1 /*
   2  * Copyright (c) 2003, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 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.hpp"
  28 #include "interpreter/bytecodeHistogram.hpp"
  29 #include "interpreter/interpreter.hpp"
  30 #include "interpreter/interpreterRuntime.hpp"
  31 #include "interpreter/interp_masm.hpp"
  32 #include "interpreter/templateInterpreterGenerator.hpp"
  33 #include "interpreter/templateTable.hpp"
  34 #include "interpreter/bytecodeTracer.hpp"
  35 #include "memory/resourceArea.hpp"
  36 #include "oops/arrayOop.hpp"
  37 #include "oops/methodData.hpp"
  38 #include "oops/method.hpp"
  39 #include "oops/oop.inline.hpp"
  40 #include "prims/jvmtiExport.hpp"
  41 #include "prims/jvmtiThreadState.hpp"
  42 #include "runtime/arguments.hpp"
  43 #include "runtime/deoptimization.hpp"
  44 #include "runtime/frame.inline.hpp"
  45 #include "runtime/sharedRuntime.hpp"
  46 #include "runtime/stubRoutines.hpp"
  47 #include "runtime/synchronizer.hpp"
  48 #include "runtime/timer.hpp"
  49 #include "runtime/vframeArray.hpp"
  50 #include "utilities/debug.hpp"
  51 #include <sys/types.h>
  52 
  53 #ifndef PRODUCT
  54 #include "oops/method.hpp"
  55 #endif // !PRODUCT
  56 
  57 #ifdef BUILTIN_SIM
  58 #include "../../../../../../simulator/simulator.hpp"
  59 #endif
  60 
  61 // Size of interpreter code.  Increase if too small.  Interpreter will
  62 // fail with a guarantee ("not enough space for interpreter generation");
  63 // if too small.
  64 // Run with +PrintInterpreter to get the VM to print out the size.
  65 // Max size with JVMTI
  66 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
  67 
  68 #define __ _masm->
  69 
  70 //-----------------------------------------------------------------------------
  71 
  72 extern "C" void entry(CodeBuffer*);
  73 
  74 //-----------------------------------------------------------------------------
  75 
  76 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  77   address entry = __ pc();
  78 
  79   __ andr(esp, esp, -16);
  80   __ mov(c_rarg3, esp);
  81   // rmethod
  82   // rlocals
  83   // c_rarg3: first stack arg - wordSize
  84 
  85   // adjust sp
  86   __ sub(sp, c_rarg3, 18 * wordSize);
  87   __ str(lr, Address(__ pre(sp, -2 * wordSize)));
  88   __ call_VM(noreg,
  89              CAST_FROM_FN_PTR(address,
  90                               InterpreterRuntime::slow_signature_handler),
  91              rmethod, rlocals, c_rarg3);
  92 
  93   // r0: result handler
  94 
  95   // Stack layout:
  96   // rsp: return address           <- sp
  97   //      1 garbage
  98   //      8 integer args (if static first is unused)
  99   //      1 float/double identifiers
 100   //      8 double args
 101   //        stack args              <- esp
 102   //        garbage
 103   //        expression stack bottom
 104   //        bcp (NULL)
 105   //        ...
 106 
 107   // Restore LR
 108   __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
 109 
 110   // Do FP first so we can use c_rarg3 as temp
 111   __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
 112 
 113   for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
 114     const FloatRegister r = as_FloatRegister(i);
 115 
 116     Label d, done;
 117 
 118     __ tbnz(c_rarg3, i, d);
 119     __ ldrs(r, Address(sp, (10 + i) * wordSize));
 120     __ b(done);
 121     __ bind(d);
 122     __ ldrd(r, Address(sp, (10 + i) * wordSize));
 123     __ bind(done);
 124   }
 125 
 126   // c_rarg0 contains the result from the call of
 127   // InterpreterRuntime::slow_signature_handler so we don't touch it
 128   // here.  It will be loaded with the JNIEnv* later.
 129   __ ldr(c_rarg1, Address(sp, 1 * wordSize));
 130   for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
 131     Register rm = as_Register(i), rn = as_Register(i+1);
 132     __ ldp(rm, rn, Address(sp, i * wordSize));
 133   }
 134 
 135   __ add(sp, sp, 18 * wordSize);
 136   __ ret(lr);
 137 
 138   return entry;
 139 }
 140 
 141 
 142 //
 143 // Various method entries
 144 //
 145 
 146 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
 147   // rmethod: Method*
 148   // r13: sender sp
 149   // esp: args
 150 
 151   if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
 152 
 153   // These don't need a safepoint check because they aren't virtually
 154   // callable. We won't enter these intrinsics from compiled code.
 155   // If in the future we added an intrinsic which was virtually callable
 156   // we'd have to worry about how to safepoint so that this code is used.
 157 
 158   // mathematical functions inlined by compiler
 159   // (interpreter must provide identical implementation
 160   // in order to avoid monotonicity bugs when switching
 161   // from interpreter to compiler in the middle of some
 162   // computation)
 163   //
 164   // stack:
 165   //        [ arg ] <-- esp
 166   //        [ arg ]
 167   // retaddr in lr
 168 
 169   address entry_point = NULL;
 170   Register continuation = lr;
 171   switch (kind) {
 172   case Interpreter::java_lang_math_abs:
 173     entry_point = __ pc();
 174     __ ldrd(v0, Address(esp));
 175     __ fabsd(v0, v0);
 176     __ mov(sp, r13); // Restore caller's SP
 177     break;
 178   case Interpreter::java_lang_math_sqrt:
 179     entry_point = __ pc();
 180     __ ldrd(v0, Address(esp));
 181     __ fsqrtd(v0, v0);
 182     __ mov(sp, r13);
 183     break;
 184   case Interpreter::java_lang_math_sin :
 185   case Interpreter::java_lang_math_cos :
 186   case Interpreter::java_lang_math_tan :
 187   case Interpreter::java_lang_math_log :
 188   case Interpreter::java_lang_math_log10 :
 189   case Interpreter::java_lang_math_exp :
 190     entry_point = __ pc();
 191     __ ldrd(v0, Address(esp));
 192     __ mov(sp, r13);
 193     __ mov(r19, lr);
 194     continuation = r19;  // The first callee-saved register
 195     generate_transcendental_entry(kind, 1);
 196     break;
 197   case Interpreter::java_lang_math_pow :
 198     entry_point = __ pc();
 199     __ mov(r19, lr);
 200     continuation = r19;
 201     __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
 202     __ ldrd(v1, Address(esp));
 203     __ mov(sp, r13);
 204     generate_transcendental_entry(kind, 2);
 205     break;
 206   case Interpreter::java_lang_math_fmaD :
 207     if (UseFMA) {
 208       entry_point = __ pc();
 209       __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
 210       __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
 211       __ ldrd(v2, Address(esp));
 212       __ fmaddd(v0, v0, v1, v2);
 213       __ mov(sp, r13); // Restore caller's SP
 214     }
 215     break;
 216   case Interpreter::java_lang_math_fmaF :
 217     if (UseFMA) {
 218       entry_point = __ pc();
 219       __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
 220       __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
 221       __ ldrs(v2, Address(esp));
 222       __ fmadds(v0, v0, v1, v2);
 223       __ mov(sp, r13); // Restore caller's SP
 224     }
 225     break;
 226   default:
 227     ;
 228   }
 229   if (entry_point) {
 230     __ br(continuation);
 231   }
 232 
 233   return entry_point;
 234 }
 235 
 236   // double trigonometrics and transcendentals
 237   // static jdouble dsin(jdouble x);
 238   // static jdouble dcos(jdouble x);
 239   // static jdouble dtan(jdouble x);
 240   // static jdouble dlog(jdouble x);
 241   // static jdouble dlog10(jdouble x);
 242   // static jdouble dexp(jdouble x);
 243   // static jdouble dpow(jdouble x, jdouble y);
 244 
 245 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
 246   address fn;
 247   switch (kind) {
 248   case Interpreter::java_lang_math_sin :
 249     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
 250     break;
 251   case Interpreter::java_lang_math_cos :
 252     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
 253     break;
 254   case Interpreter::java_lang_math_tan :
 255     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
 256     break;
 257   case Interpreter::java_lang_math_log :
 258     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
 259     break;
 260   case Interpreter::java_lang_math_log10 :
 261     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
 262     break;
 263   case Interpreter::java_lang_math_exp :
 264     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
 265     break;
 266   case Interpreter::java_lang_math_pow :
 267     fpargs = 2;
 268     fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
 269     break;
 270   default:
 271     ShouldNotReachHere();
 272     fn = NULL;  // unreachable
 273   }
 274   const int gpargs = 0, rtype = 3;
 275   __ mov(rscratch1, fn);
 276   __ blrt(rscratch1, gpargs, fpargs, rtype);
 277 }
 278 
 279 // Abstract method entry
 280 // Attempt to execute abstract method. Throw exception
 281 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 282   // rmethod: Method*
 283   // r13: sender SP
 284 
 285   address entry_point = __ pc();
 286 
 287   // abstract method entry
 288 
 289   //  pop return address, reset last_sp to NULL
 290   __ empty_expression_stack();
 291   __ restore_bcp();      // bcp must be correct for exception handler   (was destroyed)
 292   __ restore_locals();   // make sure locals pointer is correct as well (was destroyed)
 293 
 294   // throw exception
 295   __ call_VM(noreg, CAST_FROM_FN_PTR(address,
 296                              InterpreterRuntime::throw_AbstractMethodError));
 297   // the call_VM checks for exception, so we should never return here.
 298   __ should_not_reach_here();
 299 
 300   return entry_point;
 301 }
 302 
 303 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 304   address entry = __ pc();
 305 
 306 #ifdef ASSERT
 307   {
 308     Label L;
 309     __ ldr(rscratch1, Address(rfp,
 310                        frame::interpreter_frame_monitor_block_top_offset *
 311                        wordSize));
 312     __ mov(rscratch2, sp);
 313     __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
 314                            // grows negative)
 315     __ br(Assembler::HS, L); // check if frame is complete
 316     __ stop ("interpreter frame not set up");
 317     __ bind(L);
 318   }
 319 #endif // ASSERT
 320   // Restore bcp under the assumption that the current frame is still
 321   // interpreted
 322   __ restore_bcp();
 323 
 324   // expression stack must be empty before entering the VM if an
 325   // exception happened
 326   __ empty_expression_stack();
 327   // throw exception
 328   __ call_VM(noreg,
 329              CAST_FROM_FN_PTR(address,
 330                               InterpreterRuntime::throw_StackOverflowError));
 331   return entry;
 332 }
 333 
 334 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
 335         const char* name) {
 336   address entry = __ pc();
 337   // expression stack must be empty before entering the VM if an
 338   // exception happened
 339   __ empty_expression_stack();
 340   // setup parameters
 341   // ??? convention: expect aberrant index in register r1
 342   __ movw(c_rarg2, r1);
 343   __ mov(c_rarg1, (address)name);
 344   __ call_VM(noreg,
 345              CAST_FROM_FN_PTR(address,
 346                               InterpreterRuntime::
 347                               throw_ArrayIndexOutOfBoundsException),
 348              c_rarg1, c_rarg2);
 349   return entry;
 350 }
 351 
 352 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 353   address entry = __ pc();
 354 
 355   // object is at TOS
 356   __ pop(c_rarg1);
 357 
 358   // expression stack must be empty before entering the VM if an
 359   // exception happened
 360   __ empty_expression_stack();
 361 
 362   __ call_VM(noreg,
 363              CAST_FROM_FN_PTR(address,
 364                               InterpreterRuntime::
 365                               throw_ClassCastException),
 366              c_rarg1);
 367   return entry;
 368 }
 369 
 370 address TemplateInterpreterGenerator::generate_exception_handler_common(
 371         const char* name, const char* message, bool pass_oop) {
 372   assert(!pass_oop || message == NULL, "either oop or message but not both");
 373   address entry = __ pc();
 374   if (pass_oop) {
 375     // object is at TOS
 376     __ pop(c_rarg2);
 377   }
 378   // expression stack must be empty before entering the VM if an
 379   // exception happened
 380   __ empty_expression_stack();
 381   // setup parameters
 382   __ lea(c_rarg1, Address((address)name));
 383   if (pass_oop) {
 384     __ call_VM(r0, CAST_FROM_FN_PTR(address,
 385                                     InterpreterRuntime::
 386                                     create_klass_exception),
 387                c_rarg1, c_rarg2);
 388   } else {
 389     // kind of lame ExternalAddress can't take NULL because
 390     // external_word_Relocation will assert.
 391     if (message != NULL) {
 392       __ lea(c_rarg2, Address((address)message));
 393     } else {
 394       __ mov(c_rarg2, NULL_WORD);
 395     }
 396     __ call_VM(r0,
 397                CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
 398                c_rarg1, c_rarg2);
 399   }
 400   // throw exception
 401   __ b(address(Interpreter::throw_exception_entry()));
 402   return entry;
 403 }
 404 
 405 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 406   address entry = __ pc();
 407 
 408   // Restore stack bottom in case i2c adjusted stack
 409   __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 410   // and NULL it as marker that esp is now tos until next java call
 411   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 412   __ restore_bcp();
 413   __ restore_locals();
 414   __ restore_constant_pool_cache();
 415   __ get_method(rmethod);
 416 
 417   if (state == atos) {
 418     Register obj = r0;
 419     Register mdp = r1;
 420     Register tmp = r2;
 421     __ ldr(mdp, Address(rmethod, Method::method_data_offset()));
 422     __ profile_return_type(mdp, obj, tmp);
 423   }
 424 
 425   // Pop N words from the stack
 426   __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
 427   __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 428   __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
 429 
 430   __ add(esp, esp, r1, Assembler::LSL, 3);
 431 
 432   // Restore machine SP
 433   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 434   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 435   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 436   __ ldr(rscratch2,
 437          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
 438   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
 439   __ andr(sp, rscratch1, -16);
 440 
 441 #ifndef PRODUCT
 442   // tell the simulator that the method has been reentered
 443   if (NotifySimulator) {
 444     __ notify(Assembler::method_reentry);
 445   }
 446 #endif
 447 
 448  __ check_and_handle_popframe(rthread);
 449  __ check_and_handle_earlyret(rthread);
 450 
 451   __ get_dispatch();
 452   __ dispatch_next(state, step);
 453 
 454   return entry;
 455 }
 456 
 457 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 458                                                                int step,
 459                                                                address continuation) {
 460   address entry = __ pc();
 461   __ restore_bcp();
 462   __ restore_locals();
 463   __ restore_constant_pool_cache();
 464   __ get_method(rmethod);
 465   __ get_dispatch();
 466 
 467   // Calculate stack limit
 468   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 469   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 470   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 471   __ ldr(rscratch2,
 472          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
 473   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
 474   __ andr(sp, rscratch1, -16);
 475 
 476   // Restore expression stack pointer
 477   __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 478   // NULL last_sp until next java call
 479   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 480 
 481 #if INCLUDE_JVMCI
 482   // Check if we need to take lock at entry of synchronized method.  This can
 483   // only occur on method entry so emit it only for vtos with step 0.
 484   if (EnableJVMCI && state == vtos && step == 0) {
 485     Label L;
 486     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 487     __ cbz(rscratch1, L);
 488     // Clear flag.
 489     __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
 490     // Take lock.
 491     lock_method();
 492     __ bind(L);
 493   } else {
 494 #ifdef ASSERT
 495     if (EnableJVMCI) {
 496       Label L;
 497       __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 498       __ cbz(rscratch1, L);
 499       __ stop("unexpected pending monitor in deopt entry");
 500       __ bind(L);
 501     }
 502 #endif
 503   }
 504 #endif
 505   // handle exceptions
 506   {
 507     Label L;
 508     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 509     __ cbz(rscratch1, L);
 510     __ call_VM(noreg,
 511                CAST_FROM_FN_PTR(address,
 512                                 InterpreterRuntime::throw_pending_exception));
 513     __ should_not_reach_here();
 514     __ bind(L);
 515   }
 516 
 517   if (continuation == NULL) {
 518     __ dispatch_next(state, step);
 519   } else {
 520     __ jump_to_entry(continuation);
 521   }
 522   return entry;
 523 }
 524 
 525 address TemplateInterpreterGenerator::generate_result_handler_for(
 526         BasicType type) {
 527     address entry = __ pc();
 528   switch (type) {
 529   case T_BOOLEAN: __ uxtb(r0, r0);        break;
 530   case T_CHAR   : __ uxth(r0, r0);       break;
 531   case T_BYTE   : __ sxtb(r0, r0);        break;
 532   case T_SHORT  : __ sxth(r0, r0);        break;
 533   case T_INT    : __ uxtw(r0, r0);        break;  // FIXME: We almost certainly don't need this
 534   case T_LONG   : /* nothing to do */        break;
 535   case T_VOID   : /* nothing to do */        break;
 536   case T_FLOAT  : /* nothing to do */        break;
 537   case T_DOUBLE : /* nothing to do */        break;
 538   case T_OBJECT :
 539     // retrieve result from frame
 540     __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
 541     // and verify it
 542     __ verify_oop(r0);
 543     break;
 544   default       : ShouldNotReachHere();
 545   }
 546   __ ret(lr);                                  // return from result handler
 547   return entry;
 548 }
 549 
 550 address TemplateInterpreterGenerator::generate_safept_entry_for(
 551         TosState state,
 552         address runtime_entry) {
 553   address entry = __ pc();
 554   __ push(state);
 555   __ call_VM(noreg, runtime_entry);
 556   __ membar(Assembler::AnyAny);
 557   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 558   return entry;
 559 }
 560 
 561 // Helpers for commoning out cases in the various type of method entries.
 562 //
 563 
 564 
 565 // increment invocation count & check for overflow
 566 //
 567 // Note: checking for negative value instead of overflow
 568 //       so we have a 'sticky' overflow test
 569 //
 570 // rmethod: method
 571 //
 572 void TemplateInterpreterGenerator::generate_counter_incr(
 573         Label* overflow,
 574         Label* profile_method,
 575         Label* profile_method_continue) {
 576   Label done;
 577   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 578   if (TieredCompilation) {
 579     int increment = InvocationCounter::count_increment;
 580     Label no_mdo;
 581     if (ProfileInterpreter) {
 582       // Are we profiling?
 583       __ ldr(r0, Address(rmethod, Method::method_data_offset()));
 584       __ cbz(r0, no_mdo);
 585       // Increment counter in the MDO
 586       const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
 587                                                 in_bytes(InvocationCounter::counter_offset()));
 588       const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
 589       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
 590       __ b(done);
 591     }
 592     __ bind(no_mdo);
 593     // Increment counter in MethodCounters
 594     const Address invocation_counter(rscratch2,
 595                   MethodCounters::invocation_counter_offset() +
 596                   InvocationCounter::counter_offset());
 597     __ get_method_counters(rmethod, rscratch2, done);
 598     const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
 599     __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
 600     __ bind(done);
 601   } else { // not TieredCompilation
 602     const Address backedge_counter(rscratch2,
 603                   MethodCounters::backedge_counter_offset() +
 604                   InvocationCounter::counter_offset());
 605     const Address invocation_counter(rscratch2,
 606                   MethodCounters::invocation_counter_offset() +
 607                   InvocationCounter::counter_offset());
 608 
 609     __ get_method_counters(rmethod, rscratch2, done);
 610 
 611     if (ProfileInterpreter) { // %%% Merge this into MethodData*
 612       __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
 613       __ addw(r1, r1, 1);
 614       __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
 615     }
 616     // Update standard invocation counters
 617     __ ldrw(r1, invocation_counter);
 618     __ ldrw(r0, backedge_counter);
 619 
 620     __ addw(r1, r1, InvocationCounter::count_increment);
 621     __ andw(r0, r0, InvocationCounter::count_mask_value);
 622 
 623     __ strw(r1, invocation_counter);
 624     __ addw(r0, r0, r1);                // add both counters
 625 
 626     // profile_method is non-null only for interpreted method so
 627     // profile_method != NULL == !native_call
 628 
 629     if (ProfileInterpreter && profile_method != NULL) {
 630       // Test to see if we should create a method data oop
 631       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
 632       __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 633       __ cmpw(r0, rscratch2);
 634       __ br(Assembler::LT, *profile_method_continue);
 635 
 636       // if no method data exists, go to profile_method
 637       __ test_method_data_pointer(rscratch2, *profile_method);
 638     }
 639 
 640     {
 641       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
 642       __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 643       __ cmpw(r0, rscratch2);
 644       __ br(Assembler::HS, *overflow);
 645     }
 646     __ bind(done);
 647   }
 648 }
 649 
 650 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 651 
 652   // Asm interpreter on entry
 653   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 654   // Everything as it was on entry
 655 
 656   // InterpreterRuntime::frequency_counter_overflow takes two
 657   // arguments, the first (thread) is passed by call_VM, the second
 658   // indicates if the counter overflow occurs at a backwards branch
 659   // (NULL bcp).  We pass zero for it.  The call returns the address
 660   // of the verified entry point for the method or NULL if the
 661   // compilation did not complete (either went background or bailed
 662   // out).
 663   __ mov(c_rarg1, 0);
 664   __ call_VM(noreg,
 665              CAST_FROM_FN_PTR(address,
 666                               InterpreterRuntime::frequency_counter_overflow),
 667              c_rarg1);
 668 
 669   __ b(do_continue);
 670 }
 671 
 672 // See if we've got enough room on the stack for locals plus overhead
 673 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 674 // without going through the signal handler, i.e., reserved and yellow zones
 675 // will not be made usable. The shadow zone must suffice to handle the
 676 // overflow.
 677 // The expression stack grows down incrementally, so the normal guard
 678 // page mechanism will work for that.
 679 //
 680 // NOTE: Since the additional locals are also always pushed (wasn't
 681 // obvious in generate_method_entry) so the guard should work for them
 682 // too.
 683 //
 684 // Args:
 685 //      r3: number of additional locals this frame needs (what we must check)
 686 //      rmethod: Method*
 687 //
 688 // Kills:
 689 //      r0
 690 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 691 
 692   // monitor entry size: see picture of stack set
 693   // (generate_method_entry) and frame_amd64.hpp
 694   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 695 
 696   // total overhead size: entry_size + (saved rbp through expr stack
 697   // bottom).  be sure to change this if you add/subtract anything
 698   // to/from the overhead area
 699   const int overhead_size =
 700     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 701 
 702   const int page_size = os::vm_page_size();
 703 
 704   Label after_frame_check;
 705 
 706   // see if the frame is greater than one page in size. If so,
 707   // then we need to verify there is enough stack space remaining
 708   // for the additional locals.
 709   //
 710   // Note that we use SUBS rather than CMP here because the immediate
 711   // field of this instruction may overflow.  SUBS can cope with this
 712   // because it is a macro that will expand to some number of MOV
 713   // instructions and a register operation.
 714   __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
 715   __ br(Assembler::LS, after_frame_check);
 716 
 717   // compute rsp as if this were going to be the last frame on
 718   // the stack before the red zone
 719 
 720   // locals + overhead, in bytes
 721   __ mov(r0, overhead_size);
 722   __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize);  // 2 slots per parameter.
 723 
 724   const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
 725   __ ldr(rscratch1, stack_limit);
 726 
 727 #ifdef ASSERT
 728   Label limit_okay;
 729   // Verify that thread stack limit is non-zero.
 730   __ cbnz(rscratch1, limit_okay);
 731   __ stop("stack overflow limit is zero");
 732   __ bind(limit_okay);
 733 #endif
 734 
 735   // Add stack limit to locals.
 736   __ add(r0, r0, rscratch1);
 737 
 738   // Check against the current stack bottom.
 739   __ cmp(sp, r0);
 740   __ br(Assembler::HI, after_frame_check);
 741 
 742   // Remove the incoming args, peeling the machine SP back to where it
 743   // was in the caller.  This is not strictly necessary, but unless we
 744   // do so the stack frame may have a garbage FP; this ensures a
 745   // correct call stack that we can always unwind.  The ANDR should be
 746   // unnecessary because the sender SP in r13 is always aligned, but
 747   // it doesn't hurt.
 748   __ andr(sp, r13, -16);
 749 
 750   // Note: the restored frame is not necessarily interpreted.
 751   // Use the shared runtime version of the StackOverflowError.
 752   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 753   __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
 754 
 755   // all done with frame size check
 756   __ bind(after_frame_check);
 757 }
 758 
 759 // Allocate monitor and lock method (asm interpreter)
 760 //
 761 // Args:
 762 //      rmethod: Method*
 763 //      rlocals: locals
 764 //
 765 // Kills:
 766 //      r0
 767 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 768 //      rscratch1, rscratch2 (scratch regs)
 769 void TemplateInterpreterGenerator::lock_method() {
 770   // synchronize method
 771   const Address access_flags(rmethod, Method::access_flags_offset());
 772   const Address monitor_block_top(
 773         rfp,
 774         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 775   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 776 
 777 #ifdef ASSERT
 778   {
 779     Label L;
 780     __ ldrw(r0, access_flags);
 781     __ tst(r0, JVM_ACC_SYNCHRONIZED);
 782     __ br(Assembler::NE, L);
 783     __ stop("method doesn't need synchronization");
 784     __ bind(L);
 785   }
 786 #endif // ASSERT
 787 
 788   // get synchronization object
 789   {
 790     Label done;
 791     __ ldrw(r0, access_flags);
 792     __ tst(r0, JVM_ACC_STATIC);
 793     // get receiver (assume this is frequent case)
 794     __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 795     __ br(Assembler::EQ, done);
 796     __ load_mirror(r0, rmethod);
 797 
 798 #ifdef ASSERT
 799     {
 800       Label L;
 801       __ cbnz(r0, L);
 802       __ stop("synchronization object is NULL");
 803       __ bind(L);
 804     }
 805 #endif // ASSERT
 806 
 807     __ bind(done);
 808   }
 809 
 810   // add space for monitor & lock
 811   __ sub(sp, sp, entry_size); // add space for a monitor entry
 812   __ sub(esp, esp, entry_size);
 813   __ mov(rscratch1, esp);
 814   __ str(rscratch1, monitor_block_top);  // set new monitor block top
 815   // store object
 816   __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
 817   __ mov(c_rarg1, esp); // object address
 818   __ lock_object(c_rarg1);
 819 }
 820 
 821 // Generate a fixed interpreter frame. This is identical setup for
 822 // interpreted methods and for native methods hence the shared code.
 823 //
 824 // Args:
 825 //      lr: return address
 826 //      rmethod: Method*
 827 //      rlocals: pointer to locals
 828 //      rcpool: cp cache
 829 //      stack_pointer: previous sp
 830 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 831   // initialize fixed part of activation frame
 832   if (native_call) {
 833     __ sub(esp, sp, 14 *  wordSize);
 834     __ mov(rbcp, zr);
 835     __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
 836     // add 2 zero-initialized slots for native calls
 837     __ stp(zr, zr, Address(sp, 12 * wordSize));
 838   } else {
 839     __ sub(esp, sp, 12 *  wordSize);
 840     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));      // get ConstMethod
 841     __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
 842     __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
 843   }
 844 
 845   if (ProfileInterpreter) {
 846     Label method_data_continue;
 847     __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
 848     __ cbz(rscratch1, method_data_continue);
 849     __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
 850     __ bind(method_data_continue);
 851     __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize));  // save Method* and mdp (method data pointer)
 852   } else {
 853     __ stp(zr, rmethod, Address(sp, 6 * wordSize));        // save Method* (no mdp)
 854   }
 855 
 856   // Get mirror and store it in the frame as GC root for this Method*
 857   __ load_mirror(rscratch1, rmethod);
 858   __ stp(rscratch1, zr, Address(sp, 4 * wordSize));
 859 
 860   __ ldr(rcpool, Address(rmethod, Method::const_offset()));
 861   __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
 862   __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
 863   __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
 864 
 865   __ stp(rfp, lr, Address(sp, 10 * wordSize));
 866   __ lea(rfp, Address(sp, 10 * wordSize));
 867 
 868   // set sender sp
 869   // leave last_sp as null
 870   __ stp(zr, r13, Address(sp, 8 * wordSize));
 871 
 872   // Move SP out of the way
 873   if (! native_call) {
 874     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 875     __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 876     __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 877     __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
 878     __ andr(sp, rscratch1, -16);
 879   }
 880 }
 881 
 882 // End of helpers
 883 
 884 // Various method entries
 885 //------------------------------------------------------------------------------------------------------------------------
 886 //
 887 //
 888 
 889 // Method entry for java.lang.ref.Reference.get.
 890 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 891 #if INCLUDE_ALL_GCS
 892   // Code: _aload_0, _getfield, _areturn
 893   // parameter size = 1
 894   //
 895   // The code that gets generated by this routine is split into 2 parts:
 896   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 897   //    2. The slow path - which is an expansion of the regular method entry.
 898   //
 899   // Notes:-
 900   // * In the G1 code we do not check whether we need to block for
 901   //   a safepoint. If G1 is enabled then we must execute the specialized
 902   //   code for Reference.get (except when the Reference object is null)
 903   //   so that we can log the value in the referent field with an SATB
 904   //   update buffer.
 905   //   If the code for the getfield template is modified so that the
 906   //   G1 pre-barrier code is executed when the current method is
 907   //   Reference.get() then going through the normal method entry
 908   //   will be fine.
 909   // * The G1 code can, however, check the receiver object (the instance
 910   //   of java.lang.Reference) and jump to the slow path if null. If the
 911   //   Reference object is null then we obviously cannot fetch the referent
 912   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 913   //   regular method entry code to generate the NPE.
 914   //
 915   // This code is based on generate_accessor_entry.
 916   //
 917   // rmethod: Method*
 918   // r13: senderSP must preserve for slow path, set SP to it on fast path
 919 
 920   address entry = __ pc();
 921 
 922   const int referent_offset = java_lang_ref_Reference::referent_offset;
 923   guarantee(referent_offset > 0, "referent offset not initialized");
 924 
 925   if (UseG1GC) {
 926     Label slow_path;
 927     const Register local_0 = c_rarg0;
 928     // Check if local 0 != NULL
 929     // If the receiver is null then it is OK to jump to the slow path.
 930     __ ldr(local_0, Address(esp, 0));
 931     __ cbz(local_0, slow_path);
 932 
 933     // Load the value of the referent field.
 934     const Address field_address(local_0, referent_offset);
 935     __ load_heap_oop(local_0, field_address);
 936 
 937     __ mov(r19, r13);   // Move senderSP to a callee-saved register
 938     // Generate the G1 pre-barrier code to log the value of
 939     // the referent field in an SATB buffer.
 940     __ enter(); // g1_write may call runtime
 941     __ g1_write_barrier_pre(noreg /* obj */,
 942                             local_0 /* pre_val */,
 943                             rthread /* thread */,
 944                             rscratch2 /* tmp */,
 945                             true /* tosca_live */,
 946                             true /* expand_call */);
 947     __ leave();
 948     // areturn
 949     __ andr(sp, r19, -16);  // done with stack
 950     __ ret(lr);
 951 
 952     // generate a vanilla interpreter entry as the slow path
 953     __ bind(slow_path);
 954     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 955     return entry;
 956   }
 957 #endif // INCLUDE_ALL_GCS
 958 
 959   // If G1 is not enabled then attempt to go through the accessor entry point
 960   // Reference.get is an accessor
 961   return NULL;
 962 }
 963 
 964 /**
 965  * Method entry for static native methods:
 966  *   int java.util.zip.CRC32.update(int crc, int b)
 967  */
 968 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
 969   if (UseCRC32Intrinsics) {
 970     address entry = __ pc();
 971 
 972     // rmethod: Method*
 973     // r13: senderSP must preserved for slow path
 974     // esp: args
 975 
 976     Label slow_path;
 977     // If we need a safepoint check, generate full interpreter entry.
 978     __ safepoint_poll(slow_path);
 979 
 980     // We don't generate local frame and don't align stack because
 981     // we call stub code and there is no safepoint on this path.
 982 
 983     // Load parameters
 984     const Register crc = c_rarg0;  // crc
 985     const Register val = c_rarg1;  // source java byte value
 986     const Register tbl = c_rarg2;  // scratch
 987 
 988     // Arguments are reversed on java expression stack
 989     __ ldrw(val, Address(esp, 0));              // byte value
 990     __ ldrw(crc, Address(esp, wordSize));       // Initial CRC
 991 
 992     unsigned long offset;
 993     __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
 994     __ add(tbl, tbl, offset);
 995 
 996     __ mvnw(crc, crc); // ~crc
 997     __ update_byte_crc32(crc, val, tbl);
 998     __ mvnw(crc, crc); // ~crc
 999 
1000     // result in c_rarg0
1001 
1002     __ andr(sp, r13, -16);
1003     __ ret(lr);
1004 
1005     // generate a vanilla native entry as the slow path
1006     __ bind(slow_path);
1007     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1008     return entry;
1009   }
1010   return NULL;
1011 }
1012 
1013 /**
1014  * Method entry for static native methods:
1015  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1016  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1017  */
1018 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1019   if (UseCRC32Intrinsics) {
1020     address entry = __ pc();
1021 
1022     // rmethod,: Method*
1023     // r13: senderSP must preserved for slow path
1024 
1025     Label slow_path;
1026     // If we need a safepoint check, generate full interpreter entry.
1027     __ safepoint_poll(slow_path);
1028 
1029     // We don't generate local frame and don't align stack because
1030     // we call stub code and there is no safepoint on this path.
1031 
1032     // Load parameters
1033     const Register crc = c_rarg0;  // crc
1034     const Register buf = c_rarg1;  // source java byte array address
1035     const Register len = c_rarg2;  // length
1036     const Register off = len;      // offset (never overlaps with 'len')
1037 
1038     // Arguments are reversed on java expression stack
1039     // Calculate address of start element
1040     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1041       __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1042       __ ldrw(off, Address(esp, wordSize)); // offset
1043       __ add(buf, buf, off); // + offset
1044       __ ldrw(crc,   Address(esp, 4*wordSize)); // Initial CRC
1045     } else {
1046       __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1047       __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1048       __ ldrw(off, Address(esp, wordSize)); // offset
1049       __ add(buf, buf, off); // + offset
1050       __ ldrw(crc,   Address(esp, 3*wordSize)); // Initial CRC
1051     }
1052     // Can now load 'len' since we're finished with 'off'
1053     __ ldrw(len, Address(esp, 0x0)); // Length
1054 
1055     __ andr(sp, r13, -16); // Restore the caller's SP
1056 
1057     // We are frameless so we can just jump to the stub.
1058     __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1059 
1060     // generate a vanilla native entry as the slow path
1061     __ bind(slow_path);
1062     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1063     return entry;
1064   }
1065   return NULL;
1066 }
1067 
1068 /**
1069  * Method entry for intrinsic-candidate (non-native) methods:
1070  *   int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1071  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1072  * Unlike CRC32, CRC32C does not have any methods marked as native
1073  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1074  */
1075 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1076   if (UseCRC32CIntrinsics) {
1077     address entry = __ pc();
1078 
1079     // Prepare jump to stub using parameters from the stack
1080     const Register crc = c_rarg0; // initial crc
1081     const Register buf = c_rarg1; // source java byte array address
1082     const Register len = c_rarg2; // len argument to the kernel
1083 
1084     const Register end = len; // index of last element to process
1085     const Register off = crc; // offset
1086 
1087     __ ldrw(end, Address(esp)); // int end
1088     __ ldrw(off, Address(esp, wordSize)); // int offset
1089     __ sub(len, end, off);
1090     __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1091     __ add(buf, buf, off); // + offset
1092     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1093       __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1094     } else {
1095       __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1096       __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1097     }
1098 
1099     __ andr(sp, r13, -16); // Restore the caller's SP
1100 
1101     // Jump to the stub.
1102     __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1103 
1104     return entry;
1105   }
1106   return NULL;
1107 }
1108 
1109 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1110   // Bang each page in the shadow zone. We can't assume it's been done for
1111   // an interpreter frame with greater than a page of locals, so each page
1112   // needs to be checked.  Only true for non-native.
1113   if (UseStackBanging) {
1114     const int n_shadow_pages = JavaThread::stack_shadow_zone_size() / os::vm_page_size();
1115     const int start_page = native_call ? n_shadow_pages : 1;
1116     const int page_size = os::vm_page_size();
1117     for (int pages = start_page; pages <= n_shadow_pages ; pages++) {
1118       __ sub(rscratch2, sp, pages*page_size);
1119       __ str(zr, Address(rscratch2));
1120     }
1121   }
1122 }
1123 
1124 
1125 // Interpreter stub for calling a native method. (asm interpreter)
1126 // This sets up a somewhat different looking stack for calling the
1127 // native method than the typical interpreter frame setup.
1128 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1129   // determine code generation flags
1130   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1131 
1132   // r1: Method*
1133   // rscratch1: sender sp
1134 
1135   address entry_point = __ pc();
1136 
1137   const Address constMethod       (rmethod, Method::const_offset());
1138   const Address access_flags      (rmethod, Method::access_flags_offset());
1139   const Address size_of_parameters(r2, ConstMethod::
1140                                        size_of_parameters_offset());
1141 
1142   // get parameter size (always needed)
1143   __ ldr(r2, constMethod);
1144   __ load_unsigned_short(r2, size_of_parameters);
1145 
1146   // Native calls don't need the stack size check since they have no
1147   // expression stack and the arguments are already on the stack and
1148   // we only add a handful of words to the stack.
1149 
1150   // rmethod: Method*
1151   // r2: size of parameters
1152   // rscratch1: sender sp
1153 
1154   // for natives the size of locals is zero
1155 
1156   // compute beginning of parameters (rlocals)
1157   __ add(rlocals, esp, r2, ext::uxtx, 3);
1158   __ add(rlocals, rlocals, -wordSize);
1159 
1160   // Pull SP back to minimum size: this avoids holes in the stack
1161   __ andr(sp, esp, -16);
1162 
1163   // initialize fixed part of activation frame
1164   generate_fixed_frame(true);
1165 #ifndef PRODUCT
1166   // tell the simulator that a method has been entered
1167   if (NotifySimulator) {
1168     __ notify(Assembler::method_entry);
1169   }
1170 #endif
1171 
1172   // make sure method is native & not abstract
1173 #ifdef ASSERT
1174   __ ldrw(r0, access_flags);
1175   {
1176     Label L;
1177     __ tst(r0, JVM_ACC_NATIVE);
1178     __ br(Assembler::NE, L);
1179     __ stop("tried to execute non-native method as native");
1180     __ bind(L);
1181   }
1182   {
1183     Label L;
1184     __ tst(r0, JVM_ACC_ABSTRACT);
1185     __ br(Assembler::EQ, L);
1186     __ stop("tried to execute abstract method in interpreter");
1187     __ bind(L);
1188   }
1189 #endif
1190 
1191   // Since at this point in the method invocation the exception
1192   // handler would try to exit the monitor of synchronized methods
1193   // which hasn't been entered yet, we set the thread local variable
1194   // _do_not_unlock_if_synchronized to true. The remove_activation
1195   // will check this flag.
1196 
1197    const Address do_not_unlock_if_synchronized(rthread,
1198         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1199   __ mov(rscratch2, true);
1200   __ strb(rscratch2, do_not_unlock_if_synchronized);
1201 
1202   // increment invocation count & check for overflow
1203   Label invocation_counter_overflow;
1204   if (inc_counter) {
1205     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1206   }
1207 
1208   Label continue_after_compile;
1209   __ bind(continue_after_compile);
1210 
1211   bang_stack_shadow_pages(true);
1212 
1213   // reset the _do_not_unlock_if_synchronized flag
1214   __ strb(zr, do_not_unlock_if_synchronized);
1215 
1216   // check for synchronized methods
1217   // Must happen AFTER invocation_counter check and stack overflow check,
1218   // so method is not locked if overflows.
1219   if (synchronized) {
1220     lock_method();
1221   } else {
1222     // no synchronization necessary
1223 #ifdef ASSERT
1224     {
1225       Label L;
1226       __ ldrw(r0, access_flags);
1227       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1228       __ br(Assembler::EQ, L);
1229       __ stop("method needs synchronization");
1230       __ bind(L);
1231     }
1232 #endif
1233   }
1234 
1235   // start execution
1236 #ifdef ASSERT
1237   {
1238     Label L;
1239     const Address monitor_block_top(rfp,
1240                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1241     __ ldr(rscratch1, monitor_block_top);
1242     __ cmp(esp, rscratch1);
1243     __ br(Assembler::EQ, L);
1244     __ stop("broken stack frame setup in interpreter");
1245     __ bind(L);
1246   }
1247 #endif
1248 
1249   // jvmti support
1250   __ notify_method_entry();
1251 
1252   // work registers
1253   const Register t = r17;
1254   const Register result_handler = r19;
1255 
1256   // allocate space for parameters
1257   __ ldr(t, Address(rmethod, Method::const_offset()));
1258   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1259 
1260   __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1261   __ andr(sp, rscratch1, -16);
1262   __ mov(esp, rscratch1);
1263 
1264   // get signature handler
1265   {
1266     Label L;
1267     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1268     __ cbnz(t, L);
1269     __ call_VM(noreg,
1270                CAST_FROM_FN_PTR(address,
1271                                 InterpreterRuntime::prepare_native_call),
1272                rmethod);
1273     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1274     __ bind(L);
1275   }
1276 
1277   // call signature handler
1278   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1279          "adjust this code");
1280   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1281          "adjust this code");
1282   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1283           "adjust this code");
1284 
1285   // The generated handlers do not touch rmethod (the method).
1286   // However, large signatures cannot be cached and are generated
1287   // each time here.  The slow-path generator can do a GC on return,
1288   // so we must reload it after the call.
1289   __ blr(t);
1290   __ get_method(rmethod);        // slow path can do a GC, reload rmethod
1291 
1292 
1293   // result handler is in r0
1294   // set result handler
1295   __ mov(result_handler, r0);
1296   // pass mirror handle if static call
1297   {
1298     Label L;
1299     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1300     __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1301     // get mirror
1302     __ load_mirror(t, rmethod);
1303     // copy mirror into activation frame
1304     __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1305     // pass handle to mirror
1306     __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1307     __ bind(L);
1308   }
1309 
1310   // get native function entry point in r10
1311   {
1312     Label L;
1313     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1314     address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1315     __ mov(rscratch2, unsatisfied);
1316     __ ldr(rscratch2, rscratch2);
1317     __ cmp(r10, rscratch2);
1318     __ br(Assembler::NE, L);
1319     __ call_VM(noreg,
1320                CAST_FROM_FN_PTR(address,
1321                                 InterpreterRuntime::prepare_native_call),
1322                rmethod);
1323     __ get_method(rmethod);
1324     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1325     __ bind(L);
1326   }
1327 
1328   // pass JNIEnv
1329   __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1330 
1331   // It is enough that the pc() points into the right code
1332   // segment. It does not have to be the correct return pc.
1333   __ set_last_Java_frame(esp, rfp, (address)NULL, rscratch1);
1334 
1335   // change thread state
1336 #ifdef ASSERT
1337   {
1338     Label L;
1339     __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1340     __ cmp(t, _thread_in_Java);
1341     __ br(Assembler::EQ, L);
1342     __ stop("Wrong thread state in native stub");
1343     __ bind(L);
1344   }
1345 #endif
1346 
1347   // Change state to native
1348   __ mov(rscratch1, _thread_in_native);
1349   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1350   __ stlrw(rscratch1, rscratch2);
1351 
1352   // Call the native method.
1353   __ blrt(r10, rscratch1);
1354   __ maybe_isb();
1355   __ get_method(rmethod);
1356   // result potentially in r0 or v0
1357 
1358   // make room for the pushes we're about to do
1359   __ sub(rscratch1, esp, 4 * wordSize);
1360   __ andr(sp, rscratch1, -16);
1361 
1362   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1363   // in order to extract the result of a method call. If the order of these
1364   // pushes change or anything else is added to the stack then the code in
1365   // interpreter_frame_result must also change.
1366   __ push(dtos);
1367   __ push(ltos);
1368 
1369   // change thread state
1370   __ mov(rscratch1, _thread_in_native_trans);
1371   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1372   __ stlrw(rscratch1, rscratch2);
1373 
1374   if (os::is_MP()) {
1375     if (UseMembar) {
1376       // Force this write out before the read below
1377       __ dmb(Assembler::ISH);
1378     } else {
1379       // Write serialization page so VM thread can do a pseudo remote membar.
1380       // We use the current thread pointer to calculate a thread specific
1381       // offset to write to within the page. This minimizes bus traffic
1382       // due to cache line collision.
1383       __ serialize_memory(rthread, rscratch2);
1384     }
1385   }
1386 
1387   // check for safepoint operation in progress and/or pending suspend requests
1388   {
1389     Label L, Continue;
1390     __ safepoint_poll_acquire(L);
1391     __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1392     __ cbz(rscratch2, Continue);
1393     __ bind(L);
1394 
1395     // Don't use call_VM as it will see a possible pending exception
1396     // and forward it and never return here preventing us from
1397     // clearing _last_native_pc down below. So we do a runtime call by
1398     // hand.
1399     //
1400     __ mov(c_rarg0, rthread);
1401     __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1402     __ blrt(rscratch2, 1, 0, 0);
1403     __ maybe_isb();
1404     __ get_method(rmethod);
1405     __ reinit_heapbase();
1406     __ bind(Continue);
1407   }
1408 
1409   // change thread state
1410   __ mov(rscratch1, _thread_in_Java);
1411   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1412   __ stlrw(rscratch1, rscratch2);
1413 
1414   // reset_last_Java_frame
1415   __ reset_last_Java_frame(true);
1416 
1417   if (CheckJNICalls) {
1418     // clear_pending_jni_exception_check
1419     __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1420   }
1421 
1422   // reset handle block
1423   __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1424   __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1425 
1426   // If result is an oop unbox and store it in frame where gc will see it
1427   // and result handler will pick it up
1428 
1429   {
1430     Label no_oop, not_weak, store_result;
1431     __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1432     __ cmp(t, result_handler);
1433     __ br(Assembler::NE, no_oop);
1434     // Unbox oop result, e.g. JNIHandles::resolve result.
1435     __ pop(ltos);
1436     __ cbz(r0, store_result);   // Use NULL as-is.
1437     STATIC_ASSERT(JNIHandles::weak_tag_mask == 1u);
1438     __ tbz(r0, 0, not_weak);    // Test for jweak tag.
1439     // Resolve jweak.
1440     __ ldr(r0, Address(r0, -JNIHandles::weak_tag_value));
1441 #if INCLUDE_ALL_GCS
1442     if (UseG1GC) {
1443       __ enter();                   // Barrier may call runtime.
1444       __ g1_write_barrier_pre(noreg /* obj */,
1445                               r0 /* pre_val */,
1446                               rthread /* thread */,
1447                               t /* tmp */,
1448                               true /* tosca_live */,
1449                               true /* expand_call */);
1450       __ leave();
1451     }
1452 #endif // INCLUDE_ALL_GCS
1453     __ b(store_result);
1454     __ bind(not_weak);
1455     // Resolve (untagged) jobject.
1456     __ ldr(r0, Address(r0, 0));
1457     __ bind(store_result);
1458     __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1459     // keep stack depth as expected by pushing oop which will eventually be discarded
1460     __ push(ltos);
1461     __ bind(no_oop);
1462   }
1463 
1464   {
1465     Label no_reguard;
1466     __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1467     __ ldrw(rscratch1, Address(rscratch1));
1468     __ cmp(rscratch1, JavaThread::stack_guard_yellow_reserved_disabled);
1469     __ br(Assembler::NE, no_reguard);
1470 
1471     __ pusha(); // XXX only save smashed registers
1472     __ mov(c_rarg0, rthread);
1473     __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1474     __ blrt(rscratch2, 0, 0, 0);
1475     __ popa(); // XXX only restore smashed registers
1476     __ bind(no_reguard);
1477   }
1478 
1479   // The method register is junk from after the thread_in_native transition
1480   // until here.  Also can't call_VM until the bcp has been
1481   // restored.  Need bcp for throwing exception below so get it now.
1482   __ get_method(rmethod);
1483 
1484   // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1485   // rbcp == code_base()
1486   __ ldr(rbcp, Address(rmethod, Method::const_offset()));   // get ConstMethod*
1487   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));          // get codebase
1488   // handle exceptions (exception handling will handle unlocking!)
1489   {
1490     Label L;
1491     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1492     __ cbz(rscratch1, L);
1493     // Note: At some point we may want to unify this with the code
1494     // used in call_VM_base(); i.e., we should use the
1495     // StubRoutines::forward_exception code. For now this doesn't work
1496     // here because the rsp is not correctly set at this point.
1497     __ MacroAssembler::call_VM(noreg,
1498                                CAST_FROM_FN_PTR(address,
1499                                InterpreterRuntime::throw_pending_exception));
1500     __ should_not_reach_here();
1501     __ bind(L);
1502   }
1503 
1504   // do unlocking if necessary
1505   {
1506     Label L;
1507     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1508     __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1509     // the code below should be shared with interpreter macro
1510     // assembler implementation
1511     {
1512       Label unlock;
1513       // BasicObjectLock will be first in list, since this is a
1514       // synchronized method. However, need to check that the object
1515       // has not been unlocked by an explicit monitorexit bytecode.
1516 
1517       // monitor expect in c_rarg1 for slow unlock path
1518       __ lea (c_rarg1, Address(rfp,   // address of first monitor
1519                                (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1520                                           wordSize - sizeof(BasicObjectLock))));
1521 
1522       __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1523       __ cbnz(t, unlock);
1524 
1525       // Entry already unlocked, need to throw exception
1526       __ MacroAssembler::call_VM(noreg,
1527                                  CAST_FROM_FN_PTR(address,
1528                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1529       __ should_not_reach_here();
1530 
1531       __ bind(unlock);
1532       __ unlock_object(c_rarg1);
1533     }
1534     __ bind(L);
1535   }
1536 
1537   // jvmti support
1538   // Note: This must happen _after_ handling/throwing any exceptions since
1539   //       the exception handler code notifies the runtime of method exits
1540   //       too. If this happens before, method entry/exit notifications are
1541   //       not properly paired (was bug - gri 11/22/99).
1542   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1543 
1544   // restore potential result in r0:d0, call result handler to
1545   // restore potential result in ST0 & handle result
1546 
1547   __ pop(ltos);
1548   __ pop(dtos);
1549 
1550   __ blr(result_handler);
1551 
1552   // remove activation
1553   __ ldr(esp, Address(rfp,
1554                     frame::interpreter_frame_sender_sp_offset *
1555                     wordSize)); // get sender sp
1556   // remove frame anchor
1557   __ leave();
1558 
1559   // resture sender sp
1560   __ mov(sp, esp);
1561 
1562   __ ret(lr);
1563 
1564   if (inc_counter) {
1565     // Handle overflow of counter and compile method
1566     __ bind(invocation_counter_overflow);
1567     generate_counter_overflow(continue_after_compile);
1568   }
1569 
1570   return entry_point;
1571 }
1572 
1573 //
1574 // Generic interpreted method entry to (asm) interpreter
1575 //
1576 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1577   // determine code generation flags
1578   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1579 
1580   // rscratch1: sender sp
1581   address entry_point = __ pc();
1582 
1583   const Address constMethod(rmethod, Method::const_offset());
1584   const Address access_flags(rmethod, Method::access_flags_offset());
1585   const Address size_of_parameters(r3,
1586                                    ConstMethod::size_of_parameters_offset());
1587   const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1588 
1589   // get parameter size (always needed)
1590   // need to load the const method first
1591   __ ldr(r3, constMethod);
1592   __ load_unsigned_short(r2, size_of_parameters);
1593 
1594   // r2: size of parameters
1595 
1596   __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1597   __ sub(r3, r3, r2); // r3 = no. of additional locals
1598 
1599   // see if we've got enough room on the stack for locals plus overhead.
1600   generate_stack_overflow_check();
1601 
1602   // compute beginning of parameters (rlocals)
1603   __ add(rlocals, esp, r2, ext::uxtx, 3);
1604   __ sub(rlocals, rlocals, wordSize);
1605 
1606   // Make room for locals
1607   __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1608   __ andr(sp, rscratch1, -16);
1609 
1610   // r3 - # of additional locals
1611   // allocate space for locals
1612   // explicitly initialize locals
1613   {
1614     Label exit, loop;
1615     __ ands(zr, r3, r3);
1616     __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1617     __ bind(loop);
1618     __ str(zr, Address(__ post(rscratch1, wordSize)));
1619     __ sub(r3, r3, 1); // until everything initialized
1620     __ cbnz(r3, loop);
1621     __ bind(exit);
1622   }
1623 
1624   // And the base dispatch table
1625   __ get_dispatch();
1626 
1627   // initialize fixed part of activation frame
1628   generate_fixed_frame(false);
1629 #ifndef PRODUCT
1630   // tell the simulator that a method has been entered
1631   if (NotifySimulator) {
1632     __ notify(Assembler::method_entry);
1633   }
1634 #endif
1635   // make sure method is not native & not abstract
1636 #ifdef ASSERT
1637   __ ldrw(r0, access_flags);
1638   {
1639     Label L;
1640     __ tst(r0, JVM_ACC_NATIVE);
1641     __ br(Assembler::EQ, L);
1642     __ stop("tried to execute native method as non-native");
1643     __ bind(L);
1644   }
1645  {
1646     Label L;
1647     __ tst(r0, JVM_ACC_ABSTRACT);
1648     __ br(Assembler::EQ, L);
1649     __ stop("tried to execute abstract method in interpreter");
1650     __ bind(L);
1651   }
1652 #endif
1653 
1654   // Since at this point in the method invocation the exception
1655   // handler would try to exit the monitor of synchronized methods
1656   // which hasn't been entered yet, we set the thread local variable
1657   // _do_not_unlock_if_synchronized to true. The remove_activation
1658   // will check this flag.
1659 
1660    const Address do_not_unlock_if_synchronized(rthread,
1661         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1662   __ mov(rscratch2, true);
1663   __ strb(rscratch2, do_not_unlock_if_synchronized);
1664 
1665   Label no_mdp;
1666   Register mdp = r3;
1667   __ ldr(mdp, Address(rmethod, Method::method_data_offset()));
1668   __ cbz(mdp, no_mdp);
1669   __ add(mdp, mdp, in_bytes(MethodData::data_offset()));
1670   __ profile_parameters_type(mdp, r1, r2);
1671   __ bind(no_mdp);
1672 
1673   // increment invocation count & check for overflow
1674   Label invocation_counter_overflow;
1675   Label profile_method;
1676   Label profile_method_continue;
1677   if (inc_counter) {
1678     generate_counter_incr(&invocation_counter_overflow,
1679                           &profile_method,
1680                           &profile_method_continue);
1681     if (ProfileInterpreter) {
1682       __ bind(profile_method_continue);
1683     }
1684   }
1685 
1686   Label continue_after_compile;
1687   __ bind(continue_after_compile);
1688 
1689   bang_stack_shadow_pages(false);
1690 
1691   // reset the _do_not_unlock_if_synchronized flag
1692   __ strb(zr, do_not_unlock_if_synchronized);
1693 
1694   // check for synchronized methods
1695   // Must happen AFTER invocation_counter check and stack overflow check,
1696   // so method is not locked if overflows.
1697   if (synchronized) {
1698     // Allocate monitor and lock method
1699     lock_method();
1700   } else {
1701     // no synchronization necessary
1702 #ifdef ASSERT
1703     {
1704       Label L;
1705       __ ldrw(r0, access_flags);
1706       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1707       __ br(Assembler::EQ, L);
1708       __ stop("method needs synchronization");
1709       __ bind(L);
1710     }
1711 #endif
1712   }
1713 
1714   // start execution
1715 #ifdef ASSERT
1716   {
1717     Label L;
1718      const Address monitor_block_top (rfp,
1719                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1720     __ ldr(rscratch1, monitor_block_top);
1721     __ cmp(esp, rscratch1);
1722     __ br(Assembler::EQ, L);
1723     __ stop("broken stack frame setup in interpreter");
1724     __ bind(L);
1725   }
1726 #endif
1727 
1728   // jvmti support
1729   __ notify_method_entry();
1730 
1731   __ dispatch_next(vtos);
1732 
1733   // invocation counter overflow
1734   if (inc_counter) {
1735     if (ProfileInterpreter) {
1736       // We have decided to profile this method in the interpreter
1737       __ bind(profile_method);
1738       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1739       __ set_method_data_pointer_for_bcp();
1740       // don't think we need this
1741       __ get_method(r1);
1742       __ b(profile_method_continue);
1743     }
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 //-----------------------------------------------------------------------------
1753 // Exceptions
1754 
1755 void TemplateInterpreterGenerator::generate_throw_exception() {
1756   // Entry point in previous activation (i.e., if the caller was
1757   // interpreted)
1758   Interpreter::_rethrow_exception_entry = __ pc();
1759   // Restore sp to interpreter_frame_last_sp even though we are going
1760   // to empty the expression stack for the exception processing.
1761   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1762   // r0: exception
1763   // r3: return address/pc that threw exception
1764   __ restore_bcp();    // rbcp points to call/send
1765   __ restore_locals();
1766   __ restore_constant_pool_cache();
1767   __ reinit_heapbase();  // restore rheapbase as heapbase.
1768   __ get_dispatch();
1769 
1770 #ifndef PRODUCT
1771   // tell the simulator that the caller method has been reentered
1772   if (NotifySimulator) {
1773     __ get_method(rmethod);
1774     __ notify(Assembler::method_reentry);
1775   }
1776 #endif
1777   // Entry point for exceptions thrown within interpreter code
1778   Interpreter::_throw_exception_entry = __ pc();
1779   // If we came here via a NullPointerException on the receiver of a
1780   // method, rmethod may be corrupt.
1781   __ get_method(rmethod);
1782   // expression stack is undefined here
1783   // r0: exception
1784   // rbcp: exception bcp
1785   __ verify_oop(r0);
1786   __ mov(c_rarg1, r0);
1787 
1788   // expression stack must be empty before entering the VM in case of
1789   // an exception
1790   __ empty_expression_stack();
1791   // find exception handler address and preserve exception oop
1792   __ call_VM(r3,
1793              CAST_FROM_FN_PTR(address,
1794                           InterpreterRuntime::exception_handler_for_exception),
1795              c_rarg1);
1796 
1797   // Calculate stack limit
1798   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1799   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1800   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1801   __ ldr(rscratch2,
1802          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1803   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1804   __ andr(sp, rscratch1, -16);
1805 
1806   // r0: exception handler entry point
1807   // r3: preserved exception oop
1808   // rbcp: bcp for exception handler
1809   __ push_ptr(r3); // push exception which is now the only value on the stack
1810   __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1811 
1812   // If the exception is not handled in the current frame the frame is
1813   // removed and the exception is rethrown (i.e. exception
1814   // continuation is _rethrow_exception).
1815   //
1816   // Note: At this point the bci is still the bxi for the instruction
1817   // which caused the exception and the expression stack is
1818   // empty. Thus, for any VM calls at this point, GC will find a legal
1819   // oop map (with empty expression stack).
1820 
1821   //
1822   // JVMTI PopFrame support
1823   //
1824 
1825   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1826   __ empty_expression_stack();
1827   // Set the popframe_processing bit in pending_popframe_condition
1828   // indicating that we are currently handling popframe, so that
1829   // call_VMs that may happen later do not trigger new popframe
1830   // handling cycles.
1831   __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1832   __ orr(r3, r3, JavaThread::popframe_processing_bit);
1833   __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1834 
1835   {
1836     // Check to see whether we are returning to a deoptimized frame.
1837     // (The PopFrame call ensures that the caller of the popped frame is
1838     // either interpreted or compiled and deoptimizes it if compiled.)
1839     // In this case, we can't call dispatch_next() after the frame is
1840     // popped, but instead must save the incoming arguments and restore
1841     // them after deoptimization has occurred.
1842     //
1843     // Note that we don't compare the return PC against the
1844     // deoptimization blob's unpack entry because of the presence of
1845     // adapter frames in C2.
1846     Label caller_not_deoptimized;
1847     __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1848     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1849                                InterpreterRuntime::interpreter_contains), c_rarg1);
1850     __ cbnz(r0, caller_not_deoptimized);
1851 
1852     // Compute size of arguments for saving when returning to
1853     // deoptimized caller
1854     __ get_method(r0);
1855     __ ldr(r0, Address(r0, Method::const_offset()));
1856     __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1857                                                     size_of_parameters_offset())));
1858     __ lsl(r0, r0, Interpreter::logStackElementSize);
1859     __ restore_locals(); // XXX do we need this?
1860     __ sub(rlocals, rlocals, r0);
1861     __ add(rlocals, rlocals, wordSize);
1862     // Save these arguments
1863     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1864                                            Deoptimization::
1865                                            popframe_preserve_args),
1866                           rthread, r0, rlocals);
1867 
1868     __ remove_activation(vtos,
1869                          /* throw_monitor_exception */ false,
1870                          /* install_monitor_exception */ false,
1871                          /* notify_jvmdi */ false);
1872 
1873     // Inform deoptimization that it is responsible for restoring
1874     // these arguments
1875     __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1876     __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1877 
1878     // Continue in deoptimization handler
1879     __ ret(lr);
1880 
1881     __ bind(caller_not_deoptimized);
1882   }
1883 
1884   __ remove_activation(vtos,
1885                        /* throw_monitor_exception */ false,
1886                        /* install_monitor_exception */ false,
1887                        /* notify_jvmdi */ false);
1888 
1889   // Restore the last_sp and null it out
1890   __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1891   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1892 
1893   __ restore_bcp();
1894   __ restore_locals();
1895   __ restore_constant_pool_cache();
1896   __ get_method(rmethod);
1897 
1898   // The method data pointer was incremented already during
1899   // call profiling. We have to restore the mdp for the current bcp.
1900   if (ProfileInterpreter) {
1901     __ set_method_data_pointer_for_bcp();
1902   }
1903 
1904   // Clear the popframe condition flag
1905   __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1906   assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1907 
1908 #if INCLUDE_JVMTI
1909   {
1910     Label L_done;
1911 
1912     __ ldrb(rscratch1, Address(rbcp, 0));
1913     __ cmpw(r1, Bytecodes::_invokestatic);
1914     __ br(Assembler::EQ, L_done);
1915 
1916     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1917     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1918 
1919     __ ldr(c_rarg0, Address(rlocals, 0));
1920     __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1921 
1922     __ cbz(r0, L_done);
1923 
1924     __ str(r0, Address(esp, 0));
1925     __ bind(L_done);
1926   }
1927 #endif // INCLUDE_JVMTI
1928 
1929   // Restore machine SP
1930   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1931   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1932   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1933   __ ldr(rscratch2,
1934          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1935   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1936   __ andr(sp, rscratch1, -16);
1937 
1938   __ dispatch_next(vtos);
1939   // end of PopFrame support
1940 
1941   Interpreter::_remove_activation_entry = __ pc();
1942 
1943   // preserve exception over this code sequence
1944   __ pop_ptr(r0);
1945   __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1946   // remove the activation (without doing throws on illegalMonitorExceptions)
1947   __ remove_activation(vtos, false, true, false);
1948   // restore exception
1949   // restore exception
1950   __ get_vm_result(r0, rthread);
1951 
1952   // In between activations - previous activation type unknown yet
1953   // compute continuation point - the continuation point expects the
1954   // following registers set up:
1955   //
1956   // r0: exception
1957   // lr: return address/pc that threw exception
1958   // rsp: expression stack of caller
1959   // rfp: fp of caller
1960   // FIXME: There's no point saving LR here because VM calls don't trash it
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   __ push(lr);
2039   __ push(state);
2040   __ push(RegSet::range(r0, r15), sp);
2041   __ mov(c_rarg2, r0);  // Pass itos
2042   __ call_VM(noreg,
2043              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2044              c_rarg1, c_rarg2, c_rarg3);
2045   __ pop(RegSet::range(r0, r15), sp);
2046   __ pop(state);
2047   __ pop(lr);
2048   __ ret(lr);                                   // return from result handler
2049 
2050   return entry;
2051 }
2052 
2053 void TemplateInterpreterGenerator::count_bytecode() {
2054   Register rscratch3 = r0;
2055   __ push(rscratch1);
2056   __ push(rscratch2);
2057   __ push(rscratch3);
2058   __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2059   __ atomic_add(noreg, 1, rscratch3);
2060   __ pop(rscratch3);
2061   __ pop(rscratch2);
2062   __ pop(rscratch1);
2063 }
2064 
2065 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2066 
2067 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2068 
2069 
2070 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2071   // Call a little run-time stub to avoid blow-up for each bytecode.
2072   // The run-time runtime saves the right registers, depending on
2073   // the tosca in-state for the given template.
2074 
2075   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2076          "entry must have been generated");
2077   __ bl(Interpreter::trace_code(t->tos_in()));
2078   __ reinit_heapbase();
2079 }
2080 
2081 
2082 void TemplateInterpreterGenerator::stop_interpreter_at() {
2083   Label L;
2084   __ push(rscratch1);
2085   __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2086   __ ldr(rscratch1, Address(rscratch1));
2087   __ mov(rscratch2, StopInterpreterAt);
2088   __ cmpw(rscratch1, rscratch2);
2089   __ br(Assembler::NE, L);
2090   __ brk(0);
2091   __ bind(L);
2092   __ pop(rscratch1);
2093 }
2094 
2095 #ifdef BUILTIN_SIM
2096 
2097 #include <sys/mman.h>
2098 #include <unistd.h>
2099 
2100 extern "C" {
2101   static int PAGESIZE = getpagesize();
2102   int is_mapped_address(u_int64_t address)
2103   {
2104     address = (address & ~((u_int64_t)PAGESIZE - 1));
2105     if (msync((void *)address, PAGESIZE, MS_ASYNC) == 0) {
2106       return true;
2107     }
2108     if (errno != ENOMEM) {
2109       return true;
2110     }
2111     return false;
2112   }
2113 
2114   void bccheck1(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2115   {
2116     if (method != 0) {
2117       method[0] = '\0';
2118     }
2119     if (bcidx != 0) {
2120       *bcidx = -2;
2121     }
2122     if (decode != 0) {
2123       decode[0] = 0;
2124     }
2125 
2126     if (framesize != 0) {
2127       *framesize = -1;
2128     }
2129 
2130     if (Interpreter::contains((address)pc)) {
2131       AArch64Simulator *sim = AArch64Simulator::get_current(UseSimulatorCache, DisableBCCheck);
2132       Method* meth;
2133       address bcp;
2134       if (fp) {
2135 #define FRAME_SLOT_METHOD 3
2136 #define FRAME_SLOT_BCP 7
2137         meth = (Method*)sim->getMemory()->loadU64(fp - (FRAME_SLOT_METHOD << 3));
2138         bcp = (address)sim->getMemory()->loadU64(fp - (FRAME_SLOT_BCP << 3));
2139 #undef FRAME_SLOT_METHOD
2140 #undef FRAME_SLOT_BCP
2141       } else {
2142         meth = (Method*)sim->getCPUState().xreg(RMETHOD, 0);
2143         bcp = (address)sim->getCPUState().xreg(RBCP, 0);
2144       }
2145       if (meth->is_native()) {
2146         return;
2147       }
2148       if(method && meth->is_method()) {
2149         ResourceMark rm;
2150         method[0] = 'I';
2151         method[1] = ' ';
2152         meth->name_and_sig_as_C_string(method + 2, 398);
2153       }
2154       if (bcidx) {
2155         if (meth->contains(bcp)) {
2156           *bcidx = meth->bci_from(bcp);
2157         } else {
2158           *bcidx = -2;
2159         }
2160       }
2161       if (decode) {
2162         if (!BytecodeTracer::closure()) {
2163           BytecodeTracer::set_closure(BytecodeTracer::std_closure());
2164         }
2165         stringStream str(decode, 400);
2166         BytecodeTracer::trace(meth, bcp, &str);
2167       }
2168     } else {
2169       if (method) {
2170         CodeBlob *cb = CodeCache::find_blob((address)pc);
2171         if (cb != NULL) {
2172           if (cb->is_nmethod()) {
2173             ResourceMark rm;
2174             nmethod* nm = (nmethod*)cb;
2175             method[0] = 'C';
2176             method[1] = ' ';
2177             nm->method()->name_and_sig_as_C_string(method + 2, 398);
2178           } else if (cb->is_adapter_blob()) {
2179             strcpy(method, "B adapter blob");
2180           } else if (cb->is_runtime_stub()) {
2181             strcpy(method, "B runtime stub");
2182           } else if (cb->is_exception_stub()) {
2183             strcpy(method, "B exception stub");
2184           } else if (cb->is_deoptimization_stub()) {
2185             strcpy(method, "B deoptimization stub");
2186           } else if (cb->is_safepoint_stub()) {
2187             strcpy(method, "B safepoint stub");
2188           } else if (cb->is_uncommon_trap_stub()) {
2189             strcpy(method, "B uncommon trap stub");
2190           } else if (cb->contains((address)StubRoutines::call_stub())) {
2191             strcpy(method, "B call stub");
2192           } else {
2193             strcpy(method, "B unknown blob : ");
2194             strcat(method, cb->name());
2195           }
2196           if (framesize != NULL) {
2197             *framesize = cb->frame_size();
2198           }
2199         }
2200       }
2201     }
2202   }
2203 
2204 
2205   JNIEXPORT void bccheck(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2206   {
2207     bccheck1(pc, fp, method, bcidx, framesize, decode);
2208   }
2209 }
2210 
2211 #endif // BUILTIN_SIM
2212 #endif // !PRODUCT