  93                               InterpreterRuntime::slow_signature_handler),
  94              rmethod, rlocals, c_rarg3);
  95 
  96   // r0: result handler
  97 
  98   // Stack layout:
  99   // rsp: return address           <- sp
 100   //      1 garbage
 101   //      8 integer args (if static first is unused)
 102   //      1 float/double identifiers
 103   //      8 double args
 104   //        stack args              <- esp
 105   //        garbage
 106   //        expression stack bottom
 107   //        bcp (null)
 108   //        ...
 109 
 110   // Restore LR
 111   __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
 112 
 113   // Do FP first so we can use c_rarg3 as temp
 114   __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
 115 
 116   for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
 117     const FloatRegister r = as_FloatRegister(i);
 118 
 119     Label d, done;
 120 
 121     __ tbnz(c_rarg3, i, d);
 122     __ ldrs(r, Address(sp, (10 + i) * wordSize));
 123     __ b(done);
 124     __ bind(d);
 125     __ ldrd(r, Address(sp, (10 + i) * wordSize));
 126     __ bind(done);
 127   }
 128 
 129   // c_rarg0 contains the result from the call of
 130   // InterpreterRuntime::slow_signature_handler so we don't touch it
 131   // here.  It will be loaded with the JNIEnv* later.
 132   __ ldr(c_rarg1, Address(sp, 1 * wordSize));
 133   for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
 134     Register rm = as_Register(i), rn = as_Register(i+1);
 135     __ ldp(rm, rn, Address(sp, i * wordSize));
 136   }
 137 
 138   __ add(sp, sp, 18 * wordSize);
 139   __ ret(lr);
 140 
 141   return entry;
 142 }
 143 
 144 
 145 //
 146 // Various method entries
 147 //
 148 
 149 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
 150   // rmethod: Method*
 151   // r19_sender_sp: sender sp
 152   // esp: args
 153 
 154   // These don't need a safepoint check because they aren't virtually
 155   // callable. We won't enter these intrinsics from compiled code.
 156   // If in the future we added an intrinsic which was virtually callable
 157   // we'd have to worry about how to safepoint so that this code is used.
 158 
 159   // mathematical functions inlined by compiler
 160   // (interpreter must provide identical implementation
 161   // in order to avoid monotonicity bugs when switching
 162   // from interpreter to compiler in the middle of some
 163   // computation)
 164   //
 165   // stack:
 166   //        [ arg ] <-- esp
 167   //        [ arg ]
 168   // retaddr in lr
 169 
 170   address entry_point = nullptr;
 171   Register continuation = lr;
 172   switch (kind) {
 173   case Interpreter::java_lang_math_abs:
 174     entry_point = __ pc();
 175     __ ldrd(v0, Address(esp));
 176     __ fabsd(v0, v0);
 177     __ mov(sp, r19_sender_sp); // Restore caller's SP
 178     break;
 179   case Interpreter::java_lang_math_sqrt:
 180     entry_point = __ pc();
 181     __ ldrd(v0, Address(esp));
 182     __ fsqrtd(v0, v0);
 183     __ mov(sp, r19_sender_sp);
 184     break;
 185   case Interpreter::java_lang_math_sin :
 186   case Interpreter::java_lang_math_cos :
 187   case Interpreter::java_lang_math_tan :
 188   case Interpreter::java_lang_math_log :
 189   case Interpreter::java_lang_math_log10 :
 190   case Interpreter::java_lang_math_exp :
 191     entry_point = __ pc();
 192     __ ldrd(v0, Address(esp));
 193     __ mov(sp, r19_sender_sp);
 194     __ mov(r23, lr);
 195     continuation = r23;  // The first free callee-saved register
 196     generate_transcendental_entry(kind, 1);
 197     break;
 198   case Interpreter::java_lang_math_pow :
 199     entry_point = __ pc();
 200     __ mov(r23, lr);
 201     continuation = r23;
 202     __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
 203     __ ldrd(v1, Address(esp));
 204     __ mov(sp, r19_sender_sp);
 205     generate_transcendental_entry(kind, 2);
 206     break;
 207   case Interpreter::java_lang_math_fmaD :
 208     if (UseFMA) {
 209       entry_point = __ pc();
 210       __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
 211       __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
 212       __ ldrd(v2, Address(esp));
 213       __ fmaddd(v0, v0, v1, v2);
 214       __ mov(sp, r19_sender_sp); // Restore caller's SP
 215     }
 216     break;
 217   case Interpreter::java_lang_math_fmaF :
 218     if (UseFMA) {
 219       entry_point = __ pc();
 220       __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
 221       __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
 222       __ ldrs(v2, Address(esp));
 223       __ fmadds(v0, v0, v1, v2);
 224       __ mov(sp, r19_sender_sp); // Restore caller's SP
 225     }
 226     break;
 227   default:
 228     ;
 229   }
 230   if (entry_point) {
 231     __ br(continuation);
 232   }
 233 
 234   return entry_point;
 235 }
 236 
 237   // double trigonometrics and transcendentals
 238   // static jdouble dsin(jdouble x);
 239   // static jdouble dcos(jdouble x);
 240   // static jdouble dtan(jdouble x);
 241   // static jdouble dlog(jdouble x);
 242   // static jdouble dlog10(jdouble x);
 243   // static jdouble dexp(jdouble x);
 244   // static jdouble dpow(jdouble x, jdouble y);
 245 
 246 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
 247   address fn;
 248   switch (kind) {
 249   case Interpreter::java_lang_math_sin :
 250     if (StubRoutines::dsin() == nullptr) {
 251       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
 252     } else {
 253       fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
 254     }
 255     break;
 256   case Interpreter::java_lang_math_cos :
 257     if (StubRoutines::dcos() == nullptr) {
 258       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
 259     } else {
 260       fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
 261     }
 262     break;
 263   case Interpreter::java_lang_math_tan :
 264     if (StubRoutines::dtan() == nullptr) {
 265       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
 266     } else {
 267       fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
 268     }
 269     break;
 270   case Interpreter::java_lang_math_log :
 271     if (StubRoutines::dlog() == nullptr) {
 272       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
 273     } else {
 274       fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
 275     }
 276     break;
 277   case Interpreter::java_lang_math_log10 :
 278     if (StubRoutines::dlog10() == nullptr) {
 279       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
 280     } else {
 281       fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
 282     }
 283     break;
 284   case Interpreter::java_lang_math_exp :
 285     if (StubRoutines::dexp() == nullptr) {
 286       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
 287     } else {
 288       fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
 289     }
 290     break;
 291   case Interpreter::java_lang_math_pow :
 292     if (StubRoutines::dpow() == nullptr) {
 293       fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
 294     } else {
 295       fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
 296     }
 297     break;
 298   default:
 299     ShouldNotReachHere();
 300     fn = nullptr;  // unreachable
 301   }
 302   __ mov(rscratch1, fn);
 303   __ blr(rscratch1);
 304 }
 305 
 306 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
 307   assert(VM_Version::supports_float16(), "this intrinsic is not supported");
 308   // r19_sender_sp: sender sp
 309   // stack:
 310   //        [ arg ] <-- esp
 311   //        [ arg ]
 312   // retaddr in lr
 313   // result in v0
 314 
 315   address entry_point = __ pc();
 316   __ ldrw(c_rarg0, Address(esp));
 317   __ flt16_to_flt(v0, c_rarg0, v1);
 318   __ mov(sp, r19_sender_sp); // Restore caller's SP
 319   __ br(lr);
 320   return entry_point;
 321 }
 322 
 323 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
 324   assert(VM_Version::supports_float16(), "this intrinsic is not supported");
 325   // r19_sender_sp: sender sp
 326   // stack:
 327   //        [ arg ] <-- esp
 328   //        [ arg ]
 329   // retaddr in lr
 330   // result in c_rarg0
 331 
 332   address entry_point = __ pc();
 333   __ ldrs(v0, Address(esp));
 334   __ flt_to_flt16(c_rarg0, v0, v1);
 335   __ mov(sp, r19_sender_sp); // Restore caller's SP
 336   __ br(lr);
 337   return entry_point;
 338 }
 339 
 340 // Abstract method entry
 341 // Attempt to execute abstract method. Throw exception
 342 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 343   // rmethod: Method*
 344   // r19_sender_sp: sender SP
 345 
 346   address entry_point = __ pc();
 347 
 348   // abstract method entry
 349 
 350   //  pop return address, reset last_sp to null
 351   __ empty_expression_stack();
 352   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
 353   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
 354 
 355   // throw exception
 356   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
 357                                      InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
 358                                      rmethod);
 359   // the call_VM checks for exception, so we should never return here.
 360   __ should_not_reach_here();
 361 
 362   return entry_point;
 363 }
 364 
 365 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 366   address entry = __ pc();
 367 
 368 #ifdef ASSERT
 369   {
 370     Label L;
 371     __ ldr(rscratch1, Address(rfp,
 372                        frame::interpreter_frame_monitor_block_top_offset *
 373                        wordSize));
 374     __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
 375     __ mov(rscratch2, sp);
 376     __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
 377                            // grows negative)
 378     __ br(Assembler::HS, L); // check if frame is complete
 379     __ stop ("interpreter frame not set up");
 380     __ bind(L);
 381   }
 382 #endif // ASSERT
 383   // Restore bcp under the assumption that the current frame is still
 384   // interpreted
 385   __ restore_bcp();
 386 
 387   // expression stack must be empty before entering the VM if an
 388   // exception happened
 389   __ empty_expression_stack();
 390   // throw exception
 391   __ call_VM(noreg,
 392              CAST_FROM_FN_PTR(address,
 393                               InterpreterRuntime::throw_StackOverflowError));
 394   return entry;
 395 }
 396 
 397 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
 398   address entry = __ pc();
 399   // expression stack must be empty before entering the VM if an
 400   // exception happened
 401   __ empty_expression_stack();
 402   // setup parameters
 403 
 404   // ??? convention: expect aberrant index in register r1
 405   __ movw(c_rarg2, r1);
 406   // ??? convention: expect array in register r3
 407   __ mov(c_rarg1, r3);
 408   __ call_VM(noreg,
 409              CAST_FROM_FN_PTR(address,
 410                               InterpreterRuntime::
 411                               throw_ArrayIndexOutOfBoundsException),
 412              c_rarg1, c_rarg2);
 413   return entry;
 414 }
 415 
 416 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 417   address entry = __ pc();
 418 
 419   // object is at TOS
 420   __ pop(c_rarg1);
 421 
 422   // expression stack must be empty before entering the VM if an
 423   // exception happened
 424   __ empty_expression_stack();
 425 
 426   __ call_VM(noreg,
 427              CAST_FROM_FN_PTR(address,
 428                               InterpreterRuntime::
 429                               throw_ClassCastException),
 430              c_rarg1);
 431   return entry;
 432 }
 433 
 434 address TemplateInterpreterGenerator::generate_exception_handler_common(
 435         const char* name, const char* message, bool pass_oop) {
 436   assert(!pass_oop || message == nullptr, "either oop or message but not both");
 437   address entry = __ pc();
 438   if (pass_oop) {
 439     // object is at TOS
 440     __ pop(c_rarg2);
 441   }
 442   // expression stack must be empty before entering the VM if an
 443   // exception happened
 444   __ empty_expression_stack();
 445   // setup parameters
 446   __ lea(c_rarg1, Address((address)name));
 447   if (pass_oop) {
 448     __ call_VM(r0, CAST_FROM_FN_PTR(address,
 449                                     InterpreterRuntime::
 450                                     create_klass_exception),
 451                c_rarg1, c_rarg2);
 452   } else {
 453     // kind of lame ExternalAddress can't take null because
 454     // external_word_Relocation will assert.
 455     if (message != nullptr) {
 456       __ lea(c_rarg2, Address((address)message));
 457     } else {
 458       __ mov(c_rarg2, NULL_WORD);
 459     }
 460     __ call_VM(r0,
 461                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 462                c_rarg1, c_rarg2);
 463   }
 464   // throw exception
 465   __ b(address(Interpreter::throw_exception_entry()));
 466   return entry;
 467 }
 468 
 469 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 470   address entry = __ pc();
 471 
 472   // Restore stack bottom in case i2c adjusted stack
 473   __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 474   __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
 475   // and null it as marker that esp is now tos until next java call
 476   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 477 
 478   if (state == atos && InlineTypeReturnedAsFields) {
 479     __ store_inline_type_fields_to_buf(nullptr, true);
 480   }
 481 
 482   __ restore_bcp();
 483   __ restore_locals();
 484   __ restore_constant_pool_cache();
 485   __ get_method(rmethod);
 486 
 487   if (state == atos) {
 488     Register obj = r0;
 489     Register mdp = r1;
 490     Register tmp = r2;
 491     __ profile_return_type(mdp, obj, tmp);
 492   }
 493 
 494   const Register cache = r1;
 495   const Register index = r2;
 496 
 497   if (index_size == sizeof(u4)) {
 498     __ load_resolved_indy_entry(cache, index);
 499     __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
 500     __ add(esp, esp, cache, Assembler::LSL, 3);
 501   } else {
 502     // Pop N words from the stack
 503     assert(index_size == sizeof(u2), "Can only be u2");
 504     __ load_method_entry(cache, index);
 505     __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
 506     __ add(esp, esp, cache, Assembler::LSL, 3);
 507   }
 508 
 509   // Restore machine SP
 510   __ restore_sp_after_call();
 511 
 512   __ check_and_handle_popframe(rthread);
 513   __ check_and_handle_earlyret(rthread);
 514 
 515   __ get_dispatch();
 516   __ dispatch_next(state, step);
 517 
 518   return entry;
 519 }
 520 
 521 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 522                                                                int step,
 523                                                                address continuation) {
 524   address entry = __ pc();
 525   __ restore_bcp();
 526   __ restore_locals();
 527   __ restore_constant_pool_cache();
 528   __ get_method(rmethod);
 529   __ get_dispatch();
 530 
 531   __ restore_sp_after_call();  // Restore SP to extended SP
 532 
 533   // Restore expression stack pointer
 534   __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 535   __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
 536   // null last_sp until next java call
 537   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 538 
 539 #if INCLUDE_JVMCI
 540   // Check if we need to take lock at entry of synchronized method.  This can
 541   // only occur on method entry so emit it only for vtos with step 0.
 542   if (EnableJVMCI && state == vtos && step == 0) {
 543     Label L;
 544     __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
 545     __ cbz(rscratch1, L);
 546     // Clear flag.
 547     __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
 548     // Take lock.
 549     lock_method();
 550     __ bind(L);
 551   } else {
 552 #ifdef ASSERT
 553     if (EnableJVMCI) {
 554       Label L;
 555       __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
 556       __ cbz(rscratch1, L);
 557       __ stop("unexpected pending monitor in deopt entry");
 558       __ bind(L);
 559     }
 560 #endif
 561   }
 562 #endif
 563   // handle exceptions
 564   {
 565     Label L;
 566     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 567     __ cbz(rscratch1, L);
 568     __ call_VM(noreg,
 569                CAST_FROM_FN_PTR(address,
 570                                 InterpreterRuntime::throw_pending_exception));
 571     __ should_not_reach_here();
 572     __ bind(L);
 573   }
 574 
 575   if (continuation == nullptr) {
 576     __ dispatch_next(state, step);
 577   } else {
 578     __ jump_to_entry(continuation);
 579   }
 580   return entry;
 581 }
 582 
 583 address TemplateInterpreterGenerator::generate_result_handler_for(
 584         BasicType type) {
 585     address entry = __ pc();
 586   switch (type) {
 587   case T_BOOLEAN: __ c2bool(r0);         break;
 588   case T_CHAR   : __ uxth(r0, r0);       break;
 589   case T_BYTE   : __ sxtb(r0, r0);        break;
 590   case T_SHORT  : __ sxth(r0, r0);        break;
 591   case T_INT    : __ uxtw(r0, r0);        break;  // FIXME: We almost certainly don't need this
 592   case T_LONG   : /* nothing to do */        break;
 593   case T_VOID   : /* nothing to do */        break;
 594   case T_FLOAT  : /* nothing to do */        break;
 595   case T_DOUBLE : /* nothing to do */        break;
 596   case T_OBJECT :
 597     // retrieve result from frame
 598     __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
 599     // and verify it
 600     __ verify_oop(r0);
 601     break;
 602   default       : ShouldNotReachHere();
 603   }
 604   __ ret(lr);                                  // return from result handler
 605   return entry;
 606 }
 607 
 608 address TemplateInterpreterGenerator::generate_safept_entry_for(
 609         TosState state,
 610         address runtime_entry) {
 611   address entry = __ pc();
 612   __ push(state);
 613   __ push_cont_fastpath(rthread);
 614   __ call_VM(noreg, runtime_entry);
 615   __ pop_cont_fastpath(rthread);
 616   __ membar(Assembler::AnyAny);
 617   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 618   return entry;
 619 }
 620 
 621 // Helpers for commoning out cases in the various type of method entries.
 622 //
 623 
 624 
 625 // increment invocation count & check for overflow
 626 //
 627 // Note: checking for negative value instead of overflow
 628 //       so we have a 'sticky' overflow test
 629 //
 630 // rmethod: method
 631 //
 632 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
 633   Label done;
 634   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 635   int increment = InvocationCounter::count_increment;
 636   Label no_mdo;
 637   if (ProfileInterpreter) {
 638     // Are we profiling?
 639     __ ldr(r0, Address(rmethod, Method::method_data_offset()));
 640     __ cbz(r0, no_mdo);
 641     // Increment counter in the MDO
 642     const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
 643                                               in_bytes(InvocationCounter::counter_offset()));
 644     const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
 645     __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
 646     __ b(done);
 647   }
 648   __ bind(no_mdo);
 649   // Increment counter in MethodCounters
 650   const Address invocation_counter(rscratch2,
 651                 MethodCounters::invocation_counter_offset() +
 652                 InvocationCounter::counter_offset());
 653   __ get_method_counters(rmethod, rscratch2, done);
 654   const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
 655   __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
 656   __ bind(done);
 657 }
 658 
 659 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 660 
 661   // Asm interpreter on entry
 662   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 663   // Everything as it was on entry
 664 
 665   // InterpreterRuntime::frequency_counter_overflow takes two
 666   // arguments, the first (thread) is passed by call_VM, the second
 667   // indicates if the counter overflow occurs at a backwards branch
 668   // (null bcp).  We pass zero for it.  The call returns the address
 669   // of the verified entry point for the method or null if the
 670   // compilation did not complete (either went background or bailed
 671   // out).
 672   __ mov(c_rarg1, 0);
 673   __ call_VM(noreg,
 674              CAST_FROM_FN_PTR(address,
 675                               InterpreterRuntime::frequency_counter_overflow),
 676              c_rarg1);
 677 
 678   __ b(do_continue);
 679 }
 680 
 681 // See if we've got enough room on the stack for locals plus overhead
 682 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 683 // without going through the signal handler, i.e., reserved and yellow zones
 684 // will not be made usable. The shadow zone must suffice to handle the
 685 // overflow.
 686 // The expression stack grows down incrementally, so the normal guard
 687 // page mechanism will work for that.
 688 //
 689 // NOTE: Since the additional locals are also always pushed (wasn't
 690 // obvious in generate_method_entry) so the guard should work for them
 691 // too.
 692 //
 693 // Args:
 694 //      r3: number of additional locals this frame needs (what we must check)
 695 //      rmethod: Method*
 696 //
 697 // Kills:
 698 //      r0
 699 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 700 
 701   // monitor entry size: see picture of stack set
 702   // (generate_method_entry) and frame_amd64.hpp
 703   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 704 
 705   // total overhead size: entry_size + (saved rbp through expr stack
 706   // bottom).  be sure to change this if you add/subtract anything
 707   // to/from the overhead area
 708   const int overhead_size =
 709     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 710 
 711   const size_t page_size = os::vm_page_size();
 712 
 713   Label after_frame_check;
 714 
 715   // see if the frame is greater than one page in size. If so,
 716   // then we need to verify there is enough stack space remaining
 717   // for the additional locals.
 718   //
 719   // Note that we use SUBS rather than CMP here because the immediate
 720   // field of this instruction may overflow.  SUBS can cope with this
 721   // because it is a macro that will expand to some number of MOV
 722   // instructions and a register operation.
 723   __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
 724   __ br(Assembler::LS, after_frame_check);
 725 
 726   // compute rsp as if this were going to be the last frame on
 727   // the stack before the red zone
 728 
 729   // locals + overhead, in bytes
 730   __ mov(r0, overhead_size);
 731   __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize);  // 2 slots per parameter.
 732 
 733   const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
 734   __ ldr(rscratch1, stack_limit);
 735 
 736 #ifdef ASSERT
 737   Label limit_okay;
 738   // Verify that thread stack limit is non-zero.
 739   __ cbnz(rscratch1, limit_okay);
 740   __ stop("stack overflow limit is zero");
 741   __ bind(limit_okay);
 742 #endif
 743 
 744   // Add stack limit to locals.
 745   __ add(r0, r0, rscratch1);
 746 
 747   // Check against the current stack bottom.
 748   __ cmp(sp, r0);
 749   __ br(Assembler::HI, after_frame_check);
 750 
 751   // Remove the incoming args, peeling the machine SP back to where it
 752   // was in the caller.  This is not strictly necessary, but unless we
 753   // do so the stack frame may have a garbage FP; this ensures a
 754   // correct call stack that we can always unwind.  The ANDR should be
 755   // unnecessary because the sender SP in r19 is always aligned, but
 756   // it doesn't hurt.
 757   __ andr(sp, r19_sender_sp, -16);
 758 
 759   // Note: the restored frame is not necessarily interpreted.
 760   // Use the shared runtime version of the StackOverflowError.
 761   assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
 762   __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
 763 
 764   // all done with frame size check
 765   __ bind(after_frame_check);
 766 }
 767 
 768 // Allocate monitor and lock method (asm interpreter)
 769 //
 770 // Args:
 771 //      rmethod: Method*
 772 //      rlocals: locals
 773 //
 774 // Kills:
 775 //      r0
 776 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 777 //      rscratch1, rscratch2 (scratch regs)
 778 void TemplateInterpreterGenerator::lock_method() {
 779   // synchronize method
 780   const Address access_flags(rmethod, Method::access_flags_offset());
 781   const Address monitor_block_top(
 782         rfp,
 783         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 784   const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
 785 
 786 #ifdef ASSERT
 787   {
 788     Label L;
 789     __ ldrw(r0, access_flags);
 790     __ tst(r0, JVM_ACC_SYNCHRONIZED);
 791     __ br(Assembler::NE, L);
 792     __ stop("method doesn't need synchronization");
 793     __ bind(L);
 794   }
 795 #endif // ASSERT
 796 
 797   // get synchronization object
 798   {
 799     Label done;
 800     __ ldrw(r0, access_flags);
 801     __ tst(r0, JVM_ACC_STATIC);
 802     // get receiver (assume this is frequent case)
 803     __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 804     __ br(Assembler::EQ, done);
 805     __ load_mirror(r0, rmethod, r5, rscratch2);
 806 
 807 #ifdef ASSERT
 808     {
 809       Label L;
 810       __ cbnz(r0, L);
 811       __ stop("synchronization object is null");
 812       __ bind(L);
 813     }
 814 #endif // ASSERT
 815 
 816     __ bind(done);
 817   }
 818 
 819   // add space for monitor & lock
 820   __ check_extended_sp();
 821   __ sub(sp, sp, entry_size); // add space for a monitor entry
 822   __ sub(esp, esp, entry_size);
 823   __ sub(rscratch1, sp, rfp);
 824   __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
 825   __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
 826   __ sub(rscratch1, esp, rfp);
 827   __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
 828   __ str(rscratch1, monitor_block_top);  // set new monitor block top
 829 
 830   // store object
 831   __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
 832   __ mov(c_rarg1, esp); // object address
 833   __ lock_object(c_rarg1);
 834 }
 835 
 836 // Generate a fixed interpreter frame. This is identical setup for
 837 // interpreted methods and for native methods hence the shared code.
 838 //
 839 // Args:
 840 //      lr: return address
 841 //      rmethod: Method*
 842 //      rlocals: pointer to locals
 843 //      rcpool: cp cache
 844 //      stack_pointer: previous sp
 845 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 846   // initialize fixed part of activation frame
 847   if (native_call) {
 848     __ sub(esp, sp, 14 *  wordSize);
 849     __ mov(rbcp, zr);
 850     __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
 851     __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
 852     // add 2 zero-initialized slots for native calls
 853     __ stp(zr, zr, Address(sp, 12 * wordSize));
 854   } else {
 855     __ sub(esp, sp, 12 *  wordSize);
 856     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));    // get ConstMethod
 857     __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
 858     __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
 859     __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
 860   }
 861 
 862   if (ProfileInterpreter) {
 863     Label method_data_continue;
 864     __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
 865     __ cbz(rscratch1, method_data_continue);
 866     __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
 867     __ bind(method_data_continue);
 868     __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize));  // save Method* and mdp (method data pointer)
 869   } else {
 870     __ stp(zr, rmethod, Address(sp, 6 * wordSize));         // save Method* (no mdp)
 871   }
 872 
 873   __ protect_return_address();
 874   __ stp(rfp, lr, Address(sp, 10 * wordSize));
 875   __ lea(rfp, Address(sp, 10 * wordSize));
 876 
 877   __ ldr(rcpool, Address(rmethod, Method::const_offset()));
 878   __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
 879   __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
 880   __ sub(rscratch1, rlocals, rfp);
 881   __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize);   // rscratch1 = rlocals - fp();
 882   // Store relativized rlocals, see frame::interpreter_frame_locals().
 883   __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
 884 
 885   // set sender sp
 886   // leave last_sp as null
 887   __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
 888 
 889   // Get mirror
 890   __ load_mirror(r10, rmethod, r5, rscratch2);
 891   if (! native_call) {
 892     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 893     __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 894     __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
 895     __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
 896     __ andr(rscratch1, rscratch1, -16);
 897     __ sub(rscratch2, rscratch1, rfp);
 898     __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
 899     // Store extended SP and mirror
 900     __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
 901     // Move SP out of the way
 902     __ mov(sp, rscratch1);
 903   } else {
 904     // Make sure there is room for the exception oop pushed in case method throws
 905     // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
 906     __ sub(rscratch1, sp, 2 * wordSize);
 907     __ sub(rscratch2, rscratch1, rfp);
 908     __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
 909     __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
 910     __ mov(sp, rscratch1);
 911   }
 912 }
 913 
 914 // End of helpers
 915 
 916 // Various method entries
 917 //------------------------------------------------------------------------------------------------------------------------
 918 //
 919 //
 920 
 921 // Method entry for java.lang.ref.Reference.get.
 922 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 923   // Code: _aload_0, _getfield, _areturn
 924   // parameter size = 1
 925   //
 926   // The code that gets generated by this routine is split into 2 parts:
 927   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 928   //    2. The slow path - which is an expansion of the regular method entry.
 929   //
 930   // Notes:-
 931   // * In the G1 code we do not check whether we need to block for
 932   //   a safepoint. If G1 is enabled then we must execute the specialized
 933   //   code for Reference.get (except when the Reference object is null)
 934   //   so that we can log the value in the referent field with an SATB
 935   //   update buffer.
 936   //   If the code for the getfield template is modified so that the
 937   //   G1 pre-barrier code is executed when the current method is
 938   //   Reference.get() then going through the normal method entry
 939   //   will be fine.
 940   // * The G1 code can, however, check the receiver object (the instance
 941   //   of java.lang.Reference) and jump to the slow path if null. If the
 942   //   Reference object is null then we obviously cannot fetch the referent
 943   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 944   //   regular method entry code to generate the NPE.
 945   //
 946   // This code is based on generate_accessor_entry.
 947   //
 948   // rmethod: Method*
 949   // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
 950 
 951   // LR is live.  It must be saved around calls.
 952 
 953   address entry = __ pc();
 954 
 955   const int referent_offset = java_lang_ref_Reference::referent_offset();
 956 
 957   Label slow_path;
 958   const Register local_0 = c_rarg0;
 959   // Check if local 0 != null
 960   // If the receiver is null then it is OK to jump to the slow path.
 961   __ ldr(local_0, Address(esp, 0));
 962   __ cbz(local_0, slow_path);
 963 
 964   // Load the value of the referent field.
 965   const Address field_address(local_0, referent_offset);
 966   BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
 967   bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
 968 
 969   // areturn
 970   __ andr(sp, r19_sender_sp, -16);  // done with stack
 971   __ ret(lr);
 972 
 973   // generate a vanilla interpreter entry as the slow path
 974   __ bind(slow_path);
 975   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 976   return entry;
 977 
 978 }
 979 
 980 /**
 981  * Method entry for static native methods:
 982  *   int java.util.zip.CRC32.update(int crc, int b)
 983  */
 984 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
 985   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
 986   address entry = __ pc();
 987 
 988   // rmethod: Method*
 989   // r19_sender_sp: senderSP must preserved for slow path
 990   // esp: args
 991 
 992   Label slow_path;
 993   // If we need a safepoint check, generate full interpreter entry.
 994   __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
 995 
 996   // We don't generate local frame and don't align stack because
 997   // we call stub code and there is no safepoint on this path.
 998 
 999   // Load parameters
1000   const Register crc = c_rarg0;  // crc
1001   const Register val = c_rarg1;  // source java byte value
1002   const Register tbl = c_rarg2;  // scratch
1003 
1004   // Arguments are reversed on java expression stack
1005   __ ldrw(val, Address(esp, 0));              // byte value
1006   __ ldrw(crc, Address(esp, wordSize));       // Initial CRC
1007 
1008   uint64_t offset;
1009   __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1010   __ add(tbl, tbl, offset);
1011 
1012   __ mvnw(crc, crc); // ~crc
1013   __ update_byte_crc32(crc, val, tbl);
1014   __ mvnw(crc, crc); // ~crc
1015 
1016   // result in c_rarg0
1017 
1018   __ andr(sp, r19_sender_sp, -16);
1019   __ ret(lr);
1020 
1021   // generate a vanilla native entry as the slow path
1022   __ bind(slow_path);
1023   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1024   return entry;
1025 }
1026 
1027 /**
1028  * Method entry for static native methods:
1029  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1030  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1031  */
1032 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1033   assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1034   address entry = __ pc();
1035 
1036   // rmethod,: Method*
1037   // r19_sender_sp: senderSP must preserved for slow path
1038 
1039   Label slow_path;
1040   // If we need a safepoint check, generate full interpreter entry.
1041   __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1042 
1043   // We don't generate local frame and don't align stack because
1044   // we call stub code and there is no safepoint on this path.
1045 
1046   // Load parameters
1047   const Register crc = c_rarg0;  // crc
1048   const Register buf = c_rarg1;  // source java byte array address
1049   const Register len = c_rarg2;  // length
1050   const Register off = len;      // offset (never overlaps with 'len')
1051 
1052   // Arguments are reversed on java expression stack
1053   // Calculate address of start element
1054   if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1055     __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1056     __ ldrw(off, Address(esp, wordSize)); // offset
1057     __ add(buf, buf, off); // + offset
1058     __ ldrw(crc,   Address(esp, 4*wordSize)); // Initial CRC
1059   } else {
1060     __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1061     __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1062     __ ldrw(off, Address(esp, wordSize)); // offset
1063     __ add(buf, buf, off); // + offset
1064     __ ldrw(crc,   Address(esp, 3*wordSize)); // Initial CRC
1065   }
1066   // Can now load 'len' since we're finished with 'off'
1067   __ ldrw(len, Address(esp, 0x0)); // Length
1068 
1069   __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1070 
1071   // We are frameless so we can just jump to the stub.
1072   __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1073 
1074   // generate a vanilla native entry as the slow path
1075   __ bind(slow_path);
1076   __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1077   return entry;
1078 }
1079 
1080 /**
1081  * Method entry for intrinsic-candidate (non-native) methods:
1082  *   int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1083  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1084  * Unlike CRC32, CRC32C does not have any methods marked as native
1085  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1086  */
1087 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1088   assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1089   address entry = __ pc();
1090 
1091   // Prepare jump to stub using parameters from the stack
1092   const Register crc = c_rarg0; // initial crc
1093   const Register buf = c_rarg1; // source java byte array address
1094   const Register len = c_rarg2; // len argument to the kernel
1095 
1096   const Register end = len; // index of last element to process
1097   const Register off = crc; // offset
1098 
1099   __ ldrw(end, Address(esp)); // int end
1100   __ ldrw(off, Address(esp, wordSize)); // int offset
1101   __ sub(len, end, off);
1102   __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1103   __ add(buf, buf, off); // + offset
1104   if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1105     __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1106   } else {
1107     __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1108     __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1109   }
1110 
1111   __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1112 
1113   // Jump to the stub.
1114   __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1115 
1116   return entry;
1117 }
1118 
1119 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1120   // See more discussion in stackOverflow.hpp.
1121 
1122   const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1123   const int page_size = (int)os::vm_page_size();
1124   const int n_shadow_pages = shadow_zone_size / page_size;
1125 
1126 #ifdef ASSERT
1127   Label L_good_limit;
1128   __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1129   __ cbnz(rscratch1, L_good_limit);
1130   __ stop("shadow zone safe limit is not initialized");
1131   __ bind(L_good_limit);
1132 
1133   Label L_good_watermark;
1134   __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1135   __ cbnz(rscratch1, L_good_watermark);
1136   __ stop("shadow zone growth watermark is not initialized");
1137   __ bind(L_good_watermark);
1138 #endif
1139 
1140   Label L_done;
1141 
1142   __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1143   __ cmp(sp, rscratch1);
1144   __ br(Assembler::HI, L_done);
1145 
1146   for (int p = 1; p <= n_shadow_pages; p++) {
1147     __ sub(rscratch2, sp, p*page_size);
1148     __ str(zr, Address(rscratch2));
1149   }
1150 
1151   // Record the new watermark, but only if the update is above the safe limit.
1152   // Otherwise, the next time around the check above would pass the safe limit.
1153   __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1154   __ cmp(sp, rscratch1);
1155   __ br(Assembler::LS, L_done);
1156   __ mov(rscratch1, sp);
1157   __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1158 
1159   __ bind(L_done);
1160 }
1161 
1162 // Interpreter stub for calling a native method. (asm interpreter)
1163 // This sets up a somewhat different looking stack for calling the
1164 // native method than the typical interpreter frame setup.
1165 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1166   // determine code generation flags
1167   bool inc_counter  = UseCompiler || CountCompiledCalls;
1168 
1169   // r1: Method*
1170   // rscratch1: sender sp
1171 
1172   address entry_point = __ pc();
1173 
1174   const Address constMethod       (rmethod, Method::const_offset());
1175   const Address access_flags      (rmethod, Method::access_flags_offset());
1176   const Address size_of_parameters(r2, ConstMethod::
1177                                        size_of_parameters_offset());
1178 
1179   // get parameter size (always needed)
1180   __ ldr(r2, constMethod);
1181   __ load_unsigned_short(r2, size_of_parameters);
1182 
1183   // Native calls don't need the stack size check since they have no
1184   // expression stack and the arguments are already on the stack and
1185   // we only add a handful of words to the stack.
1186 
1187   // rmethod: Method*
1188   // r2: size of parameters
1189   // rscratch1: sender sp
1190 
1191   // for natives the size of locals is zero
1192 
1193   // compute beginning of parameters (rlocals)
1194   __ add(rlocals, esp, r2, ext::uxtx, 3);
1195   __ add(rlocals, rlocals, -wordSize);
1196 
1197   // Pull SP back to minimum size: this avoids holes in the stack
1198   __ andr(sp, esp, -16);
1199 
1200   // initialize fixed part of activation frame
1201   generate_fixed_frame(true);
1202 
1203   // make sure method is native & not abstract
1204 #ifdef ASSERT
1205   __ ldrw(r0, access_flags);
1206   {
1207     Label L;
1208     __ tst(r0, JVM_ACC_NATIVE);
1209     __ br(Assembler::NE, L);
1210     __ stop("tried to execute non-native method as native");
1211     __ bind(L);
1212   }
1213   {
1214     Label L;
1215     __ tst(r0, JVM_ACC_ABSTRACT);
1216     __ br(Assembler::EQ, L);
1217     __ stop("tried to execute abstract method in interpreter");
1218     __ bind(L);
1219   }
1220 #endif
1221 
1222   // Since at this point in the method invocation the exception
1223   // handler would try to exit the monitor of synchronized methods
1224   // which hasn't been entered yet, we set the thread local variable
1225   // _do_not_unlock_if_synchronized to true. The remove_activation
1226   // will check this flag.
1227 
1228    const Address do_not_unlock_if_synchronized(rthread,
1229         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1230   __ mov(rscratch2, true);
1231   __ strb(rscratch2, do_not_unlock_if_synchronized);
1232 
1233   // increment invocation count & check for overflow
1234   Label invocation_counter_overflow;
1235   if (inc_counter) {
1236     generate_counter_incr(&invocation_counter_overflow);
1237   }
1238 
1239   Label continue_after_compile;
1240   __ bind(continue_after_compile);
1241 
1242   bang_stack_shadow_pages(true);
1243 
1244   // reset the _do_not_unlock_if_synchronized flag
1245   __ strb(zr, do_not_unlock_if_synchronized);
1246 
1247   // check for synchronized methods
1248   // Must happen AFTER invocation_counter check and stack overflow check,
1249   // so method is not locked if overflows.
1250   if (synchronized) {
1251     lock_method();
1252   } else {
1253     // no synchronization necessary
1254 #ifdef ASSERT
1255     {
1256       Label L;
1257       __ ldrw(r0, access_flags);
1258       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1259       __ br(Assembler::EQ, L);
1260       __ stop("method needs synchronization");
1261       __ bind(L);
1262     }
1263 #endif
1264   }
1265 
1266   // start execution
1267 #ifdef ASSERT
1268   {
1269     Label L;
1270     const Address monitor_block_top(rfp,
1271                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1272     __ ldr(rscratch1, monitor_block_top);
1273     __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1274     __ cmp(esp, rscratch1);
1275     __ br(Assembler::EQ, L);
1276     __ stop("broken stack frame setup in interpreter 1");
1277     __ bind(L);
1278   }
1279 #endif
1280 
1281   // jvmti support
1282   __ notify_method_entry();
1283 
1284   // work registers
1285   const Register t = r17;
1286   const Register result_handler = r19;
1287 
1288   // allocate space for parameters
1289   __ ldr(t, Address(rmethod, Method::const_offset()));
1290   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1291 
1292   __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1293   __ andr(sp, rscratch1, -16);
1294   __ mov(esp, rscratch1);
1295 
1296   // get signature handler
1297   {
1298     Label L;
1299     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1300     __ cbnz(t, L);
1301     __ call_VM(noreg,
1302                CAST_FROM_FN_PTR(address,
1303                                 InterpreterRuntime::prepare_native_call),
1304                rmethod);
1305     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1306     __ bind(L);
1307   }
1308 
1309   // call signature handler
1310   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1311          "adjust this code");
1312   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1313          "adjust this code");
1314   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1315           "adjust this code");
1316 
1317   // The generated handlers do not touch rmethod (the method).
1318   // However, large signatures cannot be cached and are generated
1319   // each time here.  The slow-path generator can do a GC on return,
1320   // so we must reload it after the call.
1321   __ blr(t);
1322   __ get_method(rmethod);        // slow path can do a GC, reload rmethod
1323 
1324 
1325   // result handler is in r0
1326   // set result handler
1327   __ mov(result_handler, r0);
1328   // pass mirror handle if static call
1329   {
1330     Label L;
1331     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1332     __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1333     // get mirror
1334     __ load_mirror(t, rmethod, r10, rscratch2);
1335     // copy mirror into activation frame
1336     __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1337     // pass handle to mirror
1338     __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1339     __ bind(L);
1340   }
1341 
1342   // get native function entry point in r10
1343   {
1344     Label L;
1345     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1346     address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1347     __ mov(rscratch2, unsatisfied);
1348     __ ldr(rscratch2, rscratch2);
1349     __ cmp(r10, rscratch2);
1350     __ br(Assembler::NE, L);
1351     __ call_VM(noreg,
1352                CAST_FROM_FN_PTR(address,
1353                                 InterpreterRuntime::prepare_native_call),
1354                rmethod);
1355     __ get_method(rmethod);
1356     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1357     __ bind(L);
1358   }
1359 
1360   // pass JNIEnv
1361   __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1362 
1363   // Set the last Java PC in the frame anchor to be the return address from
1364   // the call to the native method: this will allow the debugger to
1365   // generate an accurate stack trace.
1366   Label native_return;
1367   __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1368 
1369   // change thread state
1370 #ifdef ASSERT
1371   {
1372     Label L;
1373     __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1374     __ cmp(t, (u1)_thread_in_Java);
1375     __ br(Assembler::EQ, L);
1376     __ stop("Wrong thread state in native stub");
1377     __ bind(L);
1378   }
1379 #endif
1380 
1381   // Change state to native
1382   __ mov(rscratch1, _thread_in_native);
1383   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1384   __ stlrw(rscratch1, rscratch2);
1385 
1386   // Call the native method.
1387   __ blr(r10);
1388   __ bind(native_return);
1389   __ get_method(rmethod);
1390   // result potentially in r0 or v0
1391 
1392   // Restore cpu control state after JNI call
1393   __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1394 
1395   // make room for the pushes we're about to do
1396   __ sub(rscratch1, esp, 4 * wordSize);
1397   __ andr(sp, rscratch1, -16);
1398 
1399   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1400   // in order to extract the result of a method call. If the order of these
1401   // pushes change or anything else is added to the stack then the code in
1402   // interpreter_frame_result must also change.
1403   __ push(dtos);
1404   __ push(ltos);
1405 
1406   __ verify_sve_vector_length();
1407 
1408   // change thread state
1409   __ mov(rscratch1, _thread_in_native_trans);
1410   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1411   __ stlrw(rscratch1, rscratch2);
1412 
1413   // Force this write out before the read below
1414   if (!UseSystemMemoryBarrier) {
1415     __ dmb(Assembler::ISH);
1416   }
1417 
1418   // check for safepoint operation in progress and/or pending suspend requests
1419   {
1420     Label L, Continue;
1421 
1422     // We need an acquire here to ensure that any subsequent load of the
1423     // global SafepointSynchronize::_state flag is ordered after this load
1424     // of the thread-local polling word.  We don't want this poll to
1425     // return false (i.e. not safepointing) and a later poll of the global
1426     // SafepointSynchronize::_state spuriously to return true.
1427     //
1428     // This is to avoid a race when we're in a native->Java transition
1429     // racing the code which wakes up from a safepoint.
1430     __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1431     __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1432     __ cbz(rscratch2, Continue);
1433     __ bind(L);
1434 
1435     // Don't use call_VM as it will see a possible pending exception
1436     // and forward it and never return here preventing us from
1437     // clearing _last_native_pc down below. So we do a runtime call by
1438     // hand.
1439     //
1440     __ mov(c_rarg0, rthread);
1441     __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1442     __ blr(rscratch2);
1443     __ get_method(rmethod);
1444     __ reinit_heapbase();
1445     __ bind(Continue);
1446   }
1447 
1448   // change thread state
1449   __ mov(rscratch1, _thread_in_Java);
1450   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1451   __ stlrw(rscratch1, rscratch2);
1452 
1453   // reset_last_Java_frame
1454   __ reset_last_Java_frame(true);
1455 
1456   if (CheckJNICalls) {
1457     // clear_pending_jni_exception_check
1458     __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1459   }
1460 
1461   // reset handle block
1462   __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1463   __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1464 
1465   // If result is an oop unbox and store it in frame where gc will see it
1466   // and result handler will pick it up
1467 
1468   {
1469     Label no_oop;
1470     __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1471     __ cmp(t, result_handler);
1472     __ br(Assembler::NE, no_oop);
1473     // Unbox oop result, e.g. JNIHandles::resolve result.
1474     __ pop(ltos);
1475     __ resolve_jobject(r0, t, rscratch2);
1476     __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1477     // keep stack depth as expected by pushing oop which will eventually be discarded
1478     __ push(ltos);
1479     __ bind(no_oop);
1480   }
1481 
1482   {
1483     Label no_reguard;
1484     __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1485     __ ldrw(rscratch1, Address(rscratch1));
1486     __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1487     __ br(Assembler::NE, no_reguard);
1488 
1489     __ push_call_clobbered_registers();
1490     __ mov(c_rarg0, rthread);
1491     __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1492     __ blr(rscratch2);
1493     __ pop_call_clobbered_registers();
1494 
1495     __ bind(no_reguard);
1496   }
1497 
1498   // The method register is junk from after the thread_in_native transition
1499   // until here.  Also can't call_VM until the bcp has been
1500   // restored.  Need bcp for throwing exception below so get it now.
1501   __ get_method(rmethod);
1502 
1503   // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1504   // rbcp == code_base()
1505   __ ldr(rbcp, Address(rmethod, Method::const_offset()));   // get ConstMethod*
1506   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));          // get codebase
1507   // handle exceptions (exception handling will handle unlocking!)
1508   {
1509     Label L;
1510     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1511     __ cbz(rscratch1, L);
1512     // Note: At some point we may want to unify this with the code
1513     // used in call_VM_base(); i.e., we should use the
1514     // StubRoutines::forward_exception code. For now this doesn't work
1515     // here because the rsp is not correctly set at this point.
1516     __ MacroAssembler::call_VM(noreg,
1517                                CAST_FROM_FN_PTR(address,
1518                                InterpreterRuntime::throw_pending_exception));
1519     __ should_not_reach_here();
1520     __ bind(L);
1521   }
1522 
1523   // do unlocking if necessary
1524   {
1525     Label L;
1526     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1527     __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1528     // the code below should be shared with interpreter macro
1529     // assembler implementation
1530     {
1531       Label unlock;
1532       // BasicObjectLock will be first in list, since this is a
1533       // synchronized method. However, need to check that the object
1534       // has not been unlocked by an explicit monitorexit bytecode.
1535 
1536       // monitor expect in c_rarg1 for slow unlock path
1537       __ lea (c_rarg1, Address(rfp,   // address of first monitor
1538                                (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1539                                           wordSize - sizeof(BasicObjectLock))));
1540 
1541       __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1542       __ cbnz(t, unlock);
1543 
1544       // Entry already unlocked, need to throw exception
1545       __ MacroAssembler::call_VM(noreg,
1546                                  CAST_FROM_FN_PTR(address,
1547                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1548       __ should_not_reach_here();
1549 
1550       __ bind(unlock);
1551       __ unlock_object(c_rarg1);
1552     }
1553     __ bind(L);
1554   }
1555 
1556   // jvmti support
1557   // Note: This must happen _after_ handling/throwing any exceptions since
1558   //       the exception handler code notifies the runtime of method exits
1559   //       too. If this happens before, method entry/exit notifications are
1560   //       not properly paired (was bug - gri 11/22/99).
1561   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1562 
1563   // restore potential result in r0:d0, call result handler to
1564   // restore potential result in ST0 & handle result
1565 
1566   __ pop(ltos);
1567   __ pop(dtos);
1568 
1569   __ blr(result_handler);
1570 
1571   // remove activation
1572   __ ldr(esp, Address(rfp,
1573                     frame::interpreter_frame_sender_sp_offset *
1574                     wordSize)); // get sender sp
1575   // remove frame anchor
1576   __ leave();
1577 
1578   // restore sender sp
1579   __ mov(sp, esp);
1580 
1581   __ ret(lr);
1582 
1583   if (inc_counter) {
1584     // Handle overflow of counter and compile method
1585     __ bind(invocation_counter_overflow);
1586     generate_counter_overflow(continue_after_compile);
1587   }
1588 
1589   return entry_point;
1590 }
1591 
1592 //
1593 // Generic interpreted method entry to (asm) interpreter
1594 //
1595 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1596   // determine code generation flags
1597   bool inc_counter  = UseCompiler || CountCompiledCalls;
1598 
1599   // rscratch1: sender sp
1600   address entry_point = __ pc();
1601 
1602   const Address constMethod(rmethod, Method::const_offset());
1603   const Address access_flags(rmethod, Method::access_flags_offset());
1604   const Address size_of_parameters(r3,
1605                                    ConstMethod::size_of_parameters_offset());
1606   const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1607 
1608   // get parameter size (always needed)
1609   // need to load the const method first
1610   __ ldr(r3, constMethod);
1611   __ load_unsigned_short(r2, size_of_parameters);
1612 
1613   // r2: size of parameters
1614 
1615   __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1616   __ sub(r3, r3, r2); // r3 = no. of additional locals
1617 
1618   // see if we've got enough room on the stack for locals plus overhead.
1619   generate_stack_overflow_check();
1620 
1621   // compute beginning of parameters (rlocals)
1622   __ add(rlocals, esp, r2, ext::uxtx, 3);
1623   __ sub(rlocals, rlocals, wordSize);
1624 
1625   __ mov(rscratch1, esp);
1626 
1627   // r3 - # of additional locals
1628   // allocate space for locals
1629   // explicitly initialize locals
1630   // Initializing memory allocated for locals in the same direction as
1631   // the stack grows to ensure page initialization order according
1632   // to windows-aarch64 stack page growth requirement (see
1633   // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1634   {
1635     Label exit, loop;
1636     __ ands(zr, r3, r3);
1637     __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1638     __ bind(loop);
1639     __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1640     __ sub(r3, r3, 1); // until everything initialized
1641     __ cbnz(r3, loop);
1642     __ bind(exit);
1643   }
1644 
1645   // Padding between locals and fixed part of activation frame to ensure
1646   // SP is always 16-byte aligned.
1647   __ andr(sp, rscratch1, -16);
1648 
1649   // And the base dispatch table
1650   __ get_dispatch();
1651 
1652   // initialize fixed part of activation frame
1653   generate_fixed_frame(false);
1654 
1655   // make sure method is not native & not abstract
1656 #ifdef ASSERT
1657   __ ldrw(r0, access_flags);
1658   {
1659     Label L;
1660     __ tst(r0, JVM_ACC_NATIVE);
1661     __ br(Assembler::EQ, L);
1662     __ stop("tried to execute native method as non-native");
1663     __ bind(L);
1664   }
1665  {
1666     Label L;
1667     __ tst(r0, JVM_ACC_ABSTRACT);
1668     __ br(Assembler::EQ, L);
1669     __ stop("tried to execute abstract method in interpreter");
1670     __ bind(L);
1671   }
1672 #endif
1673 
1674   // Since at this point in the method invocation the exception
1675   // handler would try to exit the monitor of synchronized methods
1676   // which hasn't been entered yet, we set the thread local variable
1677   // _do_not_unlock_if_synchronized to true. The remove_activation
1678   // will check this flag.
1679 
1680    const Address do_not_unlock_if_synchronized(rthread,
1681         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1682   __ mov(rscratch2, true);
1683   __ strb(rscratch2, do_not_unlock_if_synchronized);
1684 
1685   Register mdp = r3;
1686   __ profile_parameters_type(mdp, r1, r2);
1687 
1688   // increment invocation count & check for overflow
1689   Label invocation_counter_overflow;
1690   if (inc_counter) {
1691     generate_counter_incr(&invocation_counter_overflow);
1692   }
1693 
1694   Label continue_after_compile;
1695   __ bind(continue_after_compile);
1696 
1697   bang_stack_shadow_pages(false);
1698 
1699   // reset the _do_not_unlock_if_synchronized flag
1700   __ strb(zr, do_not_unlock_if_synchronized);
1701 
1702   // check for synchronized methods
1703   // Must happen AFTER invocation_counter check and stack overflow check,
1704   // so method is not locked if overflows.
1705   if (synchronized) {
1706     // Allocate monitor and lock method
1707     lock_method();
1708   } else {
1709     // no synchronization necessary
1710 #ifdef ASSERT
1711     {
1712       Label L;
1713       __ ldrw(r0, access_flags);
1714       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1715       __ br(Assembler::EQ, L);
1716       __ stop("method needs synchronization");
1717       __ bind(L);
1718     }
1719 #endif
1720   }
1721 
1722   // start execution
1723 #ifdef ASSERT
1724   {
1725     Label L;
1726      const Address monitor_block_top (rfp,
1727                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1728     __ ldr(rscratch1, monitor_block_top);
1729     __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1730     __ cmp(esp, rscratch1);
1731     __ br(Assembler::EQ, L);
1732     __ stop("broken stack frame setup in interpreter 2");
1733     __ bind(L);
1734   }
1735 #endif
1736 
1737   // jvmti support
1738   __ notify_method_entry();
1739 
1740   __ dispatch_next(vtos);
1741 
1742   // invocation counter overflow
1743   if (inc_counter) {
1744     // Handle overflow of counter and compile method
1745     __ bind(invocation_counter_overflow);
1746     generate_counter_overflow(continue_after_compile);
1747   }
1748 
1749   return entry_point;
1750 }
1751 
1752 // Method entry for java.lang.Thread.currentThread
1753 address TemplateInterpreterGenerator::generate_currentThread() {
1754   address entry_point = __ pc();
1755 
1756   __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1757   __ resolve_oop_handle(r0, rscratch1, rscratch2);
1758   __ ret(lr);
1759 
1760   return entry_point;
1761 }
1762 
1763 // Not supported
1764 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1765 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1766 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1767 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1768 
1769 //-----------------------------------------------------------------------------
1770 // Exceptions
1771 
1772 void TemplateInterpreterGenerator::generate_throw_exception() {
1773   // Entry point in previous activation (i.e., if the caller was
1774   // interpreted)
1775   Interpreter::_rethrow_exception_entry = __ pc();
1776   // Restore sp to interpreter_frame_last_sp even though we are going
1777   // to empty the expression stack for the exception processing.
1778   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1779   // r0: exception
1780   // r3: return address/pc that threw exception
1781   __ restore_bcp();    // rbcp points to call/send
1782   __ restore_locals();
1783   __ restore_constant_pool_cache();
1784   __ reinit_heapbase();  // restore rheapbase as heapbase.
1785   __ get_dispatch();
1786 
1787   // Entry point for exceptions thrown within interpreter code
1788   Interpreter::_throw_exception_entry = __ pc();
1789   // If we came here via a NullPointerException on the receiver of a
1790   // method, rmethod may be corrupt.
1791   __ get_method(rmethod);
1792   // expression stack is undefined here
1793   // r0: exception
1794   // rbcp: exception bcp
1795   __ verify_oop(r0);
1796   __ mov(c_rarg1, r0);
1797 
1798   // expression stack must be empty before entering the VM in case of
1799   // an exception
1800   __ empty_expression_stack();
1801   // find exception handler address and preserve exception oop
1802   __ call_VM(r3,
1803              CAST_FROM_FN_PTR(address,
1804                           InterpreterRuntime::exception_handler_for_exception),
1805              c_rarg1);
1806 
1807   // Restore machine SP
1808   __ restore_sp_after_call();
1809 
1810   // r0: exception handler entry point
1811   // r3: preserved exception oop
1812   // rbcp: bcp for exception handler
1813   __ push_ptr(r3); // push exception which is now the only value on the stack
1814   __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1815 
1816   // If the exception is not handled in the current frame the frame is
1817   // removed and the exception is rethrown (i.e. exception
1818   // continuation is _rethrow_exception).
1819   //
1820   // Note: At this point the bci is still the bxi for the instruction
1821   // which caused the exception and the expression stack is
1822   // empty. Thus, for any VM calls at this point, GC will find a legal
1823   // oop map (with empty expression stack).
1824 
1825   //
1826   // JVMTI PopFrame support
1827   //
1828 
1829   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1830   __ empty_expression_stack();
1831   // Set the popframe_processing bit in pending_popframe_condition
1832   // indicating that we are currently handling popframe, so that
1833   // call_VMs that may happen later do not trigger new popframe
1834   // handling cycles.
1835   __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1836   __ orr(r3, r3, JavaThread::popframe_processing_bit);
1837   __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1838 
1839   {
1840     // Check to see whether we are returning to a deoptimized frame.
1841     // (The PopFrame call ensures that the caller of the popped frame is
1842     // either interpreted or compiled and deoptimizes it if compiled.)
1843     // In this case, we can't call dispatch_next() after the frame is
1844     // popped, but instead must save the incoming arguments and restore
1845     // them after deoptimization has occurred.
1846     //
1847     // Note that we don't compare the return PC against the
1848     // deoptimization blob's unpack entry because of the presence of
1849     // adapter frames in C2.
1850     Label caller_not_deoptimized;
1851     __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1852     // This is a return address, so requires authenticating for PAC.
1853     __ authenticate_return_address(c_rarg1);
1854     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1855                                InterpreterRuntime::interpreter_contains), c_rarg1);
1856     __ cbnz(r0, caller_not_deoptimized);
1857 
1858     // Compute size of arguments for saving when returning to
1859     // deoptimized caller
1860     __ get_method(r0);
1861     __ ldr(r0, Address(r0, Method::const_offset()));
1862     __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1863                                                     size_of_parameters_offset())));
1864     __ lsl(r0, r0, Interpreter::logStackElementSize);
1865     __ restore_locals(); // XXX do we need this?
1866     __ sub(rlocals, rlocals, r0);
1867     __ add(rlocals, rlocals, wordSize);
1868     // Save these arguments
1869     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1870                                            Deoptimization::
1871                                            popframe_preserve_args),
1872                           rthread, r0, rlocals);
1873 
1874     __ remove_activation(vtos,
1875                          /* throw_monitor_exception */ false,
1876                          /* install_monitor_exception */ false,
1877                          /* notify_jvmdi */ false);
1878 
1879     // Inform deoptimization that it is responsible for restoring
1880     // these arguments
1881     __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1882     __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1883 
1884     // Continue in deoptimization handler
1885     __ ret(lr);
1886 
1887     __ bind(caller_not_deoptimized);
1888   }
1889 
1890   __ remove_activation(vtos,
1891                        /* throw_monitor_exception */ false,
1892                        /* install_monitor_exception */ false,
1893                        /* notify_jvmdi */ false);
1894 
1895   // Restore the last_sp and null it out
1896   __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1897   __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1898   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1899 
1900   __ restore_bcp();
1901   __ restore_locals();
1902   __ restore_constant_pool_cache();
1903   __ get_method(rmethod);
1904   __ get_dispatch();
1905 
1906   // The method data pointer was incremented already during
1907   // call profiling. We have to restore the mdp for the current bcp.
1908   if (ProfileInterpreter) {
1909     __ set_method_data_pointer_for_bcp();
1910   }
1911 
1912   // Clear the popframe condition flag
1913   __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1914   assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1915 
1916 #if INCLUDE_JVMTI
1917   {
1918     Label L_done;
1919 
1920     __ ldrb(rscratch1, Address(rbcp, 0));
1921     __ cmpw(rscratch1, Bytecodes::_invokestatic);
1922     __ br(Assembler::NE, L_done);
1923 
1924     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1925     // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1926 
1927     __ ldr(c_rarg0, Address(rlocals, 0));
1928     __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1929 
1930     __ cbz(r0, L_done);
1931 
1932     __ str(r0, Address(esp, 0));
1933     __ bind(L_done);
1934   }
1935 #endif // INCLUDE_JVMTI
1936 
1937   // Restore machine SP
1938   __ restore_sp_after_call();
1939 
1940   __ dispatch_next(vtos);
1941   // end of PopFrame support
1942 
1943   Interpreter::_remove_activation_entry = __ pc();
1944 
1945   // preserve exception over this code sequence
1946   __ pop_ptr(r0);
1947   __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1948   // remove the activation (without doing throws on illegalMonitorExceptions)
1949   __ remove_activation(vtos, false, true, false);
1950   // restore exception
1951   __ get_vm_result(r0, rthread);
1952 
1953   // In between activations - previous activation type unknown yet
1954   // compute continuation point - the continuation point expects the
1955   // following registers set up:
1956   //
1957   // r0: exception
1958   // lr: return address/pc that threw exception
1959   // esp: expression stack of caller
1960   // rfp: fp of caller
1961   __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize)));  // save exception & return address
1962   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1963                           SharedRuntime::exception_handler_for_return_address),
1964                         rthread, lr);
1965   __ mov(r1, r0);                               // save exception handler
1966   __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize)));  // restore exception & return address
1967   // We might be returning to a deopt handler that expects r3 to
1968   // contain the exception pc
1969   __ mov(r3, lr);
1970   // Note that an "issuing PC" is actually the next PC after the call
1971   __ br(r1);                                    // jump to exception
1972                                                 // handler of caller
1973 }
1974 
1975 
1976 //
1977 // JVMTI ForceEarlyReturn support
1978 //
1979 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1980   address entry = __ pc();
1981 
1982   __ restore_bcp();
1983   __ restore_locals();
1984   __ empty_expression_stack();
1985   __ load_earlyret_value(state);
1986 
1987   __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1988   Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1989 
1990   // Clear the earlyret state
1991   assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1992   __ str(zr, cond_addr);
1993 
1994   __ remove_activation(state,
1995                        false, /* throw_monitor_exception */
1996                        false, /* install_monitor_exception */
1997                        true); /* notify_jvmdi */
1998   __ ret(lr);
1999 
2000   return entry;
2001 } // end of ForceEarlyReturn support
2002 
2003 
2004 
2005 //-----------------------------------------------------------------------------
2006 // Helper for vtos entry point generation
2007 
2008 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2009                                                          address& bep,
2010                                                          address& cep,
2011                                                          address& sep,
2012                                                          address& aep,
2013                                                          address& iep,
2014                                                          address& lep,
2015                                                          address& fep,
2016                                                          address& dep,
2017                                                          address& vep) {
2018   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2019   Label L;
2020   aep = __ pc();  __ push_ptr();  __ b(L);
2021   fep = __ pc();  __ push_f();    __ b(L);
2022   dep = __ pc();  __ push_d();    __ b(L);
2023   lep = __ pc();  __ push_l();    __ b(L);
2024   bep = cep = sep =
2025   iep = __ pc();  __ push_i();
2026   vep = __ pc();
2027   __ bind(L);
2028   generate_and_dispatch(t);
2029 }
2030 
2031 //-----------------------------------------------------------------------------
2032 
2033 // Non-product code
2034 #ifndef PRODUCT
2035 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2036   address entry = __ pc();
2037 
2038   __ protect_return_address();
2039   __ push(lr);
2040   __ push(state);
2041   __ push(RegSet::range(r0, r15), sp);
2042   __ mov(c_rarg2, r0);  // Pass itos
2043   __ call_VM(noreg,
2044              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2045              c_rarg1, c_rarg2, c_rarg3);
2046   __ pop(RegSet::range(r0, r15), sp);
2047   __ pop(state);
2048   __ pop(lr);
2049   __ authenticate_return_address();
2050   __ ret(lr);                                   // return from result handler
2051 
2052   return entry;
2053 }
2054 
2055 void TemplateInterpreterGenerator::count_bytecode() {
2056   __ mov(r10, (address) &BytecodeCounter::_counter_value);
2057   __ atomic_addw(noreg, 1, r10);
2058 }
2059 
2060 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2061   __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2062   __ atomic_addw(noreg, 1, r10);
2063 }
2064 
2065 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2066   // Calculate new index for counter:
2067   //   _index = (_index >> log2_number_of_codes) |
2068   //            (bytecode << log2_number_of_codes);
2069   Register index_addr = rscratch1;
2070   Register index = rscratch2;
2071   __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2072   __ ldrw(index, index_addr);
2073   __ mov(r10,
2074          ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2075   __ orrw(index, r10, index, Assembler::LSR,
2076           BytecodePairHistogram::log2_number_of_codes);
2077   __ strw(index, index_addr);
2078 
2079   // Bump bucket contents:
2080   //   _counters[_index] ++;
2081   Register counter_addr = rscratch1;
2082   __ mov(r10, (address) &BytecodePairHistogram::_counters);
2083   __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2084   __ atomic_addw(noreg, 1, counter_addr);
2085 }
2086 
2087 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2088   // Call a little run-time stub to avoid blow-up for each bytecode.
2089   // The run-time runtime saves the right registers, depending on
2090   // the tosca in-state for the given template.
2091 
2092   assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2093          "entry must have been generated");
2094   __ bl(Interpreter::trace_code(t->tos_in()));
2095   __ reinit_heapbase();
2096 }
2097 
2098 
2099 void TemplateInterpreterGenerator::stop_interpreter_at() {
2100   Label L;
2101   __ push(rscratch1);
2102   __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2103   __ ldr(rscratch1, Address(rscratch1));
2104   __ mov(rscratch2, StopInterpreterAt);
2105   __ cmpw(rscratch1, rscratch2);
2106   __ br(Assembler::NE, L);
2107   __ brk(0);
2108   __ bind(L);
2109   __ pop(rscratch1);
2110 }
2111 
2112 #endif // !PRODUCT