src/cpu/x86/vm/templateInterpreter_x86_32.cpp

Fri, 30 Nov 2012 15:23:16 -0800

author
twisti
date
Fri, 30 Nov 2012 15:23:16 -0800
changeset 4318
cd3d6a6b95d9
parent 4037
da91efe96a93
child 4338
fd74228fd5ca
permissions
-rw-r--r--

8003240: x86: move MacroAssembler into separate file
Reviewed-by: kvn

     1 /*
     2  * Copyright (c) 1997, 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 "asm/macroAssembler.hpp"
    27 #include "interpreter/bytecodeHistogram.hpp"
    28 #include "interpreter/interpreter.hpp"
    29 #include "interpreter/interpreterGenerator.hpp"
    30 #include "interpreter/interpreterRuntime.hpp"
    31 #include "interpreter/templateTable.hpp"
    32 #include "oops/arrayOop.hpp"
    33 #include "oops/methodData.hpp"
    34 #include "oops/method.hpp"
    35 #include "oops/oop.inline.hpp"
    36 #include "prims/jvmtiExport.hpp"
    37 #include "prims/jvmtiThreadState.hpp"
    38 #include "runtime/arguments.hpp"
    39 #include "runtime/deoptimization.hpp"
    40 #include "runtime/frame.inline.hpp"
    41 #include "runtime/sharedRuntime.hpp"
    42 #include "runtime/stubRoutines.hpp"
    43 #include "runtime/synchronizer.hpp"
    44 #include "runtime/timer.hpp"
    45 #include "runtime/vframeArray.hpp"
    46 #include "utilities/debug.hpp"
    48 #define __ _masm->
    51 #ifndef CC_INTERP
    52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
    53 const int bci_offset    = frame::interpreter_frame_bcx_offset    * wordSize;
    54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
    56 //------------------------------------------------------------------------------------------------------------------------
    58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
    59   address entry = __ pc();
    61   // Note: There should be a minimal interpreter frame set up when stack
    62   // overflow occurs since we check explicitly for it now.
    63   //
    64 #ifdef ASSERT
    65   { Label L;
    66     __ lea(rax, Address(rbp,
    67                 frame::interpreter_frame_monitor_block_top_offset * wordSize));
    68     __ cmpptr(rax, rsp);  // rax, = maximal rsp for current rbp,
    69                         //  (stack grows negative)
    70     __ jcc(Assembler::aboveEqual, L); // check if frame is complete
    71     __ stop ("interpreter frame not set up");
    72     __ bind(L);
    73   }
    74 #endif // ASSERT
    75   // Restore bcp under the assumption that the current frame is still
    76   // interpreted
    77   __ restore_bcp();
    79   // expression stack must be empty before entering the VM if an exception
    80   // happened
    81   __ empty_expression_stack();
    82   __ empty_FPU_stack();
    83   // throw exception
    84   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
    85   return entry;
    86 }
    88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
    89   address entry = __ pc();
    90   // expression stack must be empty before entering the VM if an exception happened
    91   __ empty_expression_stack();
    92   __ empty_FPU_stack();
    93   // setup parameters
    94   // ??? convention: expect aberrant index in register rbx,
    95   __ lea(rax, ExternalAddress((address)name));
    96   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
    97   return entry;
    98 }
   100 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
   101   address entry = __ pc();
   102   // object is at TOS
   103   __ pop(rax);
   104   // expression stack must be empty before entering the VM if an exception
   105   // happened
   106   __ empty_expression_stack();
   107   __ empty_FPU_stack();
   108   __ call_VM(noreg,
   109              CAST_FROM_FN_PTR(address,
   110                               InterpreterRuntime::throw_ClassCastException),
   111              rax);
   112   return entry;
   113 }
   115 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
   116   assert(!pass_oop || message == NULL, "either oop or message but not both");
   117   address entry = __ pc();
   118   if (pass_oop) {
   119     // object is at TOS
   120     __ pop(rbx);
   121   }
   122   // expression stack must be empty before entering the VM if an exception happened
   123   __ empty_expression_stack();
   124   __ empty_FPU_stack();
   125   // setup parameters
   126   __ lea(rax, ExternalAddress((address)name));
   127   if (pass_oop) {
   128     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
   129   } else {
   130     if (message != NULL) {
   131       __ lea(rbx, ExternalAddress((address)message));
   132     } else {
   133       __ movptr(rbx, NULL_WORD);
   134     }
   135     __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
   136   }
   137   // throw exception
   138   __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
   139   return entry;
   140 }
   143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
   144   address entry = __ pc();
   145   // NULL last_sp until next java call
   146   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
   147   __ dispatch_next(state);
   148   return entry;
   149 }
   152 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
   153   TosState incoming_state = state;
   154   address entry = __ pc();
   156 #ifdef COMPILER2
   157   // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
   158   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
   159     for (int i = 1; i < 8; i++) {
   160         __ ffree(i);
   161     }
   162   } else if (UseSSE < 2) {
   163     __ empty_FPU_stack();
   164   }
   165 #endif
   166   if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
   167     __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
   168   } else {
   169     __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
   170   }
   172   // In SSE mode, interpreter returns FP results in xmm0 but they need
   173   // to end up back on the FPU so it can operate on them.
   174   if (incoming_state == ftos && UseSSE >= 1) {
   175     __ subptr(rsp, wordSize);
   176     __ movflt(Address(rsp, 0), xmm0);
   177     __ fld_s(Address(rsp, 0));
   178     __ addptr(rsp, wordSize);
   179   } else if (incoming_state == dtos && UseSSE >= 2) {
   180     __ subptr(rsp, 2*wordSize);
   181     __ movdbl(Address(rsp, 0), xmm0);
   182     __ fld_d(Address(rsp, 0));
   183     __ addptr(rsp, 2*wordSize);
   184   }
   186   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
   188   // Restore stack bottom in case i2c adjusted stack
   189   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
   190   // and NULL it as marker that rsp is now tos until next java call
   191   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
   193   __ restore_bcp();
   194   __ restore_locals();
   196   Label L_got_cache, L_giant_index;
   197   if (EnableInvokeDynamic) {
   198     __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
   199     __ jcc(Assembler::equal, L_giant_index);
   200   }
   201   __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
   202   __ bind(L_got_cache);
   203   __ movl(rbx, Address(rbx, rcx,
   204                     Address::times_ptr, ConstantPoolCache::base_offset() +
   205                     ConstantPoolCacheEntry::flags_offset()));
   206   __ andptr(rbx, 0xFF);
   207   __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
   208   __ dispatch_next(state, step);
   210   // out of the main line of code...
   211   if (EnableInvokeDynamic) {
   212     __ bind(L_giant_index);
   213     __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
   214     __ jmp(L_got_cache);
   215   }
   217   return entry;
   218 }
   221 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
   222   address entry = __ pc();
   224   // In SSE mode, FP results are in xmm0
   225   if (state == ftos && UseSSE > 0) {
   226     __ subptr(rsp, wordSize);
   227     __ movflt(Address(rsp, 0), xmm0);
   228     __ fld_s(Address(rsp, 0));
   229     __ addptr(rsp, wordSize);
   230   } else if (state == dtos && UseSSE >= 2) {
   231     __ subptr(rsp, 2*wordSize);
   232     __ movdbl(Address(rsp, 0), xmm0);
   233     __ fld_d(Address(rsp, 0));
   234     __ addptr(rsp, 2*wordSize);
   235   }
   237   __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
   239   // The stack is not extended by deopt but we must NULL last_sp as this
   240   // entry is like a "return".
   241   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
   242   __ restore_bcp();
   243   __ restore_locals();
   244   // handle exceptions
   245   { Label L;
   246     const Register thread = rcx;
   247     __ get_thread(thread);
   248     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
   249     __ jcc(Assembler::zero, L);
   250     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
   251     __ should_not_reach_here();
   252     __ bind(L);
   253   }
   254   __ dispatch_next(state, step);
   255   return entry;
   256 }
   259 int AbstractInterpreter::BasicType_as_index(BasicType type) {
   260   int i = 0;
   261   switch (type) {
   262     case T_BOOLEAN: i = 0; break;
   263     case T_CHAR   : i = 1; break;
   264     case T_BYTE   : i = 2; break;
   265     case T_SHORT  : i = 3; break;
   266     case T_INT    : // fall through
   267     case T_LONG   : // fall through
   268     case T_VOID   : i = 4; break;
   269     case T_FLOAT  : i = 5; break;  // have to treat float and double separately for SSE
   270     case T_DOUBLE : i = 6; break;
   271     case T_OBJECT : // fall through
   272     case T_ARRAY  : i = 7; break;
   273     default       : ShouldNotReachHere();
   274   }
   275   assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
   276   return i;
   277 }
   280 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
   281   address entry = __ pc();
   282   switch (type) {
   283     case T_BOOLEAN: __ c2bool(rax);            break;
   284     case T_CHAR   : __ andptr(rax, 0xFFFF);    break;
   285     case T_BYTE   : __ sign_extend_byte (rax); break;
   286     case T_SHORT  : __ sign_extend_short(rax); break;
   287     case T_INT    : /* nothing to do */        break;
   288     case T_DOUBLE :
   289     case T_FLOAT  :
   290       { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
   291         __ pop(t);                            // remove return address first
   292         // Must return a result for interpreter or compiler. In SSE
   293         // mode, results are returned in xmm0 and the FPU stack must
   294         // be empty.
   295         if (type == T_FLOAT && UseSSE >= 1) {
   296           // Load ST0
   297           __ fld_d(Address(rsp, 0));
   298           // Store as float and empty fpu stack
   299           __ fstp_s(Address(rsp, 0));
   300           // and reload
   301           __ movflt(xmm0, Address(rsp, 0));
   302         } else if (type == T_DOUBLE && UseSSE >= 2 ) {
   303           __ movdbl(xmm0, Address(rsp, 0));
   304         } else {
   305           // restore ST0
   306           __ fld_d(Address(rsp, 0));
   307         }
   308         // and pop the temp
   309         __ addptr(rsp, 2 * wordSize);
   310         __ push(t);                           // restore return address
   311       }
   312       break;
   313     case T_OBJECT :
   314       // retrieve result from frame
   315       __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
   316       // and verify it
   317       __ verify_oop(rax);
   318       break;
   319     default       : ShouldNotReachHere();
   320   }
   321   __ ret(0);                                   // return from result handler
   322   return entry;
   323 }
   325 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
   326   address entry = __ pc();
   327   __ push(state);
   328   __ call_VM(noreg, runtime_entry);
   329   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
   330   return entry;
   331 }
   334 // Helpers for commoning out cases in the various type of method entries.
   335 //
   337 // increment invocation count & check for overflow
   338 //
   339 // Note: checking for negative value instead of overflow
   340 //       so we have a 'sticky' overflow test
   341 //
   342 // rbx,: method
   343 // rcx: invocation counter
   344 //
   345 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
   346   const Address invocation_counter(rbx, in_bytes(Method::invocation_counter_offset()) +
   347                                         in_bytes(InvocationCounter::counter_offset()));
   348   // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
   349   if (TieredCompilation) {
   350     int increment = InvocationCounter::count_increment;
   351     int mask = ((1 << Tier0InvokeNotifyFreqLog)  - 1) << InvocationCounter::count_shift;
   352     Label no_mdo, done;
   353     if (ProfileInterpreter) {
   354       // Are we profiling?
   355       __ movptr(rax, Address(rbx, Method::method_data_offset()));
   356       __ testptr(rax, rax);
   357       __ jccb(Assembler::zero, no_mdo);
   358       // Increment counter in the MDO
   359       const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
   360                                                 in_bytes(InvocationCounter::counter_offset()));
   361       __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
   362       __ jmpb(done);
   363     }
   364     __ bind(no_mdo);
   365     // Increment counter in Method* (we don't need to load it, it's in rcx).
   366     __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
   367     __ bind(done);
   368   } else {
   369     const Address backedge_counter  (rbx, Method::backedge_counter_offset() +
   370                                           InvocationCounter::counter_offset());
   372     if (ProfileInterpreter) { // %%% Merge this into MethodData*
   373       __ incrementl(Address(rbx,Method::interpreter_invocation_counter_offset()));
   374     }
   375     // Update standard invocation counters
   376     __ movl(rax, backedge_counter);               // load backedge counter
   378     __ incrementl(rcx, InvocationCounter::count_increment);
   379     __ andl(rax, InvocationCounter::count_mask_value);  // mask out the status bits
   381     __ movl(invocation_counter, rcx);             // save invocation count
   382     __ addl(rcx, rax);                            // add both counters
   384     // profile_method is non-null only for interpreted method so
   385     // profile_method != NULL == !native_call
   386     // BytecodeInterpreter only calls for native so code is elided.
   388     if (ProfileInterpreter && profile_method != NULL) {
   389       // Test to see if we should create a method data oop
   390       __ cmp32(rcx,
   391                ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
   392       __ jcc(Assembler::less, *profile_method_continue);
   394       // if no method data exists, go to profile_method
   395       __ test_method_data_pointer(rax, *profile_method);
   396     }
   398     __ cmp32(rcx,
   399              ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
   400     __ jcc(Assembler::aboveEqual, *overflow);
   401   }
   402 }
   404 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
   406   // Asm interpreter on entry
   407   // rdi - locals
   408   // rsi - bcp
   409   // rbx, - method
   410   // rdx - cpool
   411   // rbp, - interpreter frame
   413   // C++ interpreter on entry
   414   // rsi - new interpreter state pointer
   415   // rbp - interpreter frame pointer
   416   // rbx - method
   418   // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
   419   // rbx, - method
   420   // rcx - rcvr (assuming there is one)
   421   // top of stack return address of interpreter caller
   422   // rsp - sender_sp
   424   // C++ interpreter only
   425   // rsi - previous interpreter state pointer
   427   const Address size_of_parameters(rbx, Method::size_of_parameters_offset());
   429   // InterpreterRuntime::frequency_counter_overflow takes one argument
   430   // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
   431   // The call returns the address of the verified entry point for the method or NULL
   432   // if the compilation did not complete (either went background or bailed out).
   433   __ movptr(rax, (intptr_t)false);
   434   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
   436   __ movptr(rbx, Address(rbp, method_offset));   // restore Method*
   438   // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
   439   // and jump to the interpreted entry.
   440   __ jmp(*do_continue, relocInfo::none);
   442 }
   444 void InterpreterGenerator::generate_stack_overflow_check(void) {
   445   // see if we've got enough room on the stack for locals plus overhead.
   446   // the expression stack grows down incrementally, so the normal guard
   447   // page mechanism will work for that.
   448   //
   449   // Registers live on entry:
   450   //
   451   // Asm interpreter
   452   // rdx: number of additional locals this frame needs (what we must check)
   453   // rbx,: Method*
   455   // destroyed on exit
   456   // rax,
   458   // NOTE:  since the additional locals are also always pushed (wasn't obvious in
   459   // generate_method_entry) so the guard should work for them too.
   460   //
   462   // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
   463   const int entry_size    = frame::interpreter_frame_monitor_size() * wordSize;
   465   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
   466   // be sure to change this if you add/subtract anything to/from the overhead area
   467   const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
   469   const int page_size = os::vm_page_size();
   471   Label after_frame_check;
   473   // see if the frame is greater than one page in size. If so,
   474   // then we need to verify there is enough stack space remaining
   475   // for the additional locals.
   476   __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
   477   __ jcc(Assembler::belowEqual, after_frame_check);
   479   // compute rsp as if this were going to be the last frame on
   480   // the stack before the red zone
   482   Label after_frame_check_pop;
   484   __ push(rsi);
   486   const Register thread = rsi;
   488   __ get_thread(thread);
   490   const Address stack_base(thread, Thread::stack_base_offset());
   491   const Address stack_size(thread, Thread::stack_size_offset());
   493   // locals + overhead, in bytes
   494   __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
   496 #ifdef ASSERT
   497   Label stack_base_okay, stack_size_okay;
   498   // verify that thread stack base is non-zero
   499   __ cmpptr(stack_base, (int32_t)NULL_WORD);
   500   __ jcc(Assembler::notEqual, stack_base_okay);
   501   __ stop("stack base is zero");
   502   __ bind(stack_base_okay);
   503   // verify that thread stack size is non-zero
   504   __ cmpptr(stack_size, 0);
   505   __ jcc(Assembler::notEqual, stack_size_okay);
   506   __ stop("stack size is zero");
   507   __ bind(stack_size_okay);
   508 #endif
   510   // Add stack base to locals and subtract stack size
   511   __ addptr(rax, stack_base);
   512   __ subptr(rax, stack_size);
   514   // Use the maximum number of pages we might bang.
   515   const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
   516                                                                               (StackRedPages+StackYellowPages);
   517   __ addptr(rax, max_pages * page_size);
   519   // check against the current stack bottom
   520   __ cmpptr(rsp, rax);
   521   __ jcc(Assembler::above, after_frame_check_pop);
   523   __ pop(rsi);  // get saved bcp / (c++ prev state ).
   525   // Restore sender's sp as SP. This is necessary if the sender's
   526   // frame is an extended compiled frame (see gen_c2i_adapter())
   527   // and safer anyway in case of JSR292 adaptations.
   529   __ pop(rax); // return address must be moved if SP is changed
   530   __ mov(rsp, rsi);
   531   __ push(rax);
   533   // Note: the restored frame is not necessarily interpreted.
   534   // Use the shared runtime version of the StackOverflowError.
   535   assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
   536   __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
   537   // all done with frame size check
   538   __ bind(after_frame_check_pop);
   539   __ pop(rsi);
   541   __ bind(after_frame_check);
   542 }
   544 // Allocate monitor and lock method (asm interpreter)
   545 // rbx, - Method*
   546 //
   547 void InterpreterGenerator::lock_method(void) {
   548   // synchronize method
   549   const Address access_flags      (rbx, Method::access_flags_offset());
   550   const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
   551   const int entry_size            = frame::interpreter_frame_monitor_size() * wordSize;
   553   #ifdef ASSERT
   554     { Label L;
   555       __ movl(rax, access_flags);
   556       __ testl(rax, JVM_ACC_SYNCHRONIZED);
   557       __ jcc(Assembler::notZero, L);
   558       __ stop("method doesn't need synchronization");
   559       __ bind(L);
   560     }
   561   #endif // ASSERT
   562   // get synchronization object
   563   { Label done;
   564     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
   565     __ movl(rax, access_flags);
   566     __ testl(rax, JVM_ACC_STATIC);
   567     __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0)));  // get receiver (assume this is frequent case)
   568     __ jcc(Assembler::zero, done);
   569     __ movptr(rax, Address(rbx, Method::const_offset()));
   570     __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
   571     __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
   572     __ movptr(rax, Address(rax, mirror_offset));
   573     __ bind(done);
   574   }
   575   // add space for monitor & lock
   576   __ subptr(rsp, entry_size);                                           // add space for a monitor entry
   577   __ movptr(monitor_block_top, rsp);                                    // set new monitor block top
   578   __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
   579   __ mov(rdx, rsp);                                                    // object address
   580   __ lock_object(rdx);
   581 }
   583 //
   584 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
   585 // and for native methods hence the shared code.
   587 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
   588   // initialize fixed part of activation frame
   589   __ push(rax);                                       // save return address
   590   __ enter();                                         // save old & set new rbp,
   593   __ push(rsi);                                       // set sender sp
   594   __ push((int32_t)NULL_WORD);                        // leave last_sp as null
   595   __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
   596   __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
   597   __ push(rbx);                                      // save Method*
   598   if (ProfileInterpreter) {
   599     Label method_data_continue;
   600     __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
   601     __ testptr(rdx, rdx);
   602     __ jcc(Assembler::zero, method_data_continue);
   603     __ addptr(rdx, in_bytes(MethodData::data_offset()));
   604     __ bind(method_data_continue);
   605     __ push(rdx);                                       // set the mdp (method data pointer)
   606   } else {
   607     __ push(0);
   608   }
   610   __ movptr(rdx, Address(rbx, Method::const_offset()));
   611   __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
   612   __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
   613   __ push(rdx);                                       // set constant pool cache
   614   __ push(rdi);                                       // set locals pointer
   615   if (native_call) {
   616     __ push(0);                                       // no bcp
   617   } else {
   618     __ push(rsi);                                     // set bcp
   619     }
   620   __ push(0);                                         // reserve word for pointer to expression stack bottom
   621   __ movptr(Address(rsp, 0), rsp);                    // set expression stack bottom
   622 }
   624 // End of helpers
   626 //
   627 // Various method entries
   628 //------------------------------------------------------------------------------------------------------------------------
   629 //
   630 //
   632 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
   634 address InterpreterGenerator::generate_accessor_entry(void) {
   636   // rbx,: Method*
   637   // rcx: receiver (preserve for slow entry into asm interpreter)
   639   // rsi: senderSP must preserved for slow path, set SP to it on fast path
   641   address entry_point = __ pc();
   642   Label xreturn_path;
   644   // do fastpath for resolved accessor methods
   645   if (UseFastAccessorMethods) {
   646     Label slow_path;
   647     // If we need a safepoint check, generate full interpreter entry.
   648     ExternalAddress state(SafepointSynchronize::address_of_state());
   649     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
   650              SafepointSynchronize::_not_synchronized);
   652     __ jcc(Assembler::notEqual, slow_path);
   653     // ASM/C++ Interpreter
   654     // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
   655     // Note: We can only use this code if the getfield has been resolved
   656     //       and if we don't have a null-pointer exception => check for
   657     //       these conditions first and use slow path if necessary.
   658     // rbx,: method
   659     // rcx: receiver
   660     __ movptr(rax, Address(rsp, wordSize));
   662     // check if local 0 != NULL and read field
   663     __ testptr(rax, rax);
   664     __ jcc(Assembler::zero, slow_path);
   666     // read first instruction word and extract bytecode @ 1 and index @ 2
   667     __ movptr(rdx, Address(rbx, Method::const_offset()));
   668     __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
   669     __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
   670     // Shift codes right to get the index on the right.
   671     // The bytecode fetched looks like <index><0xb4><0x2a>
   672     __ shrl(rdx, 2*BitsPerByte);
   673     __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
   674     __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
   676     // rax,: local 0
   677     // rbx,: method
   678     // rcx: receiver - do not destroy since it is needed for slow path!
   679     // rcx: scratch
   680     // rdx: constant pool cache index
   681     // rdi: constant pool cache
   682     // rsi: sender sp
   684     // check if getfield has been resolved and read constant pool cache entry
   685     // check the validity of the cache entry by testing whether _indices field
   686     // contains Bytecode::_getfield in b1 byte.
   687     assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
   688     __ movl(rcx,
   689             Address(rdi,
   690                     rdx,
   691                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
   692     __ shrl(rcx, 2*BitsPerByte);
   693     __ andl(rcx, 0xFF);
   694     __ cmpl(rcx, Bytecodes::_getfield);
   695     __ jcc(Assembler::notEqual, slow_path);
   697     // Note: constant pool entry is not valid before bytecode is resolved
   698     __ movptr(rcx,
   699               Address(rdi,
   700                       rdx,
   701                       Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
   702     __ movl(rdx,
   703             Address(rdi,
   704                     rdx,
   705                     Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
   707     Label notByte, notShort, notChar;
   708     const Address field_address (rax, rcx, Address::times_1);
   710     // Need to differentiate between igetfield, agetfield, bgetfield etc.
   711     // because they are different sizes.
   712     // Use the type from the constant pool cache
   713     __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
   714     // Make sure we don't need to mask rdx after the above shift
   715     ConstantPoolCacheEntry::verify_tos_state_shift();
   716     __ cmpl(rdx, btos);
   717     __ jcc(Assembler::notEqual, notByte);
   718     __ load_signed_byte(rax, field_address);
   719     __ jmp(xreturn_path);
   721     __ bind(notByte);
   722     __ cmpl(rdx, stos);
   723     __ jcc(Assembler::notEqual, notShort);
   724     __ load_signed_short(rax, field_address);
   725     __ jmp(xreturn_path);
   727     __ bind(notShort);
   728     __ cmpl(rdx, ctos);
   729     __ jcc(Assembler::notEqual, notChar);
   730     __ load_unsigned_short(rax, field_address);
   731     __ jmp(xreturn_path);
   733     __ bind(notChar);
   734 #ifdef ASSERT
   735     Label okay;
   736     __ cmpl(rdx, atos);
   737     __ jcc(Assembler::equal, okay);
   738     __ cmpl(rdx, itos);
   739     __ jcc(Assembler::equal, okay);
   740     __ stop("what type is this?");
   741     __ bind(okay);
   742 #endif // ASSERT
   743     // All the rest are a 32 bit wordsize
   744     // This is ok for now. Since fast accessors should be going away
   745     __ movptr(rax, field_address);
   747     __ bind(xreturn_path);
   749     // _ireturn/_areturn
   750     __ pop(rdi);                               // get return address
   751     __ mov(rsp, rsi);                          // set sp to sender sp
   752     __ jmp(rdi);
   754     // generate a vanilla interpreter entry as the slow path
   755     __ bind(slow_path);
   757     (void) generate_normal_entry(false);
   758     return entry_point;
   759   }
   760   return NULL;
   762 }
   764 // Method entry for java.lang.ref.Reference.get.
   765 address InterpreterGenerator::generate_Reference_get_entry(void) {
   766 #ifndef SERIALGC
   767   // Code: _aload_0, _getfield, _areturn
   768   // parameter size = 1
   769   //
   770   // The code that gets generated by this routine is split into 2 parts:
   771   //    1. The "intrinsified" code for G1 (or any SATB based GC),
   772   //    2. The slow path - which is an expansion of the regular method entry.
   773   //
   774   // Notes:-
   775   // * In the G1 code we do not check whether we need to block for
   776   //   a safepoint. If G1 is enabled then we must execute the specialized
   777   //   code for Reference.get (except when the Reference object is null)
   778   //   so that we can log the value in the referent field with an SATB
   779   //   update buffer.
   780   //   If the code for the getfield template is modified so that the
   781   //   G1 pre-barrier code is executed when the current method is
   782   //   Reference.get() then going through the normal method entry
   783   //   will be fine.
   784   // * The G1 code below can, however, check the receiver object (the instance
   785   //   of java.lang.Reference) and jump to the slow path if null. If the
   786   //   Reference object is null then we obviously cannot fetch the referent
   787   //   and so we don't need to call the G1 pre-barrier. Thus we can use the
   788   //   regular method entry code to generate the NPE.
   789   //
   790   // This code is based on generate_accessor_enty.
   792   // rbx,: Method*
   793   // rcx: receiver (preserve for slow entry into asm interpreter)
   795   // rsi: senderSP must preserved for slow path, set SP to it on fast path
   797   address entry = __ pc();
   799   const int referent_offset = java_lang_ref_Reference::referent_offset;
   800   guarantee(referent_offset > 0, "referent offset not initialized");
   802   if (UseG1GC) {
   803     Label slow_path;
   805     // Check if local 0 != NULL
   806     // If the receiver is null then it is OK to jump to the slow path.
   807     __ movptr(rax, Address(rsp, wordSize));
   808     __ testptr(rax, rax);
   809     __ jcc(Assembler::zero, slow_path);
   811     // rax: local 0 (must be preserved across the G1 barrier call)
   812     //
   813     // rbx: method (at this point it's scratch)
   814     // rcx: receiver (at this point it's scratch)
   815     // rdx: scratch
   816     // rdi: scratch
   817     //
   818     // rsi: sender sp
   820     // Preserve the sender sp in case the pre-barrier
   821     // calls the runtime
   822     __ push(rsi);
   824     // Load the value of the referent field.
   825     const Address field_address(rax, referent_offset);
   826     __ movptr(rax, field_address);
   828     // Generate the G1 pre-barrier code to log the value of
   829     // the referent field in an SATB buffer.
   830     __ get_thread(rcx);
   831     __ g1_write_barrier_pre(noreg /* obj */,
   832                             rax /* pre_val */,
   833                             rcx /* thread */,
   834                             rbx /* tmp */,
   835                             true /* tosca_save */,
   836                             true /* expand_call */);
   838     // _areturn
   839     __ pop(rsi);                // get sender sp
   840     __ pop(rdi);                // get return address
   841     __ mov(rsp, rsi);           // set sp to sender sp
   842     __ jmp(rdi);
   844     __ bind(slow_path);
   845     (void) generate_normal_entry(false);
   847     return entry;
   848   }
   849 #endif // SERIALGC
   851   // If G1 is not enabled then attempt to go through the accessor entry point
   852   // Reference.get is an accessor
   853   return generate_accessor_entry();
   854 }
   856 //
   857 // Interpreter stub for calling a native method. (asm interpreter)
   858 // This sets up a somewhat different looking stack for calling the native method
   859 // than the typical interpreter frame setup.
   860 //
   862 address InterpreterGenerator::generate_native_entry(bool synchronized) {
   863   // determine code generation flags
   864   bool inc_counter  = UseCompiler || CountCompiledCalls;
   866   // rbx,: Method*
   867   // rsi: sender sp
   868   // rsi: previous interpreter state (C++ interpreter) must preserve
   869   address entry_point = __ pc();
   872   const Address size_of_parameters(rbx, Method::size_of_parameters_offset());
   873   const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
   874   const Address access_flags      (rbx, Method::access_flags_offset());
   876   // get parameter size (always needed)
   877   __ load_unsigned_short(rcx, size_of_parameters);
   879   // native calls don't need the stack size check since they have no expression stack
   880   // and the arguments are already on the stack and we only add a handful of words
   881   // to the stack
   883   // rbx,: Method*
   884   // rcx: size of parameters
   885   // rsi: sender sp
   887   __ pop(rax);                                       // get return address
   888   // for natives the size of locals is zero
   890   // compute beginning of parameters (rdi)
   891   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
   894   // add 2 zero-initialized slots for native calls
   895   // NULL result handler
   896   __ push((int32_t)NULL_WORD);
   897   // NULL oop temp (mirror or jni oop result)
   898   __ push((int32_t)NULL_WORD);
   900   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
   901   // initialize fixed part of activation frame
   903   generate_fixed_frame(true);
   905   // make sure method is native & not abstract
   906 #ifdef ASSERT
   907   __ movl(rax, access_flags);
   908   {
   909     Label L;
   910     __ testl(rax, JVM_ACC_NATIVE);
   911     __ jcc(Assembler::notZero, L);
   912     __ stop("tried to execute non-native method as native");
   913     __ bind(L);
   914   }
   915   { Label L;
   916     __ testl(rax, JVM_ACC_ABSTRACT);
   917     __ jcc(Assembler::zero, L);
   918     __ stop("tried to execute abstract method in interpreter");
   919     __ bind(L);
   920   }
   921 #endif
   923   // Since at this point in the method invocation the exception handler
   924   // would try to exit the monitor of synchronized methods which hasn't
   925   // been entered yet, we set the thread local variable
   926   // _do_not_unlock_if_synchronized to true. The remove_activation will
   927   // check this flag.
   929   __ get_thread(rax);
   930   const Address do_not_unlock_if_synchronized(rax,
   931         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
   932   __ movbool(do_not_unlock_if_synchronized, true);
   934   // increment invocation count & check for overflow
   935   Label invocation_counter_overflow;
   936   if (inc_counter) {
   937     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
   938   }
   940   Label continue_after_compile;
   941   __ bind(continue_after_compile);
   943   bang_stack_shadow_pages(true);
   945   // reset the _do_not_unlock_if_synchronized flag
   946   __ get_thread(rax);
   947   __ movbool(do_not_unlock_if_synchronized, false);
   949   // check for synchronized methods
   950   // Must happen AFTER invocation_counter check and stack overflow check,
   951   // so method is not locked if overflows.
   952   //
   953   if (synchronized) {
   954     lock_method();
   955   } else {
   956     // no synchronization necessary
   957 #ifdef ASSERT
   958       { Label L;
   959         __ movl(rax, access_flags);
   960         __ testl(rax, JVM_ACC_SYNCHRONIZED);
   961         __ jcc(Assembler::zero, L);
   962         __ stop("method needs synchronization");
   963         __ bind(L);
   964       }
   965 #endif
   966   }
   968   // start execution
   969 #ifdef ASSERT
   970   { Label L;
   971     const Address monitor_block_top (rbp,
   972                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
   973     __ movptr(rax, monitor_block_top);
   974     __ cmpptr(rax, rsp);
   975     __ jcc(Assembler::equal, L);
   976     __ stop("broken stack frame setup in interpreter");
   977     __ bind(L);
   978   }
   979 #endif
   981   // jvmti/dtrace support
   982   __ notify_method_entry();
   984   // work registers
   985   const Register method = rbx;
   986   const Register thread = rdi;
   987   const Register t      = rcx;
   989   // allocate space for parameters
   990   __ get_method(method);
   991   __ load_unsigned_short(t, Address(method, Method::size_of_parameters_offset()));
   992   __ shlptr(t, Interpreter::logStackElementSize);
   993   __ addptr(t, 2*wordSize);     // allocate two more slots for JNIEnv and possible mirror
   994   __ subptr(rsp, t);
   995   __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
   997   // get signature handler
   998   { Label L;
   999     __ movptr(t, Address(method, Method::signature_handler_offset()));
  1000     __ testptr(t, t);
  1001     __ jcc(Assembler::notZero, L);
  1002     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
  1003     __ get_method(method);
  1004     __ movptr(t, Address(method, Method::signature_handler_offset()));
  1005     __ bind(L);
  1008   // call signature handler
  1009   assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
  1010   assert(InterpreterRuntime::SignatureHandlerGenerator::to  () == rsp, "adjust this code");
  1011   assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t  , "adjust this code");
  1012   // The generated handlers do not touch RBX (the method oop).
  1013   // However, large signatures cannot be cached and are generated
  1014   // each time here.  The slow-path generator will blow RBX
  1015   // sometime, so we must reload it after the call.
  1016   __ call(t);
  1017   __ get_method(method);        // slow path call blows RBX on DevStudio 5.0
  1019   // result handler is in rax,
  1020   // set result handler
  1021   __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
  1023   // pass mirror handle if static call
  1024   { Label L;
  1025     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
  1026     __ movl(t, Address(method, Method::access_flags_offset()));
  1027     __ testl(t, JVM_ACC_STATIC);
  1028     __ jcc(Assembler::zero, L);
  1029     // get mirror
  1030     __ movptr(t, Address(method, Method:: const_offset()));
  1031     __ movptr(t, Address(t, ConstMethod::constants_offset()));
  1032     __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
  1033     __ movptr(t, Address(t, mirror_offset));
  1034     // copy mirror into activation frame
  1035     __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
  1036     // pass handle to mirror
  1037     __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
  1038     __ movptr(Address(rsp, wordSize), t);
  1039     __ bind(L);
  1042   // get native function entry point
  1043   { Label L;
  1044     __ movptr(rax, Address(method, Method::native_function_offset()));
  1045     ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
  1046     __ cmpptr(rax, unsatisfied.addr());
  1047     __ jcc(Assembler::notEqual, L);
  1048     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
  1049     __ get_method(method);
  1050     __ movptr(rax, Address(method, Method::native_function_offset()));
  1051     __ bind(L);
  1054   // pass JNIEnv
  1055   __ get_thread(thread);
  1056   __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
  1057   __ movptr(Address(rsp, 0), t);
  1059   // set_last_Java_frame_before_call
  1060   // It is enough that the pc()
  1061   // points into the right code segment. It does not have to be the correct return pc.
  1062   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
  1064   // change thread state
  1065 #ifdef ASSERT
  1066   { Label L;
  1067     __ movl(t, Address(thread, JavaThread::thread_state_offset()));
  1068     __ cmpl(t, _thread_in_Java);
  1069     __ jcc(Assembler::equal, L);
  1070     __ stop("Wrong thread state in native stub");
  1071     __ bind(L);
  1073 #endif
  1075   // Change state to native
  1076   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
  1077   __ call(rax);
  1079   // result potentially in rdx:rax or ST0
  1081   // Either restore the MXCSR register after returning from the JNI Call
  1082   // or verify that it wasn't changed.
  1083   if (VM_Version::supports_sse()) {
  1084     if (RestoreMXCSROnJNICalls) {
  1085       __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
  1087     else if (CheckJNICalls ) {
  1088       __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
  1092   // Either restore the x87 floating pointer control word after returning
  1093   // from the JNI call or verify that it wasn't changed.
  1094   if (CheckJNICalls) {
  1095     __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
  1098   // save potential result in ST(0) & rdx:rax
  1099   // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
  1100   // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
  1101   // It is safe to do this push because state is _thread_in_native and return address will be found
  1102   // via _last_native_pc and not via _last_jave_sp
  1104   // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
  1105   // If the order changes or anything else is added to the stack the code in
  1106   // interpreter_frame_result will have to be changed.
  1108   { Label L;
  1109     Label push_double;
  1110     ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
  1111     ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
  1112     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
  1113               float_handler.addr());
  1114     __ jcc(Assembler::equal, push_double);
  1115     __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
  1116               double_handler.addr());
  1117     __ jcc(Assembler::notEqual, L);
  1118     __ bind(push_double);
  1119     __ push(dtos);
  1120     __ bind(L);
  1122   __ push(ltos);
  1124   // change thread state
  1125   __ get_thread(thread);
  1126   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
  1127   if(os::is_MP()) {
  1128     if (UseMembar) {
  1129       // Force this write out before the read below
  1130       __ membar(Assembler::Membar_mask_bits(
  1131            Assembler::LoadLoad | Assembler::LoadStore |
  1132            Assembler::StoreLoad | Assembler::StoreStore));
  1133     } else {
  1134       // Write serialization page so VM thread can do a pseudo remote membar.
  1135       // We use the current thread pointer to calculate a thread specific
  1136       // offset to write to within the page. This minimizes bus traffic
  1137       // due to cache line collision.
  1138       __ serialize_memory(thread, rcx);
  1142   if (AlwaysRestoreFPU) {
  1143     //  Make sure the control word is correct.
  1144     __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
  1147   // check for safepoint operation in progress and/or pending suspend requests
  1148   { Label Continue;
  1150     __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
  1151              SafepointSynchronize::_not_synchronized);
  1153     Label L;
  1154     __ jcc(Assembler::notEqual, L);
  1155     __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
  1156     __ jcc(Assembler::equal, Continue);
  1157     __ bind(L);
  1159     // Don't use call_VM as it will see a possible pending exception and forward it
  1160     // and never return here preventing us from clearing _last_native_pc down below.
  1161     // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
  1162     // preserved and correspond to the bcp/locals pointers. So we do a runtime call
  1163     // by hand.
  1164     //
  1165     __ push(thread);
  1166     __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
  1167                                             JavaThread::check_special_condition_for_native_trans)));
  1168     __ increment(rsp, wordSize);
  1169     __ get_thread(thread);
  1171     __ bind(Continue);
  1174   // change thread state
  1175   __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
  1177   __ reset_last_Java_frame(thread, true, true);
  1179   // reset handle block
  1180   __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
  1181   __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
  1183   // If result was an oop then unbox and save it in the frame
  1184   { Label L;
  1185     Label no_oop, store_result;
  1186     ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
  1187     __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
  1188               handler.addr());
  1189     __ jcc(Assembler::notEqual, no_oop);
  1190     __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
  1191     __ pop(ltos);
  1192     __ testptr(rax, rax);
  1193     __ jcc(Assembler::zero, store_result);
  1194     // unbox
  1195     __ movptr(rax, Address(rax, 0));
  1196     __ bind(store_result);
  1197     __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
  1198     // keep stack depth as expected by pushing oop which will eventually be discarded
  1199     __ push(ltos);
  1200     __ bind(no_oop);
  1204      Label no_reguard;
  1205      __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
  1206      __ jcc(Assembler::notEqual, no_reguard);
  1208      __ pusha();
  1209      __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
  1210      __ popa();
  1212      __ bind(no_reguard);
  1215   // restore rsi to have legal interpreter frame,
  1216   // i.e., bci == 0 <=> rsi == code_base()
  1217   // Can't call_VM until bcp is within reasonable.
  1218   __ get_method(method);      // method is junk from thread_in_native to now.
  1219   __ movptr(rsi, Address(method,Method::const_offset()));   // get ConstMethod*
  1220   __ lea(rsi, Address(rsi,ConstMethod::codes_offset()));    // get codebase
  1222   // handle exceptions (exception handling will handle unlocking!)
  1223   { Label L;
  1224     __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
  1225     __ jcc(Assembler::zero, L);
  1226     // Note: At some point we may want to unify this with the code used in call_VM_base();
  1227     //       i.e., we should use the StubRoutines::forward_exception code. For now this
  1228     //       doesn't work here because the rsp is not correctly set at this point.
  1229     __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
  1230     __ should_not_reach_here();
  1231     __ bind(L);
  1234   // do unlocking if necessary
  1235   { Label L;
  1236     __ movl(t, Address(method, Method::access_flags_offset()));
  1237     __ testl(t, JVM_ACC_SYNCHRONIZED);
  1238     __ jcc(Assembler::zero, L);
  1239     // the code below should be shared with interpreter macro assembler implementation
  1240     { Label unlock;
  1241       // BasicObjectLock will be first in list, since this is a synchronized method. However, need
  1242       // to check that the object has not been unlocked by an explicit monitorexit bytecode.
  1243       const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
  1245       __ lea(rdx, monitor);                   // address of first monitor
  1247       __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
  1248       __ testptr(t, t);
  1249       __ jcc(Assembler::notZero, unlock);
  1251       // Entry already unlocked, need to throw exception
  1252       __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
  1253       __ should_not_reach_here();
  1255       __ bind(unlock);
  1256       __ unlock_object(rdx);
  1258     __ bind(L);
  1261   // jvmti/dtrace support
  1262   // Note: This must happen _after_ handling/throwing any exceptions since
  1263   //       the exception handler code notifies the runtime of method exits
  1264   //       too. If this happens before, method entry/exit notifications are
  1265   //       not properly paired (was bug - gri 11/22/99).
  1266   __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
  1268   // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
  1269   __ pop(ltos);
  1270   __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
  1271   __ call(t);
  1273   // remove activation
  1274   __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
  1275   __ leave();                                // remove frame anchor
  1276   __ pop(rdi);                               // get return address
  1277   __ mov(rsp, t);                            // set sp to sender sp
  1278   __ jmp(rdi);
  1280   if (inc_counter) {
  1281     // Handle overflow of counter and compile method
  1282     __ bind(invocation_counter_overflow);
  1283     generate_counter_overflow(&continue_after_compile);
  1286   return entry_point;
  1289 //
  1290 // Generic interpreted method entry to (asm) interpreter
  1291 //
  1292 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
  1293   // determine code generation flags
  1294   bool inc_counter  = UseCompiler || CountCompiledCalls;
  1296   // rbx,: Method*
  1297   // rsi: sender sp
  1298   address entry_point = __ pc();
  1301   const Address size_of_parameters(rbx, Method::size_of_parameters_offset());
  1302   const Address size_of_locals    (rbx, Method::size_of_locals_offset());
  1303   const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
  1304   const Address access_flags      (rbx, Method::access_flags_offset());
  1306   // get parameter size (always needed)
  1307   __ load_unsigned_short(rcx, size_of_parameters);
  1309   // rbx,: Method*
  1310   // rcx: size of parameters
  1312   // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
  1314   __ load_unsigned_short(rdx, size_of_locals);       // get size of locals in words
  1315   __ subl(rdx, rcx);                                // rdx = no. of additional locals
  1317   // see if we've got enough room on the stack for locals plus overhead.
  1318   generate_stack_overflow_check();
  1320   // get return address
  1321   __ pop(rax);
  1323   // compute beginning of parameters (rdi)
  1324   __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
  1326   // rdx - # of additional locals
  1327   // allocate space for locals
  1328   // explicitly initialize locals
  1330     Label exit, loop;
  1331     __ testl(rdx, rdx);
  1332     __ jcc(Assembler::lessEqual, exit);               // do nothing if rdx <= 0
  1333     __ bind(loop);
  1334     __ push((int32_t)NULL_WORD);                      // initialize local variables
  1335     __ decrement(rdx);                                // until everything initialized
  1336     __ jcc(Assembler::greater, loop);
  1337     __ bind(exit);
  1340   if (inc_counter) __ movl(rcx, invocation_counter);  // (pre-)fetch invocation count
  1341   // initialize fixed part of activation frame
  1342   generate_fixed_frame(false);
  1344   // make sure method is not native & not abstract
  1345 #ifdef ASSERT
  1346   __ movl(rax, access_flags);
  1348     Label L;
  1349     __ testl(rax, JVM_ACC_NATIVE);
  1350     __ jcc(Assembler::zero, L);
  1351     __ stop("tried to execute native method as non-native");
  1352     __ bind(L);
  1354   { Label L;
  1355     __ testl(rax, JVM_ACC_ABSTRACT);
  1356     __ jcc(Assembler::zero, L);
  1357     __ stop("tried to execute abstract method in interpreter");
  1358     __ bind(L);
  1360 #endif
  1362   // Since at this point in the method invocation the exception handler
  1363   // would try to exit the monitor of synchronized methods which hasn't
  1364   // been entered yet, we set the thread local variable
  1365   // _do_not_unlock_if_synchronized to true. The remove_activation will
  1366   // check this flag.
  1368   __ get_thread(rax);
  1369   const Address do_not_unlock_if_synchronized(rax,
  1370         in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
  1371   __ movbool(do_not_unlock_if_synchronized, true);
  1373   // increment invocation count & check for overflow
  1374   Label invocation_counter_overflow;
  1375   Label profile_method;
  1376   Label profile_method_continue;
  1377   if (inc_counter) {
  1378     generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
  1379     if (ProfileInterpreter) {
  1380       __ bind(profile_method_continue);
  1383   Label continue_after_compile;
  1384   __ bind(continue_after_compile);
  1386   bang_stack_shadow_pages(false);
  1388   // reset the _do_not_unlock_if_synchronized flag
  1389   __ get_thread(rax);
  1390   __ movbool(do_not_unlock_if_synchronized, false);
  1392   // check for synchronized methods
  1393   // Must happen AFTER invocation_counter check and stack overflow check,
  1394   // so method is not locked if overflows.
  1395   //
  1396   if (synchronized) {
  1397     // Allocate monitor and lock method
  1398     lock_method();
  1399   } else {
  1400     // no synchronization necessary
  1401 #ifdef ASSERT
  1402       { Label L;
  1403         __ movl(rax, access_flags);
  1404         __ testl(rax, JVM_ACC_SYNCHRONIZED);
  1405         __ jcc(Assembler::zero, L);
  1406         __ stop("method needs synchronization");
  1407         __ bind(L);
  1409 #endif
  1412   // start execution
  1413 #ifdef ASSERT
  1414   { Label L;
  1415      const Address monitor_block_top (rbp,
  1416                  frame::interpreter_frame_monitor_block_top_offset * wordSize);
  1417     __ movptr(rax, monitor_block_top);
  1418     __ cmpptr(rax, rsp);
  1419     __ jcc(Assembler::equal, L);
  1420     __ stop("broken stack frame setup in interpreter");
  1421     __ bind(L);
  1423 #endif
  1425   // jvmti support
  1426   __ notify_method_entry();
  1428   __ dispatch_next(vtos);
  1430   // invocation counter overflow
  1431   if (inc_counter) {
  1432     if (ProfileInterpreter) {
  1433       // We have decided to profile this method in the interpreter
  1434       __ bind(profile_method);
  1435       __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
  1436       __ set_method_data_pointer_for_bcp();
  1437       __ get_method(rbx);
  1438       __ jmp(profile_method_continue);
  1440     // Handle overflow of counter and compile method
  1441     __ bind(invocation_counter_overflow);
  1442     generate_counter_overflow(&continue_after_compile);
  1445   return entry_point;
  1448 //------------------------------------------------------------------------------------------------------------------------
  1449 // Entry points
  1450 //
  1451 // Here we generate the various kind of entries into the interpreter.
  1452 // The two main entry type are generic bytecode methods and native call method.
  1453 // These both come in synchronized and non-synchronized versions but the
  1454 // frame layout they create is very similar. The other method entry
  1455 // types are really just special purpose entries that are really entry
  1456 // and interpretation all in one. These are for trivial methods like
  1457 // accessor, empty, or special math methods.
  1458 //
  1459 // When control flow reaches any of the entry types for the interpreter
  1460 // the following holds ->
  1461 //
  1462 // Arguments:
  1463 //
  1464 // rbx,: Method*
  1465 // rcx: receiver
  1466 //
  1467 //
  1468 // Stack layout immediately at entry
  1469 //
  1470 // [ return address     ] <--- rsp
  1471 // [ parameter n        ]
  1472 //   ...
  1473 // [ parameter 1        ]
  1474 // [ expression stack   ] (caller's java expression stack)
  1476 // Assuming that we don't go to one of the trivial specialized
  1477 // entries the stack will look like below when we are ready to execute
  1478 // the first bytecode (or call the native routine). The register usage
  1479 // will be as the template based interpreter expects (see interpreter_x86.hpp).
  1480 //
  1481 // local variables follow incoming parameters immediately; i.e.
  1482 // the return address is moved to the end of the locals).
  1483 //
  1484 // [ monitor entry      ] <--- rsp
  1485 //   ...
  1486 // [ monitor entry      ]
  1487 // [ expr. stack bottom ]
  1488 // [ saved rsi          ]
  1489 // [ current rdi        ]
  1490 // [ Method*            ]
  1491 // [ saved rbp,          ] <--- rbp,
  1492 // [ return address     ]
  1493 // [ local variable m   ]
  1494 //   ...
  1495 // [ local variable 1   ]
  1496 // [ parameter n        ]
  1497 //   ...
  1498 // [ parameter 1        ] <--- rdi
  1500 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
  1501   // determine code generation flags
  1502   bool synchronized = false;
  1503   address entry_point = NULL;
  1505   switch (kind) {
  1506     case Interpreter::zerolocals             :                                                                             break;
  1507     case Interpreter::zerolocals_synchronized: synchronized = true;                                                        break;
  1508     case Interpreter::native                 : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);  break;
  1509     case Interpreter::native_synchronized    : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);   break;
  1510     case Interpreter::empty                  : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();        break;
  1511     case Interpreter::accessor               : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();     break;
  1512     case Interpreter::abstract               : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();     break;
  1514     case Interpreter::java_lang_math_sin     : // fall thru
  1515     case Interpreter::java_lang_math_cos     : // fall thru
  1516     case Interpreter::java_lang_math_tan     : // fall thru
  1517     case Interpreter::java_lang_math_abs     : // fall thru
  1518     case Interpreter::java_lang_math_log     : // fall thru
  1519     case Interpreter::java_lang_math_log10   : // fall thru
  1520     case Interpreter::java_lang_math_sqrt    : // fall thru
  1521     case Interpreter::java_lang_math_pow     : // fall thru
  1522     case Interpreter::java_lang_math_exp     : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind);     break;
  1523     case Interpreter::java_lang_ref_reference_get
  1524                                              : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
  1525     default:
  1526       fatal(err_msg("unexpected method kind: %d", kind));
  1527       break;
  1530   if (entry_point) return entry_point;
  1532   return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
  1536 // These should never be compiled since the interpreter will prefer
  1537 // the compiled version to the intrinsic version.
  1538 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
  1539   switch (method_kind(m)) {
  1540     case Interpreter::java_lang_math_sin     : // fall thru
  1541     case Interpreter::java_lang_math_cos     : // fall thru
  1542     case Interpreter::java_lang_math_tan     : // fall thru
  1543     case Interpreter::java_lang_math_abs     : // fall thru
  1544     case Interpreter::java_lang_math_log     : // fall thru
  1545     case Interpreter::java_lang_math_log10   : // fall thru
  1546     case Interpreter::java_lang_math_sqrt    : // fall thru
  1547     case Interpreter::java_lang_math_pow     : // fall thru
  1548     case Interpreter::java_lang_math_exp     :
  1549       return false;
  1550     default:
  1551       return true;
  1555 // How much stack a method activation needs in words.
  1556 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
  1558   const int stub_code = 4;  // see generate_call_stub
  1559   // Save space for one monitor to get into the interpreted method in case
  1560   // the method is synchronized
  1561   int monitor_size    = method->is_synchronized() ?
  1562                                 1*frame::interpreter_frame_monitor_size() : 0;
  1564   // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
  1565   // be sure to change this if you add/subtract anything to/from the overhead area
  1566   const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
  1568   const int extra_stack = Method::extra_stack_entries();
  1569   const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
  1570                            Interpreter::stackElementWords;
  1571   return overhead_size + method_stack + stub_code;
  1574 // asm based interpreter deoptimization helpers
  1576 int AbstractInterpreter::layout_activation(Method* method,
  1577                                            int tempcount,
  1578                                            int popframe_extra_args,
  1579                                            int moncount,
  1580                                            int caller_actual_parameters,
  1581                                            int callee_param_count,
  1582                                            int callee_locals,
  1583                                            frame* caller,
  1584                                            frame* interpreter_frame,
  1585                                            bool is_top_frame) {
  1586   // Note: This calculation must exactly parallel the frame setup
  1587   // in AbstractInterpreterGenerator::generate_method_entry.
  1588   // If interpreter_frame!=NULL, set up the method, locals, and monitors.
  1589   // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
  1590   // as determined by a previous call to this method.
  1591   // It is also guaranteed to be walkable even though it is in a skeletal state
  1592   // NOTE: return size is in words not bytes
  1594   // fixed size of an interpreter frame:
  1595   int max_locals = method->max_locals() * Interpreter::stackElementWords;
  1596   int extra_locals = (method->max_locals() - method->size_of_parameters()) *
  1597                      Interpreter::stackElementWords;
  1599   int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
  1601   // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
  1602   // Since the callee parameters already account for the callee's params we only need to account for
  1603   // the extra locals.
  1606   int size = overhead +
  1607          ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
  1608          (moncount*frame::interpreter_frame_monitor_size()) +
  1609          tempcount*Interpreter::stackElementWords + popframe_extra_args;
  1611   if (interpreter_frame != NULL) {
  1612 #ifdef ASSERT
  1613     if (!EnableInvokeDynamic)
  1614       // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
  1615       // Probably, since deoptimization doesn't work yet.
  1616       assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
  1617     assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
  1618 #endif
  1620     interpreter_frame->interpreter_frame_set_method(method);
  1621     // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
  1622     // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
  1623     // and sender_sp is fp+8
  1624     intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
  1626 #ifdef ASSERT
  1627     if (caller->is_interpreted_frame()) {
  1628       assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
  1630 #endif
  1632     interpreter_frame->interpreter_frame_set_locals(locals);
  1633     BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
  1634     BasicObjectLock* monbot = montop - moncount;
  1635     interpreter_frame->interpreter_frame_set_monitor_end(monbot);
  1637     // Set last_sp
  1638     intptr_t*  rsp = (intptr_t*) monbot  -
  1639                      tempcount*Interpreter::stackElementWords -
  1640                      popframe_extra_args;
  1641     interpreter_frame->interpreter_frame_set_last_sp(rsp);
  1643     // All frames but the initial (oldest) interpreter frame we fill in have a
  1644     // value for sender_sp that allows walking the stack but isn't
  1645     // truly correct. Correct the value here.
  1647     if (extra_locals != 0 &&
  1648         interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
  1649       interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
  1651     *interpreter_frame->interpreter_frame_cache_addr() =
  1652       method->constants()->cache();
  1654   return size;
  1658 //------------------------------------------------------------------------------------------------------------------------
  1659 // Exceptions
  1661 void TemplateInterpreterGenerator::generate_throw_exception() {
  1662   // Entry point in previous activation (i.e., if the caller was interpreted)
  1663   Interpreter::_rethrow_exception_entry = __ pc();
  1664   const Register thread = rcx;
  1666   // Restore sp to interpreter_frame_last_sp even though we are going
  1667   // to empty the expression stack for the exception processing.
  1668   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
  1669   // rax,: exception
  1670   // rdx: return address/pc that threw exception
  1671   __ restore_bcp();                              // rsi points to call/send
  1672   __ restore_locals();
  1674   // Entry point for exceptions thrown within interpreter code
  1675   Interpreter::_throw_exception_entry = __ pc();
  1676   // expression stack is undefined here
  1677   // rax,: exception
  1678   // rsi: exception bcp
  1679   __ verify_oop(rax);
  1681   // expression stack must be empty before entering the VM in case of an exception
  1682   __ empty_expression_stack();
  1683   __ empty_FPU_stack();
  1684   // find exception handler address and preserve exception oop
  1685   __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
  1686   // rax,: exception handler entry point
  1687   // rdx: preserved exception oop
  1688   // rsi: bcp for exception handler
  1689   __ push_ptr(rdx);                              // push exception which is now the only value on the stack
  1690   __ jmp(rax);                                   // jump to exception handler (may be _remove_activation_entry!)
  1692   // If the exception is not handled in the current frame the frame is removed and
  1693   // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
  1694   //
  1695   // Note: At this point the bci is still the bxi for the instruction which caused
  1696   //       the exception and the expression stack is empty. Thus, for any VM calls
  1697   //       at this point, GC will find a legal oop map (with empty expression stack).
  1699   // In current activation
  1700   // tos: exception
  1701   // rsi: exception bcp
  1703   //
  1704   // JVMTI PopFrame support
  1705   //
  1707    Interpreter::_remove_activation_preserving_args_entry = __ pc();
  1708   __ empty_expression_stack();
  1709   __ empty_FPU_stack();
  1710   // Set the popframe_processing bit in pending_popframe_condition indicating that we are
  1711   // currently handling popframe, so that call_VMs that may happen later do not trigger new
  1712   // popframe handling cycles.
  1713   __ get_thread(thread);
  1714   __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
  1715   __ orl(rdx, JavaThread::popframe_processing_bit);
  1716   __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
  1719     // Check to see whether we are returning to a deoptimized frame.
  1720     // (The PopFrame call ensures that the caller of the popped frame is
  1721     // either interpreted or compiled and deoptimizes it if compiled.)
  1722     // In this case, we can't call dispatch_next() after the frame is
  1723     // popped, but instead must save the incoming arguments and restore
  1724     // them after deoptimization has occurred.
  1725     //
  1726     // Note that we don't compare the return PC against the
  1727     // deoptimization blob's unpack entry because of the presence of
  1728     // adapter frames in C2.
  1729     Label caller_not_deoptimized;
  1730     __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
  1731     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
  1732     __ testl(rax, rax);
  1733     __ jcc(Assembler::notZero, caller_not_deoptimized);
  1735     // Compute size of arguments for saving when returning to deoptimized caller
  1736     __ get_method(rax);
  1737     __ load_unsigned_short(rax, Address(rax, in_bytes(Method::size_of_parameters_offset())));
  1738     __ shlptr(rax, Interpreter::logStackElementSize);
  1739     __ restore_locals();
  1740     __ subptr(rdi, rax);
  1741     __ addptr(rdi, wordSize);
  1742     // Save these arguments
  1743     __ get_thread(thread);
  1744     __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
  1746     __ remove_activation(vtos, rdx,
  1747                          /* throw_monitor_exception */ false,
  1748                          /* install_monitor_exception */ false,
  1749                          /* notify_jvmdi */ false);
  1751     // Inform deoptimization that it is responsible for restoring these arguments
  1752     __ get_thread(thread);
  1753     __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
  1755     // Continue in deoptimization handler
  1756     __ jmp(rdx);
  1758     __ bind(caller_not_deoptimized);
  1761   __ remove_activation(vtos, rdx,
  1762                        /* throw_monitor_exception */ false,
  1763                        /* install_monitor_exception */ false,
  1764                        /* notify_jvmdi */ false);
  1766   // Finish with popframe handling
  1767   // A previous I2C followed by a deoptimization might have moved the
  1768   // outgoing arguments further up the stack. PopFrame expects the
  1769   // mutations to those outgoing arguments to be preserved and other
  1770   // constraints basically require this frame to look exactly as
  1771   // though it had previously invoked an interpreted activation with
  1772   // no space between the top of the expression stack (current
  1773   // last_sp) and the top of stack. Rather than force deopt to
  1774   // maintain this kind of invariant all the time we call a small
  1775   // fixup routine to move the mutated arguments onto the top of our
  1776   // expression stack if necessary.
  1777   __ mov(rax, rsp);
  1778   __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
  1779   __ get_thread(thread);
  1780   // PC must point into interpreter here
  1781   __ set_last_Java_frame(thread, noreg, rbp, __ pc());
  1782   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
  1783   __ get_thread(thread);
  1784   __ reset_last_Java_frame(thread, true, true);
  1785   // Restore the last_sp and null it out
  1786   __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
  1787   __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
  1789   __ restore_bcp();
  1790   __ restore_locals();
  1791   // The method data pointer was incremented already during
  1792   // call profiling. We have to restore the mdp for the current bcp.
  1793   if (ProfileInterpreter) {
  1794     __ set_method_data_pointer_for_bcp();
  1797   // Clear the popframe condition flag
  1798   __ get_thread(thread);
  1799   __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
  1801   __ dispatch_next(vtos);
  1802   // end of PopFrame support
  1804   Interpreter::_remove_activation_entry = __ pc();
  1806   // preserve exception over this code sequence
  1807   __ pop_ptr(rax);
  1808   __ get_thread(thread);
  1809   __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
  1810   // remove the activation (without doing throws on illegalMonitorExceptions)
  1811   __ remove_activation(vtos, rdx, false, true, false);
  1812   // restore exception
  1813   __ get_thread(thread);
  1814   __ get_vm_result(rax, thread);
  1816   // Inbetween activations - previous activation type unknown yet
  1817   // compute continuation point - the continuation point expects
  1818   // the following registers set up:
  1819   //
  1820   // rax: exception
  1821   // rdx: return address/pc that threw exception
  1822   // rsp: expression stack of caller
  1823   // rbp: rbp, of caller
  1824   __ push(rax);                                  // save exception
  1825   __ push(rdx);                                  // save return address
  1826   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
  1827   __ mov(rbx, rax);                              // save exception handler
  1828   __ pop(rdx);                                   // restore return address
  1829   __ pop(rax);                                   // restore exception
  1830   // Note that an "issuing PC" is actually the next PC after the call
  1831   __ jmp(rbx);                                   // jump to exception handler of caller
  1835 //
  1836 // JVMTI ForceEarlyReturn support
  1837 //
  1838 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
  1839   address entry = __ pc();
  1840   const Register thread = rcx;
  1842   __ restore_bcp();
  1843   __ restore_locals();
  1844   __ empty_expression_stack();
  1845   __ empty_FPU_stack();
  1846   __ load_earlyret_value(state);
  1848   __ get_thread(thread);
  1849   __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
  1850   const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
  1852   // Clear the earlyret state
  1853   __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
  1855   __ remove_activation(state, rsi,
  1856                        false, /* throw_monitor_exception */
  1857                        false, /* install_monitor_exception */
  1858                        true); /* notify_jvmdi */
  1859   __ jmp(rsi);
  1860   return entry;
  1861 } // end of ForceEarlyReturn support
  1864 //------------------------------------------------------------------------------------------------------------------------
  1865 // Helper for vtos entry point generation
  1867 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
  1868   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
  1869   Label L;
  1870   fep = __ pc(); __ push(ftos); __ jmp(L);
  1871   dep = __ pc(); __ push(dtos); __ jmp(L);
  1872   lep = __ pc(); __ push(ltos); __ jmp(L);
  1873   aep = __ pc(); __ push(atos); __ jmp(L);
  1874   bep = cep = sep =             // fall through
  1875   iep = __ pc(); __ push(itos); // fall through
  1876   vep = __ pc(); __ bind(L);    // fall through
  1877   generate_and_dispatch(t);
  1880 //------------------------------------------------------------------------------------------------------------------------
  1881 // Generation of individual instructions
  1883 // helpers for generate_and_dispatch
  1887 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
  1888  : TemplateInterpreterGenerator(code) {
  1889    generate_all(); // down here so it can be "virtual"
  1892 //------------------------------------------------------------------------------------------------------------------------
  1894 // Non-product code
  1895 #ifndef PRODUCT
  1896 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
  1897   address entry = __ pc();
  1899   // prepare expression stack
  1900   __ pop(rcx);          // pop return address so expression stack is 'pure'
  1901   __ push(state);       // save tosca
  1903   // pass tosca registers as arguments & call tracer
  1904   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
  1905   __ mov(rcx, rax);     // make sure return address is not destroyed by pop(state)
  1906   __ pop(state);        // restore tosca
  1908   // return
  1909   __ jmp(rcx);
  1911   return entry;
  1915 void TemplateInterpreterGenerator::count_bytecode() {
  1916   __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
  1920 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
  1921   __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
  1925 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
  1926   __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
  1927   __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
  1928   __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
  1929   ExternalAddress table((address) BytecodePairHistogram::_counters);
  1930   Address index(noreg, rbx, Address::times_4);
  1931   __ incrementl(ArrayAddress(table, index));
  1935 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
  1936   // Call a little run-time stub to avoid blow-up for each bytecode.
  1937   // The run-time runtime saves the right registers, depending on
  1938   // the tosca in-state for the given template.
  1939   assert(Interpreter::trace_code(t->tos_in()) != NULL,
  1940          "entry must have been generated");
  1941   __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
  1945 void TemplateInterpreterGenerator::stop_interpreter_at() {
  1946   Label L;
  1947   __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
  1948            StopInterpreterAt);
  1949   __ jcc(Assembler::notEqual, L);
  1950   __ int3();
  1951   __ bind(L);
  1953 #endif // !PRODUCT
  1954 #endif // CC_INTERP

mercurial