src/cpu/x86/vm/interp_masm_x86_64.cpp

Thu, 08 Sep 2011 10:12:25 +0200

author
bdelsart
date
Thu, 08 Sep 2011 10:12:25 +0200
changeset 3130
5432047c7db7
parent 3050
fdb992d83a87
child 3156
f08d439fab8c
permissions
-rw-r--r--

7087445: Improve platform independence of JSR292 shared code
Summary: changes necessary for some JSR292 ports
Reviewed-by: jrose, dholmes

     1 /*
     2  * Copyright (c) 2003, 2011, 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/methodDataOop.hpp"
    32 #include "oops/methodOop.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
    50 // Implementation of InterpreterMacroAssembler
    52 #ifdef CC_INTERP
    53 void InterpreterMacroAssembler::get_method(Register reg) {
    54   movptr(reg, Address(rbp, -((int)sizeof(BytecodeInterpreter) + 2 * wordSize)));
    55   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
    56 }
    57 #endif // CC_INTERP
    59 #ifndef CC_INTERP
    61 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
    62                                                   int number_of_arguments) {
    63   // interpreter specific
    64   //
    65   // Note: No need to save/restore bcp & locals (r13 & r14) pointer
    66   //       since these are callee saved registers and no blocking/
    67   //       GC can happen in leaf calls.
    68   // Further Note: DO NOT save/restore bcp/locals. If a caller has
    69   // already saved them so that it can use esi/edi as temporaries
    70   // then a save/restore here will DESTROY the copy the caller
    71   // saved! There used to be a save_bcp() that only happened in
    72   // the ASSERT path (no restore_bcp). Which caused bizarre failures
    73   // when jvm built with ASSERTs.
    74 #ifdef ASSERT
    75   {
    76     Label L;
    77     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
    78     jcc(Assembler::equal, L);
    79     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
    80          " last_sp != NULL");
    81     bind(L);
    82   }
    83 #endif
    84   // super call
    85   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
    86   // interpreter specific
    87   // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
    88   // but since they may not have been saved (and we don't want to
    89   // save thme here (see note above) the assert is invalid.
    90 }
    92 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
    93                                              Register java_thread,
    94                                              Register last_java_sp,
    95                                              address  entry_point,
    96                                              int      number_of_arguments,
    97                                              bool     check_exceptions) {
    98   // interpreter specific
    99   //
   100   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
   101   //       really make a difference for these runtime calls, since they are
   102   //       slow anyway. Btw., bcp must be saved/restored since it may change
   103   //       due to GC.
   104   // assert(java_thread == noreg , "not expecting a precomputed java thread");
   105   save_bcp();
   106 #ifdef ASSERT
   107   {
   108     Label L;
   109     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
   110     jcc(Assembler::equal, L);
   111     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
   112          " last_sp != NULL");
   113     bind(L);
   114   }
   115 #endif /* ASSERT */
   116   // super call
   117   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
   118                                entry_point, number_of_arguments,
   119                                check_exceptions);
   120   // interpreter specific
   121   restore_bcp();
   122   restore_locals();
   123 }
   126 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
   127   if (JvmtiExport::can_pop_frame()) {
   128     Label L;
   129     // Initiate popframe handling only if it is not already being
   130     // processed.  If the flag has the popframe_processing bit set, it
   131     // means that this code is called *during* popframe handling - we
   132     // don't want to reenter.
   133     // This method is only called just after the call into the vm in
   134     // call_VM_base, so the arg registers are available.
   135     movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
   136     testl(c_rarg0, JavaThread::popframe_pending_bit);
   137     jcc(Assembler::zero, L);
   138     testl(c_rarg0, JavaThread::popframe_processing_bit);
   139     jcc(Assembler::notZero, L);
   140     // Call Interpreter::remove_activation_preserving_args_entry() to get the
   141     // address of the same-named entrypoint in the generated interpreter code.
   142     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
   143     jmp(rax);
   144     bind(L);
   145   }
   146 }
   149 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
   150   movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   151   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
   152   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
   153   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
   154   switch (state) {
   155     case atos: movptr(rax, oop_addr);
   156                movptr(oop_addr, (int32_t)NULL_WORD);
   157                verify_oop(rax, state);              break;
   158     case ltos: movptr(rax, val_addr);                 break;
   159     case btos:                                   // fall through
   160     case ctos:                                   // fall through
   161     case stos:                                   // fall through
   162     case itos: movl(rax, val_addr);                 break;
   163     case ftos: movflt(xmm0, val_addr);              break;
   164     case dtos: movdbl(xmm0, val_addr);              break;
   165     case vtos: /* nothing to do */                  break;
   166     default  : ShouldNotReachHere();
   167   }
   168   // Clean up tos value in the thread object
   169   movl(tos_addr,  (int) ilgl);
   170   movl(val_addr,  (int32_t) NULL_WORD);
   171 }
   174 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
   175   if (JvmtiExport::can_force_early_return()) {
   176     Label L;
   177     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   178     testptr(c_rarg0, c_rarg0);
   179     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
   181     // Initiate earlyret handling only if it is not already being processed.
   182     // If the flag has the earlyret_processing bit set, it means that this code
   183     // is called *during* earlyret handling - we don't want to reenter.
   184     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
   185     cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
   186     jcc(Assembler::notEqual, L);
   188     // Call Interpreter::remove_activation_early_entry() to get the address of the
   189     // same-named entrypoint in the generated interpreter code.
   190     movptr(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
   191     movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
   192     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
   193     jmp(rax);
   194     bind(L);
   195   }
   196 }
   199 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
   200   Register reg,
   201   int bcp_offset) {
   202   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
   203   movl(reg, Address(r13, bcp_offset));
   204   bswapl(reg);
   205   shrl(reg, 16);
   206 }
   209 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
   210                                                        int bcp_offset,
   211                                                        size_t index_size) {
   212   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
   213   if (index_size == sizeof(u2)) {
   214     load_unsigned_short(index, Address(r13, bcp_offset));
   215   } else if (index_size == sizeof(u4)) {
   216     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
   217     movl(index, Address(r13, bcp_offset));
   218     // Check if the secondary index definition is still ~x, otherwise
   219     // we have to change the following assembler code to calculate the
   220     // plain index.
   221     assert(constantPoolCacheOopDesc::decode_secondary_index(~123) == 123, "else change next line");
   222     notl(index);  // convert to plain index
   223   } else if (index_size == sizeof(u1)) {
   224     assert(EnableInvokeDynamic, "tiny index used only for JSR 292");
   225     load_unsigned_byte(index, Address(r13, bcp_offset));
   226   } else {
   227     ShouldNotReachHere();
   228   }
   229 }
   232 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
   233                                                            Register index,
   234                                                            int bcp_offset,
   235                                                            size_t index_size) {
   236   assert_different_registers(cache, index);
   237   get_cache_index_at_bcp(index, bcp_offset, index_size);
   238   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   239   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   240   // convert from field index to ConstantPoolCacheEntry index
   241   shll(index, 2);
   242 }
   245 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
   246                                                                         Register index,
   247                                                                         Register bytecode,
   248                                                                         int byte_no,
   249                                                                         int bcp_offset,
   250                                                                         size_t index_size) {
   251   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
   252   // We use a 32-bit load here since the layout of 64-bit words on
   253   // little-endian machines allow us that.
   254   movl(bytecode, Address(cache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
   255   const int shift_count = (1 + byte_no) * BitsPerByte;
   256   shrl(bytecode, shift_count);
   257   andl(bytecode, 0xFF);
   258 }
   261 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
   262                                                                Register tmp,
   263                                                                int bcp_offset,
   264                                                                size_t index_size) {
   265   assert(cache != tmp, "must use different register");
   266   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
   267   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
   268   // convert from field index to ConstantPoolCacheEntry index
   269   // and from word offset to byte offset
   270   shll(tmp, 2 + LogBytesPerWord);
   271   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
   272   // skip past the header
   273   addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
   274   addptr(cache, tmp);  // construct pointer to cache entry
   275 }
   278 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
   279 // subtype of super_klass.
   280 //
   281 // Args:
   282 //      rax: superklass
   283 //      Rsub_klass: subklass
   284 //
   285 // Kills:
   286 //      rcx, rdi
   287 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
   288                                                   Label& ok_is_subtype) {
   289   assert(Rsub_klass != rax, "rax holds superklass");
   290   assert(Rsub_klass != r14, "r14 holds locals");
   291   assert(Rsub_klass != r13, "r13 holds bcp");
   292   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
   293   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
   295   // Profile the not-null value's klass.
   296   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
   298   // Do the check.
   299   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
   301   // Profile the failure of the check.
   302   profile_typecheck_failed(rcx); // blows rcx
   303 }
   307 // Java Expression Stack
   309 void InterpreterMacroAssembler::pop_ptr(Register r) {
   310   pop(r);
   311 }
   313 void InterpreterMacroAssembler::pop_i(Register r) {
   314   // XXX can't use pop currently, upper half non clean
   315   movl(r, Address(rsp, 0));
   316   addptr(rsp, wordSize);
   317 }
   319 void InterpreterMacroAssembler::pop_l(Register r) {
   320   movq(r, Address(rsp, 0));
   321   addptr(rsp, 2 * Interpreter::stackElementSize);
   322 }
   324 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
   325   movflt(r, Address(rsp, 0));
   326   addptr(rsp, wordSize);
   327 }
   329 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
   330   movdbl(r, Address(rsp, 0));
   331   addptr(rsp, 2 * Interpreter::stackElementSize);
   332 }
   334 void InterpreterMacroAssembler::push_ptr(Register r) {
   335   push(r);
   336 }
   338 void InterpreterMacroAssembler::push_i(Register r) {
   339   push(r);
   340 }
   342 void InterpreterMacroAssembler::push_l(Register r) {
   343   subptr(rsp, 2 * wordSize);
   344   movq(Address(rsp, 0), r);
   345 }
   347 void InterpreterMacroAssembler::push_f(XMMRegister r) {
   348   subptr(rsp, wordSize);
   349   movflt(Address(rsp, 0), r);
   350 }
   352 void InterpreterMacroAssembler::push_d(XMMRegister r) {
   353   subptr(rsp, 2 * wordSize);
   354   movdbl(Address(rsp, 0), r);
   355 }
   357 void InterpreterMacroAssembler::pop(TosState state) {
   358   switch (state) {
   359   case atos: pop_ptr();                 break;
   360   case btos:
   361   case ctos:
   362   case stos:
   363   case itos: pop_i();                   break;
   364   case ltos: pop_l();                   break;
   365   case ftos: pop_f();                   break;
   366   case dtos: pop_d();                   break;
   367   case vtos: /* nothing to do */        break;
   368   default:   ShouldNotReachHere();
   369   }
   370   verify_oop(rax, state);
   371 }
   373 void InterpreterMacroAssembler::push(TosState state) {
   374   verify_oop(rax, state);
   375   switch (state) {
   376   case atos: push_ptr();                break;
   377   case btos:
   378   case ctos:
   379   case stos:
   380   case itos: push_i();                  break;
   381   case ltos: push_l();                  break;
   382   case ftos: push_f();                  break;
   383   case dtos: push_d();                  break;
   384   case vtos: /* nothing to do */        break;
   385   default  : ShouldNotReachHere();
   386   }
   387 }
   390 // Helpers for swap and dup
   391 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
   392   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
   393 }
   395 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
   396   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
   397 }
   400 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
   401   // set sender sp
   402   lea(r13, Address(rsp, wordSize));
   403   // record last_sp
   404   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
   405 }
   408 // Jump to from_interpreted entry of a call unless single stepping is possible
   409 // in this thread in which case we must call the i2i entry
   410 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
   411   prepare_to_jump_from_interpreted();
   413   if (JvmtiExport::can_post_interpreter_events()) {
   414     Label run_compiled_code;
   415     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
   416     // compiled code in threads for which the event is enabled.  Check here for
   417     // interp_only_mode if these events CAN be enabled.
   418     // interp_only is an int, on little endian it is sufficient to test the byte only
   419     // Is a cmpl faster?
   420     cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
   421     jccb(Assembler::zero, run_compiled_code);
   422     jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
   423     bind(run_compiled_code);
   424   }
   426   jmp(Address(method, methodOopDesc::from_interpreted_offset()));
   428 }
   431 // The following two routines provide a hook so that an implementation
   432 // can schedule the dispatch in two parts.  amd64 does not do this.
   433 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
   434   // Nothing amd64 specific to be done here
   435 }
   437 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
   438   dispatch_next(state, step);
   439 }
   441 void InterpreterMacroAssembler::dispatch_base(TosState state,
   442                                               address* table,
   443                                               bool verifyoop) {
   444   verify_FPU(1, state);
   445   if (VerifyActivationFrameSize) {
   446     Label L;
   447     mov(rcx, rbp);
   448     subptr(rcx, rsp);
   449     int32_t min_frame_size =
   450       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
   451       wordSize;
   452     cmpptr(rcx, (int32_t)min_frame_size);
   453     jcc(Assembler::greaterEqual, L);
   454     stop("broken stack frame");
   455     bind(L);
   456   }
   457   if (verifyoop) {
   458     verify_oop(rax, state);
   459   }
   460   lea(rscratch1, ExternalAddress((address)table));
   461   jmp(Address(rscratch1, rbx, Address::times_8));
   462 }
   464 void InterpreterMacroAssembler::dispatch_only(TosState state) {
   465   dispatch_base(state, Interpreter::dispatch_table(state));
   466 }
   468 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
   469   dispatch_base(state, Interpreter::normal_table(state));
   470 }
   472 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
   473   dispatch_base(state, Interpreter::normal_table(state), false);
   474 }
   477 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
   478   // load next bytecode (load before advancing r13 to prevent AGI)
   479   load_unsigned_byte(rbx, Address(r13, step));
   480   // advance r13
   481   increment(r13, step);
   482   dispatch_base(state, Interpreter::dispatch_table(state));
   483 }
   485 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
   486   // load current bytecode
   487   load_unsigned_byte(rbx, Address(r13, 0));
   488   dispatch_base(state, table);
   489 }
   491 // remove activation
   492 //
   493 // Unlock the receiver if this is a synchronized method.
   494 // Unlock any Java monitors from syncronized blocks.
   495 // Remove the activation from the stack.
   496 //
   497 // If there are locked Java monitors
   498 //    If throw_monitor_exception
   499 //       throws IllegalMonitorStateException
   500 //    Else if install_monitor_exception
   501 //       installs IllegalMonitorStateException
   502 //    Else
   503 //       no error processing
   504 void InterpreterMacroAssembler::remove_activation(
   505         TosState state,
   506         Register ret_addr,
   507         bool throw_monitor_exception,
   508         bool install_monitor_exception,
   509         bool notify_jvmdi) {
   510   // Note: Registers rdx xmm0 may be in use for the
   511   // result check if synchronized method
   512   Label unlocked, unlock, no_unlock;
   514   // get the value of _do_not_unlock_if_synchronized into rdx
   515   const Address do_not_unlock_if_synchronized(r15_thread,
   516     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
   517   movbool(rdx, do_not_unlock_if_synchronized);
   518   movbool(do_not_unlock_if_synchronized, false); // reset the flag
   520  // get method access flags
   521   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
   522   movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
   523   testl(rcx, JVM_ACC_SYNCHRONIZED);
   524   jcc(Assembler::zero, unlocked);
   526   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
   527   // is set.
   528   testbool(rdx);
   529   jcc(Assembler::notZero, no_unlock);
   531   // unlock monitor
   532   push(state); // save result
   534   // BasicObjectLock will be first in list, since this is a
   535   // synchronized method. However, need to check that the object has
   536   // not been unlocked by an explicit monitorexit bytecode.
   537   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
   538                         wordSize - (int) sizeof(BasicObjectLock));
   539   // We use c_rarg1 so that if we go slow path it will be the correct
   540   // register for unlock_object to pass to VM directly
   541   lea(c_rarg1, monitor); // address of first monitor
   543   movptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
   544   testptr(rax, rax);
   545   jcc(Assembler::notZero, unlock);
   547   pop(state);
   548   if (throw_monitor_exception) {
   549     // Entry already unlocked, need to throw exception
   550     call_VM(noreg, CAST_FROM_FN_PTR(address,
   551                    InterpreterRuntime::throw_illegal_monitor_state_exception));
   552     should_not_reach_here();
   553   } else {
   554     // Monitor already unlocked during a stack unroll. If requested,
   555     // install an illegal_monitor_state_exception.  Continue with
   556     // stack unrolling.
   557     if (install_monitor_exception) {
   558       call_VM(noreg, CAST_FROM_FN_PTR(address,
   559                      InterpreterRuntime::new_illegal_monitor_state_exception));
   560     }
   561     jmp(unlocked);
   562   }
   564   bind(unlock);
   565   unlock_object(c_rarg1);
   566   pop(state);
   568   // Check that for block-structured locking (i.e., that all locked
   569   // objects has been unlocked)
   570   bind(unlocked);
   572   // rax: Might contain return value
   574   // Check that all monitors are unlocked
   575   {
   576     Label loop, exception, entry, restart;
   577     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
   578     const Address monitor_block_top(
   579         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
   580     const Address monitor_block_bot(
   581         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
   583     bind(restart);
   584     // We use c_rarg1 so that if we go slow path it will be the correct
   585     // register for unlock_object to pass to VM directly
   586     movptr(c_rarg1, monitor_block_top); // points to current entry, starting
   587                                   // with top-most entry
   588     lea(rbx, monitor_block_bot);  // points to word before bottom of
   589                                   // monitor block
   590     jmp(entry);
   592     // Entry already locked, need to throw exception
   593     bind(exception);
   595     if (throw_monitor_exception) {
   596       // Throw exception
   597       MacroAssembler::call_VM(noreg,
   598                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
   599                                    throw_illegal_monitor_state_exception));
   600       should_not_reach_here();
   601     } else {
   602       // Stack unrolling. Unlock object and install illegal_monitor_exception.
   603       // Unlock does not block, so don't have to worry about the frame.
   604       // We don't have to preserve c_rarg1 since we are going to throw an exception.
   606       push(state);
   607       unlock_object(c_rarg1);
   608       pop(state);
   610       if (install_monitor_exception) {
   611         call_VM(noreg, CAST_FROM_FN_PTR(address,
   612                                         InterpreterRuntime::
   613                                         new_illegal_monitor_state_exception));
   614       }
   616       jmp(restart);
   617     }
   619     bind(loop);
   620     // check if current entry is used
   621     cmpptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
   622     jcc(Assembler::notEqual, exception);
   624     addptr(c_rarg1, entry_size); // otherwise advance to next entry
   625     bind(entry);
   626     cmpptr(c_rarg1, rbx); // check if bottom reached
   627     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
   628   }
   630   bind(no_unlock);
   632   // jvmti support
   633   if (notify_jvmdi) {
   634     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
   635   } else {
   636     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
   637   }
   639   // remove activation
   640   // get sender sp
   641   movptr(rbx,
   642          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
   643   leave();                           // remove frame anchor
   644   pop(ret_addr);                     // get return address
   645   mov(rsp, rbx);                     // set sp to sender sp
   646 }
   648 #endif // C_INTERP
   650 // Lock object
   651 //
   652 // Args:
   653 //      c_rarg1: BasicObjectLock to be used for locking
   654 //
   655 // Kills:
   656 //      rax
   657 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
   658 //      rscratch1, rscratch2 (scratch regs)
   659 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
   660   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
   662   if (UseHeavyMonitors) {
   663     call_VM(noreg,
   664             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   665             lock_reg);
   666   } else {
   667     Label done;
   669     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
   670     const Register obj_reg = c_rarg3; // Will contain the oop
   672     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
   673     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
   674     const int mark_offset = lock_offset +
   675                             BasicLock::displaced_header_offset_in_bytes();
   677     Label slow_case;
   679     // Load object pointer into obj_reg %c_rarg3
   680     movptr(obj_reg, Address(lock_reg, obj_offset));
   682     if (UseBiasedLocking) {
   683       biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
   684     }
   686     // Load immediate 1 into swap_reg %rax
   687     movl(swap_reg, 1);
   689     // Load (object->mark() | 1) into swap_reg %rax
   690     orptr(swap_reg, Address(obj_reg, 0));
   692     // Save (object->mark() | 1) into BasicLock's displaced header
   693     movptr(Address(lock_reg, mark_offset), swap_reg);
   695     assert(lock_offset == 0,
   696            "displached header must be first word in BasicObjectLock");
   698     if (os::is_MP()) lock();
   699     cmpxchgptr(lock_reg, Address(obj_reg, 0));
   700     if (PrintBiasedLockingStatistics) {
   701       cond_inc32(Assembler::zero,
   702                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   703     }
   704     jcc(Assembler::zero, done);
   706     // Test if the oopMark is an obvious stack pointer, i.e.,
   707     //  1) (mark & 7) == 0, and
   708     //  2) rsp <= mark < mark + os::pagesize()
   709     //
   710     // These 3 tests can be done by evaluating the following
   711     // expression: ((mark - rsp) & (7 - os::vm_page_size())),
   712     // assuming both stack pointer and pagesize have their
   713     // least significant 3 bits clear.
   714     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
   715     subptr(swap_reg, rsp);
   716     andptr(swap_reg, 7 - os::vm_page_size());
   718     // Save the test result, for recursive case, the result is zero
   719     movptr(Address(lock_reg, mark_offset), swap_reg);
   721     if (PrintBiasedLockingStatistics) {
   722       cond_inc32(Assembler::zero,
   723                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
   724     }
   725     jcc(Assembler::zero, done);
   727     bind(slow_case);
   729     // Call the runtime routine for slow case
   730     call_VM(noreg,
   731             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
   732             lock_reg);
   734     bind(done);
   735   }
   736 }
   739 // Unlocks an object. Used in monitorexit bytecode and
   740 // remove_activation.  Throws an IllegalMonitorException if object is
   741 // not locked by current thread.
   742 //
   743 // Args:
   744 //      c_rarg1: BasicObjectLock for lock
   745 //
   746 // Kills:
   747 //      rax
   748 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
   749 //      rscratch1, rscratch2 (scratch regs)
   750 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
   751   assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
   753   if (UseHeavyMonitors) {
   754     call_VM(noreg,
   755             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   756             lock_reg);
   757   } else {
   758     Label done;
   760     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
   761     const Register header_reg = c_rarg2;  // Will contain the old oopMark
   762     const Register obj_reg    = c_rarg3;  // Will contain the oop
   764     save_bcp(); // Save in case of exception
   766     // Convert from BasicObjectLock structure to object and BasicLock
   767     // structure Store the BasicLock address into %rax
   768     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
   770     // Load oop into obj_reg(%c_rarg3)
   771     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
   773     // Free entry
   774     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
   776     if (UseBiasedLocking) {
   777       biased_locking_exit(obj_reg, header_reg, done);
   778     }
   780     // Load the old header from BasicLock structure
   781     movptr(header_reg, Address(swap_reg,
   782                                BasicLock::displaced_header_offset_in_bytes()));
   784     // Test for recursion
   785     testptr(header_reg, header_reg);
   787     // zero for recursive case
   788     jcc(Assembler::zero, done);
   790     // Atomic swap back the old header
   791     if (os::is_MP()) lock();
   792     cmpxchgptr(header_reg, Address(obj_reg, 0));
   794     // zero for recursive case
   795     jcc(Assembler::zero, done);
   797     // Call the runtime routine for slow case.
   798     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
   799          obj_reg); // restore obj
   800     call_VM(noreg,
   801             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
   802             lock_reg);
   804     bind(done);
   806     restore_bcp();
   807   }
   808 }
   810 #ifndef CC_INTERP
   812 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
   813                                                          Label& zero_continue) {
   814   assert(ProfileInterpreter, "must be profiling interpreter");
   815   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
   816   testptr(mdp, mdp);
   817   jcc(Assembler::zero, zero_continue);
   818 }
   821 // Set the method data pointer for the current bcp.
   822 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
   823   assert(ProfileInterpreter, "must be profiling interpreter");
   824   Label set_mdp;
   825   push(rax);
   826   push(rbx);
   828   get_method(rbx);
   829   // Test MDO to avoid the call if it is NULL.
   830   movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
   831   testptr(rax, rax);
   832   jcc(Assembler::zero, set_mdp);
   833   // rbx: method
   834   // r13: bcp
   835   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
   836   // rax: mdi
   837   // mdo is guaranteed to be non-zero here, we checked for it before the call.
   838   movptr(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
   839   addptr(rbx, in_bytes(methodDataOopDesc::data_offset()));
   840   addptr(rax, rbx);
   841   bind(set_mdp);
   842   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
   843   pop(rbx);
   844   pop(rax);
   845 }
   847 void InterpreterMacroAssembler::verify_method_data_pointer() {
   848   assert(ProfileInterpreter, "must be profiling interpreter");
   849 #ifdef ASSERT
   850   Label verify_continue;
   851   push(rax);
   852   push(rbx);
   853   push(c_rarg3);
   854   push(c_rarg2);
   855   test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
   856   get_method(rbx);
   858   // If the mdp is valid, it will point to a DataLayout header which is
   859   // consistent with the bcp.  The converse is highly probable also.
   860   load_unsigned_short(c_rarg2,
   861                       Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
   862   addptr(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
   863   lea(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
   864   cmpptr(c_rarg2, r13);
   865   jcc(Assembler::equal, verify_continue);
   866   // rbx: method
   867   // r13: bcp
   868   // c_rarg3: mdp
   869   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
   870                rbx, r13, c_rarg3);
   871   bind(verify_continue);
   872   pop(c_rarg2);
   873   pop(c_rarg3);
   874   pop(rbx);
   875   pop(rax);
   876 #endif // ASSERT
   877 }
   880 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
   881                                                 int constant,
   882                                                 Register value) {
   883   assert(ProfileInterpreter, "must be profiling interpreter");
   884   Address data(mdp_in, constant);
   885   movptr(data, value);
   886 }
   889 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   890                                                       int constant,
   891                                                       bool decrement) {
   892   // Counter address
   893   Address data(mdp_in, constant);
   895   increment_mdp_data_at(data, decrement);
   896 }
   898 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
   899                                                       bool decrement) {
   900   assert(ProfileInterpreter, "must be profiling interpreter");
   901   // %%% this does 64bit counters at best it is wasting space
   902   // at worst it is a rare bug when counters overflow
   904   if (decrement) {
   905     // Decrement the register.  Set condition codes.
   906     addptr(data, (int32_t) -DataLayout::counter_increment);
   907     // If the decrement causes the counter to overflow, stay negative
   908     Label L;
   909     jcc(Assembler::negative, L);
   910     addptr(data, (int32_t) DataLayout::counter_increment);
   911     bind(L);
   912   } else {
   913     assert(DataLayout::counter_increment == 1,
   914            "flow-free idiom only works with 1");
   915     // Increment the register.  Set carry flag.
   916     addptr(data, DataLayout::counter_increment);
   917     // If the increment causes the counter to overflow, pull back by 1.
   918     sbbptr(data, (int32_t)0);
   919   }
   920 }
   923 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
   924                                                       Register reg,
   925                                                       int constant,
   926                                                       bool decrement) {
   927   Address data(mdp_in, reg, Address::times_1, constant);
   929   increment_mdp_data_at(data, decrement);
   930 }
   932 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
   933                                                 int flag_byte_constant) {
   934   assert(ProfileInterpreter, "must be profiling interpreter");
   935   int header_offset = in_bytes(DataLayout::header_offset());
   936   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
   937   // Set the flag
   938   orl(Address(mdp_in, header_offset), header_bits);
   939 }
   943 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
   944                                                  int offset,
   945                                                  Register value,
   946                                                  Register test_value_out,
   947                                                  Label& not_equal_continue) {
   948   assert(ProfileInterpreter, "must be profiling interpreter");
   949   if (test_value_out == noreg) {
   950     cmpptr(value, Address(mdp_in, offset));
   951   } else {
   952     // Put the test value into a register, so caller can use it:
   953     movptr(test_value_out, Address(mdp_in, offset));
   954     cmpptr(test_value_out, value);
   955   }
   956   jcc(Assembler::notEqual, not_equal_continue);
   957 }
   960 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
   961                                                      int offset_of_disp) {
   962   assert(ProfileInterpreter, "must be profiling interpreter");
   963   Address disp_address(mdp_in, offset_of_disp);
   964   addptr(mdp_in, disp_address);
   965   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
   966 }
   969 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
   970                                                      Register reg,
   971                                                      int offset_of_disp) {
   972   assert(ProfileInterpreter, "must be profiling interpreter");
   973   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
   974   addptr(mdp_in, disp_address);
   975   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
   976 }
   979 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
   980                                                        int constant) {
   981   assert(ProfileInterpreter, "must be profiling interpreter");
   982   addptr(mdp_in, constant);
   983   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
   984 }
   987 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
   988   assert(ProfileInterpreter, "must be profiling interpreter");
   989   push(return_bci); // save/restore across call_VM
   990   call_VM(noreg,
   991           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
   992           return_bci);
   993   pop(return_bci);
   994 }
   997 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
   998                                                      Register bumped_count) {
   999   if (ProfileInterpreter) {
  1000     Label profile_continue;
  1002     // If no method data exists, go to profile_continue.
  1003     // Otherwise, assign to mdp
  1004     test_method_data_pointer(mdp, profile_continue);
  1006     // We are taking a branch.  Increment the taken count.
  1007     // We inline increment_mdp_data_at to return bumped_count in a register
  1008     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
  1009     Address data(mdp, in_bytes(JumpData::taken_offset()));
  1010     movptr(bumped_count, data);
  1011     assert(DataLayout::counter_increment == 1,
  1012             "flow-free idiom only works with 1");
  1013     addptr(bumped_count, DataLayout::counter_increment);
  1014     sbbptr(bumped_count, 0);
  1015     movptr(data, bumped_count); // Store back out
  1017     // The method data pointer needs to be updated to reflect the new target.
  1018     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
  1019     bind(profile_continue);
  1024 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
  1025   if (ProfileInterpreter) {
  1026     Label profile_continue;
  1028     // If no method data exists, go to profile_continue.
  1029     test_method_data_pointer(mdp, profile_continue);
  1031     // We are taking a branch.  Increment the not taken count.
  1032     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
  1034     // The method data pointer needs to be updated to correspond to
  1035     // the next bytecode
  1036     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
  1037     bind(profile_continue);
  1042 void InterpreterMacroAssembler::profile_call(Register mdp) {
  1043   if (ProfileInterpreter) {
  1044     Label profile_continue;
  1046     // If no method data exists, go to profile_continue.
  1047     test_method_data_pointer(mdp, profile_continue);
  1049     // We are making a call.  Increment the count.
  1050     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1052     // The method data pointer needs to be updated to reflect the new target.
  1053     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
  1054     bind(profile_continue);
  1059 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
  1060   if (ProfileInterpreter) {
  1061     Label profile_continue;
  1063     // If no method data exists, go to profile_continue.
  1064     test_method_data_pointer(mdp, profile_continue);
  1066     // We are making a call.  Increment the count.
  1067     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1069     // The method data pointer needs to be updated to reflect the new target.
  1070     update_mdp_by_constant(mdp,
  1071                            in_bytes(VirtualCallData::
  1072                                     virtual_call_data_size()));
  1073     bind(profile_continue);
  1078 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
  1079                                                      Register mdp,
  1080                                                      Register reg2,
  1081                                                      bool receiver_can_be_null) {
  1082   if (ProfileInterpreter) {
  1083     Label profile_continue;
  1085     // If no method data exists, go to profile_continue.
  1086     test_method_data_pointer(mdp, profile_continue);
  1088     Label skip_receiver_profile;
  1089     if (receiver_can_be_null) {
  1090       Label not_null;
  1091       testptr(receiver, receiver);
  1092       jccb(Assembler::notZero, not_null);
  1093       // We are making a call.  Increment the count for null receiver.
  1094       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1095       jmp(skip_receiver_profile);
  1096       bind(not_null);
  1099     // Record the receiver type.
  1100     record_klass_in_profile(receiver, mdp, reg2, true);
  1101     bind(skip_receiver_profile);
  1103     // The method data pointer needs to be updated to reflect the new target.
  1104     update_mdp_by_constant(mdp,
  1105                            in_bytes(VirtualCallData::
  1106                                     virtual_call_data_size()));
  1107     bind(profile_continue);
  1111 // This routine creates a state machine for updating the multi-row
  1112 // type profile at a virtual call site (or other type-sensitive bytecode).
  1113 // The machine visits each row (of receiver/count) until the receiver type
  1114 // is found, or until it runs out of rows.  At the same time, it remembers
  1115 // the location of the first empty row.  (An empty row records null for its
  1116 // receiver, and can be allocated for a newly-observed receiver type.)
  1117 // Because there are two degrees of freedom in the state, a simple linear
  1118 // search will not work; it must be a decision tree.  Hence this helper
  1119 // function is recursive, to generate the required tree structured code.
  1120 // It's the interpreter, so we are trading off code space for speed.
  1121 // See below for example code.
  1122 void InterpreterMacroAssembler::record_klass_in_profile_helper(
  1123                                         Register receiver, Register mdp,
  1124                                         Register reg2, int start_row,
  1125                                         Label& done, bool is_virtual_call) {
  1126   if (TypeProfileWidth == 0) {
  1127     if (is_virtual_call) {
  1128       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1130     return;
  1133   int last_row = VirtualCallData::row_limit() - 1;
  1134   assert(start_row <= last_row, "must be work left to do");
  1135   // Test this row for both the receiver and for null.
  1136   // Take any of three different outcomes:
  1137   //   1. found receiver => increment count and goto done
  1138   //   2. found null => keep looking for case 1, maybe allocate this cell
  1139   //   3. found something else => keep looking for cases 1 and 2
  1140   // Case 3 is handled by a recursive call.
  1141   for (int row = start_row; row <= last_row; row++) {
  1142     Label next_test;
  1143     bool test_for_null_also = (row == start_row);
  1145     // See if the receiver is receiver[n].
  1146     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
  1147     test_mdp_data_at(mdp, recvr_offset, receiver,
  1148                      (test_for_null_also ? reg2 : noreg),
  1149                      next_test);
  1150     // (Reg2 now contains the receiver from the CallData.)
  1152     // The receiver is receiver[n].  Increment count[n].
  1153     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
  1154     increment_mdp_data_at(mdp, count_offset);
  1155     jmp(done);
  1156     bind(next_test);
  1158     if (test_for_null_also) {
  1159       Label found_null;
  1160       // Failed the equality check on receiver[n]...  Test for null.
  1161       testptr(reg2, reg2);
  1162       if (start_row == last_row) {
  1163         // The only thing left to do is handle the null case.
  1164         if (is_virtual_call) {
  1165           jccb(Assembler::zero, found_null);
  1166           // Receiver did not match any saved receiver and there is no empty row for it.
  1167           // Increment total counter to indicate polymorphic case.
  1168           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1169           jmp(done);
  1170           bind(found_null);
  1171         } else {
  1172           jcc(Assembler::notZero, done);
  1174         break;
  1176       // Since null is rare, make it be the branch-taken case.
  1177       jcc(Assembler::zero, found_null);
  1179       // Put all the "Case 3" tests here.
  1180       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
  1182       // Found a null.  Keep searching for a matching receiver,
  1183       // but remember that this is an empty (unused) slot.
  1184       bind(found_null);
  1188   // In the fall-through case, we found no matching receiver, but we
  1189   // observed the receiver[start_row] is NULL.
  1191   // Fill in the receiver field and increment the count.
  1192   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
  1193   set_mdp_data_at(mdp, recvr_offset, receiver);
  1194   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
  1195   movl(reg2, DataLayout::counter_increment);
  1196   set_mdp_data_at(mdp, count_offset, reg2);
  1197   if (start_row > 0) {
  1198     jmp(done);
  1202 // Example state machine code for three profile rows:
  1203 //   // main copy of decision tree, rooted at row[1]
  1204 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
  1205 //   if (row[0].rec != NULL) {
  1206 //     // inner copy of decision tree, rooted at row[1]
  1207 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1208 //     if (row[1].rec != NULL) {
  1209 //       // degenerate decision tree, rooted at row[2]
  1210 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1211 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
  1212 //       row[2].init(rec); goto done;
  1213 //     } else {
  1214 //       // remember row[1] is empty
  1215 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
  1216 //       row[1].init(rec); goto done;
  1217 //     }
  1218 //   } else {
  1219 //     // remember row[0] is empty
  1220 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
  1221 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
  1222 //     row[0].init(rec); goto done;
  1223 //   }
  1224 //   done:
  1226 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
  1227                                                         Register mdp, Register reg2,
  1228                                                         bool is_virtual_call) {
  1229   assert(ProfileInterpreter, "must be profiling");
  1230   Label done;
  1232   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
  1234   bind (done);
  1237 void InterpreterMacroAssembler::profile_ret(Register return_bci,
  1238                                             Register mdp) {
  1239   if (ProfileInterpreter) {
  1240     Label profile_continue;
  1241     uint row;
  1243     // If no method data exists, go to profile_continue.
  1244     test_method_data_pointer(mdp, profile_continue);
  1246     // Update the total ret count.
  1247     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
  1249     for (row = 0; row < RetData::row_limit(); row++) {
  1250       Label next_test;
  1252       // See if return_bci is equal to bci[n]:
  1253       test_mdp_data_at(mdp,
  1254                        in_bytes(RetData::bci_offset(row)),
  1255                        return_bci, noreg,
  1256                        next_test);
  1258       // return_bci is equal to bci[n].  Increment the count.
  1259       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
  1261       // The method data pointer needs to be updated to reflect the new target.
  1262       update_mdp_by_offset(mdp,
  1263                            in_bytes(RetData::bci_displacement_offset(row)));
  1264       jmp(profile_continue);
  1265       bind(next_test);
  1268     update_mdp_for_ret(return_bci);
  1270     bind(profile_continue);
  1275 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
  1276   if (ProfileInterpreter) {
  1277     Label profile_continue;
  1279     // If no method data exists, go to profile_continue.
  1280     test_method_data_pointer(mdp, profile_continue);
  1282     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
  1284     // The method data pointer needs to be updated.
  1285     int mdp_delta = in_bytes(BitData::bit_data_size());
  1286     if (TypeProfileCasts) {
  1287       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1289     update_mdp_by_constant(mdp, mdp_delta);
  1291     bind(profile_continue);
  1296 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
  1297   if (ProfileInterpreter && TypeProfileCasts) {
  1298     Label profile_continue;
  1300     // If no method data exists, go to profile_continue.
  1301     test_method_data_pointer(mdp, profile_continue);
  1303     int count_offset = in_bytes(CounterData::count_offset());
  1304     // Back up the address, since we have already bumped the mdp.
  1305     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
  1307     // *Decrement* the counter.  We expect to see zero or small negatives.
  1308     increment_mdp_data_at(mdp, count_offset, true);
  1310     bind (profile_continue);
  1315 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
  1316   if (ProfileInterpreter) {
  1317     Label profile_continue;
  1319     // If no method data exists, go to profile_continue.
  1320     test_method_data_pointer(mdp, profile_continue);
  1322     // The method data pointer needs to be updated.
  1323     int mdp_delta = in_bytes(BitData::bit_data_size());
  1324     if (TypeProfileCasts) {
  1325       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
  1327       // Record the object type.
  1328       record_klass_in_profile(klass, mdp, reg2, false);
  1330     update_mdp_by_constant(mdp, mdp_delta);
  1332     bind(profile_continue);
  1337 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
  1338   if (ProfileInterpreter) {
  1339     Label profile_continue;
  1341     // If no method data exists, go to profile_continue.
  1342     test_method_data_pointer(mdp, profile_continue);
  1344     // Update the default case count
  1345     increment_mdp_data_at(mdp,
  1346                           in_bytes(MultiBranchData::default_count_offset()));
  1348     // The method data pointer needs to be updated.
  1349     update_mdp_by_offset(mdp,
  1350                          in_bytes(MultiBranchData::
  1351                                   default_displacement_offset()));
  1353     bind(profile_continue);
  1358 void InterpreterMacroAssembler::profile_switch_case(Register index,
  1359                                                     Register mdp,
  1360                                                     Register reg2) {
  1361   if (ProfileInterpreter) {
  1362     Label profile_continue;
  1364     // If no method data exists, go to profile_continue.
  1365     test_method_data_pointer(mdp, profile_continue);
  1367     // Build the base (index * per_case_size_in_bytes()) +
  1368     // case_array_offset_in_bytes()
  1369     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
  1370     imulptr(index, reg2); // XXX l ?
  1371     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
  1373     // Update the case count
  1374     increment_mdp_data_at(mdp,
  1375                           index,
  1376                           in_bytes(MultiBranchData::relative_count_offset()));
  1378     // The method data pointer needs to be updated.
  1379     update_mdp_by_offset(mdp,
  1380                          index,
  1381                          in_bytes(MultiBranchData::
  1382                                   relative_displacement_offset()));
  1384     bind(profile_continue);
  1390 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
  1391   if (state == atos) {
  1392     MacroAssembler::verify_oop(reg);
  1396 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
  1398 #endif // !CC_INTERP
  1401 void InterpreterMacroAssembler::notify_method_entry() {
  1402   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1403   // track stack depth.  If it is possible to enter interp_only_mode we add
  1404   // the code to check if the event should be sent.
  1405   if (JvmtiExport::can_post_interpreter_events()) {
  1406     Label L;
  1407     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1408     testl(rdx, rdx);
  1409     jcc(Assembler::zero, L);
  1410     call_VM(noreg, CAST_FROM_FN_PTR(address,
  1411                                     InterpreterRuntime::post_method_entry));
  1412     bind(L);
  1416     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1417     get_method(c_rarg1);
  1418     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
  1419                  r15_thread, c_rarg1);
  1422   // RedefineClasses() tracing support for obsolete method entry
  1423   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
  1424     get_method(c_rarg1);
  1425     call_VM_leaf(
  1426       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
  1427       r15_thread, c_rarg1);
  1432 void InterpreterMacroAssembler::notify_method_exit(
  1433     TosState state, NotifyMethodExitMode mode) {
  1434   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
  1435   // track stack depth.  If it is possible to enter interp_only_mode we add
  1436   // the code to check if the event should be sent.
  1437   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
  1438     Label L;
  1439     // Note: frame::interpreter_frame_result has a dependency on how the
  1440     // method result is saved across the call to post_method_exit. If this
  1441     // is changed then the interpreter_frame_result implementation will
  1442     // need to be updated too.
  1444     // For c++ interpreter the result is always stored at a known location in the frame
  1445     // template interpreter will leave it on the top of the stack.
  1446     NOT_CC_INTERP(push(state);)
  1447     movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
  1448     testl(rdx, rdx);
  1449     jcc(Assembler::zero, L);
  1450     call_VM(noreg,
  1451             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
  1452     bind(L);
  1453     NOT_CC_INTERP(pop(state));
  1457     SkipIfEqual skip(this, &DTraceMethodProbes, false);
  1458     NOT_CC_INTERP(push(state));
  1459     get_method(c_rarg1);
  1460     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
  1461                  r15_thread, c_rarg1);
  1462     NOT_CC_INTERP(pop(state));
  1466 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
  1467 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
  1468                                                         int increment, int mask,
  1469                                                         Register scratch, bool preloaded,
  1470                                                         Condition cond, Label* where) {
  1471   if (!preloaded) {
  1472     movl(scratch, counter_addr);
  1474   incrementl(scratch, increment);
  1475   movl(counter_addr, scratch);
  1476   andl(scratch, mask);
  1477   jcc(cond, *where);

mercurial