hotspot/src/cpu/ppc/vm/templateInterpreterGenerator_ppc.cpp
changeset 34651 07b1cc0f6040
parent 33070 54f3f085b165
child 35135 dd2ce9021031
child 35166 23125410af16
equal deleted inserted replaced
34648:b7ea5d095ef5 34651:07b1cc0f6040
       
     1 /*
       
     2  * Copyright (c) 2014, 2015, Oracle and/or its affiliates. All rights reserved.
       
     3  * Copyright (c) 2015 SAP AG. All rights reserved.
       
     4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
       
     5  *
       
     6  * This code is free software; you can redistribute it and/or modify it
       
     7  * under the terms of the GNU General Public License version 2 only, as
       
     8  * published by the Free Software Foundation.
       
     9  *
       
    10  * This code is distributed in the hope that it will be useful, but WITHOUT
       
    11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
       
    12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
       
    13  * version 2 for more details (a copy is included in the LICENSE file that
       
    14  * accompanied this code).
       
    15  *
       
    16  * You should have received a copy of the GNU General Public License version
       
    17  * 2 along with this work; if not, write to the Free Software Foundation,
       
    18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
       
    19  *
       
    20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
       
    21  * or visit www.oracle.com if you need additional information or have any
       
    22  * questions.
       
    23  *
       
    24  */
       
    25 
       
    26 #include "precompiled.hpp"
       
    27 #ifndef CC_INTERP
       
    28 #include "asm/macroAssembler.inline.hpp"
       
    29 #include "interpreter/bytecodeHistogram.hpp"
       
    30 #include "interpreter/interpreter.hpp"
       
    31 #include "interpreter/interpreterGenerator.hpp"
       
    32 #include "interpreter/interpreterRuntime.hpp"
       
    33 #include "interpreter/interp_masm.hpp"
       
    34 #include "interpreter/templateTable.hpp"
       
    35 #include "oops/arrayOop.hpp"
       
    36 #include "oops/methodData.hpp"
       
    37 #include "oops/method.hpp"
       
    38 #include "oops/oop.inline.hpp"
       
    39 #include "prims/jvmtiExport.hpp"
       
    40 #include "prims/jvmtiThreadState.hpp"
       
    41 #include "runtime/arguments.hpp"
       
    42 #include "runtime/deoptimization.hpp"
       
    43 #include "runtime/frame.inline.hpp"
       
    44 #include "runtime/sharedRuntime.hpp"
       
    45 #include "runtime/stubRoutines.hpp"
       
    46 #include "runtime/synchronizer.hpp"
       
    47 #include "runtime/timer.hpp"
       
    48 #include "runtime/vframeArray.hpp"
       
    49 #include "utilities/debug.hpp"
       
    50 #include "utilities/macros.hpp"
       
    51 
       
    52 #undef __
       
    53 #define __ _masm->
       
    54 
       
    55 #ifdef PRODUCT
       
    56 #define BLOCK_COMMENT(str) /* nothing */
       
    57 #else
       
    58 #define BLOCK_COMMENT(str) __ block_comment(str)
       
    59 #endif
       
    60 
       
    61 #define BIND(label)        __ bind(label); BLOCK_COMMENT(#label ":")
       
    62 
       
    63 //-----------------------------------------------------------------------------
       
    64 
       
    65 // Actually we should never reach here since we do stack overflow checks before pushing any frame.
       
    66 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
       
    67   address entry = __ pc();
       
    68   __ unimplemented("generate_StackOverflowError_handler");
       
    69   return entry;
       
    70 }
       
    71 
       
    72 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
       
    73   address entry = __ pc();
       
    74   __ empty_expression_stack();
       
    75   __ load_const_optimized(R4_ARG2, (address) name);
       
    76   // Index is in R17_tos.
       
    77   __ mr(R5_ARG3, R17_tos);
       
    78   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException));
       
    79   return entry;
       
    80 }
       
    81 
       
    82 #if 0
       
    83 // Call special ClassCastException constructor taking object to cast
       
    84 // and target class as arguments.
       
    85 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() {
       
    86   address entry = __ pc();
       
    87 
       
    88   // Expression stack must be empty before entering the VM if an
       
    89   // exception happened.
       
    90   __ empty_expression_stack();
       
    91 
       
    92   // Thread will be loaded to R3_ARG1.
       
    93   // Target class oop is in register R5_ARG3 by convention!
       
    94   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3);
       
    95   // Above call must not return here since exception pending.
       
    96   DEBUG_ONLY(__ should_not_reach_here();)
       
    97   return entry;
       
    98 }
       
    99 #endif
       
   100 
       
   101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
       
   102   address entry = __ pc();
       
   103   // Expression stack must be empty before entering the VM if an
       
   104   // exception happened.
       
   105   __ empty_expression_stack();
       
   106 
       
   107   // Load exception object.
       
   108   // Thread will be loaded to R3_ARG1.
       
   109   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos);
       
   110 #ifdef ASSERT
       
   111   // Above call must not return here since exception pending.
       
   112   __ should_not_reach_here();
       
   113 #endif
       
   114   return entry;
       
   115 }
       
   116 
       
   117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
       
   118   address entry = __ pc();
       
   119   //__ untested("generate_exception_handler_common");
       
   120   Register Rexception = R17_tos;
       
   121 
       
   122   // Expression stack must be empty before entering the VM if an exception happened.
       
   123   __ empty_expression_stack();
       
   124 
       
   125   __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1);
       
   126   if (pass_oop) {
       
   127     __ mr(R5_ARG3, Rexception);
       
   128     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false);
       
   129   } else {
       
   130     __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1);
       
   131     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false);
       
   132   }
       
   133 
       
   134   // Throw exception.
       
   135   __ mr(R3_ARG1, Rexception);
       
   136   __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2);
       
   137   __ mtctr(R11_scratch1);
       
   138   __ bctr();
       
   139 
       
   140   return entry;
       
   141 }
       
   142 
       
   143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
       
   144   address entry = __ pc();
       
   145   __ unimplemented("generate_continuation_for");
       
   146   return entry;
       
   147 }
       
   148 
       
   149 // This entry is returned to when a call returns to the interpreter.
       
   150 // When we arrive here, we expect that the callee stack frame is already popped.
       
   151 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
       
   152   address entry = __ pc();
       
   153 
       
   154   // Move the value out of the return register back to the TOS cache of current frame.
       
   155   switch (state) {
       
   156     case ltos:
       
   157     case btos:
       
   158     case ctos:
       
   159     case stos:
       
   160     case atos:
       
   161     case itos: __ mr(R17_tos, R3_RET); break;   // RET -> TOS cache
       
   162     case ftos:
       
   163     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
       
   164     case vtos: break;                           // Nothing to do, this was a void return.
       
   165     default  : ShouldNotReachHere();
       
   166   }
       
   167 
       
   168   __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
       
   169   __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
       
   170   __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
       
   171 
       
   172   // Compiled code destroys templateTableBase, reload.
       
   173   __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2);
       
   174 
       
   175   if (state == atos) {
       
   176     __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2);
       
   177   }
       
   178 
       
   179   const Register cache = R11_scratch1;
       
   180   const Register size  = R12_scratch2;
       
   181   __ get_cache_and_index_at_bcp(cache, 1, index_size);
       
   182 
       
   183   // Get least significant byte of 64 bit value:
       
   184 #if defined(VM_LITTLE_ENDIAN)
       
   185   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
       
   186 #else
       
   187   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
       
   188 #endif
       
   189   __ sldi(size, size, Interpreter::logStackElementSize);
       
   190   __ add(R15_esp, R15_esp, size);
       
   191   __ dispatch_next(state, step);
       
   192   return entry;
       
   193 }
       
   194 
       
   195 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
       
   196   address entry = __ pc();
       
   197   // If state != vtos, we're returning from a native method, which put it's result
       
   198   // into the result register. So move the value out of the return register back
       
   199   // to the TOS cache of current frame.
       
   200 
       
   201   switch (state) {
       
   202     case ltos:
       
   203     case btos:
       
   204     case ctos:
       
   205     case stos:
       
   206     case atos:
       
   207     case itos: __ mr(R17_tos, R3_RET); break;   // GR_RET -> TOS cache
       
   208     case ftos:
       
   209     case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET
       
   210     case vtos: break;                           // Nothing to do, this was a void return.
       
   211     default  : ShouldNotReachHere();
       
   212   }
       
   213 
       
   214   // Load LcpoolCache @@@ should be already set!
       
   215   __ get_constant_pool_cache(R27_constPoolCache);
       
   216 
       
   217   // Handle a pending exception, fall through if none.
       
   218   __ check_and_forward_exception(R11_scratch1, R12_scratch2);
       
   219 
       
   220   // Start executing bytecodes.
       
   221   __ dispatch_next(state, step);
       
   222 
       
   223   return entry;
       
   224 }
       
   225 
       
   226 // A result handler converts the native result into java format.
       
   227 // Use the shared code between c++ and template interpreter.
       
   228 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
       
   229   return AbstractInterpreterGenerator::generate_result_handler_for(type);
       
   230 }
       
   231 
       
   232 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
       
   233   address entry = __ pc();
       
   234 
       
   235   __ push(state);
       
   236   __ call_VM(noreg, runtime_entry);
       
   237   __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
       
   238 
       
   239   return entry;
       
   240 }
       
   241 
       
   242 // Helpers for commoning out cases in the various type of method entries.
       
   243 
       
   244 // Increment invocation count & check for overflow.
       
   245 //
       
   246 // Note: checking for negative value instead of overflow
       
   247 //       so we have a 'sticky' overflow test.
       
   248 //
       
   249 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
       
   250   // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not.
       
   251   Register Rscratch1   = R11_scratch1;
       
   252   Register Rscratch2   = R12_scratch2;
       
   253   Register R3_counters = R3_ARG1;
       
   254   Label done;
       
   255 
       
   256   if (TieredCompilation) {
       
   257     const int increment = InvocationCounter::count_increment;
       
   258     const int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
       
   259     Label no_mdo;
       
   260     if (ProfileInterpreter) {
       
   261       const Register Rmdo = Rscratch1;
       
   262       // If no method data exists, go to profile_continue.
       
   263       __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method);
       
   264       __ cmpdi(CCR0, Rmdo, 0);
       
   265       __ beq(CCR0, no_mdo);
       
   266 
       
   267       // Increment backedge counter in the MDO.
       
   268       const int mdo_bc_offs = in_bytes(MethodData::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
       
   269       __ lwz(Rscratch2, mdo_bc_offs, Rmdo);
       
   270       __ addi(Rscratch2, Rscratch2, increment);
       
   271       __ stw(Rscratch2, mdo_bc_offs, Rmdo);
       
   272       __ load_const_optimized(Rscratch1, mask, R0);
       
   273       __ and_(Rscratch1, Rscratch2, Rscratch1);
       
   274       __ bne(CCR0, done);
       
   275       __ b(*overflow);
       
   276     }
       
   277 
       
   278     // Increment counter in MethodCounters*.
       
   279     const int mo_bc_offs = in_bytes(MethodCounters::backedge_counter_offset()) + in_bytes(InvocationCounter::counter_offset());
       
   280     __ bind(no_mdo);
       
   281     __ get_method_counters(R19_method, R3_counters, done);
       
   282     __ lwz(Rscratch2, mo_bc_offs, R3_counters);
       
   283     __ addi(Rscratch2, Rscratch2, increment);
       
   284     __ stw(Rscratch2, mo_bc_offs, R3_counters);
       
   285     __ load_const_optimized(Rscratch1, mask, R0);
       
   286     __ and_(Rscratch1, Rscratch2, Rscratch1);
       
   287     __ beq(CCR0, *overflow);
       
   288 
       
   289     __ bind(done);
       
   290 
       
   291   } else {
       
   292 
       
   293     // Update standard invocation counters.
       
   294     Register Rsum_ivc_bec = R4_ARG2;
       
   295     __ get_method_counters(R19_method, R3_counters, done);
       
   296     __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2);
       
   297     // Increment interpreter invocation counter.
       
   298     if (ProfileInterpreter) {  // %%% Merge this into methodDataOop.
       
   299       __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
       
   300       __ addi(R12_scratch2, R12_scratch2, 1);
       
   301       __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters);
       
   302     }
       
   303     // Check if we must create a method data obj.
       
   304     if (ProfileInterpreter && profile_method != NULL) {
       
   305       const Register profile_limit = Rscratch1;
       
   306       int pl_offs = __ load_const_optimized(profile_limit, &InvocationCounter::InterpreterProfileLimit, R0, true);
       
   307       __ lwz(profile_limit, pl_offs, profile_limit);
       
   308       // Test to see if we should create a method data oop.
       
   309       __ cmpw(CCR0, Rsum_ivc_bec, profile_limit);
       
   310       __ blt(CCR0, *profile_method_continue);
       
   311       // If no method data exists, go to profile_method.
       
   312       __ test_method_data_pointer(*profile_method);
       
   313     }
       
   314     // Finally check for counter overflow.
       
   315     if (overflow) {
       
   316       const Register invocation_limit = Rscratch1;
       
   317       int il_offs = __ load_const_optimized(invocation_limit, &InvocationCounter::InterpreterInvocationLimit, R0, true);
       
   318       __ lwz(invocation_limit, il_offs, invocation_limit);
       
   319       assert(4 == sizeof(InvocationCounter::InterpreterInvocationLimit), "unexpected field size");
       
   320       __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit);
       
   321       __ bge(CCR0, *overflow);
       
   322     }
       
   323 
       
   324     __ bind(done);
       
   325   }
       
   326 }
       
   327 
       
   328 // Generate code to initiate compilation on invocation counter overflow.
       
   329 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) {
       
   330   // Generate code to initiate compilation on the counter overflow.
       
   331 
       
   332   // InterpreterRuntime::frequency_counter_overflow takes one arguments,
       
   333   // which indicates if the counter overflow occurs at a backwards branch (NULL bcp)
       
   334   // We pass zero in.
       
   335   // The call returns the address of the verified entry point for the method or NULL
       
   336   // if the compilation did not complete (either went background or bailed out).
       
   337   //
       
   338   // Unlike the C++ interpreter above: Check exceptions!
       
   339   // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed
       
   340   // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur.
       
   341 
       
   342   __ li(R4_ARG2, 0);
       
   343   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true);
       
   344 
       
   345   // Returns verified_entry_point or NULL.
       
   346   // We ignore it in any case.
       
   347   __ b(continue_entry);
       
   348 }
       
   349 
       
   350 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) {
       
   351   assert_different_registers(Rmem_frame_size, Rscratch1);
       
   352   __ generate_stack_overflow_check_with_compare_and_throw(Rmem_frame_size, Rscratch1);
       
   353 }
       
   354 
       
   355 void TemplateInterpreterGenerator::unlock_method(bool check_exceptions) {
       
   356   __ unlock_object(R26_monitor, check_exceptions);
       
   357 }
       
   358 
       
   359 // Lock the current method, interpreter register window must be set up!
       
   360 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) {
       
   361   const Register Robj_to_lock = Rscratch2;
       
   362 
       
   363   {
       
   364     if (!flags_preloaded) {
       
   365       __ lwz(Rflags, method_(access_flags));
       
   366     }
       
   367 
       
   368 #ifdef ASSERT
       
   369     // Check if methods needs synchronization.
       
   370     {
       
   371       Label Lok;
       
   372       __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT);
       
   373       __ btrue(CCR0,Lok);
       
   374       __ stop("method doesn't need synchronization");
       
   375       __ bind(Lok);
       
   376     }
       
   377 #endif // ASSERT
       
   378   }
       
   379 
       
   380   // Get synchronization object to Rscratch2.
       
   381   {
       
   382     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
       
   383     Label Lstatic;
       
   384     Label Ldone;
       
   385 
       
   386     __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT);
       
   387     __ btrue(CCR0, Lstatic);
       
   388 
       
   389     // Non-static case: load receiver obj from stack and we're done.
       
   390     __ ld(Robj_to_lock, R18_locals);
       
   391     __ b(Ldone);
       
   392 
       
   393     __ bind(Lstatic); // Static case: Lock the java mirror
       
   394     __ ld(Robj_to_lock, in_bytes(Method::const_offset()), R19_method);
       
   395     __ ld(Robj_to_lock, in_bytes(ConstMethod::constants_offset()), Robj_to_lock);
       
   396     __ ld(Robj_to_lock, ConstantPool::pool_holder_offset_in_bytes(), Robj_to_lock);
       
   397     __ ld(Robj_to_lock, mirror_offset, Robj_to_lock);
       
   398 
       
   399     __ bind(Ldone);
       
   400     __ verify_oop(Robj_to_lock);
       
   401   }
       
   402 
       
   403   // Got the oop to lock => execute!
       
   404   __ add_monitor_to_stack(true, Rscratch1, R0);
       
   405 
       
   406   __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor);
       
   407   __ lock_object(R26_monitor, Robj_to_lock);
       
   408 }
       
   409 
       
   410 // Generate a fixed interpreter frame for pure interpreter
       
   411 // and I2N native transition frames.
       
   412 //
       
   413 // Before (stack grows downwards):
       
   414 //
       
   415 //         |  ...         |
       
   416 //         |------------- |
       
   417 //         |  java arg0   |
       
   418 //         |  ...         |
       
   419 //         |  java argn   |
       
   420 //         |              |   <-   R15_esp
       
   421 //         |              |
       
   422 //         |--------------|
       
   423 //         | abi_112      |
       
   424 //         |              |   <-   R1_SP
       
   425 //         |==============|
       
   426 //
       
   427 //
       
   428 // After:
       
   429 //
       
   430 //         |  ...         |
       
   431 //         |  java arg0   |<-   R18_locals
       
   432 //         |  ...         |
       
   433 //         |  java argn   |
       
   434 //         |--------------|
       
   435 //         |              |
       
   436 //         |  java locals |
       
   437 //         |              |
       
   438 //         |--------------|
       
   439 //         |  abi_48      |
       
   440 //         |==============|
       
   441 //         |              |
       
   442 //         |   istate     |
       
   443 //         |              |
       
   444 //         |--------------|
       
   445 //         |   monitor    |<-   R26_monitor
       
   446 //         |--------------|
       
   447 //         |              |<-   R15_esp
       
   448 //         | expression   |
       
   449 //         | stack        |
       
   450 //         |              |
       
   451 //         |--------------|
       
   452 //         |              |
       
   453 //         | abi_112      |<-   R1_SP
       
   454 //         |==============|
       
   455 //
       
   456 // The top most frame needs an abi space of 112 bytes. This space is needed,
       
   457 // since we call to c. The c function may spill their arguments to the caller
       
   458 // frame. When we call to java, we don't need these spill slots. In order to save
       
   459 // space on the stack, we resize the caller. However, java local reside in
       
   460 // the caller frame and the frame has to be increased. The frame_size for the
       
   461 // current frame was calculated based on max_stack as size for the expression
       
   462 // stack. At the call, just a part of the expression stack might be used.
       
   463 // We don't want to waste this space and cut the frame back accordingly.
       
   464 // The resulting amount for resizing is calculated as follows:
       
   465 // resize =   (number_of_locals - number_of_arguments) * slot_size
       
   466 //          + (R1_SP - R15_esp) + 48
       
   467 //
       
   468 // The size for the callee frame is calculated:
       
   469 // framesize = 112 + max_stack + monitor + state_size
       
   470 //
       
   471 // maxstack:   Max number of slots on the expression stack, loaded from the method.
       
   472 // monitor:    We statically reserve room for one monitor object.
       
   473 // state_size: We save the current state of the interpreter to this area.
       
   474 //
       
   475 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) {
       
   476   Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes.
       
   477            top_frame_size      = R7_ARG5,
       
   478            Rconst_method       = R8_ARG6;
       
   479 
       
   480   assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size);
       
   481 
       
   482   __ ld(Rconst_method, method_(const));
       
   483   __ lhz(Rsize_of_parameters /* number of params */,
       
   484          in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method);
       
   485   if (native_call) {
       
   486     // If we're calling a native method, we reserve space for the worst-case signature
       
   487     // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2).
       
   488     // We add two slots to the parameter_count, one for the jni
       
   489     // environment and one for a possible native mirror.
       
   490     Label skip_native_calculate_max_stack;
       
   491     __ addi(top_frame_size, Rsize_of_parameters, 2);
       
   492     __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters);
       
   493     __ bge(CCR0, skip_native_calculate_max_stack);
       
   494     __ li(top_frame_size, Argument::n_register_parameters);
       
   495     __ bind(skip_native_calculate_max_stack);
       
   496     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
       
   497     __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
       
   498     __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
       
   499     assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters.
       
   500   } else {
       
   501     __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method);
       
   502     __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize);
       
   503     __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize);
       
   504     __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method);
       
   505     __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0
       
   506     __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize!
       
   507     __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize);
       
   508     __ add(parent_frame_resize, parent_frame_resize, R11_scratch1);
       
   509   }
       
   510 
       
   511   // Compute top frame size.
       
   512   __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size);
       
   513 
       
   514   // Cut back area between esp and max_stack.
       
   515   __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize);
       
   516 
       
   517   __ round_to(top_frame_size, frame::alignment_in_bytes);
       
   518   __ round_to(parent_frame_resize, frame::alignment_in_bytes);
       
   519   // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size.
       
   520   // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48.
       
   521 
       
   522   {
       
   523     // --------------------------------------------------------------------------
       
   524     // Stack overflow check
       
   525 
       
   526     Label cont;
       
   527     __ add(R11_scratch1, parent_frame_resize, top_frame_size);
       
   528     generate_stack_overflow_check(R11_scratch1, R12_scratch2);
       
   529   }
       
   530 
       
   531   // Set up interpreter state registers.
       
   532 
       
   533   __ add(R18_locals, R15_esp, Rsize_of_parameters);
       
   534   __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
       
   535   __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache);
       
   536 
       
   537   // Set method data pointer.
       
   538   if (ProfileInterpreter) {
       
   539     Label zero_continue;
       
   540     __ ld(R28_mdx, method_(method_data));
       
   541     __ cmpdi(CCR0, R28_mdx, 0);
       
   542     __ beq(CCR0, zero_continue);
       
   543     __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset()));
       
   544     __ bind(zero_continue);
       
   545   }
       
   546 
       
   547   if (native_call) {
       
   548     __ li(R14_bcp, 0); // Must initialize.
       
   549   } else {
       
   550     __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method);
       
   551   }
       
   552 
       
   553   // Resize parent frame.
       
   554   __ mflr(R12_scratch2);
       
   555   __ neg(parent_frame_resize, parent_frame_resize);
       
   556   __ resize_frame(parent_frame_resize, R11_scratch1);
       
   557   __ std(R12_scratch2, _abi(lr), R1_SP);
       
   558 
       
   559   __ addi(R26_monitor, R1_SP, - frame::ijava_state_size);
       
   560   __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize);
       
   561 
       
   562   // Store values.
       
   563   // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls
       
   564   // in InterpreterMacroAssembler::call_from_interpreter.
       
   565   __ std(R19_method, _ijava_state_neg(method), R1_SP);
       
   566   __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP);
       
   567   __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP);
       
   568   __ std(R18_locals, _ijava_state_neg(locals), R1_SP);
       
   569 
       
   570   // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only
       
   571   // be found in the frame after save_interpreter_state is done. This is always true
       
   572   // for non-top frames. But when a signal occurs, dumping the top frame can go wrong,
       
   573   // because e.g. frame::interpreter_frame_bcp() will not access the correct value
       
   574   // (Enhanced Stack Trace).
       
   575   // The signal handler does not save the interpreter state into the frame.
       
   576   __ li(R0, 0);
       
   577 #ifdef ASSERT
       
   578   // Fill remaining slots with constants.
       
   579   __ load_const_optimized(R11_scratch1, 0x5afe);
       
   580   __ load_const_optimized(R12_scratch2, 0xdead);
       
   581 #endif
       
   582   // We have to initialize some frame slots for native calls (accessed by GC).
       
   583   if (native_call) {
       
   584     __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP);
       
   585     __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP);
       
   586     if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); }
       
   587   }
       
   588 #ifdef ASSERT
       
   589   else {
       
   590     __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP);
       
   591     __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP);
       
   592     __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP);
       
   593   }
       
   594   __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP);
       
   595   __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP);
       
   596   __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP);
       
   597   __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP);
       
   598 #endif
       
   599   __ subf(R12_scratch2, top_frame_size, R1_SP);
       
   600   __ std(R0, _ijava_state_neg(oop_tmp), R1_SP);
       
   601   __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP);
       
   602 
       
   603   // Push top frame.
       
   604   __ push_frame(top_frame_size, R11_scratch1);
       
   605 }
       
   606 
       
   607 // End of helpers
       
   608 
       
   609 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
       
   610   if (!TemplateInterpreter::math_entry_available(kind)) {
       
   611     NOT_PRODUCT(__ should_not_reach_here();)
       
   612     return NULL;
       
   613   }
       
   614 
       
   615   address entry = __ pc();
       
   616 
       
   617   __ lfd(F1_RET, Interpreter::stackElementSize, R15_esp);
       
   618 
       
   619   // Pop c2i arguments (if any) off when we return.
       
   620 #ifdef ASSERT
       
   621   __ ld(R9_ARG7, 0, R1_SP);
       
   622   __ ld(R10_ARG8, 0, R21_sender_SP);
       
   623   __ cmpd(CCR0, R9_ARG7, R10_ARG8);
       
   624   __ asm_assert_eq("backlink", 0x545);
       
   625 #endif // ASSERT
       
   626   __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
       
   627 
       
   628   if (kind == Interpreter::java_lang_math_sqrt) {
       
   629     __ fsqrt(F1_RET, F1_RET);
       
   630   } else if (kind == Interpreter::java_lang_math_abs) {
       
   631     __ fabs(F1_RET, F1_RET);
       
   632   } else {
       
   633     ShouldNotReachHere();
       
   634   }
       
   635 
       
   636   // And we're done.
       
   637   __ blr();
       
   638 
       
   639   __ flush();
       
   640 
       
   641   return entry;
       
   642 }
       
   643 
       
   644 // Interpreter stub for calling a native method. (asm interpreter)
       
   645 // This sets up a somewhat different looking stack for calling the
       
   646 // native method than the typical interpreter frame setup.
       
   647 //
       
   648 // On entry:
       
   649 //   R19_method    - method
       
   650 //   R16_thread    - JavaThread*
       
   651 //   R15_esp       - intptr_t* sender tos
       
   652 //
       
   653 //   abstract stack (grows up)
       
   654 //     [  IJava (caller of JNI callee)  ]  <-- ASP
       
   655 //        ...
       
   656 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
       
   657 
       
   658   address entry = __ pc();
       
   659 
       
   660   const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
       
   661 
       
   662   // -----------------------------------------------------------------------------
       
   663   // Allocate a new frame that represents the native callee (i2n frame).
       
   664   // This is not a full-blown interpreter frame, but in particular, the
       
   665   // following registers are valid after this:
       
   666   // - R19_method
       
   667   // - R18_local (points to start of argumuments to native function)
       
   668   //
       
   669   //   abstract stack (grows up)
       
   670   //     [  IJava (caller of JNI callee)  ]  <-- ASP
       
   671   //        ...
       
   672 
       
   673   const Register signature_handler_fd = R11_scratch1;
       
   674   const Register pending_exception    = R0;
       
   675   const Register result_handler_addr  = R31;
       
   676   const Register native_method_fd     = R11_scratch1;
       
   677   const Register access_flags         = R22_tmp2;
       
   678   const Register active_handles       = R11_scratch1; // R26_monitor saved to state.
       
   679   const Register sync_state           = R12_scratch2;
       
   680   const Register sync_state_addr      = sync_state;   // Address is dead after use.
       
   681   const Register suspend_flags        = R11_scratch1;
       
   682 
       
   683   //=============================================================================
       
   684   // Allocate new frame and initialize interpreter state.
       
   685 
       
   686   Label exception_return;
       
   687   Label exception_return_sync_check;
       
   688   Label stack_overflow_return;
       
   689 
       
   690   // Generate new interpreter state and jump to stack_overflow_return in case of
       
   691   // a stack overflow.
       
   692   //generate_compute_interpreter_state(stack_overflow_return);
       
   693 
       
   694   Register size_of_parameters = R22_tmp2;
       
   695 
       
   696   generate_fixed_frame(true, size_of_parameters, noreg /* unused */);
       
   697 
       
   698   //=============================================================================
       
   699   // Increment invocation counter. On overflow, entry to JNI method
       
   700   // will be compiled.
       
   701   Label invocation_counter_overflow, continue_after_compile;
       
   702   if (inc_counter) {
       
   703     if (synchronized) {
       
   704       // Since at this point in the method invocation the exception handler
       
   705       // would try to exit the monitor of synchronized methods which hasn't
       
   706       // been entered yet, we set the thread local variable
       
   707       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
       
   708       // runtime, exception handling i.e. unlock_if_synchronized_method will
       
   709       // check this thread local flag.
       
   710       // This flag has two effects, one is to force an unwind in the topmost
       
   711       // interpreter frame and not perform an unlock while doing so.
       
   712       __ li(R0, 1);
       
   713       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
       
   714     }
       
   715     generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
       
   716 
       
   717     BIND(continue_after_compile);
       
   718     // Reset the _do_not_unlock_if_synchronized flag.
       
   719     if (synchronized) {
       
   720       __ li(R0, 0);
       
   721       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
       
   722     }
       
   723   }
       
   724 
       
   725   // access_flags = method->access_flags();
       
   726   // Load access flags.
       
   727   assert(access_flags->is_nonvolatile(),
       
   728          "access_flags must be in a non-volatile register");
       
   729   // Type check.
       
   730   assert(4 == sizeof(AccessFlags), "unexpected field size");
       
   731   __ lwz(access_flags, method_(access_flags));
       
   732 
       
   733   // We don't want to reload R19_method and access_flags after calls
       
   734   // to some helper functions.
       
   735   assert(R19_method->is_nonvolatile(),
       
   736          "R19_method must be a non-volatile register");
       
   737 
       
   738   // Check for synchronized methods. Must happen AFTER invocation counter
       
   739   // check, so method is not locked if counter overflows.
       
   740 
       
   741   if (synchronized) {
       
   742     lock_method(access_flags, R11_scratch1, R12_scratch2, true);
       
   743 
       
   744     // Update monitor in state.
       
   745     __ ld(R11_scratch1, 0, R1_SP);
       
   746     __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1);
       
   747   }
       
   748 
       
   749   // jvmti/jvmpi support
       
   750   __ notify_method_entry();
       
   751 
       
   752   //=============================================================================
       
   753   // Get and call the signature handler.
       
   754 
       
   755   __ ld(signature_handler_fd, method_(signature_handler));
       
   756   Label call_signature_handler;
       
   757 
       
   758   __ cmpdi(CCR0, signature_handler_fd, 0);
       
   759   __ bne(CCR0, call_signature_handler);
       
   760 
       
   761   // Method has never been called. Either generate a specialized
       
   762   // handler or point to the slow one.
       
   763   //
       
   764   // Pass parameter 'false' to avoid exception check in call_VM.
       
   765   __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false);
       
   766 
       
   767   // Check for an exception while looking up the target method. If we
       
   768   // incurred one, bail.
       
   769   __ ld(pending_exception, thread_(pending_exception));
       
   770   __ cmpdi(CCR0, pending_exception, 0);
       
   771   __ bne(CCR0, exception_return_sync_check); // Has pending exception.
       
   772 
       
   773   // Reload signature handler, it may have been created/assigned in the meanwhile.
       
   774   __ ld(signature_handler_fd, method_(signature_handler));
       
   775   __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below).
       
   776 
       
   777   BIND(call_signature_handler);
       
   778 
       
   779   // Before we call the signature handler we push a new frame to
       
   780   // protect the interpreter frame volatile registers when we return
       
   781   // from jni but before we can get back to Java.
       
   782 
       
   783   // First set the frame anchor while the SP/FP registers are
       
   784   // convenient and the slow signature handler can use this same frame
       
   785   // anchor.
       
   786 
       
   787   // We have a TOP_IJAVA_FRAME here, which belongs to us.
       
   788   __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/);
       
   789 
       
   790   // Now the interpreter frame (and its call chain) have been
       
   791   // invalidated and flushed. We are now protected against eager
       
   792   // being enabled in native code. Even if it goes eager the
       
   793   // registers will be reloaded as clean and we will invalidate after
       
   794   // the call so no spurious flush should be possible.
       
   795 
       
   796   // Call signature handler and pass locals address.
       
   797   //
       
   798   // Our signature handlers copy required arguments to the C stack
       
   799   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
       
   800   __ mr(R3_ARG1, R18_locals);
       
   801 #if !defined(ABI_ELFv2)
       
   802   __ ld(signature_handler_fd, 0, signature_handler_fd);
       
   803 #endif
       
   804 
       
   805   __ call_stub(signature_handler_fd);
       
   806 
       
   807   // Remove the register parameter varargs slots we allocated in
       
   808   // compute_interpreter_state. SP+16 ends up pointing to the ABI
       
   809   // outgoing argument area.
       
   810   //
       
   811   // Not needed on PPC64.
       
   812   //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord);
       
   813 
       
   814   assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register");
       
   815   // Save across call to native method.
       
   816   __ mr(result_handler_addr, R3_RET);
       
   817 
       
   818   __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror.
       
   819 
       
   820   // Set up fixed parameters and call the native method.
       
   821   // If the method is static, get mirror into R4_ARG2.
       
   822   {
       
   823     Label method_is_not_static;
       
   824     // Access_flags is non-volatile and still, no need to restore it.
       
   825 
       
   826     // Restore access flags.
       
   827     __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT);
       
   828     __ bfalse(CCR0, method_is_not_static);
       
   829 
       
   830     // constants = method->constants();
       
   831     __ ld(R11_scratch1, in_bytes(Method::const_offset()), R19_method);
       
   832     __ ld(R11_scratch1, in_bytes(ConstMethod::constants_offset()), R11_scratch1);
       
   833     // pool_holder = method->constants()->pool_holder();
       
   834     __ ld(R11_scratch1/*pool_holder*/, ConstantPool::pool_holder_offset_in_bytes(),
       
   835           R11_scratch1/*constants*/);
       
   836 
       
   837     const int mirror_offset = in_bytes(Klass::java_mirror_offset());
       
   838 
       
   839     // mirror = pool_holder->klass_part()->java_mirror();
       
   840     __ ld(R0/*mirror*/, mirror_offset, R11_scratch1/*pool_holder*/);
       
   841     // state->_native_mirror = mirror;
       
   842 
       
   843     __ ld(R11_scratch1, 0, R1_SP);
       
   844     __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1);
       
   845     // R4_ARG2 = &state->_oop_temp;
       
   846     __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp));
       
   847     BIND(method_is_not_static);
       
   848   }
       
   849 
       
   850   // At this point, arguments have been copied off the stack into
       
   851   // their JNI positions. Oops are boxed in-place on the stack, with
       
   852   // handles copied to arguments. The result handler address is in a
       
   853   // register.
       
   854 
       
   855   // Pass JNIEnv address as first parameter.
       
   856   __ addir(R3_ARG1, thread_(jni_environment));
       
   857 
       
   858   // Load the native_method entry before we change the thread state.
       
   859   __ ld(native_method_fd, method_(native_function));
       
   860 
       
   861   //=============================================================================
       
   862   // Transition from _thread_in_Java to _thread_in_native. As soon as
       
   863   // we make this change the safepoint code needs to be certain that
       
   864   // the last Java frame we established is good. The pc in that frame
       
   865   // just needs to be near here not an actual return address.
       
   866 
       
   867   // We use release_store_fence to update values like the thread state, where
       
   868   // we don't want the current thread to continue until all our prior memory
       
   869   // accesses (including the new thread state) are visible to other threads.
       
   870   __ li(R0, _thread_in_native);
       
   871   __ release();
       
   872 
       
   873   // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size");
       
   874   __ stw(R0, thread_(thread_state));
       
   875 
       
   876   if (UseMembar) {
       
   877     __ fence();
       
   878   }
       
   879 
       
   880   //=============================================================================
       
   881   // Call the native method. Argument registers must not have been
       
   882   // overwritten since "__ call_stub(signature_handler);" (except for
       
   883   // ARG1 and ARG2 for static methods).
       
   884   __ call_c(native_method_fd);
       
   885 
       
   886   __ li(R0, 0);
       
   887   __ ld(R11_scratch1, 0, R1_SP);
       
   888   __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
       
   889   __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
       
   890   __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset
       
   891 
       
   892   // Note: C++ interpreter needs the following here:
       
   893   // The frame_manager_lr field, which we use for setting the last
       
   894   // java frame, gets overwritten by the signature handler. Restore
       
   895   // it now.
       
   896   //__ get_PC_trash_LR(R11_scratch1);
       
   897   //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP);
       
   898 
       
   899   // Because of GC R19_method may no longer be valid.
       
   900 
       
   901   // Block, if necessary, before resuming in _thread_in_Java state.
       
   902   // In order for GC to work, don't clear the last_Java_sp until after
       
   903   // blocking.
       
   904 
       
   905   //=============================================================================
       
   906   // Switch thread to "native transition" state before reading the
       
   907   // synchronization state. This additional state is necessary
       
   908   // because reading and testing the synchronization state is not
       
   909   // atomic w.r.t. GC, as this scenario demonstrates: Java thread A,
       
   910   // in _thread_in_native state, loads _not_synchronized and is
       
   911   // preempted. VM thread changes sync state to synchronizing and
       
   912   // suspends threads for GC. Thread A is resumed to finish this
       
   913   // native method, but doesn't block here since it didn't see any
       
   914   // synchronization in progress, and escapes.
       
   915 
       
   916   // We use release_store_fence to update values like the thread state, where
       
   917   // we don't want the current thread to continue until all our prior memory
       
   918   // accesses (including the new thread state) are visible to other threads.
       
   919   __ li(R0/*thread_state*/, _thread_in_native_trans);
       
   920   __ release();
       
   921   __ stw(R0/*thread_state*/, thread_(thread_state));
       
   922   if (UseMembar) {
       
   923     __ fence();
       
   924   }
       
   925   // Write serialization page so that the VM thread can do a pseudo remote
       
   926   // membar. We use the current thread pointer to calculate a thread
       
   927   // specific offset to write to within the page. This minimizes bus
       
   928   // traffic due to cache line collision.
       
   929   else {
       
   930     __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2);
       
   931   }
       
   932 
       
   933   // Now before we return to java we must look for a current safepoint
       
   934   // (a new safepoint can not start since we entered native_trans).
       
   935   // We must check here because a current safepoint could be modifying
       
   936   // the callers registers right this moment.
       
   937 
       
   938   // Acquire isn't strictly necessary here because of the fence, but
       
   939   // sync_state is declared to be volatile, so we do it anyway
       
   940   // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path).
       
   941   int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
       
   942 
       
   943   // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size");
       
   944   __ lwz(sync_state, sync_state_offs, sync_state_addr);
       
   945 
       
   946   // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size");
       
   947   __ lwz(suspend_flags, thread_(suspend_flags));
       
   948 
       
   949   Label sync_check_done;
       
   950   Label do_safepoint;
       
   951   // No synchronization in progress nor yet synchronized.
       
   952   __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
       
   953   // Not suspended.
       
   954   __ cmpwi(CCR1, suspend_flags, 0);
       
   955 
       
   956   __ bne(CCR0, do_safepoint);
       
   957   __ beq(CCR1, sync_check_done);
       
   958   __ bind(do_safepoint);
       
   959   __ isync();
       
   960   // Block. We do the call directly and leave the current
       
   961   // last_Java_frame setup undisturbed. We must save any possible
       
   962   // native result across the call. No oop is present.
       
   963 
       
   964   __ mr(R3_ARG1, R16_thread);
       
   965 #if defined(ABI_ELFv2)
       
   966   __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
       
   967             relocInfo::none);
       
   968 #else
       
   969   __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
       
   970             relocInfo::none);
       
   971 #endif
       
   972 
       
   973   __ bind(sync_check_done);
       
   974 
       
   975   //=============================================================================
       
   976   // <<<<<< Back in Interpreter Frame >>>>>
       
   977 
       
   978   // We are in thread_in_native_trans here and back in the normal
       
   979   // interpreter frame. We don't have to do anything special about
       
   980   // safepoints and we can switch to Java mode anytime we are ready.
       
   981 
       
   982   // Note: frame::interpreter_frame_result has a dependency on how the
       
   983   // method result is saved across the call to post_method_exit. For
       
   984   // native methods it assumes that the non-FPU/non-void result is
       
   985   // saved in _native_lresult and a FPU result in _native_fresult. If
       
   986   // this changes then the interpreter_frame_result implementation
       
   987   // will need to be updated too.
       
   988 
       
   989   // On PPC64, we have stored the result directly after the native call.
       
   990 
       
   991   //=============================================================================
       
   992   // Back in Java
       
   993 
       
   994   // We use release_store_fence to update values like the thread state, where
       
   995   // we don't want the current thread to continue until all our prior memory
       
   996   // accesses (including the new thread state) are visible to other threads.
       
   997   __ li(R0/*thread_state*/, _thread_in_Java);
       
   998   __ release();
       
   999   __ stw(R0/*thread_state*/, thread_(thread_state));
       
  1000   if (UseMembar) {
       
  1001     __ fence();
       
  1002   }
       
  1003 
       
  1004   __ reset_last_Java_frame();
       
  1005 
       
  1006   // Jvmdi/jvmpi support. Whether we've got an exception pending or
       
  1007   // not, and whether unlocking throws an exception or not, we notify
       
  1008   // on native method exit. If we do have an exception, we'll end up
       
  1009   // in the caller's context to handle it, so if we don't do the
       
  1010   // notify here, we'll drop it on the floor.
       
  1011   __ notify_method_exit(true/*native method*/,
       
  1012                         ilgl /*illegal state (not used for native methods)*/,
       
  1013                         InterpreterMacroAssembler::NotifyJVMTI,
       
  1014                         false /*check_exceptions*/);
       
  1015 
       
  1016   //=============================================================================
       
  1017   // Handle exceptions
       
  1018 
       
  1019   if (synchronized) {
       
  1020     // Don't check for exceptions since we're still in the i2n frame. Do that
       
  1021     // manually afterwards.
       
  1022     unlock_method(false);
       
  1023   }
       
  1024 
       
  1025   // Reset active handles after returning from native.
       
  1026   // thread->active_handles()->clear();
       
  1027   __ ld(active_handles, thread_(active_handles));
       
  1028   // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size");
       
  1029   __ li(R0, 0);
       
  1030   __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles);
       
  1031 
       
  1032   Label exception_return_sync_check_already_unlocked;
       
  1033   __ ld(R0/*pending_exception*/, thread_(pending_exception));
       
  1034   __ cmpdi(CCR0, R0/*pending_exception*/, 0);
       
  1035   __ bne(CCR0, exception_return_sync_check_already_unlocked);
       
  1036 
       
  1037   //-----------------------------------------------------------------------------
       
  1038   // No exception pending.
       
  1039 
       
  1040   // Move native method result back into proper registers and return.
       
  1041   // Invoke result handler (may unbox/promote).
       
  1042   __ ld(R11_scratch1, 0, R1_SP);
       
  1043   __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1);
       
  1044   __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1);
       
  1045   __ call_stub(result_handler_addr);
       
  1046 
       
  1047   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2);
       
  1048 
       
  1049   // Must use the return pc which was loaded from the caller's frame
       
  1050   // as the VM uses return-pc-patching for deoptimization.
       
  1051   __ mtlr(R0);
       
  1052   __ blr();
       
  1053 
       
  1054   //-----------------------------------------------------------------------------
       
  1055   // An exception is pending. We call into the runtime only if the
       
  1056   // caller was not interpreted. If it was interpreted the
       
  1057   // interpreter will do the correct thing. If it isn't interpreted
       
  1058   // (call stub/compiled code) we will change our return and continue.
       
  1059 
       
  1060   BIND(exception_return_sync_check);
       
  1061 
       
  1062   if (synchronized) {
       
  1063     // Don't check for exceptions since we're still in the i2n frame. Do that
       
  1064     // manually afterwards.
       
  1065     unlock_method(false);
       
  1066   }
       
  1067   BIND(exception_return_sync_check_already_unlocked);
       
  1068 
       
  1069   const Register return_pc = R31;
       
  1070 
       
  1071   __ ld(return_pc, 0, R1_SP);
       
  1072   __ ld(return_pc, _abi(lr), return_pc);
       
  1073 
       
  1074   // Get the address of the exception handler.
       
  1075   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
       
  1076                   R16_thread,
       
  1077                   return_pc /* return pc */);
       
  1078   __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2);
       
  1079 
       
  1080   // Load the PC of the the exception handler into LR.
       
  1081   __ mtlr(R3_RET);
       
  1082 
       
  1083   // Load exception into R3_ARG1 and clear pending exception in thread.
       
  1084   __ ld(R3_ARG1/*exception*/, thread_(pending_exception));
       
  1085   __ li(R4_ARG2, 0);
       
  1086   __ std(R4_ARG2, thread_(pending_exception));
       
  1087 
       
  1088   // Load the original return pc into R4_ARG2.
       
  1089   __ mr(R4_ARG2/*issuing_pc*/, return_pc);
       
  1090 
       
  1091   // Return to exception handler.
       
  1092   __ blr();
       
  1093 
       
  1094   //=============================================================================
       
  1095   // Counter overflow.
       
  1096 
       
  1097   if (inc_counter) {
       
  1098     // Handle invocation counter overflow.
       
  1099     __ bind(invocation_counter_overflow);
       
  1100 
       
  1101     generate_counter_overflow(continue_after_compile);
       
  1102   }
       
  1103 
       
  1104   return entry;
       
  1105 }
       
  1106 
       
  1107 // Generic interpreted method entry to (asm) interpreter.
       
  1108 //
       
  1109 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
       
  1110   bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
       
  1111   address entry = __ pc();
       
  1112   // Generate the code to allocate the interpreter stack frame.
       
  1113   Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame.
       
  1114            Rsize_of_locals     = R5_ARG3; // Written by generate_fixed_frame.
       
  1115 
       
  1116   generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals);
       
  1117 
       
  1118   // --------------------------------------------------------------------------
       
  1119   // Zero out non-parameter locals.
       
  1120   // Note: *Always* zero out non-parameter locals as Sparc does. It's not
       
  1121   // worth to ask the flag, just do it.
       
  1122   Register Rslot_addr = R6_ARG4,
       
  1123            Rnum       = R7_ARG5;
       
  1124   Label Lno_locals, Lzero_loop;
       
  1125 
       
  1126   // Set up the zeroing loop.
       
  1127   __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals);
       
  1128   __ subf(Rslot_addr, Rsize_of_parameters, R18_locals);
       
  1129   __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize);
       
  1130   __ beq(CCR0, Lno_locals);
       
  1131   __ li(R0, 0);
       
  1132   __ mtctr(Rnum);
       
  1133 
       
  1134   // The zero locals loop.
       
  1135   __ bind(Lzero_loop);
       
  1136   __ std(R0, 0, Rslot_addr);
       
  1137   __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize);
       
  1138   __ bdnz(Lzero_loop);
       
  1139 
       
  1140   __ bind(Lno_locals);
       
  1141 
       
  1142   // --------------------------------------------------------------------------
       
  1143   // Counter increment and overflow check.
       
  1144   Label invocation_counter_overflow,
       
  1145         profile_method,
       
  1146         profile_method_continue;
       
  1147   if (inc_counter || ProfileInterpreter) {
       
  1148 
       
  1149     Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1;
       
  1150     if (synchronized) {
       
  1151       // Since at this point in the method invocation the exception handler
       
  1152       // would try to exit the monitor of synchronized methods which hasn't
       
  1153       // been entered yet, we set the thread local variable
       
  1154       // _do_not_unlock_if_synchronized to true. If any exception was thrown by
       
  1155       // runtime, exception handling i.e. unlock_if_synchronized_method will
       
  1156       // check this thread local flag.
       
  1157       // This flag has two effects, one is to force an unwind in the topmost
       
  1158       // interpreter frame and not perform an unlock while doing so.
       
  1159       __ li(R0, 1);
       
  1160       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
       
  1161     }
       
  1162 
       
  1163     // Argument and return type profiling.
       
  1164     __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4);
       
  1165 
       
  1166     // Increment invocation counter and check for overflow.
       
  1167     if (inc_counter) {
       
  1168       generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
       
  1169     }
       
  1170 
       
  1171     __ bind(profile_method_continue);
       
  1172 
       
  1173     // Reset the _do_not_unlock_if_synchronized flag.
       
  1174     if (synchronized) {
       
  1175       __ li(R0, 0);
       
  1176       __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread);
       
  1177     }
       
  1178   }
       
  1179 
       
  1180   // --------------------------------------------------------------------------
       
  1181   // Locking of synchronized methods. Must happen AFTER invocation_counter
       
  1182   // check and stack overflow check, so method is not locked if overflows.
       
  1183   if (synchronized) {
       
  1184     lock_method(R3_ARG1, R4_ARG2, R5_ARG3);
       
  1185   }
       
  1186 #ifdef ASSERT
       
  1187   else {
       
  1188     Label Lok;
       
  1189     __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method);
       
  1190     __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED);
       
  1191     __ asm_assert_eq("method needs synchronization", 0x8521);
       
  1192     __ bind(Lok);
       
  1193   }
       
  1194 #endif // ASSERT
       
  1195 
       
  1196   __ verify_thread();
       
  1197 
       
  1198   // --------------------------------------------------------------------------
       
  1199   // JVMTI support
       
  1200   __ notify_method_entry();
       
  1201 
       
  1202   // --------------------------------------------------------------------------
       
  1203   // Start executing instructions.
       
  1204   __ dispatch_next(vtos);
       
  1205 
       
  1206   // --------------------------------------------------------------------------
       
  1207   // Out of line counter overflow and MDO creation code.
       
  1208   if (ProfileInterpreter) {
       
  1209     // We have decided to profile this method in the interpreter.
       
  1210     __ bind(profile_method);
       
  1211     __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
       
  1212     __ set_method_data_pointer_for_bcp();
       
  1213     __ b(profile_method_continue);
       
  1214   }
       
  1215 
       
  1216   if (inc_counter) {
       
  1217     // Handle invocation counter overflow.
       
  1218     __ bind(invocation_counter_overflow);
       
  1219     generate_counter_overflow(profile_method_continue);
       
  1220   }
       
  1221   return entry;
       
  1222 }
       
  1223 
       
  1224 // CRC32 Intrinsics.
       
  1225 //
       
  1226 // Contract on scratch and work registers.
       
  1227 // =======================================
       
  1228 //
       
  1229 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers.
       
  1230 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set.
       
  1231 // You can't rely on these registers across calls.
       
  1232 //
       
  1233 // The generators for CRC32_update and for CRC32_updateBytes use the
       
  1234 // scratch/work register set internally, passing the work registers
       
  1235 // as arguments to the MacroAssembler emitters as required.
       
  1236 //
       
  1237 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments.
       
  1238 // Their contents is not constant but may change according to the requirements
       
  1239 // of the emitted code.
       
  1240 //
       
  1241 // All other registers from the scratch/work register set are used "internally"
       
  1242 // and contain garbage (i.e. unpredictable values) once blr() is reached.
       
  1243 // Basically, only R3_RET contains a defined value which is the function result.
       
  1244 //
       
  1245 /**
       
  1246  * Method entry for static native methods:
       
  1247  *   int java.util.zip.CRC32.update(int crc, int b)
       
  1248  */
       
  1249 address InterpreterGenerator::generate_CRC32_update_entry() {
       
  1250   if (UseCRC32Intrinsics) {
       
  1251     address start = __ pc();  // Remember stub start address (is rtn value).
       
  1252     Label slow_path;
       
  1253 
       
  1254     // Safepoint check
       
  1255     const Register sync_state = R11_scratch1;
       
  1256     int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
       
  1257     __ lwz(sync_state, sync_state_offs, sync_state);
       
  1258     __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
       
  1259     __ bne(CCR0, slow_path);
       
  1260 
       
  1261     // We don't generate local frame and don't align stack because
       
  1262     // we not even call stub code (we generate the code inline)
       
  1263     // and there is no safepoint on this path.
       
  1264 
       
  1265     // Load java parameters.
       
  1266     // R15_esp is callers operand stack pointer, i.e. it points to the parameters.
       
  1267     const Register argP    = R15_esp;
       
  1268     const Register crc     = R3_ARG1;  // crc value
       
  1269     const Register data    = R4_ARG2;  // address of java byte value (kernel_crc32 needs address)
       
  1270     const Register dataLen = R5_ARG3;  // source data len (1 byte). Not used because calling the single-byte emitter.
       
  1271     const Register table   = R6_ARG4;  // address of crc32 table
       
  1272     const Register tmp     = dataLen;  // Reuse unused len register to show we don't actually need a separate tmp here.
       
  1273 
       
  1274     BLOCK_COMMENT("CRC32_update {");
       
  1275 
       
  1276     // Arguments are reversed on java expression stack
       
  1277 #ifdef VM_LITTLE_ENDIAN
       
  1278     __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
       
  1279                                        // Being passed as an int, the single byte is at offset +0.
       
  1280 #else
       
  1281     __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value.
       
  1282                                        // Being passed from java as an int, the single byte is at offset +3.
       
  1283 #endif
       
  1284     __ lwz(crc,  2*wordSize, argP);    // Current crc state, zero extend to 64 bit to have a clean register.
       
  1285 
       
  1286     StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
       
  1287     __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp);
       
  1288 
       
  1289     // Restore caller sp for c2i case and return.
       
  1290     __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
       
  1291     __ blr();
       
  1292 
       
  1293     // Generate a vanilla native entry as the slow path.
       
  1294     BLOCK_COMMENT("} CRC32_update");
       
  1295     BIND(slow_path);
       
  1296     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
       
  1297     return start;
       
  1298   }
       
  1299 
       
  1300   return NULL;
       
  1301 }
       
  1302 
       
  1303 // CRC32 Intrinsics.
       
  1304 /**
       
  1305  * Method entry for static native methods:
       
  1306  *   int java.util.zip.CRC32.updateBytes(     int crc, byte[] b,  int off, int len)
       
  1307  *   int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len)
       
  1308  */
       
  1309 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
       
  1310   if (UseCRC32Intrinsics) {
       
  1311     address start = __ pc();  // Remember stub start address (is rtn value).
       
  1312     Label slow_path;
       
  1313 
       
  1314     // Safepoint check
       
  1315     const Register sync_state = R11_scratch1;
       
  1316     int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true);
       
  1317     __ lwz(sync_state, sync_state_offs, sync_state);
       
  1318     __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized);
       
  1319     __ bne(CCR0, slow_path);
       
  1320 
       
  1321     // We don't generate local frame and don't align stack because
       
  1322     // we not even call stub code (we generate the code inline)
       
  1323     // and there is no safepoint on this path.
       
  1324 
       
  1325     // Load parameters.
       
  1326     // Z_esp is callers operand stack pointer, i.e. it points to the parameters.
       
  1327     const Register argP    = R15_esp;
       
  1328     const Register crc     = R3_ARG1;  // crc value
       
  1329     const Register data    = R4_ARG2;  // address of java byte array
       
  1330     const Register dataLen = R5_ARG3;  // source data len
       
  1331     const Register table   = R6_ARG4;  // address of crc32 table
       
  1332 
       
  1333     const Register t0      = R9;       // scratch registers for crc calculation
       
  1334     const Register t1      = R10;
       
  1335     const Register t2      = R11;
       
  1336     const Register t3      = R12;
       
  1337 
       
  1338     const Register tc0     = R2;       // registers to hold pre-calculated column addresses
       
  1339     const Register tc1     = R7;
       
  1340     const Register tc2     = R8;
       
  1341     const Register tc3     = table;    // table address is reconstructed at the end of kernel_crc32_* emitters
       
  1342 
       
  1343     const Register tmp     = t0;       // Only used very locally to calculate byte buffer address.
       
  1344 
       
  1345     // Arguments are reversed on java expression stack.
       
  1346     // Calculate address of start element.
       
  1347     if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct".
       
  1348       BLOCK_COMMENT("CRC32_updateByteBuffer {");
       
  1349       // crc     @ (SP + 5W) (32bit)
       
  1350       // buf     @ (SP + 3W) (64bit ptr to long array)
       
  1351       // off     @ (SP + 2W) (32bit)
       
  1352       // dataLen @ (SP + 1W) (32bit)
       
  1353       // data = buf + off
       
  1354       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
       
  1355       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
       
  1356       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
       
  1357       __ lwz( crc,     5*wordSize, argP);  // current crc state
       
  1358       __ add( data, data, tmp);            // Add byte buffer offset.
       
  1359     } else {                                                         // Used for "updateBytes update".
       
  1360       BLOCK_COMMENT("CRC32_updateBytes {");
       
  1361       // crc     @ (SP + 4W) (32bit)
       
  1362       // buf     @ (SP + 3W) (64bit ptr to byte array)
       
  1363       // off     @ (SP + 2W) (32bit)
       
  1364       // dataLen @ (SP + 1W) (32bit)
       
  1365       // data = buf + off + base_offset
       
  1366       __ ld(  data,    3*wordSize, argP);  // start of byte buffer
       
  1367       __ lwa( tmp,     2*wordSize, argP);  // byte buffer offset
       
  1368       __ lwa( dataLen, 1*wordSize, argP);  // #bytes to process
       
  1369       __ add( data, data, tmp);            // add byte buffer offset
       
  1370       __ lwz( crc,     4*wordSize, argP);  // current crc state
       
  1371       __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE));
       
  1372     }
       
  1373 
       
  1374     StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
       
  1375 
       
  1376     // Performance measurements show the 1word and 2word variants to be almost equivalent,
       
  1377     // with very light advantages for the 1word variant. We chose the 1word variant for
       
  1378     // code compactness.
       
  1379     __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3);
       
  1380 
       
  1381     // Restore caller sp for c2i case and return.
       
  1382     __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
       
  1383     __ blr();
       
  1384 
       
  1385     // Generate a vanilla native entry as the slow path.
       
  1386     BLOCK_COMMENT("} CRC32_updateBytes(Buffer)");
       
  1387     BIND(slow_path);
       
  1388     __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1);
       
  1389     return start;
       
  1390   }
       
  1391 
       
  1392   return NULL;
       
  1393 }
       
  1394 
       
  1395 // =============================================================================
       
  1396 // Exceptions
       
  1397 
       
  1398 void TemplateInterpreterGenerator::generate_throw_exception() {
       
  1399   Register Rexception    = R17_tos,
       
  1400            Rcontinuation = R3_RET;
       
  1401 
       
  1402   // --------------------------------------------------------------------------
       
  1403   // Entry point if an method returns with a pending exception (rethrow).
       
  1404   Interpreter::_rethrow_exception_entry = __ pc();
       
  1405   {
       
  1406     __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp.
       
  1407     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
       
  1408     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
       
  1409 
       
  1410     // Compiled code destroys templateTableBase, reload.
       
  1411     __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1);
       
  1412   }
       
  1413 
       
  1414   // Entry point if a interpreted method throws an exception (throw).
       
  1415   Interpreter::_throw_exception_entry = __ pc();
       
  1416   {
       
  1417     __ mr(Rexception, R3_RET);
       
  1418 
       
  1419     __ verify_thread();
       
  1420     __ verify_oop(Rexception);
       
  1421 
       
  1422     // Expression stack must be empty before entering the VM in case of an exception.
       
  1423     __ empty_expression_stack();
       
  1424     // Find exception handler address and preserve exception oop.
       
  1425     // Call C routine to find handler and jump to it.
       
  1426     __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception);
       
  1427     __ mtctr(Rcontinuation);
       
  1428     // Push exception for exception handler bytecodes.
       
  1429     __ push_ptr(Rexception);
       
  1430 
       
  1431     // Jump to exception handler (may be remove activation entry!).
       
  1432     __ bctr();
       
  1433   }
       
  1434 
       
  1435   // If the exception is not handled in the current frame the frame is
       
  1436   // removed and the exception is rethrown (i.e. exception
       
  1437   // continuation is _rethrow_exception).
       
  1438   //
       
  1439   // Note: At this point the bci is still the bxi for the instruction
       
  1440   // which caused the exception and the expression stack is
       
  1441   // empty. Thus, for any VM calls at this point, GC will find a legal
       
  1442   // oop map (with empty expression stack).
       
  1443 
       
  1444   // In current activation
       
  1445   // tos: exception
       
  1446   // bcp: exception bcp
       
  1447 
       
  1448   // --------------------------------------------------------------------------
       
  1449   // JVMTI PopFrame support
       
  1450 
       
  1451   Interpreter::_remove_activation_preserving_args_entry = __ pc();
       
  1452   {
       
  1453     // Set the popframe_processing bit in popframe_condition indicating that we are
       
  1454     // currently handling popframe, so that call_VMs that may happen later do not
       
  1455     // trigger new popframe handling cycles.
       
  1456     __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
       
  1457     __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit);
       
  1458     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
       
  1459 
       
  1460     // Empty the expression stack, as in normal exception handling.
       
  1461     __ empty_expression_stack();
       
  1462     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false);
       
  1463 
       
  1464     // Check to see whether we are returning to a deoptimized frame.
       
  1465     // (The PopFrame call ensures that the caller of the popped frame is
       
  1466     // either interpreted or compiled and deoptimizes it if compiled.)
       
  1467     // Note that we don't compare the return PC against the
       
  1468     // deoptimization blob's unpack entry because of the presence of
       
  1469     // adapter frames in C2.
       
  1470     Label Lcaller_not_deoptimized;
       
  1471     Register return_pc = R3_ARG1;
       
  1472     __ ld(return_pc, 0, R1_SP);
       
  1473     __ ld(return_pc, _abi(lr), return_pc);
       
  1474     __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc);
       
  1475     __ cmpdi(CCR0, R3_RET, 0);
       
  1476     __ bne(CCR0, Lcaller_not_deoptimized);
       
  1477 
       
  1478     // The deoptimized case.
       
  1479     // In this case, we can't call dispatch_next() after the frame is
       
  1480     // popped, but instead must save the incoming arguments and restore
       
  1481     // them after deoptimization has occurred.
       
  1482     __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method);
       
  1483     __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2);
       
  1484     __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize);
       
  1485     __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize);
       
  1486     __ subf(R5_ARG3, R4_ARG2, R5_ARG3);
       
  1487     // Save these arguments.
       
  1488     __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3);
       
  1489 
       
  1490     // Inform deoptimization that it is responsible for restoring these arguments.
       
  1491     __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit);
       
  1492     __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
       
  1493 
       
  1494     // Return from the current method into the deoptimization blob. Will eventually
       
  1495     // end up in the deopt interpeter entry, deoptimization prepared everything that
       
  1496     // we will reexecute the call that called us.
       
  1497     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2);
       
  1498     __ mtlr(return_pc);
       
  1499     __ blr();
       
  1500 
       
  1501     // The non-deoptimized case.
       
  1502     __ bind(Lcaller_not_deoptimized);
       
  1503 
       
  1504     // Clear the popframe condition flag.
       
  1505     __ li(R0, 0);
       
  1506     __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread);
       
  1507 
       
  1508     // Get out of the current method and re-execute the call that called us.
       
  1509     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
       
  1510     __ restore_interpreter_state(R11_scratch1);
       
  1511     __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1);
       
  1512     __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0);
       
  1513     if (ProfileInterpreter) {
       
  1514       __ set_method_data_pointer_for_bcp();
       
  1515       __ ld(R11_scratch1, 0, R1_SP);
       
  1516       __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1);
       
  1517     }
       
  1518 #if INCLUDE_JVMTI
       
  1519     Label L_done;
       
  1520 
       
  1521     __ lbz(R11_scratch1, 0, R14_bcp);
       
  1522     __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic);
       
  1523     __ bne(CCR0, L_done);
       
  1524 
       
  1525     // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
       
  1526     // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
       
  1527     __ ld(R4_ARG2, 0, R18_locals);
       
  1528     __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false);
       
  1529     __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true);
       
  1530     __ cmpdi(CCR0, R4_ARG2, 0);
       
  1531     __ beq(CCR0, L_done);
       
  1532     __ std(R4_ARG2, wordSize, R15_esp);
       
  1533     __ bind(L_done);
       
  1534 #endif // INCLUDE_JVMTI
       
  1535     __ dispatch_next(vtos);
       
  1536   }
       
  1537   // end of JVMTI PopFrame support
       
  1538 
       
  1539   // --------------------------------------------------------------------------
       
  1540   // Remove activation exception entry.
       
  1541   // This is jumped to if an interpreted method can't handle an exception itself
       
  1542   // (we come from the throw/rethrow exception entry above). We're going to call
       
  1543   // into the VM to find the exception handler in the caller, pop the current
       
  1544   // frame and return the handler we calculated.
       
  1545   Interpreter::_remove_activation_entry = __ pc();
       
  1546   {
       
  1547     __ pop_ptr(Rexception);
       
  1548     __ verify_thread();
       
  1549     __ verify_oop(Rexception);
       
  1550     __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread);
       
  1551 
       
  1552     __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true);
       
  1553     __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false);
       
  1554 
       
  1555     __ get_vm_result(Rexception);
       
  1556 
       
  1557     // We are done with this activation frame; find out where to go next.
       
  1558     // The continuation point will be an exception handler, which expects
       
  1559     // the following registers set up:
       
  1560     //
       
  1561     // RET:  exception oop
       
  1562     // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled.
       
  1563 
       
  1564     Register return_pc = R31; // Needs to survive the runtime call.
       
  1565     __ ld(return_pc, 0, R1_SP);
       
  1566     __ ld(return_pc, _abi(lr), return_pc);
       
  1567     __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc);
       
  1568 
       
  1569     // Remove the current activation.
       
  1570     __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2);
       
  1571 
       
  1572     __ mr(R4_ARG2, return_pc);
       
  1573     __ mtlr(R3_RET);
       
  1574     __ mr(R3_RET, Rexception);
       
  1575     __ blr();
       
  1576   }
       
  1577 }
       
  1578 
       
  1579 // JVMTI ForceEarlyReturn support.
       
  1580 // Returns "in the middle" of a method with a "fake" return value.
       
  1581 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
       
  1582 
       
  1583   Register Rscratch1 = R11_scratch1,
       
  1584            Rscratch2 = R12_scratch2;
       
  1585 
       
  1586   address entry = __ pc();
       
  1587   __ empty_expression_stack();
       
  1588 
       
  1589   __ load_earlyret_value(state, Rscratch1);
       
  1590 
       
  1591   __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread);
       
  1592   // Clear the earlyret state.
       
  1593   __ li(R0, 0);
       
  1594   __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1);
       
  1595 
       
  1596   __ remove_activation(state, false, false);
       
  1597   // Copied from TemplateTable::_return.
       
  1598   // Restoration of lr done by remove_activation.
       
  1599   switch (state) {
       
  1600     case ltos:
       
  1601     case btos:
       
  1602     case ctos:
       
  1603     case stos:
       
  1604     case atos:
       
  1605     case itos: __ mr(R3_RET, R17_tos); break;
       
  1606     case ftos:
       
  1607     case dtos: __ fmr(F1_RET, F15_ftos); break;
       
  1608     case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need
       
  1609                // to get visible before the reference to the object gets stored anywhere.
       
  1610                __ membar(Assembler::StoreStore); break;
       
  1611     default  : ShouldNotReachHere();
       
  1612   }
       
  1613   __ blr();
       
  1614 
       
  1615   return entry;
       
  1616 } // end of ForceEarlyReturn support
       
  1617 
       
  1618 //-----------------------------------------------------------------------------
       
  1619 // Helper for vtos entry point generation
       
  1620 
       
  1621 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
       
  1622                                                          address& bep,
       
  1623                                                          address& cep,
       
  1624                                                          address& sep,
       
  1625                                                          address& aep,
       
  1626                                                          address& iep,
       
  1627                                                          address& lep,
       
  1628                                                          address& fep,
       
  1629                                                          address& dep,
       
  1630                                                          address& vep) {
       
  1631   assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
       
  1632   Label L;
       
  1633 
       
  1634   aep = __ pc();  __ push_ptr();  __ b(L);
       
  1635   fep = __ pc();  __ push_f();    __ b(L);
       
  1636   dep = __ pc();  __ push_d();    __ b(L);
       
  1637   lep = __ pc();  __ push_l();    __ b(L);
       
  1638   __ align(32, 12, 24); // align L
       
  1639   bep = cep = sep =
       
  1640   iep = __ pc();  __ push_i();
       
  1641   vep = __ pc();
       
  1642   __ bind(L);
       
  1643   generate_and_dispatch(t);
       
  1644 }
       
  1645 
       
  1646 //-----------------------------------------------------------------------------
       
  1647 // Generation of individual instructions
       
  1648 
       
  1649 // helpers for generate_and_dispatch
       
  1650 
       
  1651 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
       
  1652   : TemplateInterpreterGenerator(code) {
       
  1653   generate_all(); // Down here so it can be "virtual".
       
  1654 }
       
  1655 
       
  1656 //-----------------------------------------------------------------------------
       
  1657 
       
  1658 // Non-product code
       
  1659 #ifndef PRODUCT
       
  1660 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
       
  1661   //__ flush_bundle();
       
  1662   address entry = __ pc();
       
  1663 
       
  1664   const char *bname = NULL;
       
  1665   uint tsize = 0;
       
  1666   switch(state) {
       
  1667   case ftos:
       
  1668     bname = "trace_code_ftos {";
       
  1669     tsize = 2;
       
  1670     break;
       
  1671   case btos:
       
  1672     bname = "trace_code_btos {";
       
  1673     tsize = 2;
       
  1674     break;
       
  1675   case ctos:
       
  1676     bname = "trace_code_ctos {";
       
  1677     tsize = 2;
       
  1678     break;
       
  1679   case stos:
       
  1680     bname = "trace_code_stos {";
       
  1681     tsize = 2;
       
  1682     break;
       
  1683   case itos:
       
  1684     bname = "trace_code_itos {";
       
  1685     tsize = 2;
       
  1686     break;
       
  1687   case ltos:
       
  1688     bname = "trace_code_ltos {";
       
  1689     tsize = 3;
       
  1690     break;
       
  1691   case atos:
       
  1692     bname = "trace_code_atos {";
       
  1693     tsize = 2;
       
  1694     break;
       
  1695   case vtos:
       
  1696     // Note: In case of vtos, the topmost of stack value could be a int or doubl
       
  1697     // In case of a double (2 slots) we won't see the 2nd stack value.
       
  1698     // Maybe we simply should print the topmost 3 stack slots to cope with the problem.
       
  1699     bname = "trace_code_vtos {";
       
  1700     tsize = 2;
       
  1701 
       
  1702     break;
       
  1703   case dtos:
       
  1704     bname = "trace_code_dtos {";
       
  1705     tsize = 3;
       
  1706     break;
       
  1707   default:
       
  1708     ShouldNotReachHere();
       
  1709   }
       
  1710   BLOCK_COMMENT(bname);
       
  1711 
       
  1712   // Support short-cut for TraceBytecodesAt.
       
  1713   // Don't call into the VM if we don't want to trace to speed up things.
       
  1714   Label Lskip_vm_call;
       
  1715   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
       
  1716     int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true);
       
  1717     int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
       
  1718     __ ld(R11_scratch1, offs1, R11_scratch1);
       
  1719     __ lwa(R12_scratch2, offs2, R12_scratch2);
       
  1720     __ cmpd(CCR0, R12_scratch2, R11_scratch1);
       
  1721     __ blt(CCR0, Lskip_vm_call);
       
  1722   }
       
  1723 
       
  1724   __ push(state);
       
  1725   // Load 2 topmost expression stack values.
       
  1726   __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp);
       
  1727   __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp);
       
  1728   __ mflr(R31);
       
  1729   __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false);
       
  1730   __ mtlr(R31);
       
  1731   __ pop(state);
       
  1732 
       
  1733   if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) {
       
  1734     __ bind(Lskip_vm_call);
       
  1735   }
       
  1736   __ blr();
       
  1737   BLOCK_COMMENT("} trace_code");
       
  1738   return entry;
       
  1739 }
       
  1740 
       
  1741 void TemplateInterpreterGenerator::count_bytecode() {
       
  1742   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true);
       
  1743   __ lwz(R12_scratch2, offs, R11_scratch1);
       
  1744   __ addi(R12_scratch2, R12_scratch2, 1);
       
  1745   __ stw(R12_scratch2, offs, R11_scratch1);
       
  1746 }
       
  1747 
       
  1748 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
       
  1749   int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true);
       
  1750   __ lwz(R12_scratch2, offs, R11_scratch1);
       
  1751   __ addi(R12_scratch2, R12_scratch2, 1);
       
  1752   __ stw(R12_scratch2, offs, R11_scratch1);
       
  1753 }
       
  1754 
       
  1755 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
       
  1756   const Register addr = R11_scratch1,
       
  1757                  tmp  = R12_scratch2;
       
  1758   // Get index, shift out old bytecode, bring in new bytecode, and store it.
       
  1759   // _index = (_index >> log2_number_of_codes) |
       
  1760   //          (bytecode << log2_number_of_codes);
       
  1761   int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true);
       
  1762   __ lwz(tmp, offs1, addr);
       
  1763   __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes);
       
  1764   __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
       
  1765   __ stw(tmp, offs1, addr);
       
  1766 
       
  1767   // Bump bucket contents.
       
  1768   // _counters[_index] ++;
       
  1769   int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true);
       
  1770   __ sldi(tmp, tmp, LogBytesPerInt);
       
  1771   __ add(addr, tmp, addr);
       
  1772   __ lwz(tmp, offs2, addr);
       
  1773   __ addi(tmp, tmp, 1);
       
  1774   __ stw(tmp, offs2, addr);
       
  1775 }
       
  1776 
       
  1777 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
       
  1778   // Call a little run-time stub to avoid blow-up for each bytecode.
       
  1779   // The run-time runtime saves the right registers, depending on
       
  1780   // the tosca in-state for the given template.
       
  1781 
       
  1782   assert(Interpreter::trace_code(t->tos_in()) != NULL,
       
  1783          "entry must have been generated");
       
  1784 
       
  1785   // Note: we destroy LR here.
       
  1786   __ bl(Interpreter::trace_code(t->tos_in()));
       
  1787 }
       
  1788 
       
  1789 void TemplateInterpreterGenerator::stop_interpreter_at() {
       
  1790   Label L;
       
  1791   int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true);
       
  1792   int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true);
       
  1793   __ ld(R11_scratch1, offs1, R11_scratch1);
       
  1794   __ lwa(R12_scratch2, offs2, R12_scratch2);
       
  1795   __ cmpd(CCR0, R12_scratch2, R11_scratch1);
       
  1796   __ bne(CCR0, L);
       
  1797   __ illtrap();
       
  1798   __ bind(L);
       
  1799 }
       
  1800 
       
  1801 #endif // !PRODUCT
       
  1802 #endif // !CC_INTERP