src/cpu/x86/vm/interp_masm_x86_64.cpp

Sat, 29 Sep 2012 06:40:00 -0400

author
coleenp
date
Sat, 29 Sep 2012 06:40:00 -0400
changeset 4142
d8ce2825b193
parent 4037
da91efe96a93
child 4299
f34d701e952e
permissions
-rw-r--r--

8000213: NPG: Should have renamed arrayKlass and typeArrayKlass
Summary: Capitalize these metadata types (and objArrayKlass)
Reviewed-by: stefank, twisti, kvn

     1 /*
     2  * Copyright (c) 2003, 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  */
    25 #include "precompiled.hpp"
    26 #include "interp_masm_x86_64.hpp"
    27 #include "interpreter/interpreter.hpp"
    28 #include "interpreter/interpreterRuntime.hpp"
    29 #include "oops/arrayOop.hpp"
    30 #include "oops/markOop.hpp"
    31 #include "oops/methodData.hpp"
    32 #include "oops/method.hpp"
    33 #include "prims/jvmtiExport.hpp"
    34 #include "prims/jvmtiRedefineClassesTrace.hpp"
    35 #include "prims/jvmtiThreadState.hpp"
    36 #include "runtime/basicLock.hpp"
    37 #include "runtime/biasedLocking.hpp"
    38 #include "runtime/sharedRuntime.hpp"
    39 #ifdef TARGET_OS_FAMILY_linux
    40 # include "thread_linux.inline.hpp"
    41 #endif
    42 #ifdef TARGET_OS_FAMILY_solaris
    43 # include "thread_solaris.inline.hpp"
    44 #endif
    45 #ifdef TARGET_OS_FAMILY_windows
    46 # include "thread_windows.inline.hpp"
    47 #endif
    48 #ifdef TARGET_OS_FAMILY_bsd
    49 # include "thread_bsd.inline.hpp"
    50 #endif
    53 // Implementation of InterpreterMacroAssembler
    55 #ifdef CC_INTERP
    56 void InterpreterMacroAssembler::get_method(Register reg) {
    57   movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
    58   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
    59 }
    60 #endif // CC_INTERP
    62 #ifndef CC_INTERP
    64 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
    65                                                   int number_of_arguments) {
    66   // interpreter specific
    67   //
    68   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
    69   //       since these are callee saved registers and no blocking/
    70   //       GC can happen in leaf calls.
    71   // Further Note: DO NOT save/restore bcp/locals. If a caller has
    72   // already saved them so that it can use esi/edi as temporaries
    73   // then a save/restore here will DESTROY the copy the caller
    74   // saved! There used to be a save_bcp() that only happened in
    75   // the ASSERT path (no restore_bcp). Which caused bizarre failures
    76   // when jvm built with ASSERTs.
    77 #ifdef ASSERT
    78   {
    79     Label L;
    80     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
    81     jcc(Assembler::equal, L);
    82     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    83          " last_sp != NULL");
    84     bind(L);
    85   }
    86 #endif
    87   // super call
    88   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
    89   // interpreter specific
    90   // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
    91   // but since they may not have been saved (and we don't want to
    92   // save thme here (see note above) the assert is invalid.
    93 }
    95 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
    96                                              Register java_thread,
    97                                              Register last_java_sp,
    98                                              address  entry_point,
    99                                              int      number_of_arguments,
   100                                              bool     check_exceptions) {
   101   // interpreter specific
   102   //
   103   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
   104   //       really make a difference for these runtime calls, since they are
   105   //       slow anyway. Btw., bcp must be saved/restored since it may change
   106   //       due to GC.
   107   // assert(java_thread == noreg , "not expecting a precomputed java thread");
   108   save_bcp();
   109 #ifdef ASSERT
   110   {
   111     Label L;
   112     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
   113     jcc(Assembler::equal, L);
   114     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
   115          " last_sp != NULL");
   116     bind(L);
   117   }
   118 #endif /* ASSERT */
   119   // super call
   120   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
   121                                entry_point, number_of_arguments,
   122                                check_exceptions);
   123   // interpreter specific
   124   restore_bcp();
   125   restore_locals();
   126 }
   129 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
   130   if (JvmtiExport::can_pop_frame()) {
   131     Label L;
   132     // Initiate popframe handling only if it is not already being
   133     // processed.  If the flag has the popframe_processing bit set, it
   134     // means that this code is called *during* popframe handling - we
   135     // don't want to reenter.
   136     // This method is only called just after the call into the vm in
   137     // call_VM_base, so the arg registers are available.
   138     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
   139     testl(c_rarg0, JavaThread::popframe_pending_bit);
   140     jcc(Assembler::zero, L);
   141     testl(c_rarg0, JavaThread::popframe_processing_bit);
   142     jcc(Assembler::notZero, L);
   143     // Call Interpreter::remove_activation_preserving_args_entry() to get the
   144     // address of the same-named entrypoint in the generated interpreter code.
   145     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
   146     jmp(rax);
   147     bind(L);
   148   }
   149 }
   152 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
   153   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   154   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
   155   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
   156   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
   157   switch (state) {
   158     case atos: movptr(rax, oop_addr);
   159                movptr(oop_addr, (int32_t)NULL_WORD);
   160                verify_oop(rax, state);              break;
   161     case ltos: movptr(rax, val_addr);                 break;
   162     case btos:                                   // fall through
   163     case ctos:                                   // fall through
   164     case stos:                                   // fall through
   165     case itos: movl(rax, val_addr);                 break;
   166     case ftos: movflt(xmm0, val_addr);              break;
   167     case dtos: movdbl(xmm0, val_addr);              break;
   168     case vtos: /* nothing to do */                  break;
   169     default  : ShouldNotReachHere();
   170   }
   171   // Clean up tos value in the thread object
   172   movl(tos_addr,  (int) ilgl);
   173   movl(val_addr,  (int32_t) NULL_WORD);
   174 }
   177 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
   178   if (JvmtiExport::can_force_early_return()) {
   179     Label L;
   180     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   181     testptr(c_rarg0, c_rarg0);
   182     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
   184     // Initiate earlyret handling only if it is not already being processed.
   185     // If the flag has the earlyret_processing bit set, it means that this code
   186     // is called *during* earlyret handling - we don't want to reenter.
   187     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
   188     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
   189     jcc(Assembler::notEqual, L);
   191     // Call Interpreter::remove_activation_early_entry() to get the address of the
   192     // same-named entrypoint in the generated interpreter code.
   193     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   194     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
   195     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
   196     jmp(rax);
   197     bind(L);
   198   }
   199 }
   202 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
   203   Register reg,
   204   int bcp_offset) {
   205   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
   206   movl(reg, Address(r13, bcp_offset));
   207   bswapl(reg);
   208   shrl(reg, 16);
   209 }
   212 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
   213                                                        int bcp_offset,
   214                                                        size_t index_size) {
   215   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
   216   if (index_size == sizeof(u2)) {
   217     load_unsigned_short(index, Address(r13, bcp_offset));
   218   } else if (index_size == sizeof(u4)) {
   219     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
   220     movl(index, Address(r13, bcp_offset));
   221     // Check if the secondary index definition is still ~x, otherwise
   222     // we have to change the following assembler code to calculate the
   223     // plain index.
   224     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
   225     notl(index);  // convert to plain index
   226   } else if (index_size == sizeof(u1)) {
   227     load_unsigned_byte(index, Address(r13, bcp_offset));
   228   } else {
   229     ShouldNotReachHere();
   230   }
   231 }
   234 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
   235                                                            Register index,
   236                                                            int bcp_offset,
   237                                                            size_t index_size) {
   238   assert_different_registers(cache, index);
   239   get_cache_index_at_bcp(index, bcp_offset, index_size);
   240   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   241   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   242   // convert from field index to ConstantPoolCacheEntry index
   243   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
   244   shll(index, 2);
   245 }
   248 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
   249                                                                         Register index,
   250                                                                         Register bytecode,
   251                                                                         int byte_no,
   252                                                                         int bcp_offset,
   253                                                                         size_t index_size) {
   254   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
   255   // We use a 32-bit load here since the layout of 64-bit words on
   256   // little-endian machines allow us that.
   257   movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
   258   const int shift_count = (1 + byte_no) * BitsPerByte;
   259   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
   260          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
   261          "correct shift count");
   262   shrl(bytecode, shift_count);
   263   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
   264   andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
   265 }
   268 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
   269                                                                Register tmp,
   270                                                                int bcp_offset,
   271                                                                size_t index_size) {
   272   assert(cache != tmp, "must use different register");
   273   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
   274   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   275   // convert from field index to ConstantPoolCacheEntry index
   276   // and from word offset to byte offset
   277   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
   278   shll(tmp, 2 + LogBytesPerWord);
   279   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   280   // skip past the header
   281   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
   282   addptr(cache, tmp);  // construct pointer to cache entry
   283 }
   285 // Load object from cpool->resolved_references(index)
   286 void InterpreterMacroAssembler::load_resolved_reference_at_index(
   287                                            Register result, Register index) {
   288   assert_different_registers(result, index);
   289   // convert from field index to resolved_references() index and from
   290   // word index to byte offset. Since this is a java object, it can be compressed
   291   Register tmp = index;  // reuse
   292   shll(tmp, LogBytesPerHeapOop);
   294   get_constant_pool(result);
   295   // load pointer for resolved_references[] objArray
   296   movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
   297   // JNIHandles::resolve(obj);
   298   movptr(result, Address(result, 0));
   299   // Add in the index
   300   addptr(result, tmp);
   301   load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
   302 }
   304 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
   305 // subtype of super_klass.
   306 //
   307 // Args:
   308 //      rax: superklass
   309 //      Rsub_klass: subklass
   310 //
   311 // Kills:
   312 //      rcx, rdi
   313 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
   314                                                   Label& ok_is_subtype) {
   315   assert(Rsub_klass != rax, "rax holds superklass");
   316   assert(Rsub_klass != r14, "r14 holds locals");
   317   assert(Rsub_klass != r13, "r13 holds bcp");
   318   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
   319   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
   321   // Profile the not-null value's klass.
   322   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
   324   // Do the check.
   325   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
   327   // Profile the failure of the check.
   328   profile_typecheck_failed(rcx); // blows rcx
   329 }
   333 // Java Expression Stack
   335 void InterpreterMacroAssembler::pop_ptr(Register r) {
   336   pop(r);
   337 }
   339 void InterpreterMacroAssembler::pop_i(Register r) {
   340   // XXX can't use pop currently, upper half non clean
   341   movl(r, Address(rsp, 0));
   342   addptr(rsp, wordSize);
   343 }
   345 void InterpreterMacroAssembler::pop_l(Register r) {
   346   movq(r, Address(rsp, 0));
   347   addptr(rsp, 2 * Interpreter::stackElementSize);
   348 }
   350 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
   351   movflt(r, Address(rsp, 0));
   352   addptr(rsp, wordSize);
   353 }
   355 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
   356   movdbl(r, Address(rsp, 0));
   357   addptr(rsp, 2 * Interpreter::stackElementSize);
   358 }
   360 void InterpreterMacroAssembler::push_ptr(Register r) {
   361   push(r);
   362 }
   364 void InterpreterMacroAssembler::push_i(Register r) {
   365   push(r);
   366 }
   368 void InterpreterMacroAssembler::push_l(Register r) {
   369   subptr(rsp, 2 * wordSize);
   370   movq(Address(rsp, 0), r);
   371 }
   373 void InterpreterMacroAssembler::push_f(XMMRegister r) {
   374   subptr(rsp, wordSize);
   375   movflt(Address(rsp, 0), r);
   376 }
   378 void InterpreterMacroAssembler::push_d(XMMRegister r) {
   379   subptr(rsp, 2 * wordSize);
   380   movdbl(Address(rsp, 0), r);
   381 }
   383 void InterpreterMacroAssembler::pop(TosState state) {
   384   switch (state) {
   385   case atos: pop_ptr();                 break;
   386   case btos:
   387   case ctos:
   388   case stos:
   389   case itos: pop_i();                   break;
   390   case ltos: pop_l();                   break;
   391   case ftos: pop_f();                   break;
   392   case dtos: pop_d();                   break;
   393   case vtos: /* nothing to do */        break;
   394   default:   ShouldNotReachHere();
   395   }
   396   verify_oop(rax, state);
   397 }
   399 void InterpreterMacroAssembler::push(TosState state) {
   400   verify_oop(rax, state);
   401   switch (state) {
   402   case atos: push_ptr();                break;
   403   case btos:
   404   case ctos:
   405   case stos:
   406   case itos: push_i();                  break;
   407   case ltos: push_l();                  break;
   408   case ftos: push_f();                  break;
   409   case dtos: push_d();                  break;
   410   case vtos: /* nothing to do */        break;
   411   default  : ShouldNotReachHere();
   412   }
   413 }
   416 // Helpers for swap and dup
   417 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
   418   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
   419 }
   421 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
   422   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
   423 }
   426 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   427   // set sender sp
   428   lea(r13, Address(rsp, wordSize));
   429   // record last_sp
   430   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
   431 }
   434 // Jump to from_interpreted entry of a call unless single stepping is possible
   435 // in this thread in which case we must call the i2i entry
   436 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
   437   prepare_to_jump_from_interpreted();
   439   if (JvmtiExport::can_post_interpreter_events()) {
   440     Label run_compiled_code;
   441     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
   442     // compiled code in threads for which the event is enabled.  Check here for
   443     // interp_only_mode if these events CAN be enabled.
   444     // interp_only is an int, on little endian it is sufficient to test the byte only
   445     // Is a cmpl faster?
   446     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
   447     jccb(Assembler::zero, run_compiled_code);
   448     jmp(Address(method, Method::interpreter_entry_offset()));
   449     bind(run_compiled_code);
   450   }
   452   jmp(Address(method, Method::from_interpreted_offset()));
   454 }
   457 // The following two routines provide a hook so that an implementation
   458 // can schedule the dispatch in two parts.  amd64 does not do this.
   459 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
   460   // Nothing amd64 specific to be done here
   461 }
   463 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
   464   dispatch_next(state, step);
   465 }
   467 void InterpreterMacroAssembler::dispatch_base(TosState state,
   468                                               address* table,
   469                                               bool verifyoop) {
   470   verify_FPU(1, state);
   471   if (VerifyActivationFrameSize) {
   472     Label L;
   473     mov(rcx, rbp);
   474     subptr(rcx, rsp);
   475     int32_t min_frame_size =
   476       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
   477       wordSize;
   478     cmpptr(rcx, (int32_t)min_frame_size);
   479     jcc(Assembler::greaterEqual, L);
   480     stop("broken stack frame");
   481     bind(L);
   482   }
   483   if (verifyoop) {
   484     verify_oop(rax, state);
   485   }
   486   lea(rscratch1, ExternalAddress((address)table));
   487   jmp(Address(rscratch1, rbx, Address::times_8));
   488 }
   490 void InterpreterMacroAssembler::dispatch_only(TosState state) {
   491   dispatch_base(state, Interpreter::dispatch_table(state));
   492 }
   494 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
   495   dispatch_base(state, Interpreter::normal_table(state));
   496 }
   498 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
   499   dispatch_base(state, Interpreter::normal_table(state), false);
   500 }
   503 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
   504   // load next bytecode (load before advancing r13 to prevent AGI)
   505   load_unsigned_byte(rbx, Address(r13, step));
   506   // advance r13
   507   increment(r13, step);
   508   dispatch_base(state, Interpreter::dispatch_table(state));
   509 }
   511 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
   512   // load current bytecode
   513   load_unsigned_byte(rbx, Address(r13, 0));
   514   dispatch_base(state, table);
   515 }
   517 // remove activation
   518 //
   519 // Unlock the receiver if this is a synchronized method.
   520 // Unlock any Java monitors from syncronized blocks.
   521 // Remove the activation from the stack.
   522 //
   523 // If there are locked Java monitors
   524 //    If throw_monitor_exception
   525 //       throws IllegalMonitorStateException
   526 //    Else if install_monitor_exception
   527 //       installs IllegalMonitorStateException
   528 //    Else
   529 //       no error processing
   530 void InterpreterMacroAssembler::remove_activation(
   531         TosState state,
   532         Register ret_addr,
   533         bool throw_monitor_exception,
   534         bool install_monitor_exception,
   535         bool notify_jvmdi) {
   536   // Note: Registers rdx xmm0 may be in use for the
   537   // result check if synchronized method
   538   Label unlocked, unlock, no_unlock;
   540   // get the value of _do_not_unlock_if_synchronized into rdx
   541   const Address do_not_unlock_if_synchronized(r15_thread,
   542     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
   543   movbool(rdx, do_not_unlock_if_synchronized);
   544   movbool(do_not_unlock_if_synchronized, false); // reset the flag
   546  // get method access flags
   547   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
   548   movl(rcx, Address(rbx, Method::access_flags_offset()));
   549   testl(rcx, JVM_ACC_SYNCHRONIZED);
   550   jcc(Assembler::zero, unlocked);
   552   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
   553   // is set.
   554   testbool(rdx);
   555   jcc(Assembler::notZero, no_unlock);
   557   // unlock monitor
   558   push(state); // save result
   560   // BasicObjectLock will be first in list, since this is a
   561   // synchronized method. However, need to check that the object has
   562   // not been unlocked by an explicit monitorexit bytecode.
   563   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
   564                         wordSize - (int) sizeof(BasicObjectLock));
   565   // We use c_rarg1 so that if we go slow path it will be the correct
   566   // register for unlock_object to pass to VM directly
   567   lea(c_rarg1, monitor); // address of first monitor
   569   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
   570   testptr(rax, rax);
   571   jcc(Assembler::notZero, unlock);
   573   pop(state);
   574   if (throw_monitor_exception) {
   575     // Entry already unlocked, need to throw exception
   576     call_VM(noreg, CAST_FROM_FN_PTR(address,
   577                    InterpreterRuntime::throw_illegal_monitor_state_exception));
   578     should_not_reach_here();
   579   } else {
   580     // Monitor already unlocked during a stack unroll. If requested,
   581     // install an illegal_monitor_state_exception.  Continue with
   582     // stack unrolling.
   583     if (install_monitor_exception) {
   584       call_VM(noreg, CAST_FROM_FN_PTR(address,
   585                      InterpreterRuntime::new_illegal_monitor_state_exception));
   586     }
   587     jmp(unlocked);
   588   }
   590   bind(unlock);
   591   unlock_object(c_rarg1);
   592   pop(state);
   594   // Check that for block-structured locking (i.e., that all locked
   595   // objects has been unlocked)
   596   bind(unlocked);
   598   // rax: Might contain return value
   600   // Check that all monitors are unlocked
   601   {
   602     Label loop, exception, entry, restart;
   603     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
   604     const Address monitor_block_top(
   605         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
   606     const Address monitor_block_bot(
   607         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
   609     bind(restart);
   610     // We use c_rarg1 so that if we go slow path it will be the correct
   611     // register for unlock_object to pass to VM directly
   612     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
   613                                   // with top-most entry
   614     lea(rbx, monitor_block_bot);  // points to word before bottom of
   615                                   // monitor block
   616     jmp(entry);
   618     // Entry already locked, need to throw exception
   619     bind(exception);
   621     if (throw_monitor_exception) {
   622       // Throw exception
   623       MacroAssembler::call_VM(noreg,
   624                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
   625                                    throw_illegal_monitor_state_exception));
   626       should_not_reach_here();
   627     } else {
   628       // Stack unrolling. Unlock object and install illegal_monitor_exception.
   629       // Unlock does not block, so don't have to worry about the frame.
   630       // We don't have to preserve c_rarg1 since we are going to throw an exception.
   632       push(state);
   633       unlock_object(c_rarg1);
   634       pop(state);
   636       if (install_monitor_exception) {
   637         call_VM(noreg, CAST_FROM_FN_PTR(address,
   638                                         InterpreterRuntime::
   639                                         new_illegal_monitor_state_exception));
   640       }
   642       jmp(restart);
   643     }
   645     bind(loop);
   646     // check if current entry is used
   647     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
   648     jcc(Assembler::notEqual, exception);
   650     addptr(c_rarg1, entry_size); // otherwise advance to next entry
   651     bind(entry);
   652     cmpptr(c_rarg1, rbx); // check if bottom reached
   653     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
   654   }
   656   bind(no_unlock);
   658   // jvmti support
   659   if (notify_jvmdi) {
   660     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
   661   } else {
   662     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
   663   }
   665   // remove activation
   666   // get sender sp
   667   movptr(rbx,
   668          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
   669   leave();                           // remove frame anchor
   670   pop(ret_addr);                     // get return address
   671   mov(rsp, rbx);                     // set sp to sender sp
   672 }
   674 #endif // C_INTERP
   676 // Lock object
   677 //
   678 // Args:
   679 //      c_rarg1: BasicObjectLock to be used for locking
   680 //
   681 // Kills:
   682 //      rax
   683 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
   684 //      rscratch1, rscratch2 (scratch regs)
   685 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
   686   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
   688   if (UseHeavyMonitors) {
   689     call_VM(noreg,
   690             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   691             lock_reg);
   692   } else {
   693     Label done;
   695     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
   696     const Register obj_reg = c_rarg3; // Will contain the oop
   698     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
   699     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
   700     const int mark_offset = lock_offset +
   701                             BasicLock::displaced_header_offset_in_bytes();
   703     Label slow_case;
   705     // Load object pointer into obj_reg %c_rarg3
   706     movptr(obj_reg, Address(lock_reg, obj_offset));
   708     if (UseBiasedLocking) {
   709       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
   710     }
   712     // Load immediate 1 into swap_reg %rax
   713     movl(swap_reg, 1);
   715     // Load (object->mark() | 1) into swap_reg %rax
   716     orptr(swap_reg, Address(obj_reg, 0));
   718     // Save (object->mark() | 1) into BasicLock's displaced header
   719     movptr(Address(lock_reg, mark_offset), swap_reg);
   721     assert(lock_offset == 0,
   722            "displached header must be first word in BasicObjectLock");
   724     if (os::is_MP()) lock();
   725     cmpxchgptr(lock_reg, Address(obj_reg, 0));
   726     if (PrintBiasedLockingStatistics) {
   727       cond_inc32(Assembler::zero,
   728                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   729     }
   730     jcc(Assembler::zero, done);
   732     // Test if the oopMark is an obvious stack pointer, i.e.,
   733     //  1) (mark & 7) == 0, and
   734     //  2) rsp <= mark < mark + os::pagesize()
   735     //
   736     // These 3 tests can be done by evaluating the following
   737     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
   738     // assuming both stack pointer and pagesize have their
   739     // least significant 3 bits clear.
   740     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
   741     subptr(swap_reg, rsp);
   742     andptr(swap_reg, 7 - os::vm_page_size());
   744     // Save the test result, for recursive case, the result is zero
   745     movptr(Address(lock_reg, mark_offset), swap_reg);
   747     if (PrintBiasedLockingStatistics) {
   748       cond_inc32(Assembler::zero,
   749                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   750     }
   751     jcc(Assembler::zero, done);
   753     bind(slow_case);
   755     // Call the runtime routine for slow case
   756     call_VM(noreg,
   757             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   758             lock_reg);
   760     bind(done);
   761   }
   762 }
   765 // Unlocks an object. Used in monitorexit bytecode and
   766 // remove_activation.  Throws an IllegalMonitorException if object is
   767 // not locked by current thread.
   768 //
   769 // Args:
   770 //      c_rarg1: BasicObjectLock for lock
   771 //
   772 // Kills:
   773 //      rax
   774 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
   775 //      rscratch1, rscratch2 (scratch regs)
   776 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
   777   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
   779   if (UseHeavyMonitors) {
   780     call_VM(noreg,
   781             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   782             lock_reg);
   783   } else {
   784     Label done;
   786     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
   787     const Register header_reg = c_rarg2;  // Will contain the old oopMark
   788     const Register obj_reg    = c_rarg3;  // Will contain the oop
   790     save_bcp(); // Save in case of exception
   792     // Convert from BasicObjectLock structure to object and BasicLock
   793     // structure Store the BasicLock address into %rax
   794     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
   796     // Load oop into obj_reg(%c_rarg3)
   797     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
   799     // Free entry
   800     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
   802     if (UseBiasedLocking) {
   803       biased_locking_exit(obj_reg, header_reg, done);
   804     }
   806     // Load the old header from BasicLock structure
   807     movptr(header_reg, Address(swap_reg,
   808                                BasicLock::displaced_header_offset_in_bytes()));
   810     // Test for recursion
   811     testptr(header_reg, header_reg);
   813     // zero for recursive case
   814     jcc(Assembler::zero, done);
   816     // Atomic swap back the old header
   817     if (os::is_MP()) lock();
   818     cmpxchgptr(header_reg, Address(obj_reg, 0));
   820     // zero for recursive case
   821     jcc(Assembler::zero, done);
   823     // Call the runtime routine for slow case.
   824     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
   825          obj_reg); // restore obj
   826     call_VM(noreg,
   827             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   828             lock_reg);
   830     bind(done);
   832     restore_bcp();
   833   }
   834 }
   836 #ifndef CC_INTERP
   838 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
   839                                                          Label& zero_continue) {
   840   assert(ProfileInterpreter, "must be profiling interpreter");
   841   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
   842   testptr(mdp, mdp);
   843   jcc(Assembler::zero, zero_continue);
   844 }
   847 // Set the method data pointer for the current bcp.
   848 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
   849   assert(ProfileInterpreter, "must be profiling interpreter");
   850   Label set_mdp;
   851   push(rax);
   852   push(rbx);
   854   get_method(rbx);
   855   // Test MDO to avoid the call if it is NULL.
   856   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
   857   testptr(rax, rax);
   858   jcc(Assembler::zero, set_mdp);
   859   // rbx: method
   860   // r13: bcp
   861   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
   862   // rax: mdi
   863   // mdo is guaranteed to be non-zero here, we checked for it before the call.
   864   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
   865   addptr(rbx, in_bytes(MethodData::data_offset()));
   866   addptr(rax, rbx);
   867   bind(set_mdp);
   868   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
   869   pop(rbx);
   870   pop(rax);
   871 }
   873 void InterpreterMacroAssembler::verify_method_data_pointer() {
   874   assert(ProfileInterpreter, "must be profiling interpreter");
   875 #ifdef ASSERT
   876   Label verify_continue;
   877   push(rax);
   878   push(rbx);
   879   push(c_rarg3);
   880   push(c_rarg2);
   881   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
   882   get_method(rbx);
   884   // If the mdp is valid, it will point to a DataLayout header which is
   885   // consistent with the bcp.  The converse is highly probable also.
   886   load_unsigned_short(c_rarg2,
   887                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
   888   addptr(c_rarg2, Address(rbx, Method::const_offset()));
   889   lea(c_rarg2, Address(c_rarg2, ConstMethod::codes_offset()));
   890   cmpptr(c_rarg2, r13);
   891   jcc(Assembler::equal, verify_continue);
   892   // rbx: method
   893   // r13: bcp
   894   // c_rarg3: mdp
   895   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
   896                rbx, r13, c_rarg3);
   897   bind(verify_continue);
   898   pop(c_rarg2);
   899   pop(c_rarg3);
   900   pop(rbx);
   901   pop(rax);
   902 #endif // ASSERT
   903 }
   906 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
   907                                                 int constant,
   908                                                 Register value) {
   909   assert(ProfileInterpreter, "must be profiling interpreter");
   910   Address data(mdp_in, constant);
   911   movptr(data, value);
   912 }
   915 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   916                                                       int constant,
   917                                                       bool decrement) {
   918   // Counter address
   919   Address data(mdp_in, constant);
   921   increment_mdp_data_at(data, decrement);
   922 }
   924 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
   925                                                       bool decrement) {
   926   assert(ProfileInterpreter, "must be profiling interpreter");
   927   // %%% this does 64bit counters at best it is wasting space
   928   // at worst it is a rare bug when counters overflow
   930   if (decrement) {
   931     // Decrement the register.  Set condition codes.
   932     addptr(data, (int32_t) -DataLayout::counter_increment);
   933     // If the decrement causes the counter to overflow, stay negative
   934     Label L;
   935     jcc(Assembler::negative, L);
   936     addptr(data, (int32_t) DataLayout::counter_increment);
   937     bind(L);
   938   } else {
   939     assert(DataLayout::counter_increment == 1,
   940            "flow-free idiom only works with 1");
   941     // Increment the register.  Set carry flag.
   942     addptr(data, DataLayout::counter_increment);
   943     // If the increment causes the counter to overflow, pull back by 1.
   944     sbbptr(data, (int32_t)0);
   945   }
   946 }
   949 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   950                                                       Register reg,
   951                                                       int constant,
   952                                                       bool decrement) {
   953   Address data(mdp_in, reg, Address::times_1, constant);
   955   increment_mdp_data_at(data, decrement);
   956 }
   958 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
   959                                                 int flag_byte_constant) {
   960   assert(ProfileInterpreter, "must be profiling interpreter");
   961   int header_offset = in_bytes(DataLayout::header_offset());
   962   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
   963   // Set the flag
   964   orl(Address(mdp_in, header_offset), header_bits);
   965 }
   969 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
   970                                                  int offset,
   971                                                  Register value,
   972                                                  Register test_value_out,
   973                                                  Label& not_equal_continue) {
   974   assert(ProfileInterpreter, "must be profiling interpreter");
   975   if (test_value_out == noreg) {
   976     cmpptr(value, Address(mdp_in, offset));
   977   } else {
   978     // Put the test value into a register, so caller can use it:
   979     movptr(test_value_out, Address(mdp_in, offset));
   980     cmpptr(test_value_out, value);
   981   }
   982   jcc(Assembler::notEqual, not_equal_continue);
   983 }
   986 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
   987                                                      int offset_of_disp) {
   988   assert(ProfileInterpreter, "must be profiling interpreter");
   989   Address disp_address(mdp_in, offset_of_disp);
   990   addptr(mdp_in, disp_address);
   991   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
   992 }
   995 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
   996                                                      Register reg,
   997                                                      int offset_of_disp) {
   998   assert(ProfileInterpreter, "must be profiling interpreter");
   999   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
  1000   addptr(mdp_in, disp_address);
  1001   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1005 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
  1006                                                        int constant) {
  1007   assert(ProfileInterpreter, "must be profiling interpreter");
  1008   addptr(mdp_in, constant);
  1009   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
  1013 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
  1014   assert(ProfileInterpreter, "must be profiling interpreter");
  1015   push(return_bci); // save/restore across call_VM
  1016   call_VM(noreg,
  1017           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
  1018           return_bci);
  1019   pop(return_bci);
  1023 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
  1024                                                      Register bumped_count) {
  1025   if (ProfileInterpreter) {
  1026     Label profile_continue;
  1028     // If no method data exists, go to profile_continue.
  1029     // Otherwise, assign to mdp
  1030     test_method_data_pointer(mdp, profile_continue);
  1032     // We are taking a branch.  Increment the taken count.
  1033     // We inline increment_mdp_data_at to return bumped_count in a register
  1034     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
  1035     Address data(mdp, in_bytes(JumpData::taken_offset()));
  1036     movptr(bumped_count, data);
  1037     assert(DataLayout::counter_increment == 1,
  1038             "flow-free idiom only works with 1");
  1039     addptr(bumped_count, DataLayout::counter_increment);
  1040     sbbptr(bumped_count, 0);
  1041     movptr(data, bumped_count); // Store back out
  1043     // The method data pointer needs to be updated to reflect the new target.
  1044     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
  1045     bind(profile_continue);
  1050 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
  1051   if (ProfileInterpreter) {
  1052     Label profile_continue;
  1054     // If no method data exists, go to profile_continue.
  1055     test_method_data_pointer(mdp, profile_continue);
  1057     // We are taking a branch.  Increment the not taken count.
  1058     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
  1060     // The method data pointer needs to be updated to correspond to
  1061     // the next bytecode
  1062     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
  1063     bind(profile_continue);
  1068 void InterpreterMacroAssembler::profile_call(Register mdp) {
  1069   if (ProfileInterpreter) {
  1070     Label profile_continue;
  1072     // If no method data exists, go to profile_continue.
  1073     test_method_data_pointer(mdp, profile_continue);
  1075     // We are making a call.  Increment the count.
  1076     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1078     // The method data pointer needs to be updated to reflect the new target.
  1079     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
  1080     bind(profile_continue);
  1085 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
  1086   if (ProfileInterpreter) {
  1087     Label profile_continue;
  1089     // If no method data exists, go to profile_continue.
  1090     test_method_data_pointer(mdp, profile_continue);
  1092     // We are making a call.  Increment the count.
  1093     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1095     // The method data pointer needs to be updated to reflect the new target.
  1096     update_mdp_by_constant(mdp,
  1097                            in_bytes(VirtualCallData::
  1098                                     virtual_call_data_size()));
  1099     bind(profile_continue);
  1104 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
  1105                                                      Register mdp,
  1106                                                      Register reg2,
  1107                                                      bool receiver_can_be_null) {
  1108   if (ProfileInterpreter) {
  1109     Label profile_continue;
  1111     // If no method data exists, go to profile_continue.
  1112     test_method_data_pointer(mdp, profile_continue);
  1114     Label skip_receiver_profile;
  1115     if (receiver_can_be_null) {
  1116       Label not_null;
  1117       testptr(receiver, receiver);
  1118       jccb(Assembler::notZero, not_null);
  1119       // We are making a call.  Increment the count for null receiver.
  1120       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1121       jmp(skip_receiver_profile);
  1122       bind(not_null);
  1125     // Record the receiver type.
  1126     record_klass_in_profile(receiver, mdp, reg2, true);
  1127     bind(skip_receiver_profile);
  1129     // The method data pointer needs to be updated to reflect the new target.
  1130     update_mdp_by_constant(mdp,
  1131                            in_bytes(VirtualCallData::
  1132                                     virtual_call_data_size()));
  1133     bind(profile_continue);
  1137 // This routine creates a state machine for updating the multi-row
  1138 // type profile at a virtual call site (or other type-sensitive bytecode).
  1139 // The machine visits each row (of receiver/count) until the receiver type
  1140 // is found, or until it runs out of rows.  At the same time, it remembers
  1141 // the location of the first empty row.  (An empty row records null for its
  1142 // receiver, and can be allocated for a newly-observed receiver type.)
  1143 // Because there are two degrees of freedom in the state, a simple linear
  1144 // search will not work; it must be a decision tree.  Hence this helper
  1145 // function is recursive, to generate the required tree structured code.
  1146 // It's the interpreter, so we are trading off code space for speed.
  1147 // See below for example code.
  1148 void InterpreterMacroAssembler::record_klass_in_profile_helper(
  1149                                         Register receiver, Register mdp,
  1150                                         Register reg2, int start_row,
  1151                                         Label& done, bool is_virtual_call) {
  1152   if (TypeProfileWidth == 0) {
  1153     if (is_virtual_call) {
  1154       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1156     return;
  1159   int last_row = VirtualCallData::row_limit() - 1;
  1160   assert(start_row <= last_row, "must be work left to do");
  1161   // Test this row for both the receiver and for null.
  1162   // Take any of three different outcomes:
  1163   //   1. found receiver => increment count and goto done
  1164   //   2. found null => keep looking for case 1, maybe allocate this cell
  1165   //   3. found something else => keep looking for cases 1 and 2
  1166   // Case 3 is handled by a recursive call.
  1167   for (int row = start_row; row <= last_row; row++) {
  1168     Label next_test;
  1169     bool test_for_null_also = (row == start_row);
  1171     // See if the receiver is receiver[n].
  1172     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
  1173     test_mdp_data_at(mdp, recvr_offset, receiver,
  1174                      (test_for_null_also ? reg2 : noreg),
  1175                      next_test);
  1176     // (Reg2 now contains the receiver from the CallData.)
  1178     // The receiver is receiver[n].  Increment count[n].
  1179     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
  1180     increment_mdp_data_at(mdp, count_offset);
  1181     jmp(done);
  1182     bind(next_test);
  1184     if (test_for_null_also) {
  1185       Label found_null;
  1186       // Failed the equality check on receiver[n]...  Test for null.
  1187       testptr(reg2, reg2);
  1188       if (start_row == last_row) {
  1189         // The only thing left to do is handle the null case.
  1190         if (is_virtual_call) {
  1191           jccb(Assembler::zero, found_null);
  1192           // Receiver did not match any saved receiver and there is no empty row for it.
  1193           // Increment total counter to indicate polymorphic case.
  1194           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1195           jmp(done);
  1196           bind(found_null);
  1197         } else {
  1198           jcc(Assembler::notZero, done);
  1200         break;
  1202       // Since null is rare, make it be the branch-taken case.
  1203       jcc(Assembler::zero, found_null);
  1205       // Put all the "Case 3" tests here.
  1206       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
  1208       // Found a null.  Keep searching for a matching receiver,
  1209       // but remember that this is an empty (unused) slot.
  1210       bind(found_null);
  1214   // In the fall-through case, we found no matching receiver, but we
  1215   // observed the receiver[start_row] is NULL.
  1217   // Fill in the receiver field and increment the count.
  1218   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
  1219   set_mdp_data_at(mdp, recvr_offset, receiver);
  1220   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
  1221   movl(reg2, DataLayout::counter_increment);
  1222   set_mdp_data_at(mdp, count_offset, reg2);
  1223   if (start_row > 0) {
  1224     jmp(done);
  1228 // Example state machine code for three profile rows:
  1229 //   // main copy of decision tree, rooted at row[1]
  1230 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
  1231 //   if (row[0].rec != NULL) {
  1232 //     // inner copy of decision tree, rooted at row[1]
  1233 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1234 //     if (row[1].rec != NULL) {
  1235 //       // degenerate decision tree, rooted at row[2]
  1236 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1237 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
  1238 //       row[2].init(rec); goto done;
  1239 //     } else {
  1240 //       // remember row[1] is empty
  1241 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1242 //       row[1].init(rec); goto done;
  1243 //     }
  1244 //   } else {
  1245 //     // remember row[0] is empty
  1246 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1247 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
  1248 //     row[0].init(rec); goto done;
  1249 //   }
  1250 //   done:
  1252 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
  1253                                                         Register mdp, Register reg2,
  1254                                                         bool is_virtual_call) {
  1255   assert(ProfileInterpreter, "must be profiling");
  1256   Label done;
  1258   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
  1260   bind (done);
  1263 void InterpreterMacroAssembler::profile_ret(Register return_bci,
  1264                                             Register mdp) {
  1265   if (ProfileInterpreter) {
  1266     Label profile_continue;
  1267     uint row;
  1269     // If no method data exists, go to profile_continue.
  1270     test_method_data_pointer(mdp, profile_continue);
  1272     // Update the total ret count.
  1273     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1275     for (row = 0; row < RetData::row_limit(); row++) {
  1276       Label next_test;
  1278       // See if return_bci is equal to bci[n]:
  1279       test_mdp_data_at(mdp,
  1280                        in_bytes(RetData::bci_offset(row)),
  1281                        return_bci, noreg,
  1282                        next_test);
  1284       // return_bci is equal to bci[n].  Increment the count.
  1285       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
  1287       // The method data pointer needs to be updated to reflect the new target.
  1288       update_mdp_by_offset(mdp,
  1289                            in_bytes(RetData::bci_displacement_offset(row)));
  1290       jmp(profile_continue);
  1291       bind(next_test);
  1294     update_mdp_for_ret(return_bci);
  1296     bind(profile_continue);
  1301 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
  1302   if (ProfileInterpreter) {
  1303     Label profile_continue;
  1305     // If no method data exists, go to profile_continue.
  1306     test_method_data_pointer(mdp, profile_continue);
  1308     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
  1310     // The method data pointer needs to be updated.
  1311     int mdp_delta = in_bytes(BitData::bit_data_size());
  1312     if (TypeProfileCasts) {
  1313       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1315     update_mdp_by_constant(mdp, mdp_delta);
  1317     bind(profile_continue);
  1322 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
  1323   if (ProfileInterpreter && TypeProfileCasts) {
  1324     Label profile_continue;
  1326     // If no method data exists, go to profile_continue.
  1327     test_method_data_pointer(mdp, profile_continue);
  1329     int count_offset = in_bytes(CounterData::count_offset());
  1330     // Back up the address, since we have already bumped the mdp.
  1331     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
  1333     // *Decrement* the counter.  We expect to see zero or small negatives.
  1334     increment_mdp_data_at(mdp, count_offset, true);
  1336     bind (profile_continue);
  1341 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
  1342   if (ProfileInterpreter) {
  1343     Label profile_continue;
  1345     // If no method data exists, go to profile_continue.
  1346     test_method_data_pointer(mdp, profile_continue);
  1348     // The method data pointer needs to be updated.
  1349     int mdp_delta = in_bytes(BitData::bit_data_size());
  1350     if (TypeProfileCasts) {
  1351       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1353       // Record the object type.
  1354       record_klass_in_profile(klass, mdp, reg2, false);
  1356     update_mdp_by_constant(mdp, mdp_delta);
  1358     bind(profile_continue);
  1363 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
  1364   if (ProfileInterpreter) {
  1365     Label profile_continue;
  1367     // If no method data exists, go to profile_continue.
  1368     test_method_data_pointer(mdp, profile_continue);
  1370     // Update the default case count
  1371     increment_mdp_data_at(mdp,
  1372                           in_bytes(MultiBranchData::default_count_offset()));
  1374     // The method data pointer needs to be updated.
  1375     update_mdp_by_offset(mdp,
  1376                          in_bytes(MultiBranchData::
  1377                                   default_displacement_offset()));
  1379     bind(profile_continue);
  1384 void InterpreterMacroAssembler::profile_switch_case(Register index,
  1385                                                     Register mdp,
  1386                                                     Register reg2) {
  1387   if (ProfileInterpreter) {
  1388     Label profile_continue;
  1390     // If no method data exists, go to profile_continue.
  1391     test_method_data_pointer(mdp, profile_continue);
  1393     // Build the base (index * per_case_size_in_bytes()) +
  1394     // case_array_offset_in_bytes()
  1395     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
  1396     imulptr(index, reg2); // XXX l ?
  1397     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
  1399     // Update the case count
  1400     increment_mdp_data_at(mdp,
  1401                           index,
  1402                           in_bytes(MultiBranchData::relative_count_offset()));
  1404     // The method data pointer needs to be updated.
  1405     update_mdp_by_offset(mdp,
  1406                          index,
  1407                          in_bytes(MultiBranchData::
  1408                                   relative_displacement_offset()));
  1410     bind(profile_continue);
  1416 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
  1417   if (state == atos) {
  1418     MacroAssembler::verify_oop(reg);
  1422 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
  1424 #endif // !CC_INTERP
  1427 void InterpreterMacroAssembler::notify_method_entry() {
  1428   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1429   // track stack depth.  If it is possible to enter interp_only_mode we add
  1430   // the code to check if the event should be sent.
  1431   if (JvmtiExport::can_post_interpreter_events()) {
  1432     Label L;
  1433     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1434     testl(rdx, rdx);
  1435     jcc(Assembler::zero, L);
  1436     call_VM(noreg, CAST_FROM_FN_PTR(address,
  1437                                     InterpreterRuntime::post_method_entry));
  1438     bind(L);
  1442     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1443     get_method(c_rarg1);
  1444     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
  1445                  r15_thread, c_rarg1);
  1448   // RedefineClasses() tracing support for obsolete method entry
  1449   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
  1450     get_method(c_rarg1);
  1451     call_VM_leaf(
  1452       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
  1453       r15_thread, c_rarg1);
  1458 void InterpreterMacroAssembler::notify_method_exit(
  1459     TosState state, NotifyMethodExitMode mode) {
  1460   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1461   // track stack depth.  If it is possible to enter interp_only_mode we add
  1462   // the code to check if the event should be sent.
  1463   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
  1464     Label L;
  1465     // Note: frame::interpreter_frame_result has a dependency on how the
  1466     // method result is saved across the call to post_method_exit. If this
  1467     // is changed then the interpreter_frame_result implementation will
  1468     // need to be updated too.
  1470     // For c++ interpreter the result is always stored at a known location in the frame
  1471     // template interpreter will leave it on the top of the stack.
  1472     NOT_CC_INTERP(push(state);)
  1473     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1474     testl(rdx, rdx);
  1475     jcc(Assembler::zero, L);
  1476     call_VM(noreg,
  1477             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
  1478     bind(L);
  1479     NOT_CC_INTERP(pop(state));
  1483     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1484     NOT_CC_INTERP(push(state));
  1485     get_method(c_rarg1);
  1486     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
  1487                  r15_thread, c_rarg1);
  1488     NOT_CC_INTERP(pop(state));
  1492 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
  1493 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
  1494                                                         int increment, int mask,
  1495                                                         Register scratch, bool preloaded,
  1496                                                         Condition cond, Label* where) {
  1497   if (!preloaded) {
  1498     movl(scratch, counter_addr);
  1500   incrementl(scratch, increment);
  1501   movl(counter_addr, scratch);
  1502   andl(scratch, mask);
  1503   jcc(cond, *where);

mercurial