src/cpu/sparc/vm/c1_Runtime1_sparc.cpp

Tue, 24 Jul 2012 10:51:00 -0700

author
twisti
date
Tue, 24 Jul 2012 10:51:00 -0700
changeset 3969
1d7922586cf6
parent 3400
22cee0ee8927
child 4037
da91efe96a93
permissions
-rw-r--r--

7023639: JSR 292 method handle invocation needs a fast path for compiled code
6984705: JSR 292 method handle creation should not go through JNI
Summary: remove assembly code for JDK 7 chained method handles
Reviewed-by: jrose, twisti, kvn, mhaupt
Contributed-by: John Rose <john.r.rose@oracle.com>, Christian Thalinger <christian.thalinger@oracle.com>, Michael Haupt <michael.haupt@oracle.com>

     1 /*
     2  * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 #include "precompiled.hpp"
    26 #include "c1/c1_Defs.hpp"
    27 #include "c1/c1_MacroAssembler.hpp"
    28 #include "c1/c1_Runtime1.hpp"
    29 #include "interpreter/interpreter.hpp"
    30 #include "nativeInst_sparc.hpp"
    31 #include "oops/compiledICHolderOop.hpp"
    32 #include "oops/oop.inline.hpp"
    33 #include "prims/jvmtiExport.hpp"
    34 #include "register_sparc.hpp"
    35 #include "runtime/sharedRuntime.hpp"
    36 #include "runtime/signature.hpp"
    37 #include "runtime/vframeArray.hpp"
    38 #include "vmreg_sparc.inline.hpp"
    40 // Implementation of StubAssembler
    42 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) {
    43   // for sparc changing the number of arguments doesn't change
    44   // anything about the frame size so we'll always lie and claim that
    45   // we are only passing 1 argument.
    46   set_num_rt_args(1);
    48   assert_not_delayed();
    49   // bang stack before going to runtime
    50   set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
    51   st(G0, SP, G3_scratch);
    53   // debugging support
    54   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
    56   set_last_Java_frame(SP, noreg);
    57   if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early
    58   save_thread(L7_thread_cache);
    59   // do the call
    60   call(entry_point, relocInfo::runtime_call_type);
    61   if (!VerifyThread) {
    62     delayed()->mov(G2_thread, O0);  // pass thread as first argument
    63   } else {
    64     delayed()->nop();             // (thread already passed)
    65   }
    66   int call_offset = offset();  // offset of return address
    67   restore_thread(L7_thread_cache);
    68   reset_last_Java_frame();
    70   // check for pending exceptions
    71   { Label L;
    72     Address exception_addr(G2_thread, Thread::pending_exception_offset());
    73     ld_ptr(exception_addr, Gtemp);
    74     br_null_short(Gtemp, pt, L);
    75     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
    76     st_ptr(G0, vm_result_addr);
    77     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
    78     st_ptr(G0, vm_result_addr_2);
    80     if (frame_size() == no_frame_size) {
    81       // we use O7 linkage so that forward_exception_entry has the issuing PC
    82       call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
    83       delayed()->restore();
    84     } else if (_stub_id == Runtime1::forward_exception_id) {
    85       should_not_reach_here();
    86     } else {
    87       AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));
    88       jump_to(exc, G4);
    89       delayed()->nop();
    90     }
    91     bind(L);
    92   }
    94   // get oop result if there is one and reset the value in the thread
    95   if (oop_result1->is_valid()) {                    // get oop result if there is one and reset it in the thread
    96     get_vm_result  (oop_result1);
    97   } else {
    98     // be a little paranoid and clear the result
    99     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
   100     st_ptr(G0, vm_result_addr);
   101   }
   103   if (oop_result2->is_valid()) {
   104     get_vm_result_2(oop_result2);
   105   } else {
   106     // be a little paranoid and clear the result
   107     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
   108     st_ptr(G0, vm_result_addr_2);
   109   }
   111   return call_offset;
   112 }
   115 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
   116   // O0 is reserved for the thread
   117   mov(arg1, O1);
   118   return call_RT(oop_result1, oop_result2, entry, 1);
   119 }
   122 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
   123   // O0 is reserved for the thread
   124   mov(arg1, O1);
   125   mov(arg2, O2); assert(arg2 != O1, "smashed argument");
   126   return call_RT(oop_result1, oop_result2, entry, 2);
   127 }
   130 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
   131   // O0 is reserved for the thread
   132   mov(arg1, O1);
   133   mov(arg2, O2); assert(arg2 != O1,               "smashed argument");
   134   mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
   135   return call_RT(oop_result1, oop_result2, entry, 3);
   136 }
   139 // Implementation of Runtime1
   141 #define __ sasm->
   143 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
   144 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
   145 static int reg_save_size_in_words;
   146 static int frame_size_in_bytes = -1;
   148 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
   149   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
   150          "mismatch in calculation");
   151   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
   152   int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
   153   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
   155   int i;
   156   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   157     Register r = as_Register(i);
   158     if (r == G1 || r == G3 || r == G4 || r == G5) {
   159       int sp_offset = cpu_reg_save_offsets[i];
   160       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
   161                                 r->as_VMReg());
   162     }
   163   }
   165   if (save_fpu_registers) {
   166     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   167       FloatRegister r = as_FloatRegister(i);
   168       int sp_offset = fpu_reg_save_offsets[i];
   169       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
   170                                 r->as_VMReg());
   171     }
   172   }
   173   return oop_map;
   174 }
   176 static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
   177   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
   178          "mismatch in calculation");
   179   __ save_frame_c1(frame_size_in_bytes);
   181   // Record volatile registers as callee-save values in an OopMap so their save locations will be
   182   // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
   183   // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers
   184   // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
   185   // (as the stub's I's) when the runtime routine called by the stub creates its frame.
   186   // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
   188   int i;
   189   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   190     Register r = as_Register(i);
   191     if (r == G1 || r == G3 || r == G4 || r == G5) {
   192       int sp_offset = cpu_reg_save_offsets[i];
   193       __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
   194     }
   195   }
   197   if (save_fpu_registers) {
   198     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   199       FloatRegister r = as_FloatRegister(i);
   200       int sp_offset = fpu_reg_save_offsets[i];
   201       __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
   202     }
   203   }
   205   return generate_oop_map(sasm, save_fpu_registers);
   206 }
   208 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
   209   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
   210     Register r = as_Register(i);
   211     if (r == G1 || r == G3 || r == G4 || r == G5) {
   212       __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
   213     }
   214   }
   216   if (restore_fpu_registers) {
   217     for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
   218       FloatRegister r = as_FloatRegister(i);
   219       __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
   220     }
   221   }
   222 }
   225 void Runtime1::initialize_pd() {
   226   // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
   227   //
   228   // A stub routine will have a frame that is at least large enough to hold
   229   // a register window save area (obviously) and the volatile g registers
   230   // and floating registers. A user of save_live_registers can have a frame
   231   // that has more scratch area in it (although typically they will use L-regs).
   232   // in that case the frame will look like this (stack growing down)
   233   //
   234   // FP -> |             |
   235   //       | scratch mem |
   236   //       |   "      "  |
   237   //       --------------
   238   //       | float regs  |
   239   //       |   "    "    |
   240   //       ---------------
   241   //       | G regs      |
   242   //       | "  "        |
   243   //       ---------------
   244   //       | abi reg.    |
   245   //       | window save |
   246   //       | area        |
   247   // SP -> ---------------
   248   //
   249   int i;
   250   int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned
   252   // only G int registers are saved explicitly; others are found in register windows
   253   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
   254     Register r = as_Register(i);
   255     if (r == G1 || r == G3 || r == G4 || r == G5) {
   256       cpu_reg_save_offsets[i] = sp_offset;
   257       sp_offset++;
   258     }
   259   }
   261   // all float registers are saved explicitly
   262   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
   263   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
   264     fpu_reg_save_offsets[i] = sp_offset;
   265     sp_offset++;
   266   }
   267   reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
   268   // this should match assembler::total_frame_size_in_bytes, which
   269   // isn't callable from this context.  It's checked by an assert when
   270   // it's used though.
   271   frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
   272 }
   275 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
   276   // make a frame and preserve the caller's caller-save registers
   277   OopMap* oop_map = save_live_registers(sasm);
   278   int call_offset;
   279   if (!has_argument) {
   280     call_offset = __ call_RT(noreg, noreg, target);
   281   } else {
   282     call_offset = __ call_RT(noreg, noreg, target, G4);
   283   }
   284   OopMapSet* oop_maps = new OopMapSet();
   285   oop_maps->add_gc_map(call_offset, oop_map);
   287   __ should_not_reach_here();
   288   return oop_maps;
   289 }
   292 OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
   293                                         Register arg1, Register arg2, Register arg3) {
   294   // make a frame and preserve the caller's caller-save registers
   295   OopMap* oop_map = save_live_registers(sasm);
   297   int call_offset;
   298   if (arg1 == noreg) {
   299     call_offset = __ call_RT(result, noreg, target);
   300   } else if (arg2 == noreg) {
   301     call_offset = __ call_RT(result, noreg, target, arg1);
   302   } else if (arg3 == noreg) {
   303     call_offset = __ call_RT(result, noreg, target, arg1, arg2);
   304   } else {
   305     call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
   306   }
   307   OopMapSet* oop_maps = NULL;
   309   oop_maps = new OopMapSet();
   310   oop_maps->add_gc_map(call_offset, oop_map);
   311   restore_live_registers(sasm);
   313   __ ret();
   314   __ delayed()->restore();
   316   return oop_maps;
   317 }
   320 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
   321   // make a frame and preserve the caller's caller-save registers
   322   OopMap* oop_map = save_live_registers(sasm);
   324   // call the runtime patching routine, returns non-zero if nmethod got deopted.
   325   int call_offset = __ call_RT(noreg, noreg, target);
   326   OopMapSet* oop_maps = new OopMapSet();
   327   oop_maps->add_gc_map(call_offset, oop_map);
   329   // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
   330   // deoptimization handler entry that will cause re-execution of the current bytecode
   331   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
   332   assert(deopt_blob != NULL, "deoptimization blob must have been created");
   334   Label no_deopt;
   335   __ br_null_short(O0, Assembler::pt, no_deopt);
   337   // return to the deoptimization handler entry for unpacking and rexecute
   338   // if we simply returned the we'd deopt as if any call we patched had just
   339   // returned.
   341   restore_live_registers(sasm);
   343   AddressLiteral dest(deopt_blob->unpack_with_reexecution());
   344   __ jump_to(dest, O0);
   345   __ delayed()->restore();
   347   __ bind(no_deopt);
   348   restore_live_registers(sasm);
   349   __ ret();
   350   __ delayed()->restore();
   352   return oop_maps;
   353 }
   355 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
   357   OopMapSet* oop_maps = NULL;
   358   // for better readability
   359   const bool must_gc_arguments = true;
   360   const bool dont_gc_arguments = false;
   362   // stub code & info for the different stubs
   363   switch (id) {
   364     case forward_exception_id:
   365       {
   366         oop_maps = generate_handle_exception(id, sasm);
   367       }
   368       break;
   370     case new_instance_id:
   371     case fast_new_instance_id:
   372     case fast_new_instance_init_check_id:
   373       {
   374         Register G5_klass = G5; // Incoming
   375         Register O0_obj   = O0; // Outgoing
   377         if (id == new_instance_id) {
   378           __ set_info("new_instance", dont_gc_arguments);
   379         } else if (id == fast_new_instance_id) {
   380           __ set_info("fast new_instance", dont_gc_arguments);
   381         } else {
   382           assert(id == fast_new_instance_init_check_id, "bad StubID");
   383           __ set_info("fast new_instance init check", dont_gc_arguments);
   384         }
   386         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
   387             UseTLAB && FastTLABRefill) {
   388           Label slow_path;
   389           Register G1_obj_size = G1;
   390           Register G3_t1 = G3;
   391           Register G4_t2 = G4;
   392           assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
   394           // Push a frame since we may do dtrace notification for the
   395           // allocation which requires calling out and we don't want
   396           // to stomp the real return address.
   397           __ save_frame(0);
   399           if (id == fast_new_instance_init_check_id) {
   400             // make sure the klass is initialized
   401             __ ldub(G5_klass, in_bytes(instanceKlass::init_state_offset()), G3_t1);
   402             __ cmp_and_br_short(G3_t1, instanceKlass::fully_initialized, Assembler::notEqual, Assembler::pn, slow_path);
   403           }
   404 #ifdef ASSERT
   405           // assert object can be fast path allocated
   406           {
   407             Label ok, not_ok;
   408           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   409           // make sure it's an instance (LH > 0)
   410           __ cmp_and_br_short(G1_obj_size, 0, Assembler::lessEqual, Assembler::pn, not_ok);
   411           __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
   412           __ br(Assembler::zero, false, Assembler::pn, ok);
   413           __ delayed()->nop();
   414           __ bind(not_ok);
   415           __ stop("assert(can be fast path allocated)");
   416           __ should_not_reach_here();
   417           __ bind(ok);
   418           }
   419 #endif // ASSERT
   420           // if we got here then the TLAB allocation failed, so try
   421           // refilling the TLAB or allocating directly from eden.
   422           Label retry_tlab, try_eden;
   423           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
   425           __ bind(retry_tlab);
   427           // get the instance size
   428           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   430           __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
   432           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
   433           __ verify_oop(O0_obj);
   434           __ mov(O0, I0);
   435           __ ret();
   436           __ delayed()->restore();
   438           __ bind(try_eden);
   439           // get the instance size
   440           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
   441           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
   442           __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
   444           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
   445           __ verify_oop(O0_obj);
   446           __ mov(O0, I0);
   447           __ ret();
   448           __ delayed()->restore();
   450           __ bind(slow_path);
   452           // pop this frame so generate_stub_call can push it's own
   453           __ restore();
   454         }
   456         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
   457         // I0->O0: new instance
   458       }
   460       break;
   462     case counter_overflow_id:
   463         // G4 contains bci, G5 contains method
   464       oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
   465       break;
   467     case new_type_array_id:
   468     case new_object_array_id:
   469       {
   470         Register G5_klass = G5; // Incoming
   471         Register G4_length = G4; // Incoming
   472         Register O0_obj   = O0; // Outgoing
   474         Address klass_lh(G5_klass, Klass::layout_helper_offset());
   475         assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
   476         assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
   477         // Use this offset to pick out an individual byte of the layout_helper:
   478         const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
   479                                                  - Klass::_lh_header_size_shift / BitsPerByte);
   481         if (id == new_type_array_id) {
   482           __ set_info("new_type_array", dont_gc_arguments);
   483         } else {
   484           __ set_info("new_object_array", dont_gc_arguments);
   485         }
   487 #ifdef ASSERT
   488         // assert object type is really an array of the proper kind
   489         {
   490           Label ok;
   491           Register G3_t1 = G3;
   492           __ ld(klass_lh, G3_t1);
   493           __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
   494           int tag = ((id == new_type_array_id)
   495                      ? Klass::_lh_array_tag_type_value
   496                      : Klass::_lh_array_tag_obj_value);
   497           __ cmp_and_brx_short(G3_t1, tag, Assembler::equal, Assembler::pt, ok);
   498           __ stop("assert(is an array klass)");
   499           __ should_not_reach_here();
   500           __ bind(ok);
   501         }
   502 #endif // ASSERT
   504         if (UseTLAB && FastTLABRefill) {
   505           Label slow_path;
   506           Register G1_arr_size = G1;
   507           Register G3_t1 = G3;
   508           Register O1_t2 = O1;
   509           assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
   511           // check that array length is small enough for fast path
   512           __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
   513           __ cmp_and_br_short(G4_length, G3_t1, Assembler::greaterUnsigned, Assembler::pn, slow_path);
   515           // if we got here then the TLAB allocation failed, so try
   516           // refilling the TLAB or allocating directly from eden.
   517           Label retry_tlab, try_eden;
   518           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
   520           __ bind(retry_tlab);
   522           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
   523           __ ld(klass_lh, G3_t1);
   524           __ sll(G4_length, G3_t1, G1_arr_size);
   525           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
   526           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
   527           __ add(G1_arr_size, G3_t1, G1_arr_size);
   528           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
   529           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
   531           __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
   533           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
   534           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
   535           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
   536           __ add(O0_obj, G3_t1, G3_t1);       // body start
   537           __ initialize_body(G3_t1, O1_t2);
   538           __ verify_oop(O0_obj);
   539           __ retl();
   540           __ delayed()->nop();
   542           __ bind(try_eden);
   543           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
   544           __ ld(klass_lh, G3_t1);
   545           __ sll(G4_length, G3_t1, G1_arr_size);
   546           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
   547           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
   548           __ add(G1_arr_size, G3_t1, G1_arr_size);
   549           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
   550           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
   552           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
   553           __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
   555           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
   556           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
   557           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
   558           __ add(O0_obj, G3_t1, G3_t1);       // body start
   559           __ initialize_body(G3_t1, O1_t2);
   560           __ verify_oop(O0_obj);
   561           __ retl();
   562           __ delayed()->nop();
   564           __ bind(slow_path);
   565         }
   567         if (id == new_type_array_id) {
   568           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
   569         } else {
   570           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
   571         }
   572         // I0 -> O0: new array
   573       }
   574       break;
   576     case new_multi_array_id:
   577       { // O0: klass
   578         // O1: rank
   579         // O2: address of 1st dimension
   580         __ set_info("new_multi_array", dont_gc_arguments);
   581         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
   582         // I0 -> O0: new multi array
   583       }
   584       break;
   586     case register_finalizer_id:
   587       {
   588         __ set_info("register_finalizer", dont_gc_arguments);
   590         // load the klass and check the has finalizer flag
   591         Label register_finalizer;
   592         Register t = O1;
   593         __ load_klass(O0, t);
   594         __ ld(t, in_bytes(Klass::access_flags_offset()), t);
   595         __ set(JVM_ACC_HAS_FINALIZER, G3);
   596         __ andcc(G3, t, G0);
   597         __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
   598         __ delayed()->nop();
   600         // do a leaf return
   601         __ retl();
   602         __ delayed()->nop();
   604         __ bind(register_finalizer);
   605         OopMap* oop_map = save_live_registers(sasm);
   606         int call_offset = __ call_RT(noreg, noreg,
   607                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
   608         oop_maps = new OopMapSet();
   609         oop_maps->add_gc_map(call_offset, oop_map);
   611         // Now restore all the live registers
   612         restore_live_registers(sasm);
   614         __ ret();
   615         __ delayed()->restore();
   616       }
   617       break;
   619     case throw_range_check_failed_id:
   620       { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
   621         // G4: index
   622         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
   623       }
   624       break;
   626     case throw_index_exception_id:
   627       { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
   628         // G4: index
   629         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
   630       }
   631       break;
   633     case throw_div0_exception_id:
   634       { __ set_info("throw_div0_exception", dont_gc_arguments);
   635         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
   636       }
   637       break;
   639     case throw_null_pointer_exception_id:
   640       { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
   641         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
   642       }
   643       break;
   645     case handle_exception_id:
   646       { __ set_info("handle_exception", dont_gc_arguments);
   647         oop_maps = generate_handle_exception(id, sasm);
   648       }
   649       break;
   651     case handle_exception_from_callee_id:
   652       { __ set_info("handle_exception_from_callee", dont_gc_arguments);
   653         oop_maps = generate_handle_exception(id, sasm);
   654       }
   655       break;
   657     case unwind_exception_id:
   658       {
   659         // O0: exception
   660         // I7: address of call to this method
   662         __ set_info("unwind_exception", dont_gc_arguments);
   663         __ mov(Oexception, Oexception->after_save());
   664         __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
   666         __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
   667                         G2_thread, Oissuing_pc->after_save());
   668         __ verify_not_null_oop(Oexception->after_save());
   670         // Restore SP from L7 if the exception PC is a method handle call site.
   671         __ mov(O0, G5);  // Save the target address.
   672         __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
   673         __ tst(L0);  // Condition codes are preserved over the restore.
   674         __ restore();
   676         __ jmp(G5, 0);
   677         __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
   678       }
   679       break;
   681     case throw_array_store_exception_id:
   682       {
   683         __ set_info("throw_array_store_exception", dont_gc_arguments);
   684         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
   685       }
   686       break;
   688     case throw_class_cast_exception_id:
   689       {
   690         // G4: object
   691         __ set_info("throw_class_cast_exception", dont_gc_arguments);
   692         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
   693       }
   694       break;
   696     case throw_incompatible_class_change_error_id:
   697       {
   698         __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
   699         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
   700       }
   701       break;
   703     case slow_subtype_check_id:
   704       { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
   705         // Arguments :
   706         //
   707         //      ret  : G3
   708         //      sub  : G3, argument, destroyed
   709         //      super: G1, argument, not changed
   710         //      raddr: O7, blown by call
   711         Label miss;
   713         __ save_frame(0);               // Blow no registers!
   715         __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
   717         __ mov(1, G3);
   718         __ ret();                       // Result in G5 is 'true'
   719         __ delayed()->restore();        // free copy or add can go here
   721         __ bind(miss);
   722         __ mov(0, G3);
   723         __ ret();                       // Result in G5 is 'false'
   724         __ delayed()->restore();        // free copy or add can go here
   725       }
   727     case monitorenter_nofpu_id:
   728     case monitorenter_id:
   729       { // G4: object
   730         // G5: lock address
   731         __ set_info("monitorenter", dont_gc_arguments);
   733         int save_fpu_registers = (id == monitorenter_id);
   734         // make a frame and preserve the caller's caller-save registers
   735         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
   737         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
   739         oop_maps = new OopMapSet();
   740         oop_maps->add_gc_map(call_offset, oop_map);
   741         restore_live_registers(sasm, save_fpu_registers);
   743         __ ret();
   744         __ delayed()->restore();
   745       }
   746       break;
   748     case monitorexit_nofpu_id:
   749     case monitorexit_id:
   750       { // G4: lock address
   751         // note: really a leaf routine but must setup last java sp
   752         //       => use call_RT for now (speed can be improved by
   753         //       doing last java sp setup manually)
   754         __ set_info("monitorexit", dont_gc_arguments);
   756         int save_fpu_registers = (id == monitorexit_id);
   757         // make a frame and preserve the caller's caller-save registers
   758         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
   760         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
   762         oop_maps = new OopMapSet();
   763         oop_maps->add_gc_map(call_offset, oop_map);
   764         restore_live_registers(sasm, save_fpu_registers);
   766         __ ret();
   767         __ delayed()->restore();
   768       }
   769       break;
   771     case deoptimize_id:
   772       {
   773         __ set_info("deoptimize", dont_gc_arguments);
   774         OopMap* oop_map = save_live_registers(sasm);
   775         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
   776         oop_maps = new OopMapSet();
   777         oop_maps->add_gc_map(call_offset, oop_map);
   778         restore_live_registers(sasm);
   779         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
   780         assert(deopt_blob != NULL, "deoptimization blob must have been created");
   781         AddressLiteral dest(deopt_blob->unpack_with_reexecution());
   782         __ jump_to(dest, O0);
   783         __ delayed()->restore();
   784       }
   785       break;
   787     case access_field_patching_id:
   788       { __ set_info("access_field_patching", dont_gc_arguments);
   789         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
   790       }
   791       break;
   793     case load_klass_patching_id:
   794       { __ set_info("load_klass_patching", dont_gc_arguments);
   795         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
   796       }
   797       break;
   799     case dtrace_object_alloc_id:
   800       { // O0: object
   801         __ set_info("dtrace_object_alloc", dont_gc_arguments);
   802         // we can't gc here so skip the oopmap but make sure that all
   803         // the live registers get saved.
   804         save_live_registers(sasm);
   806         __ save_thread(L7_thread_cache);
   807         __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
   808                 relocInfo::runtime_call_type);
   809         __ delayed()->mov(I0, O0);
   810         __ restore_thread(L7_thread_cache);
   812         restore_live_registers(sasm);
   813         __ ret();
   814         __ delayed()->restore();
   815       }
   816       break;
   818 #ifndef SERIALGC
   819     case g1_pre_barrier_slow_id:
   820       { // G4: previous value of memory
   821         BarrierSet* bs = Universe::heap()->barrier_set();
   822         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
   823           __ save_frame(0);
   824           __ set((int)id, O1);
   825           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
   826           __ should_not_reach_here();
   827           break;
   828         }
   830         __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
   832         Register pre_val = G4;
   833         Register tmp  = G1_scratch;
   834         Register tmp2 = G3_scratch;
   836         Label refill, restart;
   837         bool with_frame = false; // I don't know if we can do with-frame.
   838         int satb_q_index_byte_offset =
   839           in_bytes(JavaThread::satb_mark_queue_offset() +
   840                    PtrQueue::byte_offset_of_index());
   841         int satb_q_buf_byte_offset =
   842           in_bytes(JavaThread::satb_mark_queue_offset() +
   843                    PtrQueue::byte_offset_of_buf());
   845         __ bind(restart);
   846         // Load the index into the SATB buffer. PtrQueue::_index is a
   847         // size_t so ld_ptr is appropriate
   848         __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
   850         // index == 0?
   851         __ cmp_and_brx_short(tmp, G0, Assembler::equal, Assembler::pn, refill);
   853         __ ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
   854         __ sub(tmp, oopSize, tmp);
   856         __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
   857         // Use return-from-leaf
   858         __ retl();
   859         __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
   861         __ bind(refill);
   862         __ save_frame(0);
   864         __ mov(pre_val, L0);
   865         __ mov(tmp,     L1);
   866         __ mov(tmp2,    L2);
   868         __ call_VM_leaf(L7_thread_cache,
   869                         CAST_FROM_FN_PTR(address,
   870                                          SATBMarkQueueSet::handle_zero_index_for_thread),
   871                                          G2_thread);
   873         __ mov(L0, pre_val);
   874         __ mov(L1, tmp);
   875         __ mov(L2, tmp2);
   877         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
   878         __ delayed()->restore();
   879       }
   880       break;
   882     case g1_post_barrier_slow_id:
   883       {
   884         BarrierSet* bs = Universe::heap()->barrier_set();
   885         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
   886           __ save_frame(0);
   887           __ set((int)id, O1);
   888           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
   889           __ should_not_reach_here();
   890           break;
   891         }
   893         __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
   895         Register addr = G4;
   896         Register cardtable = G5;
   897         Register tmp  = G1_scratch;
   898         Register tmp2 = G3_scratch;
   899         jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
   901         Label not_already_dirty, restart, refill;
   903 #ifdef _LP64
   904         __ srlx(addr, CardTableModRefBS::card_shift, addr);
   905 #else
   906         __ srl(addr, CardTableModRefBS::card_shift, addr);
   907 #endif
   909         AddressLiteral rs(byte_map_base);
   910         __ set(rs, cardtable);         // cardtable := <card table base>
   911         __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
   913         assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
   914         __ cmp_and_br_short(tmp, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);
   916         // We didn't take the branch, so we're already dirty: return.
   917         // Use return-from-leaf
   918         __ retl();
   919         __ delayed()->nop();
   921         // Not dirty.
   922         __ bind(not_already_dirty);
   924         // Get cardtable + tmp into a reg by itself
   925         __ add(addr, cardtable, tmp2);
   927         // First, dirty it.
   928         __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
   930         Register tmp3 = cardtable;
   931         Register tmp4 = tmp;
   933         // these registers are now dead
   934         addr = cardtable = tmp = noreg;
   936         int dirty_card_q_index_byte_offset =
   937           in_bytes(JavaThread::dirty_card_queue_offset() +
   938                    PtrQueue::byte_offset_of_index());
   939         int dirty_card_q_buf_byte_offset =
   940           in_bytes(JavaThread::dirty_card_queue_offset() +
   941                    PtrQueue::byte_offset_of_buf());
   943         __ bind(restart);
   945         // Get the index into the update buffer. PtrQueue::_index is
   946         // a size_t so ld_ptr is appropriate here.
   947         __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
   949         // index == 0?
   950         __ cmp_and_brx_short(tmp3, G0, Assembler::equal,  Assembler::pn, refill);
   952         __ ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
   953         __ sub(tmp3, oopSize, tmp3);
   955         __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
   956         // Use return-from-leaf
   957         __ retl();
   958         __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
   960         __ bind(refill);
   961         __ save_frame(0);
   963         __ mov(tmp2, L0);
   964         __ mov(tmp3, L1);
   965         __ mov(tmp4, L2);
   967         __ call_VM_leaf(L7_thread_cache,
   968                         CAST_FROM_FN_PTR(address,
   969                                          DirtyCardQueueSet::handle_zero_index_for_thread),
   970                                          G2_thread);
   972         __ mov(L0, tmp2);
   973         __ mov(L1, tmp3);
   974         __ mov(L2, tmp4);
   976         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
   977         __ delayed()->restore();
   978       }
   979       break;
   980 #endif // !SERIALGC
   982     default:
   983       { __ set_info("unimplemented entry", dont_gc_arguments);
   984         __ save_frame(0);
   985         __ set((int)id, O1);
   986         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
   987         __ should_not_reach_here();
   988       }
   989       break;
   990   }
   991   return oop_maps;
   992 }
   995 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler* sasm) {
   996   __ block_comment("generate_handle_exception");
   998   // Save registers, if required.
   999   OopMapSet* oop_maps = new OopMapSet();
  1000   OopMap* oop_map = NULL;
  1001   switch (id) {
  1002   case forward_exception_id:
  1003     // We're handling an exception in the context of a compiled frame.
  1004     // The registers have been saved in the standard places.  Perform
  1005     // an exception lookup in the caller and dispatch to the handler
  1006     // if found.  Otherwise unwind and dispatch to the callers
  1007     // exception handler.
  1008      oop_map = generate_oop_map(sasm, true);
  1010      // transfer the pending exception to the exception_oop
  1011      __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
  1012      __ ld_ptr(Oexception, 0, G0);
  1013      __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
  1014      __ add(I7, frame::pc_return_offset, Oissuing_pc);
  1015     break;
  1016   case handle_exception_id:
  1017     // At this point all registers MAY be live.
  1018     oop_map = save_live_registers(sasm);
  1019     __ mov(Oexception->after_save(),  Oexception);
  1020     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
  1021     break;
  1022   case handle_exception_from_callee_id:
  1023     // At this point all registers except exception oop (Oexception)
  1024     // and exception pc (Oissuing_pc) are dead.
  1025     oop_map = new OopMap(frame_size_in_bytes / sizeof(jint), 0);
  1026     sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
  1027     __ save_frame_c1(frame_size_in_bytes);
  1028     __ mov(Oexception->after_save(),  Oexception);
  1029     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
  1030     break;
  1031   default:  ShouldNotReachHere();
  1034   __ verify_not_null_oop(Oexception);
  1036   // save the exception and issuing pc in the thread
  1037   __ st_ptr(Oexception,  G2_thread, in_bytes(JavaThread::exception_oop_offset()));
  1038   __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
  1040   // use the throwing pc as the return address to lookup (has bci & oop map)
  1041   __ mov(Oissuing_pc, I7);
  1042   __ sub(I7, frame::pc_return_offset, I7);
  1043   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
  1044   oop_maps->add_gc_map(call_offset, oop_map);
  1046   // Note: if nmethod has been deoptimized then regardless of
  1047   // whether it had a handler or not we will deoptimize
  1048   // by entering the deopt blob with a pending exception.
  1050   // Restore the registers that were saved at the beginning, remove
  1051   // the frame and jump to the exception handler.
  1052   switch (id) {
  1053   case forward_exception_id:
  1054   case handle_exception_id:
  1055     restore_live_registers(sasm);
  1056     __ jmp(O0, 0);
  1057     __ delayed()->restore();
  1058     break;
  1059   case handle_exception_from_callee_id:
  1060     // Restore SP from L7 if the exception PC is a method handle call site.
  1061     __ mov(O0, G5);  // Save the target address.
  1062     __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
  1063     __ tst(L0);  // Condition codes are preserved over the restore.
  1064     __ restore();
  1066     __ jmp(G5, 0);  // jump to the exception handler
  1067     __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
  1068     break;
  1069   default:  ShouldNotReachHere();
  1072   return oop_maps;
  1076 #undef __
  1078 const char *Runtime1::pd_name_for_address(address entry) {
  1079   return "<unknown function>";

mercurial