1 /*
   2  * Copyright (c) 2003, 2016, 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   // Pop N words from the stack
 418   __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
 419   __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
 420   __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
 421 
 422   __ add(esp, esp, r1, Assembler::LSL, 3);
 423 
 424   // Restore machine SP
 425   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 426   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 427   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 428   __ ldr(rscratch2,
 429          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
 430   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
 431   __ andr(sp, rscratch1, -16);
 432 
 433 #ifndef PRODUCT
 434   // tell the simulator that the method has been reentered
 435   if (NotifySimulator) {
 436     __ notify(Assembler::method_reentry);
 437   }
 438 #endif
 439   __ get_dispatch();
 440   __ dispatch_next(state, step);
 441 
 442   return entry;
 443 }
 444 
 445 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
 446                                                                int step) {
 447   address entry = __ pc();
 448   __ restore_bcp();
 449   __ restore_locals();
 450   __ restore_constant_pool_cache();
 451   __ get_method(rmethod);
 452   __ get_dispatch();
 453 
 454   // Calculate stack limit
 455   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 456   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 457   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 458   __ ldr(rscratch2,
 459          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
 460   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
 461   __ andr(sp, rscratch1, -16);
 462 
 463   // Restore expression stack pointer
 464   __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 465   // NULL last_sp until next java call
 466   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
 467 
 468 #if INCLUDE_JVMCI
 469   // Check if we need to take lock at entry of synchronized method.  This can
 470   // only occur on method entry so emit it only for vtos with step 0.
 471   if (UseJVMCICompiler && state == vtos && step == 0) {
 472     Label L;
 473     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 474     __ cbz(rscratch1, L);
 475     // Clear flag.
 476     __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
 477     // Take lock.
 478     lock_method();
 479     __ bind(L);
 480   } else {
 481 #ifdef ASSERT
 482     if (UseJVMCICompiler) {
 483       Label L;
 484       __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 485       __ cbz(rscratch1, L);
 486       __ stop("unexpected pending monitor in deopt entry");
 487       __ bind(L);
 488     }
 489 #endif
 490   }
 491 #endif
 492   // handle exceptions
 493   {
 494     Label L;
 495     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 496     __ cbz(rscratch1, L);
 497     __ call_VM(noreg,
 498                CAST_FROM_FN_PTR(address,
 499                                 InterpreterRuntime::throw_pending_exception));
 500     __ should_not_reach_here();
 501     __ bind(L);
 502   }
 503 
 504   __ dispatch_next(state, step);
 505   return entry;
 506 }
 507 
 508 address TemplateInterpreterGenerator::generate_result_handler_for(
 509         BasicType type) {
 510     address entry = __ pc();
 511   switch (type) {
 512   case T_BOOLEAN: __ uxtb(r0, r0);        break;
 513   case T_CHAR   : __ uxth(r0, r0);       break;
 514   case T_BYTE   : __ sxtb(r0, r0);        break;
 515   case T_SHORT  : __ sxth(r0, r0);        break;
 516   case T_INT    : __ uxtw(r0, r0);        break;  // FIXME: We almost certainly don't need this
 517   case T_LONG   : /* nothing to do */        break;
 518   case T_VOID   : /* nothing to do */        break;
 519   case T_FLOAT  : /* nothing to do */        break;
 520   case T_DOUBLE : /* nothing to do */        break;
 521   case T_OBJECT :
 522     // retrieve result from frame
 523     __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
 524     // and verify it
 525     __ verify_oop(r0);
 526     break;
 527   default       : ShouldNotReachHere();
 528   }
 529   __ ret(lr);                                  // return from result handler
 530   return entry;
 531 }
 532 
 533 address TemplateInterpreterGenerator::generate_safept_entry_for(
 534         TosState state,
 535         address runtime_entry) {
 536   address entry = __ pc();
 537   __ push(state);
 538   __ call_VM(noreg, runtime_entry);
 539   __ membar(Assembler::AnyAny);
 540   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
 541   return entry;
 542 }
 543 
 544 // Helpers for commoning out cases in the various type of method entries.
 545 //
 546 
 547 
 548 // increment invocation count & check for overflow
 549 //
 550 // Note: checking for negative value instead of overflow
 551 //       so we have a 'sticky' overflow test
 552 //
 553 // rmethod: method
 554 //
 555 void TemplateInterpreterGenerator::generate_counter_incr(
 556         Label* overflow,
 557         Label* profile_method,
 558         Label* profile_method_continue) {
 559   Label done;
 560   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
 561   if (TieredCompilation) {
 562     int increment = InvocationCounter::count_increment;
 563     Label no_mdo;
 564     if (ProfileInterpreter) {
 565       // Are we profiling?
 566       __ ldr(r0, Address(rmethod, Method::method_data_offset()));
 567       __ cbz(r0, no_mdo);
 568       // Increment counter in the MDO
 569       const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
 570                                                 in_bytes(InvocationCounter::counter_offset()));
 571       const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
 572       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
 573       __ b(done);
 574     }
 575     __ bind(no_mdo);
 576     // Increment counter in MethodCounters
 577     const Address invocation_counter(rscratch2,
 578                   MethodCounters::invocation_counter_offset() +
 579                   InvocationCounter::counter_offset());
 580     __ get_method_counters(rmethod, rscratch2, done);
 581     const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
 582     __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
 583     __ bind(done);
 584   } else { // not TieredCompilation
 585     const Address backedge_counter(rscratch2,
 586                   MethodCounters::backedge_counter_offset() +
 587                   InvocationCounter::counter_offset());
 588     const Address invocation_counter(rscratch2,
 589                   MethodCounters::invocation_counter_offset() +
 590                   InvocationCounter::counter_offset());
 591 
 592     __ get_method_counters(rmethod, rscratch2, done);
 593 
 594     if (ProfileInterpreter) { // %%% Merge this into MethodData*
 595       __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
 596       __ addw(r1, r1, 1);
 597       __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
 598     }
 599     // Update standard invocation counters
 600     __ ldrw(r1, invocation_counter);
 601     __ ldrw(r0, backedge_counter);
 602 
 603     __ addw(r1, r1, InvocationCounter::count_increment);
 604     __ andw(r0, r0, InvocationCounter::count_mask_value);
 605 
 606     __ strw(r1, invocation_counter);
 607     __ addw(r0, r0, r1);                // add both counters
 608 
 609     // profile_method is non-null only for interpreted method so
 610     // profile_method != NULL == !native_call
 611 
 612     if (ProfileInterpreter && profile_method != NULL) {
 613       // Test to see if we should create a method data oop
 614       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
 615       __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
 616       __ cmpw(r0, rscratch2);
 617       __ br(Assembler::LT, *profile_method_continue);
 618 
 619       // if no method data exists, go to profile_method
 620       __ test_method_data_pointer(rscratch2, *profile_method);
 621     }
 622 
 623     {
 624       __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
 625       __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
 626       __ cmpw(r0, rscratch2);
 627       __ br(Assembler::HS, *overflow);
 628     }
 629     __ bind(done);
 630   }
 631 }
 632 
 633 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
 634 
 635   // Asm interpreter on entry
 636   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
 637   // Everything as it was on entry
 638 
 639   // InterpreterRuntime::frequency_counter_overflow takes two
 640   // arguments, the first (thread) is passed by call_VM, the second
 641   // indicates if the counter overflow occurs at a backwards branch
 642   // (NULL bcp).  We pass zero for it.  The call returns the address
 643   // of the verified entry point for the method or NULL if the
 644   // compilation did not complete (either went background or bailed
 645   // out).
 646   __ mov(c_rarg1, 0);
 647   __ call_VM(noreg,
 648              CAST_FROM_FN_PTR(address,
 649                               InterpreterRuntime::frequency_counter_overflow),
 650              c_rarg1);
 651 
 652   __ b(do_continue);
 653 }
 654 
 655 // See if we've got enough room on the stack for locals plus overhead
 656 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 657 // without going through the signal handler, i.e., reserved and yellow zones
 658 // will not be made usable. The shadow zone must suffice to handle the
 659 // overflow.
 660 // The expression stack grows down incrementally, so the normal guard
 661 // page mechanism will work for that.
 662 //
 663 // NOTE: Since the additional locals are also always pushed (wasn't
 664 // obvious in generate_method_entry) so the guard should work for them
 665 // too.
 666 //
 667 // Args:
 668 //      r3: number of additional locals this frame needs (what we must check)
 669 //      rmethod: Method*
 670 //
 671 // Kills:
 672 //      r0
 673 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
 674 
 675   // monitor entry size: see picture of stack set
 676   // (generate_method_entry) and frame_amd64.hpp
 677   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 678 
 679   // total overhead size: entry_size + (saved rbp through expr stack
 680   // bottom).  be sure to change this if you add/subtract anything
 681   // to/from the overhead area
 682   const int overhead_size =
 683     -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
 684 
 685   const int page_size = os::vm_page_size();
 686 
 687   Label after_frame_check;
 688 
 689   // see if the frame is greater than one page in size. If so,
 690   // then we need to verify there is enough stack space remaining
 691   // for the additional locals.
 692   //
 693   // Note that we use SUBS rather than CMP here because the immediate
 694   // field of this instruction may overflow.  SUBS can cope with this
 695   // because it is a macro that will expand to some number of MOV
 696   // instructions and a register operation.
 697   __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
 698   __ br(Assembler::LS, after_frame_check);
 699 
 700   // compute rsp as if this were going to be the last frame on
 701   // the stack before the red zone
 702 
 703   // locals + overhead, in bytes
 704   __ mov(r0, overhead_size);
 705   __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize);  // 2 slots per parameter.
 706 
 707   const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
 708   __ ldr(rscratch1, stack_limit);
 709 
 710 #ifdef ASSERT
 711   Label limit_okay;
 712   // Verify that thread stack limit is non-zero.
 713   __ cbnz(rscratch1, limit_okay);
 714   __ stop("stack overflow limit is zero");
 715   __ bind(limit_okay);
 716 #endif
 717 
 718   // Add stack limit to locals.
 719   __ add(r0, r0, rscratch1);
 720 
 721   // Check against the current stack bottom.
 722   __ cmp(sp, r0);
 723   __ br(Assembler::HI, after_frame_check);
 724 
 725   // Remove the incoming args, peeling the machine SP back to where it
 726   // was in the caller.  This is not strictly necessary, but unless we
 727   // do so the stack frame may have a garbage FP; this ensures a
 728   // correct call stack that we can always unwind.  The ANDR should be
 729   // unnecessary because the sender SP in r13 is always aligned, but
 730   // it doesn't hurt.
 731   __ andr(sp, r13, -16);
 732 
 733   // Note: the restored frame is not necessarily interpreted.
 734   // Use the shared runtime version of the StackOverflowError.
 735   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 736   __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
 737 
 738   // all done with frame size check
 739   __ bind(after_frame_check);
 740 }
 741 
 742 // Allocate monitor and lock method (asm interpreter)
 743 //
 744 // Args:
 745 //      rmethod: Method*
 746 //      rlocals: locals
 747 //
 748 // Kills:
 749 //      r0
 750 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
 751 //      rscratch1, rscratch2 (scratch regs)
 752 void TemplateInterpreterGenerator::lock_method() {
 753   // synchronize method
 754   const Address access_flags(rmethod, Method::access_flags_offset());
 755   const Address monitor_block_top(
 756         rfp,
 757         frame::interpreter_frame_monitor_block_top_offset * wordSize);
 758   const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
 759 
 760 #ifdef ASSERT
 761   {
 762     Label L;
 763     __ ldrw(r0, access_flags);
 764     __ tst(r0, JVM_ACC_SYNCHRONIZED);
 765     __ br(Assembler::NE, L);
 766     __ stop("method doesn't need synchronization");
 767     __ bind(L);
 768   }
 769 #endif // ASSERT
 770 
 771   // get synchronization object
 772   {
 773     Label done;
 774     __ ldrw(r0, access_flags);
 775     __ tst(r0, JVM_ACC_STATIC);
 776     // get receiver (assume this is frequent case)
 777     __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
 778     __ br(Assembler::EQ, done);
 779     __ load_mirror(r0, rmethod);
 780 
 781 #ifdef ASSERT
 782     {
 783       Label L;
 784       __ cbnz(r0, L);
 785       __ stop("synchronization object is NULL");
 786       __ bind(L);
 787     }
 788 #endif // ASSERT
 789 
 790     __ bind(done);
 791   }
 792 
 793   // add space for monitor & lock
 794   __ sub(sp, sp, entry_size); // add space for a monitor entry
 795   __ sub(esp, esp, entry_size);
 796   __ mov(rscratch1, esp);
 797   __ str(rscratch1, monitor_block_top);  // set new monitor block top
 798   // store object
 799   __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
 800   __ mov(c_rarg1, esp); // object address
 801   __ lock_object(c_rarg1);
 802 }
 803 
 804 // Generate a fixed interpreter frame. This is identical setup for
 805 // interpreted methods and for native methods hence the shared code.
 806 //
 807 // Args:
 808 //      lr: return address
 809 //      rmethod: Method*
 810 //      rlocals: pointer to locals
 811 //      rcpool: cp cache
 812 //      stack_pointer: previous sp
 813 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 814   // initialize fixed part of activation frame
 815   if (native_call) {
 816     __ sub(esp, sp, 14 *  wordSize);
 817     __ mov(rbcp, zr);
 818     __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
 819     // add 2 zero-initialized slots for native calls
 820     __ stp(zr, zr, Address(sp, 12 * wordSize));
 821   } else {
 822     __ sub(esp, sp, 12 *  wordSize);
 823     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));      // get ConstMethod
 824     __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
 825     __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
 826   }
 827 
 828   if (ProfileInterpreter) {
 829     Label method_data_continue;
 830     __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
 831     __ cbz(rscratch1, method_data_continue);
 832     __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
 833     __ bind(method_data_continue);
 834     __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize));  // save Method* and mdp (method data pointer)
 835   } else {
 836     __ stp(zr, rmethod, Address(sp, 6 * wordSize));        // save Method* (no mdp)
 837   }
 838 
 839   // Get mirror and store it in the frame as GC root for this Method*
 840   __ load_mirror(rscratch1, rmethod);
 841   __ stp(rscratch1, zr, Address(sp, 4 * wordSize));
 842 
 843   __ ldr(rcpool, Address(rmethod, Method::const_offset()));
 844   __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
 845   __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
 846   __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
 847 
 848   __ stp(rfp, lr, Address(sp, 10 * wordSize));
 849   __ lea(rfp, Address(sp, 10 * wordSize));
 850 
 851   // set sender sp
 852   // leave last_sp as null
 853   __ stp(zr, r13, Address(sp, 8 * wordSize));
 854 
 855   // Move SP out of the way
 856   if (! native_call) {
 857     __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
 858     __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
 859     __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
 860     __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
 861     __ andr(sp, rscratch1, -16);
 862   }
 863 }
 864 
 865 // End of helpers
 866 
 867 // Various method entries
 868 //------------------------------------------------------------------------------------------------------------------------
 869 //
 870 //
 871 
 872 // Method entry for java.lang.ref.Reference.get.
 873 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 874 #if INCLUDE_ALL_GCS
 875   // Code: _aload_0, _getfield, _areturn
 876   // parameter size = 1
 877   //
 878   // The code that gets generated by this routine is split into 2 parts:
 879   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 880   //    2. The slow path - which is an expansion of the regular method entry.
 881   //
 882   // Notes:-
 883   // * In the G1 code we do not check whether we need to block for
 884   //   a safepoint. If G1 is enabled then we must execute the specialized
 885   //   code for Reference.get (except when the Reference object is null)
 886   //   so that we can log the value in the referent field with an SATB
 887   //   update buffer.
 888   //   If the code for the getfield template is modified so that the
 889   //   G1 pre-barrier code is executed when the current method is
 890   //   Reference.get() then going through the normal method entry
 891   //   will be fine.
 892   // * The G1 code can, however, check the receiver object (the instance
 893   //   of java.lang.Reference) and jump to the slow path if null. If the
 894   //   Reference object is null then we obviously cannot fetch the referent
 895   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 896   //   regular method entry code to generate the NPE.
 897   //
 898   // This code is based on generate_accessor_entry.
 899   //
 900   // rmethod: Method*
 901   // r13: senderSP must preserve for slow path, set SP to it on fast path
 902 
 903   address entry = __ pc();
 904 
 905   const int referent_offset = java_lang_ref_Reference::referent_offset;
 906   guarantee(referent_offset > 0, "referent offset not initialized");
 907 
 908   if (UseG1GC) {
 909     Label slow_path;
 910     const Register local_0 = c_rarg0;
 911     // Check if local 0 != NULL
 912     // If the receiver is null then it is OK to jump to the slow path.
 913     __ ldr(local_0, Address(esp, 0));
 914     __ cbz(local_0, slow_path);
 915 
 916     // Load the value of the referent field.
 917     const Address field_address(local_0, referent_offset);
 918     __ load_heap_oop(local_0, field_address);
 919 
 920     __ mov(r19, r13);   // Move senderSP to a callee-saved register
 921     // Generate the G1 pre-barrier code to log the value of
 922     // the referent field in an SATB buffer.
 923     __ enter(); // g1_write may call runtime
 924     __ g1_write_barrier_pre(noreg /* obj */,
 925                             local_0 /* pre_val */,
 926                             rthread /* thread */,
 927                             rscratch2 /* tmp */,
 928                             true /* tosca_live */,
 929                             true /* expand_call */);
 930     __ leave();
 931     // areturn
 932     __ andr(sp, r19, -16);  // done with stack
 933     __ ret(lr);
 934 
 935     // generate a vanilla interpreter entry as the slow path
 936     __ bind(slow_path);
 937     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 938     return entry;
 939   }
 940 #endif // INCLUDE_ALL_GCS
 941 
 942   // If G1 is not enabled then attempt to go through the accessor entry point
 943   // Reference.get is an accessor
 944   return NULL;
 945 }
 946 
 947 /**
 948  * Method entry for static native methods:
 949  *   int java.util.zip.CRC32.update(int crc, int b)
 950  */
 951 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
 952   if (UseCRC32Intrinsics) {
 953     address entry = __ pc();
 954 
 955     // rmethod: Method*
 956     // r13: senderSP must preserved for slow path
 957     // esp: args
 958 
 959     Label slow_path;
 960     // If we need a safepoint check, generate full interpreter entry.
 961     ExternalAddress state(SafepointSynchronize::address_of_state());
 962     unsigned long offset;
 963     __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
 964     __ ldrw(rscratch1, Address(rscratch1, offset));
 965     assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
 966     __ cbnz(rscratch1, slow_path);
 967 
 968     // We don't generate local frame and don't align stack because
 969     // we call stub code and there is no safepoint on this path.
 970 
 971     // Load parameters
 972     const Register crc = c_rarg0;  // crc
 973     const Register val = c_rarg1;  // source java byte value
 974     const Register tbl = c_rarg2;  // scratch
 975 
 976     // Arguments are reversed on java expression stack
 977     __ ldrw(val, Address(esp, 0));              // byte value
 978     __ ldrw(crc, Address(esp, wordSize));       // Initial CRC
 979 
 980     __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
 981     __ add(tbl, tbl, offset);
 982 
 983     __ ornw(crc, zr, crc); // ~crc
 984     __ update_byte_crc32(crc, val, tbl);
 985     __ ornw(crc, zr, crc); // ~crc
 986 
 987     // result in c_rarg0
 988 
 989     __ andr(sp, r13, -16);
 990     __ ret(lr);
 991 
 992     // generate a vanilla native entry as the slow path
 993     __ bind(slow_path);
 994     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
 995     return entry;
 996   }
 997   return NULL;
 998 }
 999 
1000 /**
1001  * Method entry for static native methods:
1002  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1003  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1004  */
1005 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1006   if (UseCRC32Intrinsics) {
1007     address entry = __ pc();
1008 
1009     // rmethod,: Method*
1010     // r13: senderSP must preserved for slow path
1011 
1012     Label slow_path;
1013     // If we need a safepoint check, generate full interpreter entry.
1014     ExternalAddress state(SafepointSynchronize::address_of_state());
1015     unsigned long offset;
1016     __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
1017     __ ldrw(rscratch1, Address(rscratch1, offset));
1018     assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
1019     __ cbnz(rscratch1, slow_path);
1020 
1021     // We don't generate local frame and don't align stack because
1022     // we call stub code and there is no safepoint on this path.
1023 
1024     // Load parameters
1025     const Register crc = c_rarg0;  // crc
1026     const Register buf = c_rarg1;  // source java byte array address
1027     const Register len = c_rarg2;  // length
1028     const Register off = len;      // offset (never overlaps with 'len')
1029 
1030     // Arguments are reversed on java expression stack
1031     // Calculate address of start element
1032     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1033       __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1034       __ ldrw(off, Address(esp, wordSize)); // offset
1035       __ add(buf, buf, off); // + offset
1036       __ ldrw(crc,   Address(esp, 4*wordSize)); // Initial CRC
1037     } else {
1038       __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1039       __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1040       __ ldrw(off, Address(esp, wordSize)); // offset
1041       __ add(buf, buf, off); // + offset
1042       __ ldrw(crc,   Address(esp, 3*wordSize)); // Initial CRC
1043     }
1044     // Can now load 'len' since we're finished with 'off'
1045     __ ldrw(len, Address(esp, 0x0)); // Length
1046 
1047     __ andr(sp, r13, -16); // Restore the caller's SP
1048 
1049     // We are frameless so we can just jump to the stub.
1050     __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1051 
1052     // generate a vanilla native entry as the slow path
1053     __ bind(slow_path);
1054     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1055     return entry;
1056   }
1057   return NULL;
1058 }
1059 
1060 // Not supported
1061 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1062   return NULL;
1063 }
1064 
1065 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1066   // Bang each page in the shadow zone. We can't assume it's been done for
1067   // an interpreter frame with greater than a page of locals, so each page
1068   // needs to be checked.  Only true for non-native.
1069   if (UseStackBanging) {
1070     const int n_shadow_pages = JavaThread::stack_shadow_zone_size() / os::vm_page_size();
1071     const int start_page = native_call ? n_shadow_pages : 1;
1072     const int page_size = os::vm_page_size();
1073     for (int pages = start_page; pages <= n_shadow_pages ; pages++) {
1074       __ sub(rscratch2, sp, pages*page_size);
1075       __ str(zr, Address(rscratch2));
1076     }
1077   }
1078 }
1079 
1080 
1081 // Interpreter stub for calling a native method. (asm interpreter)
1082 // This sets up a somewhat different looking stack for calling the
1083 // native method than the typical interpreter frame setup.
1084 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1085   // determine code generation flags
1086   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1087 
1088   // r1: Method*
1089   // rscratch1: sender sp
1090 
1091   address entry_point = __ pc();
1092 
1093   const Address constMethod       (rmethod, Method::const_offset());
1094   const Address access_flags      (rmethod, Method::access_flags_offset());
1095   const Address size_of_parameters(r2, ConstMethod::
1096                                        size_of_parameters_offset());
1097 
1098   // get parameter size (always needed)
1099   __ ldr(r2, constMethod);
1100   __ load_unsigned_short(r2, size_of_parameters);
1101 
1102   // Native calls don't need the stack size check since they have no
1103   // expression stack and the arguments are already on the stack and
1104   // we only add a handful of words to the stack.
1105 
1106   // rmethod: Method*
1107   // r2: size of parameters
1108   // rscratch1: sender sp
1109 
1110   // for natives the size of locals is zero
1111 
1112   // compute beginning of parameters (rlocals)
1113   __ add(rlocals, esp, r2, ext::uxtx, 3);
1114   __ add(rlocals, rlocals, -wordSize);
1115 
1116   // Pull SP back to minimum size: this avoids holes in the stack
1117   __ andr(sp, esp, -16);
1118 
1119   // initialize fixed part of activation frame
1120   generate_fixed_frame(true);
1121 #ifndef PRODUCT
1122   // tell the simulator that a method has been entered
1123   if (NotifySimulator) {
1124     __ notify(Assembler::method_entry);
1125   }
1126 #endif
1127 
1128   // make sure method is native & not abstract
1129 #ifdef ASSERT
1130   __ ldrw(r0, access_flags);
1131   {
1132     Label L;
1133     __ tst(r0, JVM_ACC_NATIVE);
1134     __ br(Assembler::NE, L);
1135     __ stop("tried to execute non-native method as native");
1136     __ bind(L);
1137   }
1138   {
1139     Label L;
1140     __ tst(r0, JVM_ACC_ABSTRACT);
1141     __ br(Assembler::EQ, L);
1142     __ stop("tried to execute abstract method in interpreter");
1143     __ bind(L);
1144   }
1145 #endif
1146 
1147   // Since at this point in the method invocation the exception
1148   // handler would try to exit the monitor of synchronized methods
1149   // which hasn't been entered yet, we set the thread local variable
1150   // _do_not_unlock_if_synchronized to true. The remove_activation
1151   // will check this flag.
1152 
1153    const Address do_not_unlock_if_synchronized(rthread,
1154         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1155   __ mov(rscratch2, true);
1156   __ strb(rscratch2, do_not_unlock_if_synchronized);
1157 
1158   // increment invocation count & check for overflow
1159   Label invocation_counter_overflow;
1160   if (inc_counter) {
1161     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1162   }
1163 
1164   Label continue_after_compile;
1165   __ bind(continue_after_compile);
1166 
1167   bang_stack_shadow_pages(true);
1168 
1169   // reset the _do_not_unlock_if_synchronized flag
1170   __ strb(zr, do_not_unlock_if_synchronized);
1171 
1172   // check for synchronized methods
1173   // Must happen AFTER invocation_counter check and stack overflow check,
1174   // so method is not locked if overflows.
1175   if (synchronized) {
1176     lock_method();
1177   } else {
1178     // no synchronization necessary
1179 #ifdef ASSERT
1180     {
1181       Label L;
1182       __ ldrw(r0, access_flags);
1183       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1184       __ br(Assembler::EQ, L);
1185       __ stop("method needs synchronization");
1186       __ bind(L);
1187     }
1188 #endif
1189   }
1190 
1191   // start execution
1192 #ifdef ASSERT
1193   {
1194     Label L;
1195     const Address monitor_block_top(rfp,
1196                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1197     __ ldr(rscratch1, monitor_block_top);
1198     __ cmp(esp, rscratch1);
1199     __ br(Assembler::EQ, L);
1200     __ stop("broken stack frame setup in interpreter");
1201     __ bind(L);
1202   }
1203 #endif
1204 
1205   // jvmti support
1206   __ notify_method_entry();
1207 
1208   // work registers
1209   const Register t = r17;
1210   const Register result_handler = r19;
1211 
1212   // allocate space for parameters
1213   __ ldr(t, Address(rmethod, Method::const_offset()));
1214   __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1215 
1216   __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1217   __ andr(sp, rscratch1, -16);
1218   __ mov(esp, rscratch1);
1219 
1220   // get signature handler
1221   {
1222     Label L;
1223     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1224     __ cbnz(t, L);
1225     __ call_VM(noreg,
1226                CAST_FROM_FN_PTR(address,
1227                                 InterpreterRuntime::prepare_native_call),
1228                rmethod);
1229     __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1230     __ bind(L);
1231   }
1232 
1233   // call signature handler
1234   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1235          "adjust this code");
1236   assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1237          "adjust this code");
1238   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1239           "adjust this code");
1240 
1241   // The generated handlers do not touch rmethod (the method).
1242   // However, large signatures cannot be cached and are generated
1243   // each time here.  The slow-path generator can do a GC on return,
1244   // so we must reload it after the call.
1245   __ blr(t);
1246   __ get_method(rmethod);        // slow path can do a GC, reload rmethod
1247 
1248 
1249   // result handler is in r0
1250   // set result handler
1251   __ mov(result_handler, r0);
1252   // pass mirror handle if static call
1253   {
1254     Label L;
1255     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1256     __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1257     // get mirror
1258     __ load_mirror(t, rmethod);
1259     // copy mirror into activation frame
1260     __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1261     // pass handle to mirror
1262     __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1263     __ bind(L);
1264   }
1265 
1266   // get native function entry point in r10
1267   {
1268     Label L;
1269     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1270     address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1271     __ mov(rscratch2, unsatisfied);
1272     __ ldr(rscratch2, rscratch2);
1273     __ cmp(r10, rscratch2);
1274     __ br(Assembler::NE, L);
1275     __ call_VM(noreg,
1276                CAST_FROM_FN_PTR(address,
1277                                 InterpreterRuntime::prepare_native_call),
1278                rmethod);
1279     __ get_method(rmethod);
1280     __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1281     __ bind(L);
1282   }
1283 
1284   // pass JNIEnv
1285   __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1286 
1287   // It is enough that the pc() points into the right code
1288   // segment. It does not have to be the correct return pc.
1289   __ set_last_Java_frame(esp, rfp, (address)NULL, rscratch1);
1290 
1291   // change thread state
1292 #ifdef ASSERT
1293   {
1294     Label L;
1295     __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1296     __ cmp(t, _thread_in_Java);
1297     __ br(Assembler::EQ, L);
1298     __ stop("Wrong thread state in native stub");
1299     __ bind(L);
1300   }
1301 #endif
1302 
1303   // Change state to native
1304   __ mov(rscratch1, _thread_in_native);
1305   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1306   __ stlrw(rscratch1, rscratch2);
1307 
1308   // Call the native method.
1309   __ blrt(r10, rscratch1);
1310   __ maybe_isb();
1311   __ get_method(rmethod);
1312   // result potentially in r0 or v0
1313 
1314   // make room for the pushes we're about to do
1315   __ sub(rscratch1, esp, 4 * wordSize);
1316   __ andr(sp, rscratch1, -16);
1317 
1318   // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1319   // in order to extract the result of a method call. If the order of these
1320   // pushes change or anything else is added to the stack then the code in
1321   // interpreter_frame_result must also change.
1322   __ push(dtos);
1323   __ push(ltos);
1324 
1325   // change thread state
1326   __ mov(rscratch1, _thread_in_native_trans);
1327   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1328   __ stlrw(rscratch1, rscratch2);
1329 
1330   if (os::is_MP()) {
1331     if (UseMembar) {
1332       // Force this write out before the read below
1333       __ dsb(Assembler::SY);
1334     } else {
1335       // Write serialization page so VM thread can do a pseudo remote membar.
1336       // We use the current thread pointer to calculate a thread specific
1337       // offset to write to within the page. This minimizes bus traffic
1338       // due to cache line collision.
1339       __ serialize_memory(rthread, rscratch2);
1340     }
1341   }
1342 
1343   // check for safepoint operation in progress and/or pending suspend requests
1344   {
1345     Label Continue;
1346     {
1347       unsigned long offset;
1348       __ adrp(rscratch2, SafepointSynchronize::address_of_state(), offset);
1349       __ ldrw(rscratch2, Address(rscratch2, offset));
1350     }
1351     assert(SafepointSynchronize::_not_synchronized == 0,
1352            "SafepointSynchronize::_not_synchronized");
1353     Label L;
1354     __ cbnz(rscratch2, L);
1355     __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1356     __ cbz(rscratch2, Continue);
1357     __ bind(L);
1358 
1359     // Don't use call_VM as it will see a possible pending exception
1360     // and forward it and never return here preventing us from
1361     // clearing _last_native_pc down below. So we do a runtime call by
1362     // hand.
1363     //
1364     __ mov(c_rarg0, rthread);
1365     __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1366     __ blrt(rscratch2, 1, 0, 0);
1367     __ maybe_isb();
1368     __ get_method(rmethod);
1369     __ reinit_heapbase();
1370     __ bind(Continue);
1371   }
1372 
1373   // change thread state
1374   __ mov(rscratch1, _thread_in_Java);
1375   __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1376   __ stlrw(rscratch1, rscratch2);
1377 
1378   // reset_last_Java_frame
1379   __ reset_last_Java_frame(true);
1380 
1381   if (CheckJNICalls) {
1382     // clear_pending_jni_exception_check
1383     __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1384   }
1385 
1386   // reset handle block
1387   __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1388   __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1389 
1390   // If result is an oop unbox and store it in frame where gc will see it
1391   // and result handler will pick it up
1392 
1393   {
1394     Label no_oop, store_result;
1395     __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1396     __ cmp(t, result_handler);
1397     __ br(Assembler::NE, no_oop);
1398     // retrieve result
1399     __ pop(ltos);
1400     __ cbz(r0, store_result);
1401     __ ldr(r0, Address(r0, 0));
1402     __ bind(store_result);
1403     __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1404     // keep stack depth as expected by pushing oop which will eventually be discarded
1405     __ push(ltos);
1406     __ bind(no_oop);
1407   }
1408 
1409   {
1410     Label no_reguard;
1411     __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1412     __ ldrw(rscratch1, Address(rscratch1));
1413     __ cmp(rscratch1, JavaThread::stack_guard_yellow_reserved_disabled);
1414     __ br(Assembler::NE, no_reguard);
1415 
1416     __ pusha(); // XXX only save smashed registers
1417     __ mov(c_rarg0, rthread);
1418     __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1419     __ blrt(rscratch2, 0, 0, 0);
1420     __ popa(); // XXX only restore smashed registers
1421     __ bind(no_reguard);
1422   }
1423 
1424   // The method register is junk from after the thread_in_native transition
1425   // until here.  Also can't call_VM until the bcp has been
1426   // restored.  Need bcp for throwing exception below so get it now.
1427   __ get_method(rmethod);
1428 
1429   // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1430   // rbcp == code_base()
1431   __ ldr(rbcp, Address(rmethod, Method::const_offset()));   // get ConstMethod*
1432   __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));          // get codebase
1433   // handle exceptions (exception handling will handle unlocking!)
1434   {
1435     Label L;
1436     __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1437     __ cbz(rscratch1, L);
1438     // Note: At some point we may want to unify this with the code
1439     // used in call_VM_base(); i.e., we should use the
1440     // StubRoutines::forward_exception code. For now this doesn't work
1441     // here because the rsp is not correctly set at this point.
1442     __ MacroAssembler::call_VM(noreg,
1443                                CAST_FROM_FN_PTR(address,
1444                                InterpreterRuntime::throw_pending_exception));
1445     __ should_not_reach_here();
1446     __ bind(L);
1447   }
1448 
1449   // do unlocking if necessary
1450   {
1451     Label L;
1452     __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1453     __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1454     // the code below should be shared with interpreter macro
1455     // assembler implementation
1456     {
1457       Label unlock;
1458       // BasicObjectLock will be first in list, since this is a
1459       // synchronized method. However, need to check that the object
1460       // has not been unlocked by an explicit monitorexit bytecode.
1461 
1462       // monitor expect in c_rarg1 for slow unlock path
1463       __ lea (c_rarg1, Address(rfp,   // address of first monitor
1464                                (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1465                                           wordSize - sizeof(BasicObjectLock))));
1466 
1467       __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1468       __ cbnz(t, unlock);
1469 
1470       // Entry already unlocked, need to throw exception
1471       __ MacroAssembler::call_VM(noreg,
1472                                  CAST_FROM_FN_PTR(address,
1473                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1474       __ should_not_reach_here();
1475 
1476       __ bind(unlock);
1477       __ unlock_object(c_rarg1);
1478     }
1479     __ bind(L);
1480   }
1481 
1482   // jvmti support
1483   // Note: This must happen _after_ handling/throwing any exceptions since
1484   //       the exception handler code notifies the runtime of method exits
1485   //       too. If this happens before, method entry/exit notifications are
1486   //       not properly paired (was bug - gri 11/22/99).
1487   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1488 
1489   // restore potential result in r0:d0, call result handler to
1490   // restore potential result in ST0 & handle result
1491 
1492   __ pop(ltos);
1493   __ pop(dtos);
1494 
1495   __ blr(result_handler);
1496 
1497   // remove activation
1498   __ ldr(esp, Address(rfp,
1499                     frame::interpreter_frame_sender_sp_offset *
1500                     wordSize)); // get sender sp
1501   // remove frame anchor
1502   __ leave();
1503 
1504   // resture sender sp
1505   __ mov(sp, esp);
1506 
1507   __ ret(lr);
1508 
1509   if (inc_counter) {
1510     // Handle overflow of counter and compile method
1511     __ bind(invocation_counter_overflow);
1512     generate_counter_overflow(continue_after_compile);
1513   }
1514 
1515   return entry_point;
1516 }
1517 
1518 //
1519 // Generic interpreted method entry to (asm) interpreter
1520 //
1521 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1522   // determine code generation flags
1523   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1524 
1525   // rscratch1: sender sp
1526   address entry_point = __ pc();
1527 
1528   const Address constMethod(rmethod, Method::const_offset());
1529   const Address access_flags(rmethod, Method::access_flags_offset());
1530   const Address size_of_parameters(r3,
1531                                    ConstMethod::size_of_parameters_offset());
1532   const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1533 
1534   // get parameter size (always needed)
1535   // need to load the const method first
1536   __ ldr(r3, constMethod);
1537   __ load_unsigned_short(r2, size_of_parameters);
1538 
1539   // r2: size of parameters
1540 
1541   __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1542   __ sub(r3, r3, r2); // r3 = no. of additional locals
1543 
1544   // see if we've got enough room on the stack for locals plus overhead.
1545   generate_stack_overflow_check();
1546 
1547   // compute beginning of parameters (rlocals)
1548   __ add(rlocals, esp, r2, ext::uxtx, 3);
1549   __ sub(rlocals, rlocals, wordSize);
1550 
1551   // Make room for locals
1552   __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1553   __ andr(sp, rscratch1, -16);
1554 
1555   // r3 - # of additional locals
1556   // allocate space for locals
1557   // explicitly initialize locals
1558   {
1559     Label exit, loop;
1560     __ ands(zr, r3, r3);
1561     __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1562     __ bind(loop);
1563     __ str(zr, Address(__ post(rscratch1, wordSize)));
1564     __ sub(r3, r3, 1); // until everything initialized
1565     __ cbnz(r3, loop);
1566     __ bind(exit);
1567   }
1568 
1569   // And the base dispatch table
1570   __ get_dispatch();
1571 
1572   // initialize fixed part of activation frame
1573   generate_fixed_frame(false);
1574 #ifndef PRODUCT
1575   // tell the simulator that a method has been entered
1576   if (NotifySimulator) {
1577     __ notify(Assembler::method_entry);
1578   }
1579 #endif
1580   // make sure method is not native & not abstract
1581 #ifdef ASSERT
1582   __ ldrw(r0, access_flags);
1583   {
1584     Label L;
1585     __ tst(r0, JVM_ACC_NATIVE);
1586     __ br(Assembler::EQ, L);
1587     __ stop("tried to execute native method as non-native");
1588     __ bind(L);
1589   }
1590  {
1591     Label L;
1592     __ tst(r0, JVM_ACC_ABSTRACT);
1593     __ br(Assembler::EQ, L);
1594     __ stop("tried to execute abstract method in interpreter");
1595     __ bind(L);
1596   }
1597 #endif
1598 
1599   // Since at this point in the method invocation the exception
1600   // handler would try to exit the monitor of synchronized methods
1601   // which hasn't been entered yet, we set the thread local variable
1602   // _do_not_unlock_if_synchronized to true. The remove_activation
1603   // will check this flag.
1604 
1605    const Address do_not_unlock_if_synchronized(rthread,
1606         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1607   __ mov(rscratch2, true);
1608   __ strb(rscratch2, do_not_unlock_if_synchronized);
1609 
1610   // increment invocation count & check for overflow
1611   Label invocation_counter_overflow;
1612   Label profile_method;
1613   Label profile_method_continue;
1614   if (inc_counter) {
1615     generate_counter_incr(&invocation_counter_overflow,
1616                           &profile_method,
1617                           &profile_method_continue);
1618     if (ProfileInterpreter) {
1619       __ bind(profile_method_continue);
1620     }
1621   }
1622 
1623   Label continue_after_compile;
1624   __ bind(continue_after_compile);
1625 
1626   bang_stack_shadow_pages(false);
1627 
1628   // reset the _do_not_unlock_if_synchronized flag
1629   __ strb(zr, do_not_unlock_if_synchronized);
1630 
1631   // check for synchronized methods
1632   // Must happen AFTER invocation_counter check and stack overflow check,
1633   // so method is not locked if overflows.
1634   if (synchronized) {
1635     // Allocate monitor and lock method
1636     lock_method();
1637   } else {
1638     // no synchronization necessary
1639 #ifdef ASSERT
1640     {
1641       Label L;
1642       __ ldrw(r0, access_flags);
1643       __ tst(r0, JVM_ACC_SYNCHRONIZED);
1644       __ br(Assembler::EQ, L);
1645       __ stop("method needs synchronization");
1646       __ bind(L);
1647     }
1648 #endif
1649   }
1650 
1651   // start execution
1652 #ifdef ASSERT
1653   {
1654     Label L;
1655      const Address monitor_block_top (rfp,
1656                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
1657     __ ldr(rscratch1, monitor_block_top);
1658     __ cmp(esp, rscratch1);
1659     __ br(Assembler::EQ, L);
1660     __ stop("broken stack frame setup in interpreter");
1661     __ bind(L);
1662   }
1663 #endif
1664 
1665   // jvmti support
1666   __ notify_method_entry();
1667 
1668   __ dispatch_next(vtos);
1669 
1670   // invocation counter overflow
1671   if (inc_counter) {
1672     if (ProfileInterpreter) {
1673       // We have decided to profile this method in the interpreter
1674       __ bind(profile_method);
1675       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1676       __ set_method_data_pointer_for_bcp();
1677       // don't think we need this
1678       __ get_method(r1);
1679       __ b(profile_method_continue);
1680     }
1681     // Handle overflow of counter and compile method
1682     __ bind(invocation_counter_overflow);
1683     generate_counter_overflow(continue_after_compile);
1684   }
1685 
1686   return entry_point;
1687 }
1688 
1689 //-----------------------------------------------------------------------------
1690 // Exceptions
1691 
1692 void TemplateInterpreterGenerator::generate_throw_exception() {
1693   // Entry point in previous activation (i.e., if the caller was
1694   // interpreted)
1695   Interpreter::_rethrow_exception_entry = __ pc();
1696   // Restore sp to interpreter_frame_last_sp even though we are going
1697   // to empty the expression stack for the exception processing.
1698   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1699   // r0: exception
1700   // r3: return address/pc that threw exception
1701   __ restore_bcp();    // rbcp points to call/send
1702   __ restore_locals();
1703   __ restore_constant_pool_cache();
1704   __ reinit_heapbase();  // restore rheapbase as heapbase.
1705   __ get_dispatch();
1706 
1707 #ifndef PRODUCT
1708   // tell the simulator that the caller method has been reentered
1709   if (NotifySimulator) {
1710     __ get_method(rmethod);
1711     __ notify(Assembler::method_reentry);
1712   }
1713 #endif
1714   // Entry point for exceptions thrown within interpreter code
1715   Interpreter::_throw_exception_entry = __ pc();
1716   // If we came here via a NullPointerException on the receiver of a
1717   // method, rmethod may be corrupt.
1718   __ get_method(rmethod);
1719   // expression stack is undefined here
1720   // r0: exception
1721   // rbcp: exception bcp
1722   __ verify_oop(r0);
1723   __ mov(c_rarg1, r0);
1724 
1725   // expression stack must be empty before entering the VM in case of
1726   // an exception
1727   __ empty_expression_stack();
1728   // find exception handler address and preserve exception oop
1729   __ call_VM(r3,
1730              CAST_FROM_FN_PTR(address,
1731                           InterpreterRuntime::exception_handler_for_exception),
1732              c_rarg1);
1733 
1734   // Calculate stack limit
1735   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1736   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1737   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1738   __ ldr(rscratch2,
1739          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1740   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1741   __ andr(sp, rscratch1, -16);
1742 
1743   // r0: exception handler entry point
1744   // r3: preserved exception oop
1745   // rbcp: bcp for exception handler
1746   __ push_ptr(r3); // push exception which is now the only value on the stack
1747   __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1748 
1749   // If the exception is not handled in the current frame the frame is
1750   // removed and the exception is rethrown (i.e. exception
1751   // continuation is _rethrow_exception).
1752   //
1753   // Note: At this point the bci is still the bxi for the instruction
1754   // which caused the exception and the expression stack is
1755   // empty. Thus, for any VM calls at this point, GC will find a legal
1756   // oop map (with empty expression stack).
1757 
1758   //
1759   // JVMTI PopFrame support
1760   //
1761 
1762   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1763   __ empty_expression_stack();
1764   // Set the popframe_processing bit in pending_popframe_condition
1765   // indicating that we are currently handling popframe, so that
1766   // call_VMs that may happen later do not trigger new popframe
1767   // handling cycles.
1768   __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1769   __ orr(r3, r3, JavaThread::popframe_processing_bit);
1770   __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1771 
1772   {
1773     // Check to see whether we are returning to a deoptimized frame.
1774     // (The PopFrame call ensures that the caller of the popped frame is
1775     // either interpreted or compiled and deoptimizes it if compiled.)
1776     // In this case, we can't call dispatch_next() after the frame is
1777     // popped, but instead must save the incoming arguments and restore
1778     // them after deoptimization has occurred.
1779     //
1780     // Note that we don't compare the return PC against the
1781     // deoptimization blob's unpack entry because of the presence of
1782     // adapter frames in C2.
1783     Label caller_not_deoptimized;
1784     __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1785     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1786                                InterpreterRuntime::interpreter_contains), c_rarg1);
1787     __ cbnz(r0, caller_not_deoptimized);
1788 
1789     // Compute size of arguments for saving when returning to
1790     // deoptimized caller
1791     __ get_method(r0);
1792     __ ldr(r0, Address(r0, Method::const_offset()));
1793     __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1794                                                     size_of_parameters_offset())));
1795     __ lsl(r0, r0, Interpreter::logStackElementSize);
1796     __ restore_locals(); // XXX do we need this?
1797     __ sub(rlocals, rlocals, r0);
1798     __ add(rlocals, rlocals, wordSize);
1799     // Save these arguments
1800     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1801                                            Deoptimization::
1802                                            popframe_preserve_args),
1803                           rthread, r0, rlocals);
1804 
1805     __ remove_activation(vtos,
1806                          /* throw_monitor_exception */ false,
1807                          /* install_monitor_exception */ false,
1808                          /* notify_jvmdi */ false);
1809 
1810     // Inform deoptimization that it is responsible for restoring
1811     // these arguments
1812     __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1813     __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1814 
1815     // Continue in deoptimization handler
1816     __ ret(lr);
1817 
1818     __ bind(caller_not_deoptimized);
1819   }
1820 
1821   __ remove_activation(vtos,
1822                        /* throw_monitor_exception */ false,
1823                        /* install_monitor_exception */ false,
1824                        /* notify_jvmdi */ false);
1825 
1826   // Restore the last_sp and null it out
1827   __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1828   __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1829 
1830   __ restore_bcp();
1831   __ restore_locals();
1832   __ restore_constant_pool_cache();
1833   __ get_method(rmethod);
1834 
1835   // The method data pointer was incremented already during
1836   // call profiling. We have to restore the mdp for the current bcp.
1837   if (ProfileInterpreter) {
1838     __ set_method_data_pointer_for_bcp();
1839   }
1840 
1841   // Clear the popframe condition flag
1842   __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1843   assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1844 
1845 #if INCLUDE_JVMTI
1846   {
1847     Label L_done;
1848 
1849     __ ldrb(rscratch1, Address(rbcp, 0));
1850     __ cmpw(r1, Bytecodes::_invokestatic);
1851     __ br(Assembler::EQ, L_done);
1852 
1853     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1854     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1855 
1856     __ ldr(c_rarg0, Address(rlocals, 0));
1857     __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1858 
1859     __ cbz(r0, L_done);
1860 
1861     __ str(r0, Address(esp, 0));
1862     __ bind(L_done);
1863   }
1864 #endif // INCLUDE_JVMTI
1865 
1866   // Restore machine SP
1867   __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1868   __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1869   __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1870   __ ldr(rscratch2,
1871          Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1872   __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1873   __ andr(sp, rscratch1, -16);
1874 
1875   __ dispatch_next(vtos);
1876   // end of PopFrame support
1877 
1878   Interpreter::_remove_activation_entry = __ pc();
1879 
1880   // preserve exception over this code sequence
1881   __ pop_ptr(r0);
1882   __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1883   // remove the activation (without doing throws on illegalMonitorExceptions)
1884   __ remove_activation(vtos, false, true, false);
1885   // restore exception
1886   // restore exception
1887   __ get_vm_result(r0, rthread);
1888 
1889   // In between activations - previous activation type unknown yet
1890   // compute continuation point - the continuation point expects the
1891   // following registers set up:
1892   //
1893   // r0: exception
1894   // lr: return address/pc that threw exception
1895   // rsp: expression stack of caller
1896   // rfp: fp of caller
1897   // FIXME: There's no point saving LR here because VM calls don't trash it
1898   __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize)));  // save exception & return address
1899   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1900                           SharedRuntime::exception_handler_for_return_address),
1901                         rthread, lr);
1902   __ mov(r1, r0);                               // save exception handler
1903   __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize)));  // restore exception & return address
1904   // We might be returning to a deopt handler that expects r3 to
1905   // contain the exception pc
1906   __ mov(r3, lr);
1907   // Note that an "issuing PC" is actually the next PC after the call
1908   __ br(r1);                                    // jump to exception
1909                                                 // handler of caller
1910 }
1911 
1912 
1913 //
1914 // JVMTI ForceEarlyReturn support
1915 //
1916 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1917   address entry = __ pc();
1918 
1919   __ restore_bcp();
1920   __ restore_locals();
1921   __ empty_expression_stack();
1922   __ load_earlyret_value(state);
1923 
1924   __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1925   Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1926 
1927   // Clear the earlyret state
1928   assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1929   __ str(zr, cond_addr);
1930 
1931   __ remove_activation(state,
1932                        false, /* throw_monitor_exception */
1933                        false, /* install_monitor_exception */
1934                        true); /* notify_jvmdi */
1935   __ ret(lr);
1936 
1937   return entry;
1938 } // end of ForceEarlyReturn support
1939 
1940 
1941 
1942 //-----------------------------------------------------------------------------
1943 // Helper for vtos entry point generation
1944 
1945 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1946                                                          address& bep,
1947                                                          address& cep,
1948                                                          address& sep,
1949                                                          address& aep,
1950                                                          address& iep,
1951                                                          address& lep,
1952                                                          address& fep,
1953                                                          address& dep,
1954                                                          address& vep) {
1955   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1956   Label L;
1957   aep = __ pc();  __ push_ptr();  __ b(L);
1958   fep = __ pc();  __ push_f();    __ b(L);
1959   dep = __ pc();  __ push_d();    __ b(L);
1960   lep = __ pc();  __ push_l();    __ b(L);
1961   bep = cep = sep =
1962   iep = __ pc();  __ push_i();
1963   vep = __ pc();
1964   __ bind(L);
1965   generate_and_dispatch(t);
1966 }
1967 
1968 //-----------------------------------------------------------------------------
1969 
1970 // Non-product code
1971 #ifndef PRODUCT
1972 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1973   address entry = __ pc();
1974 
1975   __ push(lr);
1976   __ push(state);
1977   __ push(RegSet::range(r0, r15), sp);
1978   __ mov(c_rarg2, r0);  // Pass itos
1979   __ call_VM(noreg,
1980              CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1981              c_rarg1, c_rarg2, c_rarg3);
1982   __ pop(RegSet::range(r0, r15), sp);
1983   __ pop(state);
1984   __ pop(lr);
1985   __ ret(lr);                                   // return from result handler
1986 
1987   return entry;
1988 }
1989 
1990 void TemplateInterpreterGenerator::count_bytecode() {
1991   Register rscratch3 = r0;
1992   __ push(rscratch1);
1993   __ push(rscratch2);
1994   __ push(rscratch3);
1995   __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
1996   __ atomic_add(noreg, 1, rscratch3);
1997   __ pop(rscratch3);
1998   __ pop(rscratch2);
1999   __ pop(rscratch1);
2000 }
2001 
2002 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2003 
2004 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2005 
2006 
2007 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2008   // Call a little run-time stub to avoid blow-up for each bytecode.
2009   // The run-time runtime saves the right registers, depending on
2010   // the tosca in-state for the given template.
2011 
2012   assert(Interpreter::trace_code(t->tos_in()) != NULL,
2013          "entry must have been generated");
2014   __ bl(Interpreter::trace_code(t->tos_in()));
2015   __ reinit_heapbase();
2016 }
2017 
2018 
2019 void TemplateInterpreterGenerator::stop_interpreter_at() {
2020   Label L;
2021   __ push(rscratch1);
2022   __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2023   __ ldr(rscratch1, Address(rscratch1));
2024   __ mov(rscratch2, StopInterpreterAt);
2025   __ cmpw(rscratch1, rscratch2);
2026   __ br(Assembler::NE, L);
2027   __ brk(0);
2028   __ bind(L);
2029   __ pop(rscratch1);
2030 }
2031 
2032 #ifdef BUILTIN_SIM
2033 
2034 #include <sys/mman.h>
2035 #include <unistd.h>
2036 
2037 extern "C" {
2038   static int PAGESIZE = getpagesize();
2039   int is_mapped_address(u_int64_t address)
2040   {
2041     address = (address & ~((u_int64_t)PAGESIZE - 1));
2042     if (msync((void *)address, PAGESIZE, MS_ASYNC) == 0) {
2043       return true;
2044     }
2045     if (errno != ENOMEM) {
2046       return true;
2047     }
2048     return false;
2049   }
2050 
2051   void bccheck1(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2052   {
2053     if (method != 0) {
2054       method[0] = '\0';
2055     }
2056     if (bcidx != 0) {
2057       *bcidx = -2;
2058     }
2059     if (decode != 0) {
2060       decode[0] = 0;
2061     }
2062 
2063     if (framesize != 0) {
2064       *framesize = -1;
2065     }
2066 
2067     if (Interpreter::contains((address)pc)) {
2068       AArch64Simulator *sim = AArch64Simulator::get_current(UseSimulatorCache, DisableBCCheck);
2069       Method* meth;
2070       address bcp;
2071       if (fp) {
2072 #define FRAME_SLOT_METHOD 3
2073 #define FRAME_SLOT_BCP 7
2074         meth = (Method*)sim->getMemory()->loadU64(fp - (FRAME_SLOT_METHOD << 3));
2075         bcp = (address)sim->getMemory()->loadU64(fp - (FRAME_SLOT_BCP << 3));
2076 #undef FRAME_SLOT_METHOD
2077 #undef FRAME_SLOT_BCP
2078       } else {
2079         meth = (Method*)sim->getCPUState().xreg(RMETHOD, 0);
2080         bcp = (address)sim->getCPUState().xreg(RBCP, 0);
2081       }
2082       if (meth->is_native()) {
2083         return;
2084       }
2085       if(method && meth->is_method()) {
2086         ResourceMark rm;
2087         method[0] = 'I';
2088         method[1] = ' ';
2089         meth->name_and_sig_as_C_string(method + 2, 398);
2090       }
2091       if (bcidx) {
2092         if (meth->contains(bcp)) {
2093           *bcidx = meth->bci_from(bcp);
2094         } else {
2095           *bcidx = -2;
2096         }
2097       }
2098       if (decode) {
2099         if (!BytecodeTracer::closure()) {
2100           BytecodeTracer::set_closure(BytecodeTracer::std_closure());
2101         }
2102         stringStream str(decode, 400);
2103         BytecodeTracer::trace(meth, bcp, &str);
2104       }
2105     } else {
2106       if (method) {
2107         CodeBlob *cb = CodeCache::find_blob((address)pc);
2108         if (cb != NULL) {
2109           if (cb->is_nmethod()) {
2110             ResourceMark rm;
2111             nmethod* nm = (nmethod*)cb;
2112             method[0] = 'C';
2113             method[1] = ' ';
2114             nm->method()->name_and_sig_as_C_string(method + 2, 398);
2115           } else if (cb->is_adapter_blob()) {
2116             strcpy(method, "B adapter blob");
2117           } else if (cb->is_runtime_stub()) {
2118             strcpy(method, "B runtime stub");
2119           } else if (cb->is_exception_stub()) {
2120             strcpy(method, "B exception stub");
2121           } else if (cb->is_deoptimization_stub()) {
2122             strcpy(method, "B deoptimization stub");
2123           } else if (cb->is_safepoint_stub()) {
2124             strcpy(method, "B safepoint stub");
2125           } else if (cb->is_uncommon_trap_stub()) {
2126             strcpy(method, "B uncommon trap stub");
2127           } else if (cb->contains((address)StubRoutines::call_stub())) {
2128             strcpy(method, "B call stub");
2129           } else {
2130             strcpy(method, "B unknown blob : ");
2131             strcat(method, cb->name());
2132           }
2133           if (framesize != NULL) {
2134             *framesize = cb->frame_size();
2135           }
2136         }
2137       }
2138     }
2139   }
2140 
2141 
2142   JNIEXPORT void bccheck(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2143   {
2144     bccheck1(pc, fp, method, bcidx, framesize, decode);
2145   }
2146 }
2147 
2148 #endif // BUILTIN_SIM
2149 #endif // !PRODUCT