1 /*
   2  * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "asm/macroAssembler.hpp"
  27 #include "interpreter/bytecodeHistogram.hpp"
  28 #include "interpreter/interpreter.hpp"
  29 #include "interpreter/interpreterRuntime.hpp"
  30 #include "interpreter/interp_masm.hpp"
  31 #include "interpreter/templateInterpreterGenerator.hpp"
  32 #include "interpreter/templateTable.hpp"
  33 #include "oops/arrayOop.hpp"
  34 #include "oops/methodData.hpp"
  35 #include "oops/method.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "prims/jvmtiExport.hpp"
  38 #include "prims/jvmtiThreadState.hpp"
  39 #include "runtime/arguments.hpp"
  40 #include "runtime/deoptimization.hpp"
  41 #include "runtime/frame.inline.hpp"
  42 #include "runtime/sharedRuntime.hpp"
  43 #include "runtime/stubRoutines.hpp"
  44 #include "runtime/synchronizer.hpp"
  45 #include "runtime/timer.hpp"
  46 #include "runtime/vframeArray.hpp"
  47 #include "utilities/debug.hpp"
  48 #include "utilities/macros.hpp"
  49 
  50 #ifndef FAST_DISPATCH
  51 #define FAST_DISPATCH 1
  52 #endif
  53 #undef FAST_DISPATCH
  54 
  55 // Size of interpreter code.  Increase if too small.  Interpreter will
  56 // fail with a guarantee ("not enough space for interpreter generation");
  57 // if too small.
  58 // Run with +PrintInterpreter to get the VM to print out the size.
  59 // Max size with JVMTI
  60 #ifdef _LP64
  61   // The sethi() instruction generates lots more instructions when shell
  62   // stack limit is unlimited, so that's why this is much bigger.
  63 int TemplateInterpreter::InterpreterCodeSize = 260 * K;
  64 #else
  65 int TemplateInterpreter::InterpreterCodeSize = 230 * K;
  66 #endif
  67 
  68 // Generation of Interpreter
  69 //
  70 // The TemplateInterpreterGenerator generates the interpreter into Interpreter::_code.
  71 
  72 
  73 #define __ _masm->
  74 
  75 
  76 //----------------------------------------------------------------------------------------------------
  77 
  78 #ifndef _LP64
  79 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
  80   address entry = __ pc();
  81   Argument argv(0, true);
  82 
  83   // We are in the jni transition frame. Save the last_java_frame corresponding to the
  84   // outer interpreter frame
  85   //
  86   __ set_last_Java_frame(FP, noreg);
  87   // make sure the interpreter frame we've pushed has a valid return pc
  88   __ mov(O7, I7);
  89   __ mov(Lmethod, G3_scratch);
  90   __ mov(Llocals, G4_scratch);
  91   __ save_frame(0);
  92   __ mov(G2_thread, L7_thread_cache);
  93   __ add(argv.address_in_frame(), O3);
  94   __ mov(G2_thread, O0);
  95   __ mov(G3_scratch, O1);
  96   __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
  97   __ delayed()->mov(G4_scratch, O2);
  98   __ mov(L7_thread_cache, G2_thread);
  99   __ reset_last_Java_frame();
 100 
 101   // load the register arguments (the C code packed them as varargs)
 102   for (Argument ldarg = argv.successor(); ldarg.is_register(); ldarg = ldarg.successor()) {
 103       __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
 104   }
 105   __ ret();
 106   __ delayed()->
 107      restore(O0, 0, Lscratch);  // caller's Lscratch gets the result handler
 108   return entry;
 109 }
 110 
 111 
 112 #else
 113 // LP64 passes floating point arguments in F1, F3, F5, etc. instead of
 114 // O0, O1, O2 etc..
 115 // Doubles are passed in D0, D2, D4
 116 // We store the signature of the first 16 arguments in the first argument
 117 // slot because it will be overwritten prior to calling the native
 118 // function, with the pointer to the JNIEnv.
 119 // If LP64 there can be up to 16 floating point arguments in registers
 120 // or 6 integer registers.
 121 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
 122 
 123   enum {
 124     non_float  = 0,
 125     float_sig  = 1,
 126     double_sig = 2,
 127     sig_mask   = 3
 128   };
 129 
 130   address entry = __ pc();
 131   Argument argv(0, true);
 132 
 133   // We are in the jni transition frame. Save the last_java_frame corresponding to the
 134   // outer interpreter frame
 135   //
 136   __ set_last_Java_frame(FP, noreg);
 137   // make sure the interpreter frame we've pushed has a valid return pc
 138   __ mov(O7, I7);
 139   __ mov(Lmethod, G3_scratch);
 140   __ mov(Llocals, G4_scratch);
 141   __ save_frame(0);
 142   __ mov(G2_thread, L7_thread_cache);
 143   __ add(argv.address_in_frame(), O3);
 144   __ mov(G2_thread, O0);
 145   __ mov(G3_scratch, O1);
 146   __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
 147   __ delayed()->mov(G4_scratch, O2);
 148   __ mov(L7_thread_cache, G2_thread);
 149   __ reset_last_Java_frame();
 150 
 151 
 152   // load the register arguments (the C code packed them as varargs)
 153   Address Sig = argv.address_in_frame();        // Argument 0 holds the signature
 154   __ ld_ptr( Sig, G3_scratch );                   // Get register argument signature word into G3_scratch
 155   __ mov( G3_scratch, G4_scratch);
 156   __ srl( G4_scratch, 2, G4_scratch);             // Skip Arg 0
 157   Label done;
 158   for (Argument ldarg = argv.successor(); ldarg.is_float_register(); ldarg = ldarg.successor()) {
 159     Label NonFloatArg;
 160     Label LoadFloatArg;
 161     Label LoadDoubleArg;
 162     Label NextArg;
 163     Address a = ldarg.address_in_frame();
 164     __ andcc(G4_scratch, sig_mask, G3_scratch);
 165     __ br(Assembler::zero, false, Assembler::pt, NonFloatArg);
 166     __ delayed()->nop();
 167 
 168     __ cmp(G3_scratch, float_sig );
 169     __ br(Assembler::equal, false, Assembler::pt, LoadFloatArg);
 170     __ delayed()->nop();
 171 
 172     __ cmp(G3_scratch, double_sig );
 173     __ br(Assembler::equal, false, Assembler::pt, LoadDoubleArg);
 174     __ delayed()->nop();
 175 
 176     __ bind(NonFloatArg);
 177     // There are only 6 integer register arguments!
 178     if ( ldarg.is_register() )
 179       __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
 180     else {
 181     // Optimization, see if there are any more args and get out prior to checking
 182     // all 16 float registers.  My guess is that this is rare.
 183     // If is_register is false, then we are done the first six integer args.
 184       __ br_null_short(G4_scratch, Assembler::pt, done);
 185     }
 186     __ ba(NextArg);
 187     __ delayed()->srl( G4_scratch, 2, G4_scratch );
 188 
 189     __ bind(LoadFloatArg);
 190     __ ldf( FloatRegisterImpl::S, a, ldarg.as_float_register(), 4);
 191     __ ba(NextArg);
 192     __ delayed()->srl( G4_scratch, 2, G4_scratch );
 193 
 194     __ bind(LoadDoubleArg);
 195     __ ldf( FloatRegisterImpl::D, a, ldarg.as_double_register() );
 196     __ ba(NextArg);
 197     __ delayed()->srl( G4_scratch, 2, G4_scratch );
 198 
 199     __ bind(NextArg);
 200 
 201   }
 202 
 203   __ bind(done);
 204   __ ret();
 205   __ delayed()->
 206      restore(O0, 0, Lscratch);  // caller's Lscratch gets the result handler
 207   return entry;
 208 }
 209 #endif
 210 
 211 void TemplateInterpreterGenerator::generate_counter_overflow(Label& Lcontinue) {
 212 
 213   // Generate code to initiate compilation on the counter overflow.
 214 
 215   // InterpreterRuntime::frequency_counter_overflow takes two arguments,
 216   // the first indicates if the counter overflow occurs at a backwards branch (NULL bcp)
 217   // and the second is only used when the first is true.  We pass zero for both.
 218   // The call returns the address of the verified entry point for the method or NULL
 219   // if the compilation did not complete (either went background or bailed out).
 220   __ set((int)false, O2);
 221   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O2, O2, true);
 222   // returns verified_entry_point or NULL
 223   // we ignore it in any case
 224   __ ba_short(Lcontinue);
 225 
 226 }
 227 
 228 
 229 // End of helpers
 230 
 231 // Various method entries
 232 
 233 // Abstract method entry
 234 // Attempt to execute abstract method. Throw exception
 235 //
 236 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
 237   address entry = __ pc();
 238   // abstract method entry
 239   // throw exception
 240   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
 241   // the call_VM checks for exception, so we should never return here.
 242   __ should_not_reach_here();
 243   return entry;
 244 
 245 }
 246 
 247 void TemplateInterpreterGenerator::save_native_result(void) {
 248   // result potentially in O0/O1: save it across calls
 249   const Address& l_tmp = InterpreterMacroAssembler::l_tmp;
 250 
 251   // result potentially in F0/F1: save it across calls
 252   const Address& d_tmp = InterpreterMacroAssembler::d_tmp;
 253 
 254   // save and restore any potential method result value around the unlocking operation
 255   __ stf(FloatRegisterImpl::D, F0, d_tmp);
 256 #ifdef _LP64
 257   __ stx(O0, l_tmp);
 258 #else
 259   __ std(O0, l_tmp);
 260 #endif
 261 }
 262 
 263 void TemplateInterpreterGenerator::restore_native_result(void) {
 264   const Address& l_tmp = InterpreterMacroAssembler::l_tmp;
 265   const Address& d_tmp = InterpreterMacroAssembler::d_tmp;
 266 
 267   // Restore any method result value
 268   __ ldf(FloatRegisterImpl::D, d_tmp, F0);
 269 #ifdef _LP64
 270   __ ldx(l_tmp, O0);
 271 #else
 272   __ ldd(l_tmp, O0);
 273 #endif
 274 }
 275 
 276 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
 277   assert(!pass_oop || message == NULL, "either oop or message but not both");
 278   address entry = __ pc();
 279   // expression stack must be empty before entering the VM if an exception happened
 280   __ empty_expression_stack();
 281   // load exception object
 282   __ set((intptr_t)name, G3_scratch);
 283   if (pass_oop) {
 284     __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), G3_scratch, Otos_i);
 285   } else {
 286     __ set((intptr_t)message, G4_scratch);
 287     __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), G3_scratch, G4_scratch);
 288   }
 289   // throw exception
 290   assert(Interpreter::throw_exception_entry() != NULL, "generate it first");
 291   AddressLiteral thrower(Interpreter::throw_exception_entry());
 292   __ jump_to(thrower, G3_scratch);
 293   __ delayed()->nop();
 294   return entry;
 295 }
 296 
 297 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
 298   address entry = __ pc();
 299   // expression stack must be empty before entering the VM if an exception
 300   // happened
 301   __ empty_expression_stack();
 302   // load exception object
 303   __ call_VM(Oexception,
 304              CAST_FROM_FN_PTR(address,
 305                               InterpreterRuntime::throw_ClassCastException),
 306              Otos_i);
 307   __ should_not_reach_here();
 308   return entry;
 309 }
 310 
 311 
 312 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
 313   address entry = __ pc();
 314   // expression stack must be empty before entering the VM if an exception happened
 315   __ empty_expression_stack();
 316   // convention: expect aberrant index in register G3_scratch, then shuffle the
 317   // index to G4_scratch for the VM call
 318   __ mov(G3_scratch, G4_scratch);
 319   __ set((intptr_t)name, G3_scratch);
 320   __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), G3_scratch, G4_scratch);
 321   __ should_not_reach_here();
 322   return entry;
 323 }
 324 
 325 
 326 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
 327   address entry = __ pc();
 328   // expression stack must be empty before entering the VM if an exception happened
 329   __ empty_expression_stack();
 330   __ call_VM(Oexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
 331   __ should_not_reach_here();
 332   return entry;
 333 }
 334 
 335 
 336 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
 337   address entry = __ pc();
 338 
 339   if (state == atos) {
 340     __ profile_return_type(O0, G3_scratch, G1_scratch);
 341   }
 342 
 343 #if !defined(_LP64) && defined(COMPILER2)
 344   // All return values are where we want them, except for Longs.  C2 returns
 345   // longs in G1 in the 32-bit build whereas the interpreter wants them in O0/O1.
 346   // Since the interpreter will return longs in G1 and O0/O1 in the 32bit
 347   // build even if we are returning from interpreted we just do a little
 348   // stupid shuffing.
 349   // Note: I tried to make c2 return longs in O0/O1 and G1 so we wouldn't have to
 350   // do this here. Unfortunately if we did a rethrow we'd see an machepilog node
 351   // first which would move g1 -> O0/O1 and destroy the exception we were throwing.
 352 
 353   if (state == ltos) {
 354     __ srl (G1,  0, O1);
 355     __ srlx(G1, 32, O0);
 356   }
 357 #endif // !_LP64 && COMPILER2
 358 
 359   // The callee returns with the stack possibly adjusted by adapter transition
 360   // We remove that possible adjustment here.
 361   // All interpreter local registers are untouched. Any result is passed back
 362   // in the O0/O1 or float registers. Before continuing, the arguments must be
 363   // popped from the java expression stack; i.e., Lesp must be adjusted.
 364 
 365   __ mov(Llast_SP, SP);   // Remove any adapter added stack space.
 366 
 367   const Register cache = G3_scratch;
 368   const Register index  = G1_scratch;
 369   __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
 370 
 371   const Register flags = cache;
 372   __ ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset(), flags);
 373   const Register parameter_size = flags;
 374   __ and3(flags, ConstantPoolCacheEntry::parameter_size_mask, parameter_size);  // argument size in words
 375   __ sll(parameter_size, Interpreter::logStackElementSize, parameter_size);     // each argument size in bytes
 376   __ add(Lesp, parameter_size, Lesp);                                           // pop arguments
 377 
 378   __ check_and_handle_popframe(Gtemp);
 379   __ check_and_handle_earlyret(Gtemp);
 380 
 381   __ dispatch_next(state, step);
 382 
 383   return entry;
 384 }
 385 
 386 
 387 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
 388   address entry = __ pc();
 389   __ get_constant_pool_cache(LcpoolCache); // load LcpoolCache
 390 #if INCLUDE_JVMCI
 391   // Check if we need to take lock at entry of synchronized method.  This can
 392   // only occur on method entry so emit it only for vtos with step 0.
 393   if (UseJVMCICompiler && state == vtos && step == 0) {
 394     Label L;
 395     Address pending_monitor_enter_addr(G2_thread, JavaThread::pending_monitorenter_offset());
 396     __ ldbool(pending_monitor_enter_addr, Gtemp);  // Load if pending monitor enter
 397     __ cmp_and_br_short(Gtemp, G0, Assembler::equal, Assembler::pn, L);
 398     // Clear flag.
 399     __ stbool(G0, pending_monitor_enter_addr);
 400     // Take lock.
 401     lock_method();
 402     __ bind(L);
 403   } else {
 404 #ifdef ASSERT
 405     if (UseJVMCICompiler) {
 406       Label L;
 407       Address pending_monitor_enter_addr(G2_thread, JavaThread::pending_monitorenter_offset());
 408       __ ldbool(pending_monitor_enter_addr, Gtemp);  // Load if pending monitor enter
 409       __ cmp_and_br_short(Gtemp, G0, Assembler::equal, Assembler::pn, L);
 410       __ stop("unexpected pending monitor in deopt entry");
 411       __ bind(L);
 412     }
 413 #endif
 414   }
 415 #endif
 416   { Label L;
 417     Address exception_addr(G2_thread, Thread::pending_exception_offset());
 418     __ ld_ptr(exception_addr, Gtemp);  // Load pending exception.
 419     __ br_null_short(Gtemp, Assembler::pt, L);
 420     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
 421     __ should_not_reach_here();
 422     __ bind(L);
 423   }
 424   __ dispatch_next(state, step);
 425   return entry;
 426 }
 427 
 428 // A result handler converts/unboxes a native call result into
 429 // a java interpreter/compiler result. The current frame is an
 430 // interpreter frame. The activation frame unwind code must be
 431 // consistent with that of TemplateTable::_return(...). In the
 432 // case of native methods, the caller's SP was not modified.
 433 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
 434   address entry = __ pc();
 435   Register Itos_i  = Otos_i ->after_save();
 436   Register Itos_l  = Otos_l ->after_save();
 437   Register Itos_l1 = Otos_l1->after_save();
 438   Register Itos_l2 = Otos_l2->after_save();
 439   switch (type) {
 440     case T_BOOLEAN: __ subcc(G0, O0, G0); __ addc(G0, 0, Itos_i); break; // !0 => true; 0 => false
 441     case T_CHAR   : __ sll(O0, 16, O0); __ srl(O0, 16, Itos_i);   break; // cannot use and3, 0xFFFF too big as immediate value!
 442     case T_BYTE   : __ sll(O0, 24, O0); __ sra(O0, 24, Itos_i);   break;
 443     case T_SHORT  : __ sll(O0, 16, O0); __ sra(O0, 16, Itos_i);   break;
 444     case T_LONG   :
 445 #ifndef _LP64
 446                     __ mov(O1, Itos_l2);  // move other half of long
 447 #endif              // ifdef or no ifdef, fall through to the T_INT case
 448     case T_INT    : __ mov(O0, Itos_i);                         break;
 449     case T_VOID   : /* nothing to do */                         break;
 450     case T_FLOAT  : assert(F0 == Ftos_f, "fix this code" );     break;
 451     case T_DOUBLE : assert(F0 == Ftos_d, "fix this code" );     break;
 452     case T_OBJECT :
 453       __ ld_ptr(FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS, Itos_i);
 454       __ verify_oop(Itos_i);
 455       break;
 456     default       : ShouldNotReachHere();
 457   }
 458   __ ret();                           // return from interpreter activation
 459   __ delayed()->restore(I5_savedSP, G0, SP);  // remove interpreter frame
 460   NOT_PRODUCT(__ emit_int32(0);)       // marker for disassembly
 461   return entry;
 462 }
 463 
 464 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
 465   address entry = __ pc();
 466   __ push(state);
 467   __ call_VM(noreg, runtime_entry);
 468   __ dispatch_via(vtos, Interpreter::normal_table(vtos));
 469   return entry;
 470 }
 471 
 472 
 473 //
 474 // Helpers for commoning out cases in the various type of method entries.
 475 //
 476 
 477 // increment invocation count & check for overflow
 478 //
 479 // Note: checking for negative value instead of overflow
 480 //       so we have a 'sticky' overflow test
 481 //
 482 // Lmethod: method
 483 // ??: invocation counter
 484 //
 485 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
 486   // Note: In tiered we increment either counters in MethodCounters* or in
 487   // MDO depending if we're profiling or not.
 488   const Register G3_method_counters = G3_scratch;
 489   Label done;
 490 
 491   if (TieredCompilation) {
 492     const int increment = InvocationCounter::count_increment;
 493     Label no_mdo;
 494     if (ProfileInterpreter) {
 495       // If no method data exists, go to profile_continue.
 496       __ ld_ptr(Lmethod, Method::method_data_offset(), G4_scratch);
 497       __ br_null_short(G4_scratch, Assembler::pn, no_mdo);
 498       // Increment counter
 499       Address mdo_invocation_counter(G4_scratch,
 500                                      in_bytes(MethodData::invocation_counter_offset()) +
 501                                      in_bytes(InvocationCounter::counter_offset()));
 502       Address mask(G4_scratch, in_bytes(MethodData::invoke_mask_offset()));
 503       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask,
 504                                  G3_scratch, Lscratch,
 505                                  Assembler::zero, overflow);
 506       __ ba_short(done);
 507     }
 508 
 509     // Increment counter in MethodCounters*
 510     __ bind(no_mdo);
 511     Address invocation_counter(G3_method_counters,
 512             in_bytes(MethodCounters::invocation_counter_offset()) +
 513             in_bytes(InvocationCounter::counter_offset()));
 514     __ get_method_counters(Lmethod, G3_method_counters, done);
 515     Address mask(G3_method_counters, in_bytes(MethodCounters::invoke_mask_offset()));
 516     __ increment_mask_and_jump(invocation_counter, increment, mask,
 517                                G4_scratch, Lscratch,
 518                                Assembler::zero, overflow);
 519     __ bind(done);
 520   } else { // not TieredCompilation
 521     // Update standard invocation counters
 522     __ get_method_counters(Lmethod, G3_method_counters, done);
 523     __ increment_invocation_counter(G3_method_counters, O0, G4_scratch);
 524     if (ProfileInterpreter) {
 525       Address interpreter_invocation_counter(G3_method_counters,
 526             in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
 527       __ ld(interpreter_invocation_counter, G4_scratch);
 528       __ inc(G4_scratch);
 529       __ st(G4_scratch, interpreter_invocation_counter);
 530     }
 531 
 532     if (ProfileInterpreter && profile_method != NULL) {
 533       // Test to see if we should create a method data oop
 534       Address profile_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_profile_limit_offset()));
 535       __ ld(profile_limit, G1_scratch);
 536       __ cmp_and_br_short(O0, G1_scratch, Assembler::lessUnsigned, Assembler::pn, *profile_method_continue);
 537 
 538       // if no method data exists, go to profile_method
 539       __ test_method_data_pointer(*profile_method);
 540     }
 541 
 542     Address invocation_limit(G3_method_counters, in_bytes(MethodCounters::interpreter_invocation_limit_offset()));
 543     __ ld(invocation_limit, G3_scratch);
 544     __ cmp(O0, G3_scratch);
 545     __ br(Assembler::greaterEqualUnsigned, false, Assembler::pn, *overflow); // Far distance
 546     __ delayed()->nop();
 547     __ bind(done);
 548   }
 549 
 550 }
 551 
 552 // Allocate monitor and lock method (asm interpreter)
 553 // ebx - Method*
 554 //
 555 void TemplateInterpreterGenerator::lock_method() {
 556   __ ld(Lmethod, in_bytes(Method::access_flags_offset()), O0);  // Load access flags.
 557 
 558 #ifdef ASSERT
 559  { Label ok;
 560    __ btst(JVM_ACC_SYNCHRONIZED, O0);
 561    __ br( Assembler::notZero, false, Assembler::pt, ok);
 562    __ delayed()->nop();
 563    __ stop("method doesn't need synchronization");
 564    __ bind(ok);
 565   }
 566 #endif // ASSERT
 567 
 568   // get synchronization object to O0
 569   { Label done;
 570     __ btst(JVM_ACC_STATIC, O0);
 571     __ br( Assembler::zero, true, Assembler::pt, done);
 572     __ delayed()->ld_ptr(Llocals, Interpreter::local_offset_in_bytes(0), O0); // get receiver for not-static case
 573 
 574     // lock the mirror, not the Klass*
 575     __ load_mirror(O0, Lmethod);
 576 
 577 #ifdef ASSERT
 578     __ tst(O0);
 579     __ breakpoint_trap(Assembler::zero, Assembler::ptr_cc);
 580 #endif // ASSERT
 581 
 582     __ bind(done);
 583   }
 584 
 585   __ add_monitor_to_stack(true, noreg, noreg);  // allocate monitor elem
 586   __ st_ptr( O0, Lmonitors, BasicObjectLock::obj_offset_in_bytes());   // store object
 587   // __ untested("lock_object from method entry");
 588   __ lock_object(Lmonitors, O0);
 589 }
 590 
 591 // See if we've got enough room on the stack for locals plus overhead below
 592 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
 593 // without going through the signal handler, i.e., reserved and yellow zones
 594 // will not be made usable. The shadow zone must suffice to handle the
 595 // overflow.
 596 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rframe_size,
 597                                                                  Register Rscratch) {
 598   const int page_size = os::vm_page_size();
 599   Label after_frame_check;
 600 
 601   assert_different_registers(Rframe_size, Rscratch);
 602 
 603   __ set(page_size, Rscratch);
 604   __ cmp_and_br_short(Rframe_size, Rscratch, Assembler::lessEqual, Assembler::pt, after_frame_check);
 605 
 606   // Get the stack overflow limit, and in debug, verify it is non-zero.
 607   __ ld_ptr(G2_thread, JavaThread::stack_overflow_limit_offset(), Rscratch);
 608 #ifdef ASSERT
 609   Label limit_ok;
 610   __ br_notnull_short(Rscratch, Assembler::pn, limit_ok);
 611   __ stop("stack overflow limit is zero in generate_stack_overflow_check");
 612   __ bind(limit_ok);
 613 #endif
 614 
 615   // Add in the size of the frame (which is the same as subtracting it from the
 616   // SP, which would take another register.
 617   __ add(Rscratch, Rframe_size, Rscratch);
 618 
 619   // The frame is greater than one page in size, so check against
 620   // the bottom of the stack.
 621   __ cmp_and_brx_short(SP, Rscratch, Assembler::greaterUnsigned, Assembler::pt, after_frame_check);
 622 
 623   // The stack will overflow, throw an exception.
 624 
 625   // Note that SP is restored to sender's sp (in the delay slot). This
 626   // is necessary if the sender's frame is an extended compiled frame
 627   // (see gen_c2i_adapter()) and safer anyway in case of JSR292
 628   // adaptations.
 629 
 630   // Note also that the restored frame is not necessarily interpreted.
 631   // Use the shared runtime version of the StackOverflowError.
 632   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
 633   AddressLiteral stub(StubRoutines::throw_StackOverflowError_entry());
 634   __ jump_to(stub, Rscratch);
 635   __ delayed()->mov(O5_savedSP, SP);
 636 
 637   // If you get to here, then there is enough stack space.
 638   __ bind(after_frame_check);
 639 }
 640 
 641 
 642 //
 643 // Generate a fixed interpreter frame. This is identical setup for interpreted
 644 // methods and for native methods hence the shared code.
 645 
 646 
 647 //----------------------------------------------------------------------------------------------------
 648 // Stack frame layout
 649 //
 650 // When control flow reaches any of the entry types for the interpreter
 651 // the following holds ->
 652 //
 653 // C2 Calling Conventions:
 654 //
 655 // The entry code below assumes that the following registers are set
 656 // when coming in:
 657 //    G5_method: holds the Method* of the method to call
 658 //    Lesp:    points to the TOS of the callers expression stack
 659 //             after having pushed all the parameters
 660 //
 661 // The entry code does the following to setup an interpreter frame
 662 //   pop parameters from the callers stack by adjusting Lesp
 663 //   set O0 to Lesp
 664 //   compute X = (max_locals - num_parameters)
 665 //   bump SP up by X to accomadate the extra locals
 666 //   compute X = max_expression_stack
 667 //               + vm_local_words
 668 //               + 16 words of register save area
 669 //   save frame doing a save sp, -X, sp growing towards lower addresses
 670 //   set Lbcp, Lmethod, LcpoolCache
 671 //   set Llocals to i0
 672 //   set Lmonitors to FP - rounded_vm_local_words
 673 //   set Lesp to Lmonitors - 4
 674 //
 675 //  The frame has now been setup to do the rest of the entry code
 676 
 677 // Try this optimization:  Most method entries could live in a
 678 // "one size fits all" stack frame without all the dynamic size
 679 // calculations.  It might be profitable to do all this calculation
 680 // statically and approximately for "small enough" methods.
 681 
 682 //-----------------------------------------------------------------------------------------------
 683 
 684 // C1 Calling conventions
 685 //
 686 // Upon method entry, the following registers are setup:
 687 //
 688 // g2 G2_thread: current thread
 689 // g5 G5_method: method to activate
 690 // g4 Gargs  : pointer to last argument
 691 //
 692 //
 693 // Stack:
 694 //
 695 // +---------------+ <--- sp
 696 // |               |
 697 // : reg save area :
 698 // |               |
 699 // +---------------+ <--- sp + 0x40
 700 // |               |
 701 // : extra 7 slots :      note: these slots are not really needed for the interpreter (fix later)
 702 // |               |
 703 // +---------------+ <--- sp + 0x5c
 704 // |               |
 705 // :     free      :
 706 // |               |
 707 // +---------------+ <--- Gargs
 708 // |               |
 709 // :   arguments   :
 710 // |               |
 711 // +---------------+
 712 // |               |
 713 //
 714 //
 715 //
 716 // AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like:
 717 //
 718 // +---------------+ <--- sp
 719 // |               |
 720 // : reg save area :
 721 // |               |
 722 // +---------------+ <--- sp + 0x40
 723 // |               |
 724 // : extra 7 slots :      note: these slots are not really needed for the interpreter (fix later)
 725 // |               |
 726 // +---------------+ <--- sp + 0x5c
 727 // |               |
 728 // :               :
 729 // |               | <--- Lesp
 730 // +---------------+ <--- Lmonitors (fp - 0x18)
 731 // |   VM locals   |
 732 // +---------------+ <--- fp
 733 // |               |
 734 // : reg save area :
 735 // |               |
 736 // +---------------+ <--- fp + 0x40
 737 // |               |
 738 // : extra 7 slots :      note: these slots are not really needed for the interpreter (fix later)
 739 // |               |
 740 // +---------------+ <--- fp + 0x5c
 741 // |               |
 742 // :     free      :
 743 // |               |
 744 // +---------------+
 745 // |               |
 746 // : nonarg locals :
 747 // |               |
 748 // +---------------+
 749 // |               |
 750 // :   arguments   :
 751 // |               | <--- Llocals
 752 // +---------------+ <--- Gargs
 753 // |               |
 754 
 755 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
 756   //
 757   //
 758   // The entry code sets up a new interpreter frame in 4 steps:
 759   //
 760   // 1) Increase caller's SP by for the extra local space needed:
 761   //    (check for overflow)
 762   //    Efficient implementation of xload/xstore bytecodes requires
 763   //    that arguments and non-argument locals are in a contigously
 764   //    addressable memory block => non-argument locals must be
 765   //    allocated in the caller's frame.
 766   //
 767   // 2) Create a new stack frame and register window:
 768   //    The new stack frame must provide space for the standard
 769   //    register save area, the maximum java expression stack size,
 770   //    the monitor slots (0 slots initially), and some frame local
 771   //    scratch locations.
 772   //
 773   // 3) The following interpreter activation registers must be setup:
 774   //    Lesp       : expression stack pointer
 775   //    Lbcp       : bytecode pointer
 776   //    Lmethod    : method
 777   //    Llocals    : locals pointer
 778   //    Lmonitors  : monitor pointer
 779   //    LcpoolCache: constant pool cache
 780   //
 781   // 4) Initialize the non-argument locals if necessary:
 782   //    Non-argument locals may need to be initialized to NULL
 783   //    for GC to work. If the oop-map information is accurate
 784   //    (in the absence of the JSR problem), no initialization
 785   //    is necessary.
 786   //
 787   // (gri - 2/25/2000)
 788 
 789 
 790   int rounded_vm_local_words = round_to( frame::interpreter_frame_vm_local_words, WordsPerLong );
 791 
 792   const int extra_space =
 793     rounded_vm_local_words +                   // frame local scratch space
 794     Method::extra_stack_entries() +            // extra stack for jsr 292
 795     frame::memory_parameter_word_sp_offset +   // register save area
 796     (native_call ? frame::interpreter_frame_extra_outgoing_argument_words : 0);
 797 
 798   const Register Glocals_size = G3;
 799   const Register RconstMethod = Glocals_size;
 800   const Register Otmp1 = O3;
 801   const Register Otmp2 = O4;
 802   // Lscratch can't be used as a temporary because the call_stub uses
 803   // it to assert that the stack frame was setup correctly.
 804   const Address constMethod       (G5_method, Method::const_offset());
 805   const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
 806 
 807   __ ld_ptr( constMethod, RconstMethod );
 808   __ lduh( size_of_parameters, Glocals_size);
 809 
 810   // Gargs points to first local + BytesPerWord
 811   // Set the saved SP after the register window save
 812   //
 813   assert_different_registers(Gargs, Glocals_size, Gframe_size, O5_savedSP);
 814   __ sll(Glocals_size, Interpreter::logStackElementSize, Otmp1);
 815   __ add(Gargs, Otmp1, Gargs);
 816 
 817   if (native_call) {
 818     __ calc_mem_param_words( Glocals_size, Gframe_size );
 819     __ add( Gframe_size,  extra_space, Gframe_size);
 820     __ round_to( Gframe_size, WordsPerLong );
 821     __ sll( Gframe_size, LogBytesPerWord, Gframe_size );
 822 
 823     // Native calls don't need the stack size check since they have no
 824     // expression stack and the arguments are already on the stack and
 825     // we only add a handful of words to the stack.
 826   } else {
 827 
 828     //
 829     // Compute number of locals in method apart from incoming parameters
 830     //
 831     const Address size_of_locals(Otmp1, ConstMethod::size_of_locals_offset());
 832     __ ld_ptr(constMethod, Otmp1);
 833     __ lduh(size_of_locals, Otmp1);
 834     __ sub(Otmp1, Glocals_size, Glocals_size);
 835     __ round_to(Glocals_size, WordsPerLong);
 836     __ sll(Glocals_size, Interpreter::logStackElementSize, Glocals_size);
 837 
 838     // See if the frame is greater than one page in size. If so,
 839     // then we need to verify there is enough stack space remaining.
 840     // Frame_size = (max_stack + extra_space) * BytesPerWord;
 841     __ ld_ptr(constMethod, Gframe_size);
 842     __ lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size);
 843     __ add(Gframe_size, extra_space, Gframe_size);
 844     __ round_to(Gframe_size, WordsPerLong);
 845     __ sll(Gframe_size, Interpreter::logStackElementSize, Gframe_size);
 846 
 847     // Add in java locals size for stack overflow check only
 848     __ add(Gframe_size, Glocals_size, Gframe_size);
 849 
 850     const Register Otmp2 = O4;
 851     assert_different_registers(Otmp1, Otmp2, O5_savedSP);
 852     generate_stack_overflow_check(Gframe_size, Otmp1);
 853 
 854     __ sub(Gframe_size, Glocals_size, Gframe_size);
 855 
 856     //
 857     // bump SP to accomodate the extra locals
 858     //
 859     __ sub(SP, Glocals_size, SP);
 860   }
 861 
 862   //
 863   // now set up a stack frame with the size computed above
 864   //
 865   __ neg( Gframe_size );
 866   __ save( SP, Gframe_size, SP );
 867 
 868   //
 869   // now set up all the local cache registers
 870   //
 871   // NOTE: At this point, Lbyte_code/Lscratch has been modified. Note
 872   // that all present references to Lbyte_code initialize the register
 873   // immediately before use
 874   if (native_call) {
 875     __ mov(G0, Lbcp);
 876   } else {
 877     __ ld_ptr(G5_method, Method::const_offset(), Lbcp);
 878     __ add(Lbcp, in_bytes(ConstMethod::codes_offset()), Lbcp);
 879   }
 880   __ mov( G5_method, Lmethod);                 // set Lmethod
 881   // Get mirror and store it in the frame as GC root for this Method*
 882   Register mirror = LcpoolCache;
 883   __ load_mirror(mirror, Lmethod);
 884   __ st_ptr(mirror, FP, (frame::interpreter_frame_mirror_offset * wordSize) + STACK_BIAS);
 885   __ get_constant_pool_cache( LcpoolCache );   // set LcpoolCache
 886   __ sub(FP, rounded_vm_local_words * BytesPerWord, Lmonitors ); // set Lmonitors
 887 #ifdef _LP64
 888   __ add( Lmonitors, STACK_BIAS, Lmonitors );   // Account for 64 bit stack bias
 889 #endif
 890   __ sub(Lmonitors, BytesPerWord, Lesp);       // set Lesp
 891 
 892   // setup interpreter activation registers
 893   __ sub(Gargs, BytesPerWord, Llocals);        // set Llocals
 894 
 895   if (ProfileInterpreter) {
 896 #ifdef FAST_DISPATCH
 897     // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
 898     // they both use I2.
 899     assert(0, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
 900 #endif // FAST_DISPATCH
 901     __ set_method_data_pointer();
 902   }
 903 
 904 }
 905 
 906 // Method entry for java.lang.ref.Reference.get.
 907 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
 908 #if INCLUDE_ALL_GCS
 909   // Code: _aload_0, _getfield, _areturn
 910   // parameter size = 1
 911   //
 912   // The code that gets generated by this routine is split into 2 parts:
 913   //    1. The "intrinsified" code for G1 (or any SATB based GC),
 914   //    2. The slow path - which is an expansion of the regular method entry.
 915   //
 916   // Notes:-
 917   // * In the G1 code we do not check whether we need to block for
 918   //   a safepoint. If G1 is enabled then we must execute the specialized
 919   //   code for Reference.get (except when the Reference object is null)
 920   //   so that we can log the value in the referent field with an SATB
 921   //   update buffer.
 922   //   If the code for the getfield template is modified so that the
 923   //   G1 pre-barrier code is executed when the current method is
 924   //   Reference.get() then going through the normal method entry
 925   //   will be fine.
 926   // * The G1 code can, however, check the receiver object (the instance
 927   //   of java.lang.Reference) and jump to the slow path if null. If the
 928   //   Reference object is null then we obviously cannot fetch the referent
 929   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
 930   //   regular method entry code to generate the NPE.
 931   //
 932   // This code is based on generate_accessor_enty.
 933 
 934   address entry = __ pc();
 935 
 936   const int referent_offset = java_lang_ref_Reference::referent_offset;
 937   guarantee(referent_offset > 0, "referent offset not initialized");
 938 
 939   if (UseG1GC) {
 940      Label slow_path;
 941 
 942     // In the G1 code we don't check if we need to reach a safepoint. We
 943     // continue and the thread will safepoint at the next bytecode dispatch.
 944 
 945     // Check if local 0 != NULL
 946     // If the receiver is null then it is OK to jump to the slow path.
 947     __ ld_ptr(Gargs, G0, Otos_i ); // get local 0
 948     // check if local 0 == NULL and go the slow path
 949     __ cmp_and_brx_short(Otos_i, 0, Assembler::equal, Assembler::pn, slow_path);
 950 
 951 
 952     // Load the value of the referent field.
 953     if (Assembler::is_simm13(referent_offset)) {
 954       __ load_heap_oop(Otos_i, referent_offset, Otos_i);
 955     } else {
 956       __ set(referent_offset, G3_scratch);
 957       __ load_heap_oop(Otos_i, G3_scratch, Otos_i);
 958     }
 959 
 960     // Generate the G1 pre-barrier code to log the value of
 961     // the referent field in an SATB buffer. Note with
 962     // these parameters the pre-barrier does not generate
 963     // the load of the previous value
 964 
 965     __ g1_write_barrier_pre(noreg /* obj */, noreg /* index */, 0 /* offset */,
 966                             Otos_i /* pre_val */,
 967                             G3_scratch /* tmp */,
 968                             true /* preserve_o_regs */);
 969 
 970     // _areturn
 971     __ retl();                      // return from leaf routine
 972     __ delayed()->mov(O5_savedSP, SP);
 973 
 974     // Generate regular method entry
 975     __ bind(slow_path);
 976     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
 977     return entry;
 978   }
 979 #endif // INCLUDE_ALL_GCS
 980 
 981   // If G1 is not enabled then attempt to go through the accessor entry point
 982   // Reference.get is an accessor
 983   return NULL;
 984 }
 985 
 986 /**
 987  * Method entry for static native methods:
 988  *   int java.util.zip.CRC32.update(int crc, int b)
 989  */
 990 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
 991 
 992   if (UseCRC32Intrinsics) {
 993     address entry = __ pc();
 994 
 995     Label L_slow_path;
 996     // If we need a safepoint check, generate full interpreter entry.
 997     ExternalAddress state(SafepointSynchronize::address_of_state());
 998     __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2);
 999     __ set(SafepointSynchronize::_not_synchronized, O3);
1000     __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path);
1001 
1002     // Load parameters
1003     const Register crc   = O0; // initial crc
1004     const Register val   = O1; // byte to update with
1005     const Register table = O2; // address of 256-entry lookup table
1006 
1007     __ ldub(Gargs, 3, val);
1008     __ lduw(Gargs, 8, crc);
1009 
1010     __ set(ExternalAddress(StubRoutines::crc_table_addr()), table);
1011 
1012     __ not1(crc); // ~crc
1013     __ clruwu(crc);
1014     __ update_byte_crc32(crc, val, table);
1015     __ not1(crc); // ~crc
1016 
1017     // result in O0
1018     __ retl();
1019     __ delayed()->nop();
1020 
1021     // generate a vanilla native entry as the slow path
1022     __ bind(L_slow_path);
1023     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1024     return entry;
1025   }
1026   return NULL;
1027 }
1028 
1029 /**
1030  * Method entry for static native methods:
1031  *   int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1032  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1033  */
1034 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1035 
1036   if (UseCRC32Intrinsics) {
1037     address entry = __ pc();
1038 
1039     Label L_slow_path;
1040     // If we need a safepoint check, generate full interpreter entry.
1041     ExternalAddress state(SafepointSynchronize::address_of_state());
1042     __ set(ExternalAddress(SafepointSynchronize::address_of_state()), O2);
1043     __ set(SafepointSynchronize::_not_synchronized, O3);
1044     __ cmp_and_br_short(O2, O3, Assembler::notEqual, Assembler::pt, L_slow_path);
1045 
1046     // Load parameters from the stack
1047     const Register crc    = O0; // initial crc
1048     const Register buf    = O1; // source java byte array address
1049     const Register len    = O2; // len
1050     const Register offset = O3; // offset
1051 
1052     // Arguments are reversed on java expression stack
1053     // Calculate address of start element
1054     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1055       __ lduw(Gargs, 0,  len);
1056       __ lduw(Gargs, 8,  offset);
1057       __ ldx( Gargs, 16, buf);
1058       __ lduw(Gargs, 32, crc);
1059       __ add(buf, offset, buf);
1060     } else {
1061       __ lduw(Gargs, 0,  len);
1062       __ lduw(Gargs, 8,  offset);
1063       __ ldx( Gargs, 16, buf);
1064       __ lduw(Gargs, 24, crc);
1065       __ add(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE), buf); // account for the header size
1066       __ add(buf ,offset, buf);
1067     }
1068 
1069     // Call the crc32 kernel
1070     __ MacroAssembler::save_thread(L7_thread_cache);
1071     __ kernel_crc32(crc, buf, len, O3);
1072     __ MacroAssembler::restore_thread(L7_thread_cache);
1073 
1074     // result in O0
1075     __ retl();
1076     __ delayed()->nop();
1077 
1078     // generate a vanilla native entry as the slow path
1079     __ bind(L_slow_path);
1080     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1081     return entry;
1082   }
1083   return NULL;
1084 }
1085 
1086 /**
1087  * Method entry for intrinsic-candidate (non-native) methods:
1088  *   int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1089  *   int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1090  * Unlike CRC32, CRC32C does not have any methods marked as native
1091  * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1092  */
1093 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1094 
1095   if (UseCRC32CIntrinsics) {
1096     address entry = __ pc();
1097 
1098     // Load parameters from the stack
1099     const Register crc    = O0; // initial crc
1100     const Register buf    = O1; // source java byte array address
1101     const Register offset = O2; // offset
1102     const Register end    = O3; // index of last element to process
1103     const Register len    = O2; // len argument to the kernel
1104     const Register table  = O3; // crc32c lookup table address
1105 
1106     // Arguments are reversed on java expression stack
1107     // Calculate address of start element
1108     if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1109       __ lduw(Gargs, 0,  end);
1110       __ lduw(Gargs, 8,  offset);
1111       __ ldx( Gargs, 16, buf);
1112       __ lduw(Gargs, 32, crc);
1113       __ add(buf, offset, buf);
1114       __ sub(end, offset, len);
1115     } else {
1116       __ lduw(Gargs, 0,  end);
1117       __ lduw(Gargs, 8,  offset);
1118       __ ldx( Gargs, 16, buf);
1119       __ lduw(Gargs, 24, crc);
1120       __ add(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE), buf); // account for the header size
1121       __ add(buf, offset, buf);
1122       __ sub(end, offset, len);
1123     }
1124 
1125     // Call the crc32c kernel
1126     __ MacroAssembler::save_thread(L7_thread_cache);
1127     __ kernel_crc32c(crc, buf, len, table);
1128     __ MacroAssembler::restore_thread(L7_thread_cache);
1129 
1130     // result in O0
1131     __ retl();
1132     __ delayed()->nop();
1133 
1134     return entry;
1135   }
1136   return NULL;
1137 }
1138 
1139 // Not supported
1140 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
1141   return NULL;
1142 }
1143 
1144 // TODO: rather than touching all pages, check against stack_overflow_limit and bang yellow page to
1145 // generate exception
1146 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1147   // Quick & dirty stack overflow checking: bang the stack & handle trap.
1148   // Note that we do the banging after the frame is setup, since the exception
1149   // handling code expects to find a valid interpreter frame on the stack.
1150   // Doing the banging earlier fails if the caller frame is not an interpreter
1151   // frame.
1152   // (Also, the exception throwing code expects to unlock any synchronized
1153   // method receiever, so do the banging after locking the receiver.)
1154 
1155   // Bang each page in the shadow zone. We can't assume it's been done for
1156   // an interpreter frame with greater than a page of locals, so each page
1157   // needs to be checked.  Only true for non-native.
1158   if (UseStackBanging) {
1159     const int page_size = os::vm_page_size();
1160     const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size;
1161     const int start_page = native_call ? n_shadow_pages : 1;
1162     for (int pages = start_page; pages <= n_shadow_pages; pages++) {
1163       __ bang_stack_with_offset(pages*page_size);
1164     }
1165   }
1166 }
1167 
1168 //
1169 // Interpreter stub for calling a native method. (asm interpreter)
1170 // This sets up a somewhat different looking stack for calling the native method
1171 // than the typical interpreter frame setup.
1172 //
1173 
1174 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1175   address entry = __ pc();
1176 
1177   // the following temporary registers are used during frame creation
1178   const Register Gtmp1 = G3_scratch ;
1179   const Register Gtmp2 = G1_scratch;
1180   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1181 
1182   // make sure registers are different!
1183   assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2);
1184 
1185   const Address Laccess_flags(Lmethod, Method::access_flags_offset());
1186 
1187   const Register Glocals_size = G3;
1188   assert_different_registers(Glocals_size, G4_scratch, Gframe_size);
1189 
1190   // make sure method is native & not abstract
1191   // rethink these assertions - they can be simplified and shared (gri 2/25/2000)
1192 #ifdef ASSERT
1193   __ ld(G5_method, Method::access_flags_offset(), Gtmp1);
1194   {
1195     Label L;
1196     __ btst(JVM_ACC_NATIVE, Gtmp1);
1197     __ br(Assembler::notZero, false, Assembler::pt, L);
1198     __ delayed()->nop();
1199     __ stop("tried to execute non-native method as native");
1200     __ bind(L);
1201   }
1202   { Label L;
1203     __ btst(JVM_ACC_ABSTRACT, Gtmp1);
1204     __ br(Assembler::zero, false, Assembler::pt, L);
1205     __ delayed()->nop();
1206     __ stop("tried to execute abstract method as non-abstract");
1207     __ bind(L);
1208   }
1209 #endif // ASSERT
1210 
1211  // generate the code to allocate the interpreter stack frame
1212   generate_fixed_frame(true);
1213 
1214   //
1215   // No locals to initialize for native method
1216   //
1217 
1218   // this slot will be set later, we initialize it to null here just in
1219   // case we get a GC before the actual value is stored later
1220   __ st_ptr(G0, FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS);
1221 
1222   const Address do_not_unlock_if_synchronized(G2_thread,
1223     JavaThread::do_not_unlock_if_synchronized_offset());
1224   // Since at this point in the method invocation the exception handler
1225   // would try to exit the monitor of synchronized methods which hasn't
1226   // been entered yet, we set the thread local variable
1227   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1228   // runtime, exception handling i.e. unlock_if_synchronized_method will
1229   // check this thread local flag.
1230   // This flag has two effects, one is to force an unwind in the topmost
1231   // interpreter frame and not perform an unlock while doing so.
1232 
1233   __ movbool(true, G3_scratch);
1234   __ stbool(G3_scratch, do_not_unlock_if_synchronized);
1235 
1236   // increment invocation counter and check for overflow
1237   //
1238   // Note: checking for negative value instead of overflow
1239   //       so we have a 'sticky' overflow test (may be of
1240   //       importance as soon as we have true MT/MP)
1241   Label invocation_counter_overflow;
1242   Label Lcontinue;
1243   if (inc_counter) {
1244     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1245 
1246   }
1247   __ bind(Lcontinue);
1248 
1249   bang_stack_shadow_pages(true);
1250 
1251   // reset the _do_not_unlock_if_synchronized flag
1252   __ stbool(G0, do_not_unlock_if_synchronized);
1253 
1254   // check for synchronized methods
1255   // Must happen AFTER invocation_counter check and stack overflow check,
1256   // so method is not locked if overflows.
1257 
1258   if (synchronized) {
1259     lock_method();
1260   } else {
1261 #ifdef ASSERT
1262     { Label ok;
1263       __ ld(Laccess_flags, O0);
1264       __ btst(JVM_ACC_SYNCHRONIZED, O0);
1265       __ br( Assembler::zero, false, Assembler::pt, ok);
1266       __ delayed()->nop();
1267       __ stop("method needs synchronization");
1268       __ bind(ok);
1269     }
1270 #endif // ASSERT
1271   }
1272 
1273 
1274   // start execution
1275   __ verify_thread();
1276 
1277   // JVMTI support
1278   __ notify_method_entry();
1279 
1280   // native call
1281 
1282   // (note that O0 is never an oop--at most it is a handle)
1283   // It is important not to smash any handles created by this call,
1284   // until any oop handle in O0 is dereferenced.
1285 
1286   // (note that the space for outgoing params is preallocated)
1287 
1288   // get signature handler
1289   { Label L;
1290     Address signature_handler(Lmethod, Method::signature_handler_offset());
1291     __ ld_ptr(signature_handler, G3_scratch);
1292     __ br_notnull_short(G3_scratch, Assembler::pt, L);
1293     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), Lmethod);
1294     __ ld_ptr(signature_handler, G3_scratch);
1295     __ bind(L);
1296   }
1297 
1298   // Push a new frame so that the args will really be stored in
1299   // Copy a few locals across so the new frame has the variables
1300   // we need but these values will be dead at the jni call and
1301   // therefore not gc volatile like the values in the current
1302   // frame (Lmethod in particular)
1303 
1304   // Flush the method pointer to the register save area
1305   __ st_ptr(Lmethod, SP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS);
1306   __ mov(Llocals, O1);
1307 
1308   // calculate where the mirror handle body is allocated in the interpreter frame:
1309   __ add(FP, (frame::interpreter_frame_oop_temp_offset * wordSize) + STACK_BIAS, O2);
1310 
1311   // Calculate current frame size
1312   __ sub(SP, FP, O3);         // Calculate negative of current frame size
1313   __ save(SP, O3, SP);        // Allocate an identical sized frame
1314 
1315   // Note I7 has leftover trash. Slow signature handler will fill it in
1316   // should we get there. Normal jni call will set reasonable last_Java_pc
1317   // below (and fix I7 so the stack trace doesn't have a meaningless frame
1318   // in it).
1319 
1320   // Load interpreter frame's Lmethod into same register here
1321 
1322   __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod);
1323 
1324   __ mov(I1, Llocals);
1325   __ mov(I2, Lscratch2);     // save the address of the mirror
1326 
1327 
1328   // ONLY Lmethod and Llocals are valid here!
1329 
1330   // call signature handler, It will move the arg properly since Llocals in current frame
1331   // matches that in outer frame
1332 
1333   __ callr(G3_scratch, 0);
1334   __ delayed()->nop();
1335 
1336   // Result handler is in Lscratch
1337 
1338   // Reload interpreter frame's Lmethod since slow signature handler may block
1339   __ ld_ptr(FP, (Lmethod->sp_offset_in_saved_window() * wordSize) + STACK_BIAS, Lmethod);
1340 
1341   { Label not_static;
1342 
1343     __ ld(Laccess_flags, O0);
1344     __ btst(JVM_ACC_STATIC, O0);
1345     __ br( Assembler::zero, false, Assembler::pt, not_static);
1346     // get native function entry point(O0 is a good temp until the very end)
1347     __ delayed()->ld_ptr(Lmethod, in_bytes(Method::native_function_offset()), O0);
1348     // for static methods insert the mirror argument
1349     __ load_mirror(O1, Lmethod);
1350 #ifdef ASSERT
1351     if (!PrintSignatureHandlers)  // do not dirty the output with this
1352     { Label L;
1353       __ br_notnull_short(O1, Assembler::pt, L);
1354       __ stop("mirror is missing");
1355       __ bind(L);
1356     }
1357 #endif // ASSERT
1358     __ st_ptr(O1, Lscratch2, 0);
1359     __ mov(Lscratch2, O1);
1360     __ bind(not_static);
1361   }
1362 
1363   // At this point, arguments have been copied off of stack into
1364   // their JNI positions, which are O1..O5 and SP[68..].
1365   // Oops are boxed in-place on the stack, with handles copied to arguments.
1366   // The result handler is in Lscratch.  O0 will shortly hold the JNIEnv*.
1367 
1368 #ifdef ASSERT
1369   { Label L;
1370     __ br_notnull_short(O0, Assembler::pt, L);
1371     __ stop("native entry point is missing");
1372     __ bind(L);
1373   }
1374 #endif // ASSERT
1375 
1376   //
1377   // setup the frame anchor
1378   //
1379   // The scavenge function only needs to know that the PC of this frame is
1380   // in the interpreter method entry code, it doesn't need to know the exact
1381   // PC and hence we can use O7 which points to the return address from the
1382   // previous call in the code stream (signature handler function)
1383   //
1384   // The other trick is we set last_Java_sp to FP instead of the usual SP because
1385   // we have pushed the extra frame in order to protect the volatile register(s)
1386   // in that frame when we return from the jni call
1387   //
1388 
1389   __ set_last_Java_frame(FP, O7);
1390   __ mov(O7, I7);  // make dummy interpreter frame look like one above,
1391                    // not meaningless information that'll confuse me.
1392 
1393   // flush the windows now. We don't care about the current (protection) frame
1394   // only the outer frames
1395 
1396   __ flushw();
1397 
1398   // mark windows as flushed
1399   Address flags(G2_thread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::flags_offset());
1400   __ set(JavaFrameAnchor::flushed, G3_scratch);
1401   __ st(G3_scratch, flags);
1402 
1403   // Transition from _thread_in_Java to _thread_in_native. We are already safepoint ready.
1404 
1405   Address thread_state(G2_thread, JavaThread::thread_state_offset());
1406 #ifdef ASSERT
1407   { Label L;
1408     __ ld(thread_state, G3_scratch);
1409     __ cmp_and_br_short(G3_scratch, _thread_in_Java, Assembler::equal, Assembler::pt, L);
1410     __ stop("Wrong thread state in native stub");
1411     __ bind(L);
1412   }
1413 #endif // ASSERT
1414   __ set(_thread_in_native, G3_scratch);
1415   __ st(G3_scratch, thread_state);
1416 
1417   // Call the jni method, using the delay slot to set the JNIEnv* argument.
1418   __ save_thread(L7_thread_cache); // save Gthread
1419   __ callr(O0, 0);
1420   __ delayed()->
1421      add(L7_thread_cache, in_bytes(JavaThread::jni_environment_offset()), O0);
1422 
1423   // Back from jni method Lmethod in this frame is DEAD, DEAD, DEAD
1424 
1425   __ restore_thread(L7_thread_cache); // restore G2_thread
1426   __ reinit_heapbase();
1427 
1428   // must we block?
1429 
1430   // Block, if necessary, before resuming in _thread_in_Java state.
1431   // In order for GC to work, don't clear the last_Java_sp until after blocking.
1432   { Label no_block;
1433     AddressLiteral sync_state(SafepointSynchronize::address_of_state());
1434 
1435     // Switch thread to "native transition" state before reading the synchronization state.
1436     // This additional state is necessary because reading and testing the synchronization
1437     // state is not atomic w.r.t. GC, as this scenario demonstrates:
1438     //     Java thread A, in _thread_in_native state, loads _not_synchronized and is preempted.
1439     //     VM thread changes sync state to synchronizing and suspends threads for GC.
1440     //     Thread A is resumed to finish this native method, but doesn't block here since it
1441     //     didn't see any synchronization is progress, and escapes.
1442     __ set(_thread_in_native_trans, G3_scratch);
1443     __ st(G3_scratch, thread_state);
1444     if(os::is_MP()) {
1445       if (UseMembar) {
1446         // Force this write out before the read below
1447         __ membar(Assembler::StoreLoad);
1448       } else {
1449         // Write serialization page so VM thread can do a pseudo remote membar.
1450         // We use the current thread pointer to calculate a thread specific
1451         // offset to write to within the page. This minimizes bus traffic
1452         // due to cache line collision.
1453         __ serialize_memory(G2_thread, G1_scratch, G3_scratch);
1454       }
1455     }
1456     __ load_contents(sync_state, G3_scratch);
1457     __ cmp(G3_scratch, SafepointSynchronize::_not_synchronized);
1458 
1459     Label L;
1460     __ br(Assembler::notEqual, false, Assembler::pn, L);
1461     __ delayed()->ld(G2_thread, JavaThread::suspend_flags_offset(), G3_scratch);
1462     __ cmp_and_br_short(G3_scratch, 0, Assembler::equal, Assembler::pt, no_block);
1463     __ bind(L);
1464 
1465     // Block.  Save any potential method result value before the operation and
1466     // use a leaf call to leave the last_Java_frame setup undisturbed.
1467     save_native_result();
1468     __ call_VM_leaf(L7_thread_cache,
1469                     CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1470                     G2_thread);
1471 
1472     // Restore any method result value
1473     restore_native_result();
1474     __ bind(no_block);
1475   }
1476 
1477   // Clear the frame anchor now
1478 
1479   __ reset_last_Java_frame();
1480 
1481   // Move the result handler address
1482   __ mov(Lscratch, G3_scratch);
1483   // return possible result to the outer frame
1484 #ifndef __LP64
1485   __ mov(O0, I0);
1486   __ restore(O1, G0, O1);
1487 #else
1488   __ restore(O0, G0, O0);
1489 #endif /* __LP64 */
1490 
1491   // Move result handler to expected register
1492   __ mov(G3_scratch, Lscratch);
1493 
1494   // Back in normal (native) interpreter frame. State is thread_in_native_trans
1495   // switch to thread_in_Java.
1496 
1497   __ set(_thread_in_Java, G3_scratch);
1498   __ st(G3_scratch, thread_state);
1499 
1500   if (CheckJNICalls) {
1501     // clear_pending_jni_exception_check
1502     __ st_ptr(G0, G2_thread, JavaThread::pending_jni_exception_check_fn_offset());
1503   }
1504 
1505   // reset handle block
1506   __ ld_ptr(G2_thread, JavaThread::active_handles_offset(), G3_scratch);
1507   __ st(G0, G3_scratch, JNIHandleBlock::top_offset_in_bytes());
1508 
1509   // If we have an oop result store it where it will be safe for any further gc
1510   // until we return now that we've released the handle it might be protected by
1511 
1512   {
1513     Label no_oop, store_result;
1514 
1515     __ set((intptr_t)AbstractInterpreter::result_handler(T_OBJECT), G3_scratch);
1516     __ cmp_and_brx_short(G3_scratch, Lscratch, Assembler::notEqual, Assembler::pt, no_oop);
1517     // Unbox oop result, e.g. JNIHandles::resolve value in O0.
1518     __ br_null(O0, false, Assembler::pn, store_result); // Use NULL as-is.
1519     __ delayed()->andcc(O0, JNIHandles::weak_tag_mask, G0); // Test for jweak
1520     __ brx(Assembler::zero, true, Assembler::pt, store_result);
1521     __ delayed()->ld_ptr(O0, 0, O0); // Maybe resolve (untagged) jobject.
1522     // Resolve jweak.
1523     __ ld_ptr(O0, -JNIHandles::weak_tag_value, O0);
1524 #if INCLUDE_ALL_GCS
1525     if (UseG1GC) {
1526       __ g1_write_barrier_pre(noreg /* obj */,
1527                               noreg /* index */,
1528                               0 /* offset */,
1529                               O0 /* pre_val */,
1530                               G3_scratch /* tmp */,
1531                               true /* preserve_o_regs */);
1532     }
1533 #endif // INCLUDE_ALL_GCS
1534     __ bind(store_result);
1535     // Store it where gc will look for it and result handler expects it.
1536     __ st_ptr(O0, FP, (frame::interpreter_frame_oop_temp_offset*wordSize) + STACK_BIAS);
1537 
1538     __ bind(no_oop);
1539 
1540   }
1541 
1542 
1543   // handle exceptions (exception handling will handle unlocking!)
1544   { Label L;
1545     Address exception_addr(G2_thread, Thread::pending_exception_offset());
1546     __ ld_ptr(exception_addr, Gtemp);
1547     __ br_null_short(Gtemp, Assembler::pt, L);
1548     // Note: This could be handled more efficiently since we know that the native
1549     //       method doesn't have an exception handler. We could directly return
1550     //       to the exception handler for the caller.
1551     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1552     __ should_not_reach_here();
1553     __ bind(L);
1554   }
1555 
1556   // JVMTI support (preserves thread register)
1557   __ notify_method_exit(true, ilgl, InterpreterMacroAssembler::NotifyJVMTI);
1558 
1559   if (synchronized) {
1560     // save and restore any potential method result value around the unlocking operation
1561     save_native_result();
1562 
1563     __ add( __ top_most_monitor(), O1);
1564     __ unlock_object(O1);
1565 
1566     restore_native_result();
1567   }
1568 
1569 #if defined(COMPILER2) && !defined(_LP64)
1570 
1571   // C2 expects long results in G1 we can't tell if we're returning to interpreted
1572   // or compiled so just be safe.
1573 
1574   __ sllx(O0, 32, G1);          // Shift bits into high G1
1575   __ srl (O1, 0, O1);           // Zero extend O1
1576   __ or3 (O1, G1, G1);          // OR 64 bits into G1
1577 
1578 #endif /* COMPILER2 && !_LP64 */
1579 
1580   // dispose of return address and remove activation
1581 #ifdef ASSERT
1582   {
1583     Label ok;
1584     __ cmp_and_brx_short(I5_savedSP, FP, Assembler::greaterEqualUnsigned, Assembler::pt, ok);
1585     __ stop("bad I5_savedSP value");
1586     __ should_not_reach_here();
1587     __ bind(ok);
1588   }
1589 #endif
1590   __ jmp(Lscratch, 0);
1591   __ delayed()->nop();
1592 
1593 
1594   if (inc_counter) {
1595     // handle invocation counter overflow
1596     __ bind(invocation_counter_overflow);
1597     generate_counter_overflow(Lcontinue);
1598   }
1599 
1600 
1601 
1602   return entry;
1603 }
1604 
1605 
1606 // Generic method entry to (asm) interpreter
1607 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1608   address entry = __ pc();
1609 
1610   bool inc_counter  = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1611 
1612   // the following temporary registers are used during frame creation
1613   const Register Gtmp1 = G3_scratch ;
1614   const Register Gtmp2 = G1_scratch;
1615 
1616   // make sure registers are different!
1617   assert_different_registers(G2_thread, G5_method, Gargs, Gtmp1, Gtmp2);
1618 
1619   const Address constMethod       (G5_method, Method::const_offset());
1620   // Seems like G5_method is live at the point this is used. So we could make this look consistent
1621   // and use in the asserts.
1622   const Address access_flags      (Lmethod,   Method::access_flags_offset());
1623 
1624   const Register Glocals_size = G3;
1625   assert_different_registers(Glocals_size, G4_scratch, Gframe_size);
1626 
1627   // make sure method is not native & not abstract
1628   // rethink these assertions - they can be simplified and shared (gri 2/25/2000)
1629 #ifdef ASSERT
1630   __ ld(G5_method, Method::access_flags_offset(), Gtmp1);
1631   {
1632     Label L;
1633     __ btst(JVM_ACC_NATIVE, Gtmp1);
1634     __ br(Assembler::zero, false, Assembler::pt, L);
1635     __ delayed()->nop();
1636     __ stop("tried to execute native method as non-native");
1637     __ bind(L);
1638   }
1639   { Label L;
1640     __ btst(JVM_ACC_ABSTRACT, Gtmp1);
1641     __ br(Assembler::zero, false, Assembler::pt, L);
1642     __ delayed()->nop();
1643     __ stop("tried to execute abstract method as non-abstract");
1644     __ bind(L);
1645   }
1646 #endif // ASSERT
1647 
1648   // generate the code to allocate the interpreter stack frame
1649 
1650   generate_fixed_frame(false);
1651 
1652 #ifdef FAST_DISPATCH
1653   __ set((intptr_t)Interpreter::dispatch_table(), IdispatchTables);
1654                                           // set bytecode dispatch table base
1655 #endif
1656 
1657   //
1658   // Code to initialize the extra (i.e. non-parm) locals
1659   //
1660   Register init_value = noreg;    // will be G0 if we must clear locals
1661   // The way the code was setup before zerolocals was always true for vanilla java entries.
1662   // It could only be false for the specialized entries like accessor or empty which have
1663   // no extra locals so the testing was a waste of time and the extra locals were always
1664   // initialized. We removed this extra complication to already over complicated code.
1665 
1666   init_value = G0;
1667   Label clear_loop;
1668 
1669   const Register RconstMethod = O1;
1670   const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
1671   const Address size_of_locals    (RconstMethod, ConstMethod::size_of_locals_offset());
1672 
1673   // NOTE: If you change the frame layout, this code will need to
1674   // be updated!
1675   __ ld_ptr( constMethod, RconstMethod );
1676   __ lduh( size_of_locals, O2 );
1677   __ lduh( size_of_parameters, O1 );
1678   __ sll( O2, Interpreter::logStackElementSize, O2);
1679   __ sll( O1, Interpreter::logStackElementSize, O1 );
1680   __ sub( Llocals, O2, O2 );
1681   __ sub( Llocals, O1, O1 );
1682 
1683   __ bind( clear_loop );
1684   __ inc( O2, wordSize );
1685 
1686   __ cmp( O2, O1 );
1687   __ brx( Assembler::lessEqualUnsigned, true, Assembler::pt, clear_loop );
1688   __ delayed()->st_ptr( init_value, O2, 0 );
1689 
1690   const Address do_not_unlock_if_synchronized(G2_thread,
1691     JavaThread::do_not_unlock_if_synchronized_offset());
1692   // Since at this point in the method invocation the exception handler
1693   // would try to exit the monitor of synchronized methods which hasn't
1694   // been entered yet, we set the thread local variable
1695   // _do_not_unlock_if_synchronized to true. If any exception was thrown by
1696   // runtime, exception handling i.e. unlock_if_synchronized_method will
1697   // check this thread local flag.
1698   __ movbool(true, G3_scratch);
1699   __ stbool(G3_scratch, do_not_unlock_if_synchronized);
1700 
1701   __ profile_parameters_type(G1_scratch, G3_scratch, G4_scratch, Lscratch);
1702   // increment invocation counter and check for overflow
1703   //
1704   // Note: checking for negative value instead of overflow
1705   //       so we have a 'sticky' overflow test (may be of
1706   //       importance as soon as we have true MT/MP)
1707   Label invocation_counter_overflow;
1708   Label profile_method;
1709   Label profile_method_continue;
1710   Label Lcontinue;
1711   if (inc_counter) {
1712     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1713     if (ProfileInterpreter) {
1714       __ bind(profile_method_continue);
1715     }
1716   }
1717   __ bind(Lcontinue);
1718 
1719   bang_stack_shadow_pages(false);
1720 
1721   // reset the _do_not_unlock_if_synchronized flag
1722   __ stbool(G0, do_not_unlock_if_synchronized);
1723 
1724   // check for synchronized methods
1725   // Must happen AFTER invocation_counter check and stack overflow check,
1726   // so method is not locked if overflows.
1727 
1728   if (synchronized) {
1729     lock_method();
1730   } else {
1731 #ifdef ASSERT
1732     { Label ok;
1733       __ ld(access_flags, O0);
1734       __ btst(JVM_ACC_SYNCHRONIZED, O0);
1735       __ br( Assembler::zero, false, Assembler::pt, ok);
1736       __ delayed()->nop();
1737       __ stop("method needs synchronization");
1738       __ bind(ok);
1739     }
1740 #endif // ASSERT
1741   }
1742 
1743   // start execution
1744 
1745   __ verify_thread();
1746 
1747   // jvmti support
1748   __ notify_method_entry();
1749 
1750   // start executing instructions
1751   __ dispatch_next(vtos);
1752 
1753 
1754   if (inc_counter) {
1755     if (ProfileInterpreter) {
1756       // We have decided to profile this method in the interpreter
1757       __ bind(profile_method);
1758 
1759       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1760       __ set_method_data_pointer_for_bcp();
1761       __ ba_short(profile_method_continue);
1762     }
1763 
1764     // handle invocation counter overflow
1765     __ bind(invocation_counter_overflow);
1766     generate_counter_overflow(Lcontinue);
1767   }
1768 
1769 
1770   return entry;
1771 }
1772 
1773 //----------------------------------------------------------------------------------------------------
1774 // Exceptions
1775 void TemplateInterpreterGenerator::generate_throw_exception() {
1776 
1777   // Entry point in previous activation (i.e., if the caller was interpreted)
1778   Interpreter::_rethrow_exception_entry = __ pc();
1779   // O0: exception
1780 
1781   // entry point for exceptions thrown within interpreter code
1782   Interpreter::_throw_exception_entry = __ pc();
1783   __ verify_thread();
1784   // expression stack is undefined here
1785   // O0: exception, i.e. Oexception
1786   // Lbcp: exception bcp
1787   __ verify_oop(Oexception);
1788 
1789 
1790   // expression stack must be empty before entering the VM in case of an exception
1791   __ empty_expression_stack();
1792   // find exception handler address and preserve exception oop
1793   // call C routine to find handler and jump to it
1794   __ call_VM(O1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Oexception);
1795   __ push_ptr(O1); // push exception for exception handler bytecodes
1796 
1797   __ JMP(O0, 0); // jump to exception handler (may be remove activation entry!)
1798   __ delayed()->nop();
1799 
1800 
1801   // if the exception is not handled in the current frame
1802   // the frame is removed and the exception is rethrown
1803   // (i.e. exception continuation is _rethrow_exception)
1804   //
1805   // Note: At this point the bci is still the bxi for the instruction which caused
1806   //       the exception and the expression stack is empty. Thus, for any VM calls
1807   //       at this point, GC will find a legal oop map (with empty expression stack).
1808 
1809   // in current activation
1810   // tos: exception
1811   // Lbcp: exception bcp
1812 
1813   //
1814   // JVMTI PopFrame support
1815   //
1816 
1817   Interpreter::_remove_activation_preserving_args_entry = __ pc();
1818   Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset());
1819   // Set the popframe_processing bit in popframe_condition indicating that we are
1820   // currently handling popframe, so that call_VMs that may happen later do not trigger new
1821   // popframe handling cycles.
1822 
1823   __ ld(popframe_condition_addr, G3_scratch);
1824   __ or3(G3_scratch, JavaThread::popframe_processing_bit, G3_scratch);
1825   __ stw(G3_scratch, popframe_condition_addr);
1826 
1827   // Empty the expression stack, as in normal exception handling
1828   __ empty_expression_stack();
1829   __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
1830 
1831   {
1832     // Check to see whether we are returning to a deoptimized frame.
1833     // (The PopFrame call ensures that the caller of the popped frame is
1834     // either interpreted or compiled and deoptimizes it if compiled.)
1835     // In this case, we can't call dispatch_next() after the frame is
1836     // popped, but instead must save the incoming arguments and restore
1837     // them after deoptimization has occurred.
1838     //
1839     // Note that we don't compare the return PC against the
1840     // deoptimization blob's unpack entry because of the presence of
1841     // adapter frames in C2.
1842     Label caller_not_deoptimized;
1843     __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), I7);
1844     __ br_notnull_short(O0, Assembler::pt, caller_not_deoptimized);
1845 
1846     const Register Gtmp1 = G3_scratch;
1847     const Register Gtmp2 = G1_scratch;
1848     const Register RconstMethod = Gtmp1;
1849     const Address constMethod(Lmethod, Method::const_offset());
1850     const Address size_of_parameters(RconstMethod, ConstMethod::size_of_parameters_offset());
1851 
1852     // Compute size of arguments for saving when returning to deoptimized caller
1853     __ ld_ptr(constMethod, RconstMethod);
1854     __ lduh(size_of_parameters, Gtmp1);
1855     __ sll(Gtmp1, Interpreter::logStackElementSize, Gtmp1);
1856     __ sub(Llocals, Gtmp1, Gtmp2);
1857     __ add(Gtmp2, wordSize, Gtmp2);
1858     // Save these arguments
1859     __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), G2_thread, Gtmp1, Gtmp2);
1860     // Inform deoptimization that it is responsible for restoring these arguments
1861     __ set(JavaThread::popframe_force_deopt_reexecution_bit, Gtmp1);
1862     Address popframe_condition_addr(G2_thread, JavaThread::popframe_condition_offset());
1863     __ st(Gtmp1, popframe_condition_addr);
1864 
1865     // Return from the current method
1866     // The caller's SP was adjusted upon method entry to accomodate
1867     // the callee's non-argument locals. Undo that adjustment.
1868     __ ret();
1869     __ delayed()->restore(I5_savedSP, G0, SP);
1870 
1871     __ bind(caller_not_deoptimized);
1872   }
1873 
1874   // Clear the popframe condition flag
1875   __ stw(G0 /* popframe_inactive */, popframe_condition_addr);
1876 
1877   // Get out of the current method (how this is done depends on the particular compiler calling
1878   // convention that the interpreter currently follows)
1879   // The caller's SP was adjusted upon method entry to accomodate
1880   // the callee's non-argument locals. Undo that adjustment.
1881   __ restore(I5_savedSP, G0, SP);
1882   // The method data pointer was incremented already during
1883   // call profiling. We have to restore the mdp for the current bcp.
1884   if (ProfileInterpreter) {
1885     __ set_method_data_pointer_for_bcp();
1886   }
1887 
1888 #if INCLUDE_JVMTI
1889   {
1890     Label L_done;
1891 
1892     __ ldub(Address(Lbcp, 0), G1_scratch);  // Load current bytecode
1893     __ cmp_and_br_short(G1_scratch, Bytecodes::_invokestatic, Assembler::notEqual, Assembler::pn, L_done);
1894 
1895     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1896     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1897 
1898     __ call_VM(G1_scratch, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), I0, Lmethod, Lbcp);
1899 
1900     __ br_null(G1_scratch, false, Assembler::pn, L_done);
1901     __ delayed()->nop();
1902 
1903     __ st_ptr(G1_scratch, Lesp, wordSize);
1904     __ bind(L_done);
1905   }
1906 #endif // INCLUDE_JVMTI
1907 
1908   // Resume bytecode interpretation at the current bcp
1909   __ dispatch_next(vtos);
1910   // end of JVMTI PopFrame support
1911 
1912   Interpreter::_remove_activation_entry = __ pc();
1913 
1914   // preserve exception over this code sequence (remove activation calls the vm, but oopmaps are not correct here)
1915   __ pop_ptr(Oexception);                                  // get exception
1916 
1917   // Intel has the following comment:
1918   //// remove the activation (without doing throws on illegalMonitorExceptions)
1919   // They remove the activation without checking for bad monitor state.
1920   // %%% We should make sure this is the right semantics before implementing.
1921 
1922   __ set_vm_result(Oexception);
1923   __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false);
1924 
1925   __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI);
1926 
1927   __ get_vm_result(Oexception);
1928   __ verify_oop(Oexception);
1929 
1930     const int return_reg_adjustment = frame::pc_return_offset;
1931   Address issuing_pc_addr(I7, return_reg_adjustment);
1932 
1933   // We are done with this activation frame; find out where to go next.
1934   // The continuation point will be an exception handler, which expects
1935   // the following registers set up:
1936   //
1937   // Oexception: exception
1938   // Oissuing_pc: the local call that threw exception
1939   // Other On: garbage
1940   // In/Ln:  the contents of the caller's register window
1941   //
1942   // We do the required restore at the last possible moment, because we
1943   // need to preserve some state across a runtime call.
1944   // (Remember that the caller activation is unknown--it might not be
1945   // interpreted, so things like Lscratch are useless in the caller.)
1946 
1947   // Although the Intel version uses call_C, we can use the more
1948   // compact call_VM.  (The only real difference on SPARC is a
1949   // harmlessly ignored [re]set_last_Java_frame, compared with
1950   // the Intel code which lacks this.)
1951   __ mov(Oexception,      Oexception ->after_save());  // get exception in I0 so it will be on O0 after restore
1952   __ add(issuing_pc_addr, Oissuing_pc->after_save());  // likewise set I1 to a value local to the caller
1953   __ super_call_VM_leaf(L7_thread_cache,
1954                         CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
1955                         G2_thread, Oissuing_pc->after_save());
1956 
1957   // The caller's SP was adjusted upon method entry to accomodate
1958   // the callee's non-argument locals. Undo that adjustment.
1959   __ JMP(O0, 0);                         // return exception handler in caller
1960   __ delayed()->restore(I5_savedSP, G0, SP);
1961 
1962   // (same old exception object is already in Oexception; see above)
1963   // Note that an "issuing PC" is actually the next PC after the call
1964 }
1965 
1966 
1967 //
1968 // JVMTI ForceEarlyReturn support
1969 //
1970 
1971 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1972   address entry = __ pc();
1973 
1974   __ empty_expression_stack();
1975   __ load_earlyret_value(state);
1976 
1977   __ ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), G3_scratch);
1978   Address cond_addr(G3_scratch, JvmtiThreadState::earlyret_state_offset());
1979 
1980   // Clear the earlyret state
1981   __ stw(G0 /* JvmtiThreadState::earlyret_inactive */, cond_addr);
1982 
1983   __ remove_activation(state,
1984                        /* throw_monitor_exception */ false,
1985                        /* install_monitor_exception */ false);
1986 
1987   // The caller's SP was adjusted upon method entry to accomodate
1988   // the callee's non-argument locals. Undo that adjustment.
1989   __ ret();                             // return to caller
1990   __ delayed()->restore(I5_savedSP, G0, SP);
1991 
1992   return entry;
1993 } // end of JVMTI ForceEarlyReturn support
1994 
1995 
1996 //------------------------------------------------------------------------------------------------------------------------
1997 // Helper for vtos entry point generation
1998 
1999 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
2000   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2001   Label L;
2002   aep = __ pc(); __ push_ptr(); __ ba_short(L);
2003   fep = __ pc(); __ push_f();   __ ba_short(L);
2004   dep = __ pc(); __ push_d();   __ ba_short(L);
2005   lep = __ pc(); __ push_l();   __ ba_short(L);
2006   iep = __ pc(); __ push_i();
2007   bep = cep = sep = iep;                        // there aren't any
2008   vep = __ pc(); __ bind(L);                    // fall through
2009   generate_and_dispatch(t);
2010 }
2011 
2012 // --------------------------------------------------------------------------------
2013 
2014 // Non-product code
2015 #ifndef PRODUCT
2016 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2017   address entry = __ pc();
2018 
2019   __ push(state);
2020   __ mov(O7, Lscratch); // protect return address within interpreter
2021 
2022   // Pass a 0 (not used in sparc) and the top of stack to the bytecode tracer
2023   __ mov( Otos_l2, G3_scratch );
2024   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), G0, Otos_l1, G3_scratch);
2025   __ mov(Lscratch, O7); // restore return address
2026   __ pop(state);
2027   __ retl();
2028   __ delayed()->nop();
2029 
2030   return entry;
2031 }
2032 
2033 
2034 // helpers for generate_and_dispatch
2035 
2036 void TemplateInterpreterGenerator::count_bytecode() {
2037   __ inc_counter(&BytecodeCounter::_counter_value, G3_scratch, G4_scratch);
2038 }
2039 
2040 
2041 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2042   __ inc_counter(&BytecodeHistogram::_counters[t->bytecode()], G3_scratch, G4_scratch);
2043 }
2044 
2045 
2046 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2047   AddressLiteral index   (&BytecodePairHistogram::_index);
2048   AddressLiteral counters((address) &BytecodePairHistogram::_counters);
2049 
2050   // get index, shift out old bytecode, bring in new bytecode, and store it
2051   // _index = (_index >> log2_number_of_codes) |
2052   //          (bytecode << log2_number_of_codes);
2053 
2054   __ load_contents(index, G4_scratch);
2055   __ srl( G4_scratch, BytecodePairHistogram::log2_number_of_codes, G4_scratch );
2056   __ set( ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes,  G3_scratch );
2057   __ or3( G3_scratch,  G4_scratch, G4_scratch );
2058   __ store_contents(G4_scratch, index, G3_scratch);
2059 
2060   // bump bucket contents
2061   // _counters[_index] ++;
2062 
2063   __ set(counters, G3_scratch);                       // loads into G3_scratch
2064   __ sll( G4_scratch, LogBytesPerWord, G4_scratch );  // Index is word address
2065   __ add (G3_scratch, G4_scratch, G3_scratch);        // Add in index
2066   __ ld (G3_scratch, 0, G4_scratch);
2067   __ inc (G4_scratch);
2068   __ st (G4_scratch, 0, G3_scratch);
2069 }
2070 
2071 
2072 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2073   // Call a little run-time stub to avoid blow-up for each bytecode.
2074   // The run-time runtime saves the right registers, depending on
2075   // the tosca in-state for the given template.
2076   address entry = Interpreter::trace_code(t->tos_in());
2077   guarantee(entry != NULL, "entry must have been generated");
2078   __ call(entry, relocInfo::none);
2079   __ delayed()->nop();
2080 }
2081 
2082 
2083 void TemplateInterpreterGenerator::stop_interpreter_at() {
2084   AddressLiteral counter(&BytecodeCounter::_counter_value);
2085   __ load_contents(counter, G3_scratch);
2086   AddressLiteral stop_at(&StopInterpreterAt);
2087   __ load_ptr_contents(stop_at, G4_scratch);
2088   __ cmp(G3_scratch, G4_scratch);
2089   __ breakpoint_trap(Assembler::equal, Assembler::icc);
2090 }
2091 #endif // not PRODUCT