1 /*
   2  * Copyright (c) 1999, 2012, 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/codeBuffer.hpp"
  27 #include "c1/c1_CodeStubs.hpp"
  28 #include "c1/c1_Defs.hpp"
  29 #include "c1/c1_FrameMap.hpp"
  30 #include "c1/c1_LIRAssembler.hpp"
  31 #include "c1/c1_MacroAssembler.hpp"
  32 #include "c1/c1_Runtime1.hpp"
  33 #include "classfile/systemDictionary.hpp"
  34 #include "classfile/vmSymbols.hpp"
  35 #include "code/codeBlob.hpp"
  36 #include "code/compiledIC.hpp"
  37 #include "code/pcDesc.hpp"
  38 #include "code/scopeDesc.hpp"
  39 #include "code/vtableStubs.hpp"
  40 #include "compiler/disassembler.hpp"
  41 #include "gc_interface/collectedHeap.hpp"
  42 #include "interpreter/bytecode.hpp"
  43 #include "interpreter/interpreter.hpp"
  44 #include "memory/allocation.inline.hpp"
  45 #include "memory/barrierSet.hpp"
  46 #include "memory/oopFactory.hpp"
  47 #include "memory/resourceArea.hpp"
  48 #include "oops/objArrayKlass.hpp"
  49 #include "oops/oop.inline.hpp"
  50 #include "runtime/biasedLocking.hpp"
  51 #include "runtime/compilationPolicy.hpp"
  52 #include "runtime/interfaceSupport.hpp"
  53 #include "runtime/javaCalls.hpp"
  54 #include "runtime/sharedRuntime.hpp"
  55 #include "runtime/threadCritical.hpp"
  56 #include "runtime/vframe.hpp"
  57 #include "runtime/vframeArray.hpp"
  58 #include "utilities/copy.hpp"
  59 #include "utilities/events.hpp"
  60 
  61 
  62 // Implementation of StubAssembler
  63 
  64 StubAssembler::StubAssembler(CodeBuffer* code, const char * name, int stub_id) : C1_MacroAssembler(code) {
  65   _name = name;
  66   _must_gc_arguments = false;
  67   _frame_size = no_frame_size;
  68   _num_rt_args = 0;
  69   _stub_id = stub_id;
  70 }
  71 
  72 
  73 void StubAssembler::set_info(const char* name, bool must_gc_arguments) {
  74   _name = name;
  75   _must_gc_arguments = must_gc_arguments;
  76 }
  77 
  78 
  79 void StubAssembler::set_frame_size(int size) {
  80   if (_frame_size == no_frame_size) {
  81     _frame_size = size;
  82   }
  83   assert(_frame_size == size, "can't change the frame size");
  84 }
  85 
  86 
  87 void StubAssembler::set_num_rt_args(int args) {
  88   if (_num_rt_args == 0) {
  89     _num_rt_args = args;
  90   }
  91   assert(_num_rt_args == args, "can't change the number of args");
  92 }
  93 
  94 // Implementation of Runtime1
  95 
  96 CodeBlob* Runtime1::_blobs[Runtime1::number_of_ids];
  97 const char *Runtime1::_blob_names[] = {
  98   RUNTIME1_STUBS(STUB_NAME, LAST_STUB_NAME)
  99 };
 100 
 101 #ifndef PRODUCT
 102 // statistics
 103 int Runtime1::_generic_arraycopy_cnt = 0;
 104 int Runtime1::_primitive_arraycopy_cnt = 0;
 105 int Runtime1::_oop_arraycopy_cnt = 0;
 106 int Runtime1::_generic_arraycopystub_cnt = 0;
 107 int Runtime1::_arraycopy_slowcase_cnt = 0;
 108 int Runtime1::_arraycopy_checkcast_cnt = 0;
 109 int Runtime1::_arraycopy_checkcast_attempt_cnt = 0;
 110 int Runtime1::_new_type_array_slowcase_cnt = 0;
 111 int Runtime1::_new_object_array_slowcase_cnt = 0;
 112 int Runtime1::_new_instance_slowcase_cnt = 0;
 113 int Runtime1::_new_multi_array_slowcase_cnt = 0;
 114 int Runtime1::_monitorenter_slowcase_cnt = 0;
 115 int Runtime1::_monitorexit_slowcase_cnt = 0;
 116 int Runtime1::_patch_code_slowcase_cnt = 0;
 117 int Runtime1::_throw_range_check_exception_count = 0;
 118 int Runtime1::_throw_index_exception_count = 0;
 119 int Runtime1::_throw_div0_exception_count = 0;
 120 int Runtime1::_throw_null_pointer_exception_count = 0;
 121 int Runtime1::_throw_class_cast_exception_count = 0;
 122 int Runtime1::_throw_incompatible_class_change_error_count = 0;
 123 int Runtime1::_throw_array_store_exception_count = 0;
 124 int Runtime1::_throw_count = 0;
 125 
 126 static int _byte_arraycopy_cnt = 0;
 127 static int _short_arraycopy_cnt = 0;
 128 static int _int_arraycopy_cnt = 0;
 129 static int _long_arraycopy_cnt = 0;
 130 static int _oop_arraycopy_cnt = 0;
 131 
 132 address Runtime1::arraycopy_count_address(BasicType type) {
 133   switch (type) {
 134   case T_BOOLEAN:
 135   case T_BYTE:   return (address)&_byte_arraycopy_cnt;
 136   case T_CHAR:
 137   case T_SHORT:  return (address)&_short_arraycopy_cnt;
 138   case T_FLOAT:
 139   case T_INT:    return (address)&_int_arraycopy_cnt;
 140   case T_DOUBLE:
 141   case T_LONG:   return (address)&_long_arraycopy_cnt;
 142   case T_ARRAY:
 143   case T_OBJECT: return (address)&_oop_arraycopy_cnt;
 144   default:
 145     ShouldNotReachHere();
 146     return NULL;
 147   }
 148 }
 149 
 150 
 151 #endif
 152 
 153 // Simple helper to see if the caller of a runtime stub which
 154 // entered the VM has been deoptimized
 155 
 156 static bool caller_is_deopted() {
 157   JavaThread* thread = JavaThread::current();
 158   RegisterMap reg_map(thread, false);
 159   frame runtime_frame = thread->last_frame();
 160   frame caller_frame = runtime_frame.sender(&reg_map);
 161   assert(caller_frame.is_compiled_frame(), "must be compiled");
 162   return caller_frame.is_deoptimized_frame();
 163 }
 164 
 165 // Stress deoptimization
 166 static void deopt_caller() {
 167   if ( !caller_is_deopted()) {
 168     JavaThread* thread = JavaThread::current();
 169     RegisterMap reg_map(thread, false);
 170     frame runtime_frame = thread->last_frame();
 171     frame caller_frame = runtime_frame.sender(&reg_map);
 172     Deoptimization::deoptimize_frame(thread, caller_frame.id());
 173     assert(caller_is_deopted(), "Must be deoptimized");
 174   }
 175 }
 176 
 177 
 178 void Runtime1::generate_blob_for(BufferBlob* buffer_blob, StubID id) {
 179   assert(0 <= id && id < number_of_ids, "illegal stub id");
 180   ResourceMark rm;
 181   // create code buffer for code storage
 182   CodeBuffer code(buffer_blob);
 183 
 184   Compilation::setup_code_buffer(&code, 0);
 185 
 186   // create assembler for code generation
 187   StubAssembler* sasm = new StubAssembler(&code, name_for(id), id);
 188   // generate code for runtime stub
 189   OopMapSet* oop_maps;
 190   oop_maps = generate_code_for(id, sasm);
 191   assert(oop_maps == NULL || sasm->frame_size() != no_frame_size,
 192          "if stub has an oop map it must have a valid frame size");
 193 
 194 #ifdef ASSERT
 195   // Make sure that stubs that need oopmaps have them
 196   switch (id) {
 197     // These stubs don't need to have an oopmap
 198     case dtrace_object_alloc_id:
 199     case g1_pre_barrier_slow_id:
 200     case g1_post_barrier_slow_id:
 201     case slow_subtype_check_id:
 202     case fpu2long_stub_id:
 203     case unwind_exception_id:
 204     case counter_overflow_id:
 205 #if defined(SPARC) || defined(PPC)
 206     case handle_exception_nofpu_id:  // Unused on sparc
 207 #endif
 208       break;
 209 
 210     // All other stubs should have oopmaps
 211     default:
 212       assert(oop_maps != NULL, "must have an oopmap");
 213   }
 214 #endif
 215 
 216   // align so printing shows nop's instead of random code at the end (SimpleStubs are aligned)
 217   sasm->align(BytesPerWord);
 218   // make sure all code is in code buffer
 219   sasm->flush();
 220   // create blob - distinguish a few special cases
 221   CodeBlob* blob = RuntimeStub::new_runtime_stub(name_for(id),
 222                                                  &code,
 223                                                  CodeOffsets::frame_never_safe,
 224                                                  sasm->frame_size(),
 225                                                  oop_maps,
 226                                                  sasm->must_gc_arguments());
 227   // install blob
 228   assert(blob != NULL, "blob must exist");
 229   _blobs[id] = blob;
 230 }
 231 
 232 
 233 void Runtime1::initialize(BufferBlob* blob) {
 234   // platform-dependent initialization
 235   initialize_pd();
 236   // generate stubs
 237   for (int id = 0; id < number_of_ids; id++) generate_blob_for(blob, (StubID)id);
 238   // printing
 239 #ifndef PRODUCT
 240   if (PrintSimpleStubs) {
 241     ResourceMark rm;
 242     for (int id = 0; id < number_of_ids; id++) {
 243       _blobs[id]->print();
 244       if (_blobs[id]->oop_maps() != NULL) {
 245         _blobs[id]->oop_maps()->print();
 246       }
 247     }
 248   }
 249 #endif
 250 }
 251 
 252 
 253 CodeBlob* Runtime1::blob_for(StubID id) {
 254   assert(0 <= id && id < number_of_ids, "illegal stub id");
 255   return _blobs[id];
 256 }
 257 
 258 
 259 const char* Runtime1::name_for(StubID id) {
 260   assert(0 <= id && id < number_of_ids, "illegal stub id");
 261   return _blob_names[id];
 262 }
 263 
 264 const char* Runtime1::name_for_address(address entry) {
 265   for (int id = 0; id < number_of_ids; id++) {
 266     if (entry == entry_for((StubID)id)) return name_for((StubID)id);
 267   }
 268 
 269 #define FUNCTION_CASE(a, f) \
 270   if ((intptr_t)a == CAST_FROM_FN_PTR(intptr_t, f))  return #f
 271 
 272   FUNCTION_CASE(entry, os::javaTimeMillis);
 273   FUNCTION_CASE(entry, os::javaTimeNanos);
 274   FUNCTION_CASE(entry, SharedRuntime::OSR_migration_end);
 275   FUNCTION_CASE(entry, SharedRuntime::d2f);
 276   FUNCTION_CASE(entry, SharedRuntime::d2i);
 277   FUNCTION_CASE(entry, SharedRuntime::d2l);
 278   FUNCTION_CASE(entry, SharedRuntime::dcos);
 279   FUNCTION_CASE(entry, SharedRuntime::dexp);
 280   FUNCTION_CASE(entry, SharedRuntime::dlog);
 281   FUNCTION_CASE(entry, SharedRuntime::dlog10);
 282   FUNCTION_CASE(entry, SharedRuntime::dpow);
 283   FUNCTION_CASE(entry, SharedRuntime::drem);
 284   FUNCTION_CASE(entry, SharedRuntime::dsin);
 285   FUNCTION_CASE(entry, SharedRuntime::dtan);
 286   FUNCTION_CASE(entry, SharedRuntime::f2i);
 287   FUNCTION_CASE(entry, SharedRuntime::f2l);
 288   FUNCTION_CASE(entry, SharedRuntime::frem);
 289   FUNCTION_CASE(entry, SharedRuntime::l2d);
 290   FUNCTION_CASE(entry, SharedRuntime::l2f);
 291   FUNCTION_CASE(entry, SharedRuntime::ldiv);
 292   FUNCTION_CASE(entry, SharedRuntime::lmul);
 293   FUNCTION_CASE(entry, SharedRuntime::lrem);
 294   FUNCTION_CASE(entry, SharedRuntime::lrem);
 295   FUNCTION_CASE(entry, SharedRuntime::dtrace_method_entry);
 296   FUNCTION_CASE(entry, SharedRuntime::dtrace_method_exit);
 297   FUNCTION_CASE(entry, is_instance_of);
 298   FUNCTION_CASE(entry, trace_block_entry);
 299 #ifdef TRACE_HAVE_INTRINSICS
 300   FUNCTION_CASE(entry, TRACE_TIME_METHOD);
 301 #endif
 302 
 303 #undef FUNCTION_CASE
 304 
 305   // Soft float adds more runtime names.
 306   return pd_name_for_address(entry);
 307 }
 308 
 309 
 310 JRT_ENTRY(void, Runtime1::new_instance(JavaThread* thread, Klass* klass))
 311   NOT_PRODUCT(_new_instance_slowcase_cnt++;)
 312 
 313   assert(klass->is_klass(), "not a class");
 314   instanceKlassHandle h(thread, klass);
 315   h->check_valid_for_instantiation(true, CHECK);
 316   // make sure klass is initialized
 317   h->initialize(CHECK);
 318   // allocate instance and return via TLS
 319   oop obj = h->allocate_instance(CHECK);
 320   thread->set_vm_result(obj);
 321 JRT_END
 322 
 323 
 324 JRT_ENTRY(void, Runtime1::new_type_array(JavaThread* thread, Klass* klass, jint length))
 325   NOT_PRODUCT(_new_type_array_slowcase_cnt++;)
 326   // Note: no handle for klass needed since they are not used
 327   //       anymore after new_typeArray() and no GC can happen before.
 328   //       (This may have to change if this code changes!)
 329   assert(klass->is_klass(), "not a class");
 330   BasicType elt_type = TypeArrayKlass::cast(klass)->element_type();
 331   oop obj = oopFactory::new_typeArray(elt_type, length, CHECK);
 332   thread->set_vm_result(obj);
 333   // This is pretty rare but this runtime patch is stressful to deoptimization
 334   // if we deoptimize here so force a deopt to stress the path.
 335   if (DeoptimizeALot) {
 336     deopt_caller();
 337   }
 338 
 339 JRT_END
 340 
 341 
 342 JRT_ENTRY(void, Runtime1::new_object_array(JavaThread* thread, Klass* array_klass, jint length))
 343   NOT_PRODUCT(_new_object_array_slowcase_cnt++;)
 344 
 345   // Note: no handle for klass needed since they are not used
 346   //       anymore after new_objArray() and no GC can happen before.
 347   //       (This may have to change if this code changes!)
 348   assert(array_klass->is_klass(), "not a class");
 349   Klass* elem_klass = ObjArrayKlass::cast(array_klass)->element_klass();
 350   objArrayOop obj = oopFactory::new_objArray(elem_klass, length, CHECK);
 351   thread->set_vm_result(obj);
 352   // This is pretty rare but this runtime patch is stressful to deoptimization
 353   // if we deoptimize here so force a deopt to stress the path.
 354   if (DeoptimizeALot) {
 355     deopt_caller();
 356   }
 357 JRT_END
 358 
 359 
 360 JRT_ENTRY(void, Runtime1::new_multi_array(JavaThread* thread, Klass* klass, int rank, jint* dims))
 361   NOT_PRODUCT(_new_multi_array_slowcase_cnt++;)
 362 
 363   assert(klass->is_klass(), "not a class");
 364   assert(rank >= 1, "rank must be nonzero");
 365   oop obj = ArrayKlass::cast(klass)->multi_allocate(rank, dims, CHECK);
 366   thread->set_vm_result(obj);
 367 JRT_END
 368 
 369 
 370 JRT_ENTRY(void, Runtime1::unimplemented_entry(JavaThread* thread, StubID id))
 371   tty->print_cr("Runtime1::entry_for(%d) returned unimplemented entry point", id);
 372 JRT_END
 373 
 374 
 375 JRT_ENTRY(void, Runtime1::throw_array_store_exception(JavaThread* thread, oopDesc* obj))
 376   ResourceMark rm(thread);
 377   const char* klass_name = obj->klass()->external_name();
 378   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArrayStoreException(), klass_name);
 379 JRT_END
 380 
 381 
 382 // counter_overflow() is called from within C1-compiled methods. The enclosing method is the method
 383 // associated with the top activation record. The inlinee (that is possibly included in the enclosing
 384 // method) method oop is passed as an argument. In order to do that it is embedded in the code as
 385 // a constant.
 386 static nmethod* counter_overflow_helper(JavaThread* THREAD, int branch_bci, Method* m) {
 387   nmethod* osr_nm = NULL;
 388   methodHandle method(THREAD, m);
 389 
 390   RegisterMap map(THREAD, false);
 391   frame fr =  THREAD->last_frame().sender(&map);
 392   nmethod* nm = (nmethod*) fr.cb();
 393   assert(nm!= NULL && nm->is_nmethod(), "Sanity check");
 394   methodHandle enclosing_method(THREAD, nm->method());
 395 
 396   CompLevel level = (CompLevel)nm->comp_level();
 397   int bci = InvocationEntryBci;
 398   if (branch_bci != InvocationEntryBci) {
 399     // Compute desination bci
 400     address pc = method()->code_base() + branch_bci;
 401     Bytecodes::Code branch = Bytecodes::code_at(method(), pc);
 402     int offset = 0;
 403     switch (branch) {
 404       case Bytecodes::_if_icmplt: case Bytecodes::_iflt:
 405       case Bytecodes::_if_icmpgt: case Bytecodes::_ifgt:
 406       case Bytecodes::_if_icmple: case Bytecodes::_ifle:
 407       case Bytecodes::_if_icmpge: case Bytecodes::_ifge:
 408       case Bytecodes::_if_icmpeq: case Bytecodes::_if_acmpeq: case Bytecodes::_ifeq:
 409       case Bytecodes::_if_icmpne: case Bytecodes::_if_acmpne: case Bytecodes::_ifne:
 410       case Bytecodes::_ifnull: case Bytecodes::_ifnonnull: case Bytecodes::_goto:
 411         offset = (int16_t)Bytes::get_Java_u2(pc + 1);
 412         break;
 413       case Bytecodes::_goto_w:
 414         offset = Bytes::get_Java_u4(pc + 1);
 415         break;
 416       default: ;
 417     }
 418     bci = branch_bci + offset;
 419   }
 420   assert(!HAS_PENDING_EXCEPTION, "Should not have any exceptions pending");
 421   osr_nm = CompilationPolicy::policy()->event(enclosing_method, method, branch_bci, bci, level, nm, THREAD);
 422   assert(!HAS_PENDING_EXCEPTION, "Event handler should not throw any exceptions");
 423   return osr_nm;
 424 }
 425 
 426 JRT_BLOCK_ENTRY(address, Runtime1::counter_overflow(JavaThread* thread, int bci, Method* method))
 427   nmethod* osr_nm;
 428   JRT_BLOCK
 429     osr_nm = counter_overflow_helper(thread, bci, method);
 430     if (osr_nm != NULL) {
 431       RegisterMap map(thread, false);
 432       frame fr =  thread->last_frame().sender(&map);
 433       Deoptimization::deoptimize_frame(thread, fr.id());
 434     }
 435   JRT_BLOCK_END
 436   return NULL;
 437 JRT_END
 438 
 439 extern void vm_exit(int code);
 440 
 441 // Enter this method from compiled code handler below. This is where we transition
 442 // to VM mode. This is done as a helper routine so that the method called directly
 443 // from compiled code does not have to transition to VM. This allows the entry
 444 // method to see if the nmethod that we have just looked up a handler for has
 445 // been deoptimized while we were in the vm. This simplifies the assembly code
 446 // cpu directories.
 447 //
 448 // We are entering here from exception stub (via the entry method below)
 449 // If there is a compiled exception handler in this method, we will continue there;
 450 // otherwise we will unwind the stack and continue at the caller of top frame method
 451 // Note: we enter in Java using a special JRT wrapper. This wrapper allows us to
 452 // control the area where we can allow a safepoint. After we exit the safepoint area we can
 453 // check to see if the handler we are going to return is now in a nmethod that has
 454 // been deoptimized. If that is the case we return the deopt blob
 455 // unpack_with_exception entry instead. This makes life for the exception blob easier
 456 // because making that same check and diverting is painful from assembly language.
 457 JRT_ENTRY_NO_ASYNC(static address, exception_handler_for_pc_helper(JavaThread* thread, oopDesc* ex, address pc, nmethod*& nm))
 458   // Reset method handle flag.
 459   thread->set_is_method_handle_return(false);
 460 
 461   Handle exception(thread, ex);
 462   nm = CodeCache::find_nmethod(pc);
 463   assert(nm != NULL, "this is not an nmethod");
 464   // Adjust the pc as needed/
 465   if (nm->is_deopt_pc(pc)) {
 466     RegisterMap map(thread, false);
 467     frame exception_frame = thread->last_frame().sender(&map);
 468     // if the frame isn't deopted then pc must not correspond to the caller of last_frame
 469     assert(exception_frame.is_deoptimized_frame(), "must be deopted");
 470     pc = exception_frame.pc();
 471   }
 472 #ifdef ASSERT
 473   assert(exception.not_null(), "NULL exceptions should be handled by throw_exception");
 474   assert(exception->is_oop(), "just checking");
 475   // Check that exception is a subclass of Throwable, otherwise we have a VerifyError
 476   if (!(exception->is_a(SystemDictionary::Throwable_klass()))) {
 477     if (ExitVMOnVerifyError) vm_exit(-1);
 478     ShouldNotReachHere();
 479   }
 480 #endif
 481 
 482   // Check the stack guard pages and reenable them if necessary and there is
 483   // enough space on the stack to do so.  Use fast exceptions only if the guard
 484   // pages are enabled.
 485   bool guard_pages_enabled = thread->stack_yellow_zone_enabled();
 486   if (!guard_pages_enabled) guard_pages_enabled = thread->reguard_stack();
 487 
 488   if (JvmtiExport::can_post_on_exceptions()) {
 489     // To ensure correct notification of exception catches and throws
 490     // we have to deoptimize here.  If we attempted to notify the
 491     // catches and throws during this exception lookup it's possible
 492     // we could deoptimize on the way out of the VM and end back in
 493     // the interpreter at the throw site.  This would result in double
 494     // notifications since the interpreter would also notify about
 495     // these same catches and throws as it unwound the frame.
 496 
 497     RegisterMap reg_map(thread);
 498     frame stub_frame = thread->last_frame();
 499     frame caller_frame = stub_frame.sender(&reg_map);
 500 
 501     // We don't really want to deoptimize the nmethod itself since we
 502     // can actually continue in the exception handler ourselves but I
 503     // don't see an easy way to have the desired effect.
 504     Deoptimization::deoptimize_frame(thread, caller_frame.id());
 505     assert(caller_is_deopted(), "Must be deoptimized");
 506 
 507     return SharedRuntime::deopt_blob()->unpack_with_exception_in_tls();
 508   }
 509 
 510   // ExceptionCache is used only for exceptions at call sites and not for implicit exceptions
 511   if (guard_pages_enabled) {
 512     address fast_continuation = nm->handler_for_exception_and_pc(exception, pc);
 513     if (fast_continuation != NULL) {
 514       // Set flag if return address is a method handle call site.
 515       thread->set_is_method_handle_return(nm->is_method_handle_return(pc));
 516       return fast_continuation;
 517     }
 518   }
 519 
 520   // If the stack guard pages are enabled, check whether there is a handler in
 521   // the current method.  Otherwise (guard pages disabled), force an unwind and
 522   // skip the exception cache update (i.e., just leave continuation==NULL).
 523   address continuation = NULL;
 524   if (guard_pages_enabled) {
 525 
 526     // New exception handling mechanism can support inlined methods
 527     // with exception handlers since the mappings are from PC to PC
 528 
 529     // debugging support
 530     // tracing
 531     if (TraceExceptions) {
 532       ttyLocker ttyl;
 533       ResourceMark rm;
 534       tty->print_cr("Exception <%s> (0x%x) thrown in compiled method <%s> at PC " PTR_FORMAT " for thread 0x%x",
 535                     exception->print_value_string(), (address)exception(), nm->method()->print_value_string(), pc, thread);
 536     }
 537     // for AbortVMOnException flag
 538     NOT_PRODUCT(Exceptions::debug_check_abort(exception));
 539 
 540     // Clear out the exception oop and pc since looking up an
 541     // exception handler can cause class loading, which might throw an
 542     // exception and those fields are expected to be clear during
 543     // normal bytecode execution.
 544     thread->set_exception_oop(NULL);
 545     thread->set_exception_pc(NULL);
 546 
 547     continuation = SharedRuntime::compute_compiled_exc_handler(nm, pc, exception, false, false);
 548     // If an exception was thrown during exception dispatch, the exception oop may have changed
 549     thread->set_exception_oop(exception());
 550     thread->set_exception_pc(pc);
 551 
 552     // the exception cache is used only by non-implicit exceptions
 553     if (continuation != NULL) {
 554       nm->add_handler_for_exception_and_pc(exception, pc, continuation);
 555     }
 556   }
 557 
 558   thread->set_vm_result(exception());
 559   // Set flag if return address is a method handle call site.
 560   thread->set_is_method_handle_return(nm->is_method_handle_return(pc));
 561 
 562   if (TraceExceptions) {
 563     ttyLocker ttyl;
 564     ResourceMark rm;
 565     tty->print_cr("Thread " PTR_FORMAT " continuing at PC " PTR_FORMAT " for exception thrown at PC " PTR_FORMAT,
 566                   thread, continuation, pc);
 567   }
 568 
 569   return continuation;
 570 JRT_END
 571 
 572 // Enter this method from compiled code only if there is a Java exception handler
 573 // in the method handling the exception.
 574 // We are entering here from exception stub. We don't do a normal VM transition here.
 575 // We do it in a helper. This is so we can check to see if the nmethod we have just
 576 // searched for an exception handler has been deoptimized in the meantime.
 577 address Runtime1::exception_handler_for_pc(JavaThread* thread) {
 578   oop exception = thread->exception_oop();
 579   address pc = thread->exception_pc();
 580   // Still in Java mode
 581   DEBUG_ONLY(ResetNoHandleMark rnhm);
 582   nmethod* nm = NULL;
 583   address continuation = NULL;
 584   {
 585     // Enter VM mode by calling the helper
 586     ResetNoHandleMark rnhm;
 587     continuation = exception_handler_for_pc_helper(thread, exception, pc, nm);
 588   }
 589   // Back in JAVA, use no oops DON'T safepoint
 590 
 591   // Now check to see if the nmethod we were called from is now deoptimized.
 592   // If so we must return to the deopt blob and deoptimize the nmethod
 593   if (nm != NULL && caller_is_deopted()) {
 594     continuation = SharedRuntime::deopt_blob()->unpack_with_exception_in_tls();
 595   }
 596 
 597   assert(continuation != NULL, "no handler found");
 598   return continuation;
 599 }
 600 
 601 
 602 JRT_ENTRY(void, Runtime1::throw_range_check_exception(JavaThread* thread, int index))
 603   NOT_PRODUCT(_throw_range_check_exception_count++;)
 604   char message[jintAsStringSize];
 605   sprintf(message, "%d", index);
 606   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArrayIndexOutOfBoundsException(), message);
 607 JRT_END
 608 
 609 
 610 JRT_ENTRY(void, Runtime1::throw_index_exception(JavaThread* thread, int index))
 611   NOT_PRODUCT(_throw_index_exception_count++;)
 612   char message[16];
 613   sprintf(message, "%d", index);
 614   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IndexOutOfBoundsException(), message);
 615 JRT_END
 616 
 617 
 618 JRT_ENTRY(void, Runtime1::throw_div0_exception(JavaThread* thread))
 619   NOT_PRODUCT(_throw_div0_exception_count++;)
 620   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_ArithmeticException(), "/ by zero");
 621 JRT_END
 622 
 623 
 624 JRT_ENTRY(void, Runtime1::throw_null_pointer_exception(JavaThread* thread))
 625   NOT_PRODUCT(_throw_null_pointer_exception_count++;)
 626   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_NullPointerException());
 627 JRT_END
 628 
 629 
 630 JRT_ENTRY(void, Runtime1::throw_class_cast_exception(JavaThread* thread, oopDesc* object))
 631   NOT_PRODUCT(_throw_class_cast_exception_count++;)
 632   ResourceMark rm(thread);
 633   char* message = SharedRuntime::generate_class_cast_message(
 634     thread, object->klass()->external_name());
 635   SharedRuntime::throw_and_post_jvmti_exception(
 636     thread, vmSymbols::java_lang_ClassCastException(), message);
 637 JRT_END
 638 
 639 
 640 JRT_ENTRY(void, Runtime1::throw_incompatible_class_change_error(JavaThread* thread))
 641   NOT_PRODUCT(_throw_incompatible_class_change_error_count++;)
 642   ResourceMark rm(thread);
 643   SharedRuntime::throw_and_post_jvmti_exception(thread, vmSymbols::java_lang_IncompatibleClassChangeError());
 644 JRT_END
 645 
 646 
 647 JRT_ENTRY_NO_ASYNC(void, Runtime1::monitorenter(JavaThread* thread, oopDesc* obj, BasicObjectLock* lock))
 648   NOT_PRODUCT(_monitorenter_slowcase_cnt++;)
 649   if (PrintBiasedLockingStatistics) {
 650     Atomic::inc(BiasedLocking::slow_path_entry_count_addr());
 651   }
 652   Handle h_obj(thread, obj);
 653   assert(h_obj()->is_oop(), "must be NULL or an object");
 654   if (UseBiasedLocking) {
 655     // Retry fast entry if bias is revoked to avoid unnecessary inflation
 656     ObjectSynchronizer::fast_enter(h_obj, lock->lock(), true, CHECK);
 657   } else {
 658     if (UseFastLocking) {
 659       // When using fast locking, the compiled code has already tried the fast case
 660       assert(obj == lock->obj(), "must match");
 661       ObjectSynchronizer::slow_enter(h_obj, lock->lock(), THREAD);
 662     } else {
 663       lock->set_obj(obj);
 664       ObjectSynchronizer::fast_enter(h_obj, lock->lock(), false, THREAD);
 665     }
 666   }
 667 JRT_END
 668 
 669 
 670 JRT_LEAF(void, Runtime1::monitorexit(JavaThread* thread, BasicObjectLock* lock))
 671   NOT_PRODUCT(_monitorexit_slowcase_cnt++;)
 672   assert(thread == JavaThread::current(), "threads must correspond");
 673   assert(thread->last_Java_sp(), "last_Java_sp must be set");
 674   // monitorexit is non-blocking (leaf routine) => no exceptions can be thrown
 675   EXCEPTION_MARK;
 676 
 677   oop obj = lock->obj();
 678   assert(obj->is_oop(), "must be NULL or an object");
 679   if (UseFastLocking) {
 680     // When using fast locking, the compiled code has already tried the fast case
 681     ObjectSynchronizer::slow_exit(obj, lock->lock(), THREAD);
 682   } else {
 683     ObjectSynchronizer::fast_exit(obj, lock->lock(), THREAD);
 684   }
 685 JRT_END
 686 
 687 // Cf. OptoRuntime::deoptimize_caller_frame
 688 JRT_ENTRY(void, Runtime1::deoptimize(JavaThread* thread))
 689   // Called from within the owner thread, so no need for safepoint
 690   RegisterMap reg_map(thread, false);
 691   frame stub_frame = thread->last_frame();
 692   assert(stub_frame.is_runtime_frame(), "sanity check");
 693   frame caller_frame = stub_frame.sender(&reg_map);
 694 
 695   // We are coming from a compiled method; check this is true.
 696   assert(CodeCache::find_nmethod(caller_frame.pc()) != NULL, "sanity");
 697 
 698   // Deoptimize the caller frame.
 699   Deoptimization::deoptimize_frame(thread, caller_frame.id());
 700 
 701   // Return to the now deoptimized frame.
 702 JRT_END
 703 
 704 
 705 static Klass* resolve_field_return_klass(methodHandle caller, int bci, TRAPS) {
 706   Bytecode_field field_access(caller, bci);
 707   // This can be static or non-static field access
 708   Bytecodes::Code code       = field_access.code();
 709 
 710   // We must load class, initialize class and resolvethe field
 711   FieldAccessInfo result; // initialize class if needed
 712   constantPoolHandle constants(THREAD, caller->constants());
 713   LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK_NULL);
 714   return result.klass()();
 715 }
 716 
 717 
 718 //
 719 // This routine patches sites where a class wasn't loaded or
 720 // initialized at the time the code was generated.  It handles
 721 // references to classes, fields and forcing of initialization.  Most
 722 // of the cases are straightforward and involving simply forcing
 723 // resolution of a class, rewriting the instruction stream with the
 724 // needed constant and replacing the call in this function with the
 725 // patched code.  The case for static field is more complicated since
 726 // the thread which is in the process of initializing a class can
 727 // access it's static fields but other threads can't so the code
 728 // either has to deoptimize when this case is detected or execute a
 729 // check that the current thread is the initializing thread.  The
 730 // current
 731 //
 732 // Patches basically look like this:
 733 //
 734 //
 735 // patch_site: jmp patch stub     ;; will be patched
 736 // continue:   ...
 737 //             ...
 738 //             ...
 739 //             ...
 740 //
 741 // They have a stub which looks like this:
 742 //
 743 //             ;; patch body
 744 //             movl <const>, reg           (for class constants)
 745 //        <or> movl [reg1 + <const>], reg  (for field offsets)
 746 //        <or> movl reg, [reg1 + <const>]  (for field offsets)
 747 //             <being_init offset> <bytes to copy> <bytes to skip>
 748 // patch_stub: call Runtime1::patch_code (through a runtime stub)
 749 //             jmp patch_site
 750 //
 751 //
 752 // A normal patch is done by rewriting the patch body, usually a move,
 753 // and then copying it into place over top of the jmp instruction
 754 // being careful to flush caches and doing it in an MP-safe way.  The
 755 // constants following the patch body are used to find various pieces
 756 // of the patch relative to the call site for Runtime1::patch_code.
 757 // The case for getstatic and putstatic is more complicated because
 758 // getstatic and putstatic have special semantics when executing while
 759 // the class is being initialized.  getstatic/putstatic on a class
 760 // which is being_initialized may be executed by the initializing
 761 // thread but other threads have to block when they execute it.  This
 762 // is accomplished in compiled code by executing a test of the current
 763 // thread against the initializing thread of the class.  It's emitted
 764 // as boilerplate in their stub which allows the patched code to be
 765 // executed before it's copied back into the main body of the nmethod.
 766 //
 767 // being_init: get_thread(<tmp reg>
 768 //             cmpl [reg1 + <init_thread_offset>], <tmp reg>
 769 //             jne patch_stub
 770 //             movl [reg1 + <const>], reg  (for field offsets)  <or>
 771 //             movl reg, [reg1 + <const>]  (for field offsets)
 772 //             jmp continue
 773 //             <being_init offset> <bytes to copy> <bytes to skip>
 774 // patch_stub: jmp Runtim1::patch_code (through a runtime stub)
 775 //             jmp patch_site
 776 //
 777 // If the class is being initialized the patch body is rewritten and
 778 // the patch site is rewritten to jump to being_init, instead of
 779 // patch_stub.  Whenever this code is executed it checks the current
 780 // thread against the intializing thread so other threads will enter
 781 // the runtime and end up blocked waiting the class to finish
 782 // initializing inside the calls to resolve_field below.  The
 783 // initializing class will continue on it's way.  Once the class is
 784 // fully_initialized, the intializing_thread of the class becomes
 785 // NULL, so the next thread to execute this code will fail the test,
 786 // call into patch_code and complete the patching process by copying
 787 // the patch body back into the main part of the nmethod and resume
 788 // executing.
 789 //
 790 //
 791 
 792 JRT_ENTRY(void, Runtime1::patch_code(JavaThread* thread, Runtime1::StubID stub_id ))
 793   NOT_PRODUCT(_patch_code_slowcase_cnt++;)
 794 
 795   ResourceMark rm(thread);
 796   RegisterMap reg_map(thread, false);
 797   frame runtime_frame = thread->last_frame();
 798   frame caller_frame = runtime_frame.sender(&reg_map);
 799 
 800   // last java frame on stack
 801   vframeStream vfst(thread, true);
 802   assert(!vfst.at_end(), "Java frame must exist");
 803 
 804   methodHandle caller_method(THREAD, vfst.method());
 805   // Note that caller_method->code() may not be same as caller_code because of OSR's
 806   // Note also that in the presence of inlining it is not guaranteed
 807   // that caller_method() == caller_code->method()
 808 
 809   int bci = vfst.bci();
 810   Bytecodes::Code code = caller_method()->java_code_at(bci);
 811 
 812 #ifndef PRODUCT
 813   // this is used by assertions in the access_field_patching_id
 814   BasicType patch_field_type = T_ILLEGAL;
 815 #endif // PRODUCT
 816   bool deoptimize_for_volatile = false;
 817   int patch_field_offset = -1;
 818   KlassHandle init_klass(THREAD, NULL); // klass needed by load_klass_patching code
 819   KlassHandle load_klass(THREAD, NULL); // klass needed by load_klass_patching code
 820   Handle mirror(THREAD, NULL);                    // oop needed by load_mirror_patching code
 821   bool load_klass_or_mirror_patch_id =
 822     (stub_id == Runtime1::load_klass_patching_id || stub_id == Runtime1::load_mirror_patching_id);
 823 
 824   if (stub_id == Runtime1::access_field_patching_id) {
 825 
 826     Bytecode_field field_access(caller_method, bci);
 827     FieldAccessInfo result; // initialize class if needed
 828     Bytecodes::Code code = field_access.code();
 829     constantPoolHandle constants(THREAD, caller_method->constants());
 830     LinkResolver::resolve_field(result, constants, field_access.index(), Bytecodes::java_code(code), false, CHECK);
 831     patch_field_offset = result.field_offset();
 832 
 833     // If we're patching a field which is volatile then at compile it
 834     // must not have been know to be volatile, so the generated code
 835     // isn't correct for a volatile reference.  The nmethod has to be
 836     // deoptimized so that the code can be regenerated correctly.
 837     // This check is only needed for access_field_patching since this
 838     // is the path for patching field offsets.  load_klass is only
 839     // used for patching references to oops which don't need special
 840     // handling in the volatile case.
 841     deoptimize_for_volatile = result.access_flags().is_volatile();
 842 
 843 #ifndef PRODUCT
 844     patch_field_type = result.field_type();
 845 #endif
 846   } else if (load_klass_or_mirror_patch_id) {
 847     Klass* k = NULL;
 848     switch (code) {
 849       case Bytecodes::_putstatic:
 850       case Bytecodes::_getstatic:
 851         { Klass* klass = resolve_field_return_klass(caller_method, bci, CHECK);
 852           init_klass = KlassHandle(THREAD, klass);
 853           mirror = Handle(THREAD, klass->java_mirror());
 854         }
 855         break;
 856       case Bytecodes::_new:
 857         { Bytecode_new bnew(caller_method(), caller_method->bcp_from(bci));
 858           k = caller_method->constants()->klass_at(bnew.index(), CHECK);
 859         }
 860         break;
 861       case Bytecodes::_multianewarray:
 862         { Bytecode_multianewarray mna(caller_method(), caller_method->bcp_from(bci));
 863           k = caller_method->constants()->klass_at(mna.index(), CHECK);
 864         }
 865         break;
 866       case Bytecodes::_instanceof:
 867         { Bytecode_instanceof io(caller_method(), caller_method->bcp_from(bci));
 868           k = caller_method->constants()->klass_at(io.index(), CHECK);
 869         }
 870         break;
 871       case Bytecodes::_checkcast:
 872         { Bytecode_checkcast cc(caller_method(), caller_method->bcp_from(bci));
 873           k = caller_method->constants()->klass_at(cc.index(), CHECK);
 874         }
 875         break;
 876       case Bytecodes::_anewarray:
 877         { Bytecode_anewarray anew(caller_method(), caller_method->bcp_from(bci));
 878           Klass* ek = caller_method->constants()->klass_at(anew.index(), CHECK);
 879           k = ek->array_klass(CHECK);
 880         }
 881         break;
 882       case Bytecodes::_ldc:
 883       case Bytecodes::_ldc_w:
 884         {
 885           Bytecode_loadconstant cc(caller_method, bci);
 886           oop m = cc.resolve_constant(CHECK);
 887           mirror = Handle(THREAD, m);
 888         }
 889         break;
 890       default: Unimplemented();
 891     }
 892     // convert to handle
 893     load_klass = KlassHandle(THREAD, k);
 894   } else {
 895     ShouldNotReachHere();
 896   }
 897 
 898   if (deoptimize_for_volatile) {
 899     // At compile time we assumed the field wasn't volatile but after
 900     // loading it turns out it was volatile so we have to throw the
 901     // compiled code out and let it be regenerated.
 902     if (TracePatching) {
 903       tty->print_cr("Deoptimizing for patching volatile field reference");
 904     }
 905     // It's possible the nmethod was invalidated in the last
 906     // safepoint, but if it's still alive then make it not_entrant.
 907     nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
 908     if (nm != NULL) {
 909       nm->make_not_entrant();
 910     }
 911 
 912     Deoptimization::deoptimize_frame(thread, caller_frame.id());
 913 
 914     // Return to the now deoptimized frame.
 915   }
 916 
 917   // If we are patching in a non-perm oop, make sure the nmethod
 918   // is on the right list.
 919   if (ScavengeRootsInCode && mirror.not_null() && mirror()->is_scavengable()) {
 920     MutexLockerEx ml_code (CodeCache_lock, Mutex::_no_safepoint_check_flag);
 921     nmethod* nm = CodeCache::find_nmethod(caller_frame.pc());
 922     guarantee(nm != NULL, "only nmethods can contain non-perm oops");
 923     if (!nm->on_scavenge_root_list())
 924       CodeCache::add_scavenge_root_nmethod(nm);
 925   }
 926 
 927   // Now copy code back
 928 
 929   {
 930     MutexLockerEx ml_patch (Patching_lock, Mutex::_no_safepoint_check_flag);
 931     //
 932     // Deoptimization may have happened while we waited for the lock.
 933     // In that case we don't bother to do any patching we just return
 934     // and let the deopt happen
 935     if (!caller_is_deopted()) {
 936       NativeGeneralJump* jump = nativeGeneralJump_at(caller_frame.pc());
 937       address instr_pc = jump->jump_destination();
 938       NativeInstruction* ni = nativeInstruction_at(instr_pc);
 939       if (ni->is_jump() ) {
 940         // the jump has not been patched yet
 941         // The jump destination is slow case and therefore not part of the stubs
 942         // (stubs are only for StaticCalls)
 943 
 944         // format of buffer
 945         //    ....
 946         //    instr byte 0     <-- copy_buff
 947         //    instr byte 1
 948         //    ..
 949         //    instr byte n-1
 950         //      n
 951         //    ....             <-- call destination
 952 
 953         address stub_location = caller_frame.pc() + PatchingStub::patch_info_offset();
 954         unsigned char* byte_count = (unsigned char*) (stub_location - 1);
 955         unsigned char* byte_skip = (unsigned char*) (stub_location - 2);
 956         unsigned char* being_initialized_entry_offset = (unsigned char*) (stub_location - 3);
 957         address copy_buff = stub_location - *byte_skip - *byte_count;
 958         address being_initialized_entry = stub_location - *being_initialized_entry_offset;
 959         if (TracePatching) {
 960           tty->print_cr(" Patching %s at bci %d at address 0x%x  (%s)", Bytecodes::name(code), bci,
 961                         instr_pc, (stub_id == Runtime1::access_field_patching_id) ? "field" : "klass");
 962           nmethod* caller_code = CodeCache::find_nmethod(caller_frame.pc());
 963           assert(caller_code != NULL, "nmethod not found");
 964 
 965           // NOTE we use pc() not original_pc() because we already know they are
 966           // identical otherwise we'd have never entered this block of code
 967 
 968           OopMap* map = caller_code->oop_map_for_return_address(caller_frame.pc());
 969           assert(map != NULL, "null check");
 970           map->print();
 971           tty->cr();
 972 
 973           Disassembler::decode(copy_buff, copy_buff + *byte_count, tty);
 974         }
 975         // depending on the code below, do_patch says whether to copy the patch body back into the nmethod
 976         bool do_patch = true;
 977         if (stub_id == Runtime1::access_field_patching_id) {
 978           // The offset may not be correct if the class was not loaded at code generation time.
 979           // Set it now.
 980           NativeMovRegMem* n_move = nativeMovRegMem_at(copy_buff);
 981           assert(n_move->offset() == 0 || (n_move->offset() == 4 && (patch_field_type == T_DOUBLE || patch_field_type == T_LONG)), "illegal offset for type");
 982           assert(patch_field_offset >= 0, "illegal offset");
 983           n_move->add_offset_in_bytes(patch_field_offset);
 984         } else if (load_klass_or_mirror_patch_id) {
 985           // If a getstatic or putstatic is referencing a klass which
 986           // isn't fully initialized, the patch body isn't copied into
 987           // place until initialization is complete.  In this case the
 988           // patch site is setup so that any threads besides the
 989           // initializing thread are forced to come into the VM and
 990           // block.
 991           do_patch = (code != Bytecodes::_getstatic && code != Bytecodes::_putstatic) ||
 992                      InstanceKlass::cast(init_klass())->is_initialized();
 993           NativeGeneralJump* jump = nativeGeneralJump_at(instr_pc);
 994           if (jump->jump_destination() == being_initialized_entry) {
 995             assert(do_patch == true, "initialization must be complete at this point");
 996           } else {
 997             // patch the instruction <move reg, klass>
 998             NativeMovConstReg* n_copy = nativeMovConstReg_at(copy_buff);
 999 
1000             assert(n_copy->data() == 0 ||
1001                    n_copy->data() == (intptr_t)Universe::non_oop_word(),
1002                    "illegal init value");
1003             if (stub_id == Runtime1::load_klass_patching_id) {
1004             assert(load_klass() != NULL, "klass not set");
1005             n_copy->set_data((intx) (load_klass()));
1006             } else {
1007               assert(mirror() != NULL, "klass not set");
1008               n_copy->set_data((intx) (mirror()));
1009             }
1010 
1011             if (TracePatching) {
1012               Disassembler::decode(copy_buff, copy_buff + *byte_count, tty);
1013             }
1014 
1015 #if defined(SPARC) || defined(PPC)
1016             // Update the location in the nmethod with the proper
1017             // metadata.  When the code was generated, a NULL was stuffed
1018             // in the metadata table and that table needs to be update to
1019             // have the right value.  On intel the value is kept
1020             // directly in the instruction instead of in the metadata
1021             // table, so set_data above effectively updated the value.
1022             nmethod* nm = CodeCache::find_nmethod(instr_pc);
1023             assert(nm != NULL, "invalid nmethod_pc");
1024             RelocIterator mds(nm, copy_buff, copy_buff + 1);
1025             bool found = false;
1026             while (mds.next() && !found) {
1027               if (mds.type() == relocInfo::oop_type) {
1028                 assert(stub_id == Runtime1::load_mirror_patching_id, "wrong stub id");
1029                 oop_Relocation* r = mds.oop_reloc();
1030                 oop* oop_adr = r->oop_addr();
1031                 *oop_adr = mirror();
1032                 r->fix_oop_relocation();
1033                 found = true;
1034               } else if (mds.type() == relocInfo::metadata_type) {
1035                 assert(stub_id == Runtime1::load_klass_patching_id, "wrong stub id");
1036                 metadata_Relocation* r = mds.metadata_reloc();
1037                 Metadata** metadata_adr = r->metadata_addr();
1038                 *metadata_adr = load_klass();
1039                 r->fix_metadata_relocation();
1040                 found = true;
1041               }
1042             }
1043             assert(found, "the metadata must exist!");
1044 #endif
1045 
1046           }
1047         } else {
1048           ShouldNotReachHere();
1049         }
1050 
1051         if (do_patch) {
1052           // replace instructions
1053           // first replace the tail, then the call
1054 #ifdef ARM
1055           if(load_klass_or_mirror_patch_id && !VM_Version::supports_movw()) {
1056             nmethod* nm = CodeCache::find_nmethod(instr_pc);
1057             address addr = NULL;
1058             assert(nm != NULL, "invalid nmethod_pc");
1059             RelocIterator mds(nm, copy_buff, copy_buff + 1);
1060             while (mds.next()) {
1061               if (mds.type() == relocInfo::oop_type) {
1062                 assert(stub_id == Runtime1::load_mirror_patching_id, "wrong stub id");
1063                 oop_Relocation* r = mds.oop_reloc();
1064                 addr = (address)r->oop_addr();
1065                 break;
1066               } else if (mds.type() == relocInfo::metadata_type) {
1067                 assert(stub_id == Runtime1::load_klass_patching_id, "wrong stub id");
1068                 metadata_Relocation* r = mds.metadata_reloc();
1069                 addr = (address)r->metadata_addr();
1070                 break;
1071               }
1072             }
1073             assert(addr != NULL, "metadata relocation must exist");
1074             copy_buff -= *byte_count;
1075             NativeMovConstReg* n_copy2 = nativeMovConstReg_at(copy_buff);
1076             n_copy2->set_pc_relative_offset(addr, instr_pc);
1077           }
1078 #endif
1079 
1080           for (int i = NativeCall::instruction_size; i < *byte_count; i++) {
1081             address ptr = copy_buff + i;
1082             int a_byte = (*ptr) & 0xFF;
1083             address dst = instr_pc + i;
1084             *(unsigned char*)dst = (unsigned char) a_byte;
1085           }
1086           ICache::invalidate_range(instr_pc, *byte_count);
1087           NativeGeneralJump::replace_mt_safe(instr_pc, copy_buff);
1088 
1089           if (load_klass_or_mirror_patch_id) {
1090             relocInfo::relocType rtype =
1091               (stub_id == Runtime1::load_klass_patching_id) ?
1092                                    relocInfo::metadata_type :
1093                                    relocInfo::oop_type;
1094             // update relocInfo to metadata
1095             nmethod* nm = CodeCache::find_nmethod(instr_pc);
1096             assert(nm != NULL, "invalid nmethod_pc");
1097 
1098             // The old patch site is now a move instruction so update
1099             // the reloc info so that it will get updated during
1100             // future GCs.
1101             RelocIterator iter(nm, (address)instr_pc, (address)(instr_pc + 1));
1102             relocInfo::change_reloc_info_for_address(&iter, (address) instr_pc,
1103                                                      relocInfo::none, rtype);
1104 #ifdef SPARC
1105             // Sparc takes two relocations for an metadata so update the second one.
1106             address instr_pc2 = instr_pc + NativeMovConstReg::add_offset;
1107             RelocIterator iter2(nm, instr_pc2, instr_pc2 + 1);
1108             relocInfo::change_reloc_info_for_address(&iter2, (address) instr_pc2,
1109                                                      relocInfo::none, rtype);
1110 #endif
1111 #ifdef PPC
1112           { address instr_pc2 = instr_pc + NativeMovConstReg::lo_offset;
1113             RelocIterator iter2(nm, instr_pc2, instr_pc2 + 1);
1114             relocInfo::change_reloc_info_for_address(&iter2, (address) instr_pc2,
1115                                                      relocInfo::none, rtype);
1116           }
1117 #endif
1118           }
1119 
1120         } else {
1121           ICache::invalidate_range(copy_buff, *byte_count);
1122           NativeGeneralJump::insert_unconditional(instr_pc, being_initialized_entry);
1123         }
1124       }
1125     }
1126   }
1127 JRT_END
1128 
1129 //
1130 // Entry point for compiled code. We want to patch a nmethod.
1131 // We don't do a normal VM transition here because we want to
1132 // know after the patching is complete and any safepoint(s) are taken
1133 // if the calling nmethod was deoptimized. We do this by calling a
1134 // helper method which does the normal VM transition and when it
1135 // completes we can check for deoptimization. This simplifies the
1136 // assembly code in the cpu directories.
1137 //
1138 int Runtime1::move_klass_patching(JavaThread* thread) {
1139 //
1140 // NOTE: we are still in Java
1141 //
1142   Thread* THREAD = thread;
1143   debug_only(NoHandleMark nhm;)
1144   {
1145     // Enter VM mode
1146 
1147     ResetNoHandleMark rnhm;
1148     patch_code(thread, load_klass_patching_id);
1149   }
1150   // Back in JAVA, use no oops DON'T safepoint
1151 
1152   // Return true if calling code is deoptimized
1153 
1154   return caller_is_deopted();
1155 }
1156 
1157 int Runtime1::move_mirror_patching(JavaThread* thread) {
1158 //
1159 // NOTE: we are still in Java
1160 //
1161   Thread* THREAD = thread;
1162   debug_only(NoHandleMark nhm;)
1163   {
1164     // Enter VM mode
1165 
1166     ResetNoHandleMark rnhm;
1167     patch_code(thread, load_mirror_patching_id);
1168   }
1169   // Back in JAVA, use no oops DON'T safepoint
1170 
1171   // Return true if calling code is deoptimized
1172 
1173   return caller_is_deopted();
1174 }
1175 
1176 //
1177 // Entry point for compiled code. We want to patch a nmethod.
1178 // We don't do a normal VM transition here because we want to
1179 // know after the patching is complete and any safepoint(s) are taken
1180 // if the calling nmethod was deoptimized. We do this by calling a
1181 // helper method which does the normal VM transition and when it
1182 // completes we can check for deoptimization. This simplifies the
1183 // assembly code in the cpu directories.
1184 //
1185 
1186 int Runtime1::access_field_patching(JavaThread* thread) {
1187 //
1188 // NOTE: we are still in Java
1189 //
1190   Thread* THREAD = thread;
1191   debug_only(NoHandleMark nhm;)
1192   {
1193     // Enter VM mode
1194 
1195     ResetNoHandleMark rnhm;
1196     patch_code(thread, access_field_patching_id);
1197   }
1198   // Back in JAVA, use no oops DON'T safepoint
1199 
1200   // Return true if calling code is deoptimized
1201 
1202   return caller_is_deopted();
1203 JRT_END
1204 
1205 
1206 JRT_LEAF(void, Runtime1::trace_block_entry(jint block_id))
1207   // for now we just print out the block id
1208   tty->print("%d ", block_id);
1209 JRT_END
1210 
1211 
1212 // Array copy return codes.
1213 enum {
1214   ac_failed = -1, // arraycopy failed
1215   ac_ok = 0       // arraycopy succeeded
1216 };
1217 
1218 
1219 // Below length is the # elements copied.
1220 template <class T> int obj_arraycopy_work(oopDesc* src, T* src_addr,
1221                                           oopDesc* dst, T* dst_addr,
1222                                           int length) {
1223 
1224   // For performance reasons, we assume we are using a card marking write
1225   // barrier. The assert will fail if this is not the case.
1226   // Note that we use the non-virtual inlineable variant of write_ref_array.
1227   BarrierSet* bs = Universe::heap()->barrier_set();
1228   assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt");
1229   assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well.");
1230   if (src == dst) {
1231     // same object, no check
1232     bs->write_ref_array_pre(dst_addr, length);
1233     Copy::conjoint_oops_atomic(src_addr, dst_addr, length);
1234     bs->write_ref_array((HeapWord*)dst_addr, length);
1235     return ac_ok;
1236   } else {
1237     Klass* bound = ObjArrayKlass::cast(dst->klass())->element_klass();
1238     Klass* stype = ObjArrayKlass::cast(src->klass())->element_klass();
1239     if (stype == bound || stype->is_subtype_of(bound)) {
1240       // Elements are guaranteed to be subtypes, so no check necessary
1241       bs->write_ref_array_pre(dst_addr, length);
1242       Copy::conjoint_oops_atomic(src_addr, dst_addr, length);
1243       bs->write_ref_array((HeapWord*)dst_addr, length);
1244       return ac_ok;
1245     }
1246   }
1247   return ac_failed;
1248 }
1249 
1250 // fast and direct copy of arrays; returning -1, means that an exception may be thrown
1251 // and we did not copy anything
1252 JRT_LEAF(int, Runtime1::arraycopy(oopDesc* src, int src_pos, oopDesc* dst, int dst_pos, int length))
1253 #ifndef PRODUCT
1254   _generic_arraycopy_cnt++;        // Slow-path oop array copy
1255 #endif
1256 
1257   if (src == NULL || dst == NULL || src_pos < 0 || dst_pos < 0 || length < 0) return ac_failed;
1258   if (!dst->is_array() || !src->is_array()) return ac_failed;
1259   if ((unsigned int) arrayOop(src)->length() < (unsigned int)src_pos + (unsigned int)length) return ac_failed;
1260   if ((unsigned int) arrayOop(dst)->length() < (unsigned int)dst_pos + (unsigned int)length) return ac_failed;
1261 
1262   if (length == 0) return ac_ok;
1263   if (src->is_typeArray()) {
1264     Klass* const klass_oop = src->klass();
1265     if (klass_oop != dst->klass()) return ac_failed;
1266     TypeArrayKlass* klass = TypeArrayKlass::cast(klass_oop);
1267     const int l2es = klass->log2_element_size();
1268     const int ihs = klass->array_header_in_bytes() / wordSize;
1269     char* src_addr = (char*) ((oopDesc**)src + ihs) + (src_pos << l2es);
1270     char* dst_addr = (char*) ((oopDesc**)dst + ihs) + (dst_pos << l2es);
1271     // Potential problem: memmove is not guaranteed to be word atomic
1272     // Revisit in Merlin
1273     memmove(dst_addr, src_addr, length << l2es);
1274     return ac_ok;
1275   } else if (src->is_objArray() && dst->is_objArray()) {
1276     if (UseCompressedOops) {
1277       narrowOop *src_addr  = objArrayOop(src)->obj_at_addr<narrowOop>(src_pos);
1278       narrowOop *dst_addr  = objArrayOop(dst)->obj_at_addr<narrowOop>(dst_pos);
1279       return obj_arraycopy_work(src, src_addr, dst, dst_addr, length);
1280     } else {
1281       oop *src_addr  = objArrayOop(src)->obj_at_addr<oop>(src_pos);
1282       oop *dst_addr  = objArrayOop(dst)->obj_at_addr<oop>(dst_pos);
1283       return obj_arraycopy_work(src, src_addr, dst, dst_addr, length);
1284     }
1285   }
1286   return ac_failed;
1287 JRT_END
1288 
1289 
1290 JRT_LEAF(void, Runtime1::primitive_arraycopy(HeapWord* src, HeapWord* dst, int length))
1291 #ifndef PRODUCT
1292   _primitive_arraycopy_cnt++;
1293 #endif
1294 
1295   if (length == 0) return;
1296   // Not guaranteed to be word atomic, but that doesn't matter
1297   // for anything but an oop array, which is covered by oop_arraycopy.
1298   Copy::conjoint_jbytes(src, dst, length);
1299 JRT_END
1300 
1301 JRT_LEAF(void, Runtime1::oop_arraycopy(HeapWord* src, HeapWord* dst, int num))
1302 #ifndef PRODUCT
1303   _oop_arraycopy_cnt++;
1304 #endif
1305 
1306   if (num == 0) return;
1307   BarrierSet* bs = Universe::heap()->barrier_set();
1308   assert(bs->has_write_ref_array_opt(), "Barrier set must have ref array opt");
1309   assert(bs->has_write_ref_array_pre_opt(), "For pre-barrier as well.");
1310   if (UseCompressedOops) {
1311     bs->write_ref_array_pre((narrowOop*)dst, num);
1312     Copy::conjoint_oops_atomic((narrowOop*) src, (narrowOop*) dst, num);
1313   } else {
1314     bs->write_ref_array_pre((oop*)dst, num);
1315     Copy::conjoint_oops_atomic((oop*) src, (oop*) dst, num);
1316   }
1317   bs->write_ref_array(dst, num);
1318 JRT_END
1319 
1320 
1321 JRT_LEAF(int, Runtime1::is_instance_of(oopDesc* mirror, oopDesc* obj))
1322   // had to return int instead of bool, otherwise there may be a mismatch
1323   // between the C calling convention and the Java one.
1324   // e.g., on x86, GCC may clear only %al when returning a bool false, but
1325   // JVM takes the whole %eax as the return value, which may misinterpret
1326   // the return value as a boolean true.
1327 
1328   assert(mirror != NULL, "should null-check on mirror before calling");
1329   Klass* k = java_lang_Class::as_Klass(mirror);
1330   return (k != NULL && obj != NULL && obj->is_a(k)) ? 1 : 0;
1331 JRT_END
1332 
1333 
1334 #ifndef PRODUCT
1335 void Runtime1::print_statistics() {
1336   tty->print_cr("C1 Runtime statistics:");
1337   tty->print_cr(" _resolve_invoke_virtual_cnt:     %d", SharedRuntime::_resolve_virtual_ctr);
1338   tty->print_cr(" _resolve_invoke_opt_virtual_cnt: %d", SharedRuntime::_resolve_opt_virtual_ctr);
1339   tty->print_cr(" _resolve_invoke_static_cnt:      %d", SharedRuntime::_resolve_static_ctr);
1340   tty->print_cr(" _handle_wrong_method_cnt:        %d", SharedRuntime::_wrong_method_ctr);
1341   tty->print_cr(" _ic_miss_cnt:                    %d", SharedRuntime::_ic_miss_ctr);
1342   tty->print_cr(" _generic_arraycopy_cnt:          %d", _generic_arraycopy_cnt);
1343   tty->print_cr(" _generic_arraycopystub_cnt:      %d", _generic_arraycopystub_cnt);
1344   tty->print_cr(" _byte_arraycopy_cnt:             %d", _byte_arraycopy_cnt);
1345   tty->print_cr(" _short_arraycopy_cnt:            %d", _short_arraycopy_cnt);
1346   tty->print_cr(" _int_arraycopy_cnt:              %d", _int_arraycopy_cnt);
1347   tty->print_cr(" _long_arraycopy_cnt:             %d", _long_arraycopy_cnt);
1348   tty->print_cr(" _primitive_arraycopy_cnt:        %d", _primitive_arraycopy_cnt);
1349   tty->print_cr(" _oop_arraycopy_cnt (C):          %d", Runtime1::_oop_arraycopy_cnt);
1350   tty->print_cr(" _oop_arraycopy_cnt (stub):       %d", _oop_arraycopy_cnt);
1351   tty->print_cr(" _arraycopy_slowcase_cnt:         %d", _arraycopy_slowcase_cnt);
1352   tty->print_cr(" _arraycopy_checkcast_cnt:        %d", _arraycopy_checkcast_cnt);
1353   tty->print_cr(" _arraycopy_checkcast_attempt_cnt:%d", _arraycopy_checkcast_attempt_cnt);
1354 
1355   tty->print_cr(" _new_type_array_slowcase_cnt:    %d", _new_type_array_slowcase_cnt);
1356   tty->print_cr(" _new_object_array_slowcase_cnt:  %d", _new_object_array_slowcase_cnt);
1357   tty->print_cr(" _new_instance_slowcase_cnt:      %d", _new_instance_slowcase_cnt);
1358   tty->print_cr(" _new_multi_array_slowcase_cnt:   %d", _new_multi_array_slowcase_cnt);
1359   tty->print_cr(" _monitorenter_slowcase_cnt:      %d", _monitorenter_slowcase_cnt);
1360   tty->print_cr(" _monitorexit_slowcase_cnt:       %d", _monitorexit_slowcase_cnt);
1361   tty->print_cr(" _patch_code_slowcase_cnt:        %d", _patch_code_slowcase_cnt);
1362 
1363   tty->print_cr(" _throw_range_check_exception_count:            %d:", _throw_range_check_exception_count);
1364   tty->print_cr(" _throw_index_exception_count:                  %d:", _throw_index_exception_count);
1365   tty->print_cr(" _throw_div0_exception_count:                   %d:", _throw_div0_exception_count);
1366   tty->print_cr(" _throw_null_pointer_exception_count:           %d:", _throw_null_pointer_exception_count);
1367   tty->print_cr(" _throw_class_cast_exception_count:             %d:", _throw_class_cast_exception_count);
1368   tty->print_cr(" _throw_incompatible_class_change_error_count:  %d:", _throw_incompatible_class_change_error_count);
1369   tty->print_cr(" _throw_array_store_exception_count:            %d:", _throw_array_store_exception_count);
1370   tty->print_cr(" _throw_count:                                  %d:", _throw_count);
1371 
1372   SharedRuntime::print_ic_miss_histogram();
1373   tty->cr();
1374 }
1375 #endif // PRODUCT