1 /*
   2  * Copyright 1999-2010 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  20  * CA 95054 USA or visit www.sun.com if you need additional information or
  21  * have any questions.
  22  *
  23  */
  24 
  25 #include "incls/_precompiled.incl"
  26 #include "incls/_c1_Runtime1_sparc.cpp.incl"
  27 
  28 // Implementation of StubAssembler
  29 
  30 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry_point, int number_of_arguments) {
  31   // for sparc changing the number of arguments doesn't change
  32   // anything about the frame size so we'll always lie and claim that
  33   // we are only passing 1 argument.
  34   set_num_rt_args(1);
  35 
  36   assert_not_delayed();
  37   // bang stack before going to runtime
  38   set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
  39   st(G0, SP, G3_scratch);
  40 
  41   // debugging support
  42   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
  43 
  44   set_last_Java_frame(SP, noreg);
  45   if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early
  46   save_thread(L7_thread_cache);
  47   // do the call
  48   call(entry_point, relocInfo::runtime_call_type);
  49   if (!VerifyThread) {
  50     delayed()->mov(G2_thread, O0);  // pass thread as first argument
  51   } else {
  52     delayed()->nop();             // (thread already passed)
  53   }
  54   int call_offset = offset();  // offset of return address
  55   restore_thread(L7_thread_cache);
  56   reset_last_Java_frame();
  57 
  58   // check for pending exceptions
  59   { Label L;
  60     Address exception_addr(G2_thread, Thread::pending_exception_offset());
  61     ld_ptr(exception_addr, Gtemp);
  62     br_null(Gtemp, false, pt, L);
  63     delayed()->nop();
  64     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
  65     st_ptr(G0, vm_result_addr);
  66     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
  67     st_ptr(G0, vm_result_addr_2);
  68 
  69     if (frame_size() == no_frame_size) {
  70       // we use O7 linkage so that forward_exception_entry has the issuing PC
  71       call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
  72       delayed()->restore();
  73     } else if (_stub_id == Runtime1::forward_exception_id) {
  74       should_not_reach_here();
  75     } else {
  76       AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));
  77       jump_to(exc, G4);
  78       delayed()->nop();
  79     }
  80     bind(L);
  81   }
  82 
  83   // get oop result if there is one and reset the value in the thread
  84   if (oop_result1->is_valid()) {                    // get oop result if there is one and reset it in the thread
  85     get_vm_result  (oop_result1);
  86   } else {
  87     // be a little paranoid and clear the result
  88     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
  89     st_ptr(G0, vm_result_addr);
  90   }
  91 
  92   if (oop_result2->is_valid()) {
  93     get_vm_result_2(oop_result2);
  94   } else {
  95     // be a little paranoid and clear the result
  96     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
  97     st_ptr(G0, vm_result_addr_2);
  98   }
  99 
 100   return call_offset;
 101 }
 102 
 103 
 104 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
 105   // O0 is reserved for the thread
 106   mov(arg1, O1);
 107   return call_RT(oop_result1, oop_result2, entry, 1);
 108 }
 109 
 110 
 111 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
 112   // O0 is reserved for the thread
 113   mov(arg1, O1);
 114   mov(arg2, O2); assert(arg2 != O1, "smashed argument");
 115   return call_RT(oop_result1, oop_result2, entry, 2);
 116 }
 117 
 118 
 119 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
 120   // O0 is reserved for the thread
 121   mov(arg1, O1);
 122   mov(arg2, O2); assert(arg2 != O1,               "smashed argument");
 123   mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
 124   return call_RT(oop_result1, oop_result2, entry, 3);
 125 }
 126 
 127 
 128 // Implementation of Runtime1
 129 
 130 #define __ sasm->
 131 
 132 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
 133 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
 134 static int reg_save_size_in_words;
 135 static int frame_size_in_bytes = -1;
 136 
 137 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
 138   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 139          " mismatch in calculation");
 140   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 141   int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
 142   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
 143 
 144   int i;
 145   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 146     Register r = as_Register(i);
 147     if (r == G1 || r == G3 || r == G4 || r == G5) {
 148       int sp_offset = cpu_reg_save_offsets[i];
 149       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 150                                 r->as_VMReg());
 151     }
 152   }
 153 
 154   if (save_fpu_registers) {
 155     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 156       FloatRegister r = as_FloatRegister(i);
 157       int sp_offset = fpu_reg_save_offsets[i];
 158       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 159                                 r->as_VMReg());
 160     }
 161   }
 162   return oop_map;
 163 }
 164 
 165 static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
 166   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 167          " mismatch in calculation");
 168   __ save_frame_c1(frame_size_in_bytes);
 169   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 170 
 171   // Record volatile registers as callee-save values in an OopMap so their save locations will be
 172   // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
 173   // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers
 174   // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
 175   // (as the stub's I's) when the runtime routine called by the stub creates its frame.
 176   // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
 177 
 178   int i;
 179   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 180     Register r = as_Register(i);
 181     if (r == G1 || r == G3 || r == G4 || r == G5) {
 182       int sp_offset = cpu_reg_save_offsets[i];
 183       __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 184     }
 185   }
 186 
 187   if (save_fpu_registers) {
 188     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 189       FloatRegister r = as_FloatRegister(i);
 190       int sp_offset = fpu_reg_save_offsets[i];
 191       __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 192     }
 193   }
 194 
 195   return generate_oop_map(sasm, save_fpu_registers);
 196 }
 197 
 198 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 199   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
 200     Register r = as_Register(i);
 201     if (r == G1 || r == G3 || r == G4 || r == G5) {
 202       __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 203     }
 204   }
 205 
 206   if (restore_fpu_registers) {
 207     for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
 208       FloatRegister r = as_FloatRegister(i);
 209       __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 210     }
 211   }
 212 }
 213 
 214 
 215 void Runtime1::initialize_pd() {
 216   // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
 217   //
 218   // A stub routine will have a frame that is at least large enough to hold
 219   // a register window save area (obviously) and the volatile g registers
 220   // and floating registers. A user of save_live_registers can have a frame
 221   // that has more scratch area in it (although typically they will use L-regs).
 222   // in that case the frame will look like this (stack growing down)
 223   //
 224   // FP -> |             |
 225   //       | scratch mem |
 226   //       |   "      "  |
 227   //       --------------
 228   //       | float regs  |
 229   //       |   "    "    |
 230   //       ---------------
 231   //       | G regs      |
 232   //       | "  "        |
 233   //       ---------------
 234   //       | abi reg.    |
 235   //       | window save |
 236   //       | area        |
 237   // SP -> ---------------
 238   //
 239   int i;
 240   int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned
 241 
 242   // only G int registers are saved explicitly; others are found in register windows
 243   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 244     Register r = as_Register(i);
 245     if (r == G1 || r == G3 || r == G4 || r == G5) {
 246       cpu_reg_save_offsets[i] = sp_offset;
 247       sp_offset++;
 248     }
 249   }
 250 
 251   // all float registers are saved explicitly
 252   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
 253   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 254     fpu_reg_save_offsets[i] = sp_offset;
 255     sp_offset++;
 256   }
 257   reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
 258   // this should match assembler::total_frame_size_in_bytes, which
 259   // isn't callable from this context.  It's checked by an assert when
 260   // it's used though.
 261   frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
 262 }
 263 
 264 
 265 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 266   // make a frame and preserve the caller's caller-save registers
 267   OopMap* oop_map = save_live_registers(sasm);
 268   int call_offset;
 269   if (!has_argument) {
 270     call_offset = __ call_RT(noreg, noreg, target);
 271   } else {
 272     call_offset = __ call_RT(noreg, noreg, target, G4);
 273   }
 274   OopMapSet* oop_maps = new OopMapSet();
 275   oop_maps->add_gc_map(call_offset, oop_map);
 276 
 277   __ should_not_reach_here();
 278   return oop_maps;
 279 }
 280 
 281 
 282 OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
 283                                         Register arg1, Register arg2, Register arg3) {
 284   // make a frame and preserve the caller's caller-save registers
 285   OopMap* oop_map = save_live_registers(sasm);
 286 
 287   int call_offset;
 288   if (arg1 == noreg) {
 289     call_offset = __ call_RT(result, noreg, target);
 290   } else if (arg2 == noreg) {
 291     call_offset = __ call_RT(result, noreg, target, arg1);
 292   } else if (arg3 == noreg) {
 293     call_offset = __ call_RT(result, noreg, target, arg1, arg2);
 294   } else {
 295     call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
 296   }
 297   OopMapSet* oop_maps = NULL;
 298 
 299   oop_maps = new OopMapSet();
 300   oop_maps->add_gc_map(call_offset, oop_map);
 301   restore_live_registers(sasm);
 302 
 303   __ ret();
 304   __ delayed()->restore();
 305 
 306   return oop_maps;
 307 }
 308 
 309 
 310 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 311   // make a frame and preserve the caller's caller-save registers
 312   OopMap* oop_map = save_live_registers(sasm);
 313 
 314   // call the runtime patching routine, returns non-zero if nmethod got deopted.
 315   int call_offset = __ call_RT(noreg, noreg, target);
 316   OopMapSet* oop_maps = new OopMapSet();
 317   oop_maps->add_gc_map(call_offset, oop_map);
 318 
 319   // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
 320   // deoptimization handler entry that will cause re-execution of the current bytecode
 321   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 322   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 323 
 324   Label no_deopt;
 325   __ tst(O0);
 326   __ brx(Assembler::equal, false, Assembler::pt, no_deopt);
 327   __ delayed()->nop();
 328 
 329   // return to the deoptimization handler entry for unpacking and rexecute
 330   // if we simply returned the we'd deopt as if any call we patched had just
 331   // returned.
 332 
 333   restore_live_registers(sasm);
 334   __ restore();
 335   __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
 336   __ delayed()->nop();
 337 
 338   __ bind(no_deopt);
 339   restore_live_registers(sasm);
 340   __ ret();
 341   __ delayed()->restore();
 342 
 343   return oop_maps;
 344 }
 345 
 346 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 347 
 348   OopMapSet* oop_maps = NULL;
 349   // for better readability
 350   const bool must_gc_arguments = true;
 351   const bool dont_gc_arguments = false;
 352 
 353   // stub code & info for the different stubs
 354   switch (id) {
 355     case forward_exception_id:
 356       {
 357         // we're handling an exception in the context of a compiled
 358         // frame.  The registers have been saved in the standard
 359         // places.  Perform an exception lookup in the caller and
 360         // dispatch to the handler if found.  Otherwise unwind and
 361         // dispatch to the callers exception handler.
 362 
 363         oop_maps = new OopMapSet();
 364         OopMap* oop_map = generate_oop_map(sasm, true);
 365 
 366         // transfer the pending exception to the exception_oop
 367         __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
 368         __ ld_ptr(Oexception, 0, G0);
 369         __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
 370         __ add(I7, frame::pc_return_offset, Oissuing_pc);
 371 
 372         generate_handle_exception(sasm, oop_maps, oop_map);
 373         __ should_not_reach_here();
 374       }
 375       break;
 376 
 377     case new_instance_id:
 378     case fast_new_instance_id:
 379     case fast_new_instance_init_check_id:
 380       {
 381         Register G5_klass = G5; // Incoming
 382         Register O0_obj   = O0; // Outgoing
 383 
 384         if (id == new_instance_id) {
 385           __ set_info("new_instance", dont_gc_arguments);
 386         } else if (id == fast_new_instance_id) {
 387           __ set_info("fast new_instance", dont_gc_arguments);
 388         } else {
 389           assert(id == fast_new_instance_init_check_id, "bad StubID");
 390           __ set_info("fast new_instance init check", dont_gc_arguments);
 391         }
 392 
 393         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 394             UseTLAB && FastTLABRefill) {
 395           Label slow_path;
 396           Register G1_obj_size = G1;
 397           Register G3_t1 = G3;
 398           Register G4_t2 = G4;
 399           assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
 400 
 401           // Push a frame since we may do dtrace notification for the
 402           // allocation which requires calling out and we don't want
 403           // to stomp the real return address.
 404           __ save_frame(0);
 405 
 406           if (id == fast_new_instance_init_check_id) {
 407             // make sure the klass is initialized
 408             __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);
 409             __ cmp(G3_t1, instanceKlass::fully_initialized);
 410             __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
 411             __ delayed()->nop();
 412           }
 413 #ifdef ASSERT
 414           // assert object can be fast path allocated
 415           {
 416             Label ok, not_ok;
 417           __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);
 418           __ cmp(G1_obj_size, 0);  // make sure it's an instance (LH > 0)
 419           __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);
 420           __ delayed()->nop();
 421           __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
 422           __ br(Assembler::zero, false, Assembler::pn, ok);
 423           __ delayed()->nop();
 424           __ bind(not_ok);
 425           __ stop("assert(can be fast path allocated)");
 426           __ should_not_reach_here();
 427           __ bind(ok);
 428           }
 429 #endif // ASSERT
 430           // if we got here then the TLAB allocation failed, so try
 431           // refilling the TLAB or allocating directly from eden.
 432           Label retry_tlab, try_eden;
 433           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
 434 
 435           __ bind(retry_tlab);
 436 
 437           // get the instance size
 438           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 439           __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
 440           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 441           __ verify_oop(O0_obj);
 442           __ mov(O0, I0);
 443           __ ret();
 444           __ delayed()->restore();
 445 
 446           __ bind(try_eden);
 447           // get the instance size
 448           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 449           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
 450           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 451           __ verify_oop(O0_obj);
 452           __ mov(O0, I0);
 453           __ ret();
 454           __ delayed()->restore();
 455 
 456           __ bind(slow_path);
 457 
 458           // pop this frame so generate_stub_call can push it's own
 459           __ restore();
 460         }
 461 
 462         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
 463         // I0->O0: new instance
 464       }
 465 
 466       break;
 467 
 468 #ifdef TIERED
 469     case counter_overflow_id:
 470         // G4 contains bci
 471       oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4);
 472       break;
 473 #endif // TIERED
 474 
 475     case new_type_array_id:
 476     case new_object_array_id:
 477       {
 478         Register G5_klass = G5; // Incoming
 479         Register G4_length = G4; // Incoming
 480         Register O0_obj   = O0; // Outgoing
 481 
 482         Address klass_lh(G5_klass, ((klassOopDesc::header_size() * HeapWordSize)
 483                                     + Klass::layout_helper_offset_in_bytes()));
 484         assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 485         assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
 486         // Use this offset to pick out an individual byte of the layout_helper:
 487         const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
 488                                                  - Klass::_lh_header_size_shift / BitsPerByte);
 489 
 490         if (id == new_type_array_id) {
 491           __ set_info("new_type_array", dont_gc_arguments);
 492         } else {
 493           __ set_info("new_object_array", dont_gc_arguments);
 494         }
 495 
 496 #ifdef ASSERT
 497         // assert object type is really an array of the proper kind
 498         {
 499           Label ok;
 500           Register G3_t1 = G3;
 501           __ ld(klass_lh, G3_t1);
 502           __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
 503           int tag = ((id == new_type_array_id)
 504                      ? Klass::_lh_array_tag_type_value
 505                      : Klass::_lh_array_tag_obj_value);
 506           __ cmp(G3_t1, tag);
 507           __ brx(Assembler::equal, false, Assembler::pt, ok);
 508           __ delayed()->nop();
 509           __ stop("assert(is an array klass)");
 510           __ should_not_reach_here();
 511           __ bind(ok);
 512         }
 513 #endif // ASSERT
 514 
 515         if (UseTLAB && FastTLABRefill) {
 516           Label slow_path;
 517           Register G1_arr_size = G1;
 518           Register G3_t1 = G3;
 519           Register O1_t2 = O1;
 520           assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
 521 
 522           // check that array length is small enough for fast path
 523           __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
 524           __ cmp(G4_length, G3_t1);
 525           __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
 526           __ delayed()->nop();
 527 
 528           // if we got here then the TLAB allocation failed, so try
 529           // refilling the TLAB or allocating directly from eden.
 530           Label retry_tlab, try_eden;
 531           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
 532 
 533           __ bind(retry_tlab);
 534 
 535           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 536           __ ld(klass_lh, G3_t1);
 537           __ sll(G4_length, G3_t1, G1_arr_size);
 538           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 539           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 540           __ add(G1_arr_size, G3_t1, G1_arr_size);
 541           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
 542           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 543 
 544           __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
 545 
 546           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 547           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 548           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 549           __ add(O0_obj, G3_t1, G3_t1);       // body start
 550           __ initialize_body(G3_t1, O1_t2);
 551           __ verify_oop(O0_obj);
 552           __ retl();
 553           __ delayed()->nop();
 554 
 555           __ bind(try_eden);
 556           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 557           __ ld(klass_lh, G3_t1);
 558           __ sll(G4_length, G3_t1, G1_arr_size);
 559           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 560           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 561           __ add(G1_arr_size, G3_t1, G1_arr_size);
 562           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
 563           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 564 
 565           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
 566 
 567           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 568           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 569           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 570           __ add(O0_obj, G3_t1, G3_t1);       // body start
 571           __ initialize_body(G3_t1, O1_t2);
 572           __ verify_oop(O0_obj);
 573           __ retl();
 574           __ delayed()->nop();
 575 
 576           __ bind(slow_path);
 577         }
 578 
 579         if (id == new_type_array_id) {
 580           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
 581         } else {
 582           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
 583         }
 584         // I0 -> O0: new array
 585       }
 586       break;
 587 
 588     case new_multi_array_id:
 589       { // O0: klass
 590         // O1: rank
 591         // O2: address of 1st dimension
 592         __ set_info("new_multi_array", dont_gc_arguments);
 593         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
 594         // I0 -> O0: new multi array
 595       }
 596       break;
 597 
 598     case register_finalizer_id:
 599       {
 600         __ set_info("register_finalizer", dont_gc_arguments);
 601 
 602         // load the klass and check the has finalizer flag
 603         Label register_finalizer;
 604         Register t = O1;
 605         __ ld_ptr(O0, oopDesc::klass_offset_in_bytes(), t);
 606         __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t);
 607         __ set(JVM_ACC_HAS_FINALIZER, G3);
 608         __ andcc(G3, t, G0);
 609         __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
 610         __ delayed()->nop();
 611 
 612         // do a leaf return
 613         __ retl();
 614         __ delayed()->nop();
 615 
 616         __ bind(register_finalizer);
 617         OopMap* oop_map = save_live_registers(sasm);
 618         int call_offset = __ call_RT(noreg, noreg,
 619                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
 620         oop_maps = new OopMapSet();
 621         oop_maps->add_gc_map(call_offset, oop_map);
 622 
 623         // Now restore all the live registers
 624         restore_live_registers(sasm);
 625 
 626         __ ret();
 627         __ delayed()->restore();
 628       }
 629       break;
 630 
 631     case throw_range_check_failed_id:
 632       { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
 633         // G4: index
 634         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
 635       }
 636       break;
 637 
 638     case throw_index_exception_id:
 639       { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
 640         // G4: index
 641         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
 642       }
 643       break;
 644 
 645     case throw_div0_exception_id:
 646       { __ set_info("throw_div0_exception", dont_gc_arguments);
 647         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 648       }
 649       break;
 650 
 651     case throw_null_pointer_exception_id:
 652       { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
 653         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 654       }
 655       break;
 656 
 657     case handle_exception_id:
 658       {
 659         __ set_info("handle_exception", dont_gc_arguments);
 660         // make a frame and preserve the caller's caller-save registers
 661 
 662         oop_maps = new OopMapSet();
 663         OopMap* oop_map = save_live_registers(sasm);
 664         __ mov(Oexception->after_save(),  Oexception);
 665         __ mov(Oissuing_pc->after_save(), Oissuing_pc);
 666         generate_handle_exception(sasm, oop_maps, oop_map);
 667       }
 668       break;
 669 
 670     case unwind_exception_id:
 671       {
 672         // O0: exception
 673         // I7: address of call to this method
 674 
 675         __ set_info("unwind_exception", dont_gc_arguments);
 676         __ mov(Oexception, Oexception->after_save());
 677         __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
 678 
 679         __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
 680                         G2_thread, Oissuing_pc->after_save());
 681         __ verify_not_null_oop(Oexception->after_save());
 682 
 683         // Restore SP from L7 if the exception PC is a MethodHandle call site.
 684         __ mov(O0, G5);  // Save the target address.
 685         __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
 686         __ tst(L0);  // Condition codes are preserved over the restore.
 687         __ restore();
 688 
 689         __ jmp(G5, 0);
 690         __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7, SP);  // Restore SP if required.
 691       }
 692       break;
 693 
 694     case throw_array_store_exception_id:
 695       {
 696         __ set_info("throw_array_store_exception", dont_gc_arguments);
 697         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false);
 698       }
 699       break;
 700 
 701     case throw_class_cast_exception_id:
 702       {
 703         // G4: object
 704         __ set_info("throw_class_cast_exception", dont_gc_arguments);
 705         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 706       }
 707       break;
 708 
 709     case throw_incompatible_class_change_error_id:
 710       {
 711         __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
 712         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 713       }
 714       break;
 715 
 716     case slow_subtype_check_id:
 717       { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
 718         // Arguments :
 719         //
 720         //      ret  : G3
 721         //      sub  : G3, argument, destroyed
 722         //      super: G1, argument, not changed
 723         //      raddr: O7, blown by call
 724         Label miss;
 725 
 726         __ save_frame(0);               // Blow no registers!
 727 
 728         __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
 729 
 730         __ mov(1, G3);
 731         __ ret();                       // Result in G5 is 'true'
 732         __ delayed()->restore();        // free copy or add can go here
 733 
 734         __ bind(miss);
 735         __ mov(0, G3);
 736         __ ret();                       // Result in G5 is 'false'
 737         __ delayed()->restore();        // free copy or add can go here
 738       }
 739 
 740     case monitorenter_nofpu_id:
 741     case monitorenter_id:
 742       { // G4: object
 743         // G5: lock address
 744         __ set_info("monitorenter", dont_gc_arguments);
 745 
 746         int save_fpu_registers = (id == monitorenter_id);
 747         // make a frame and preserve the caller's caller-save registers
 748         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 749 
 750         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
 751 
 752         oop_maps = new OopMapSet();
 753         oop_maps->add_gc_map(call_offset, oop_map);
 754         restore_live_registers(sasm, save_fpu_registers);
 755 
 756         __ ret();
 757         __ delayed()->restore();
 758       }
 759       break;
 760 
 761     case monitorexit_nofpu_id:
 762     case monitorexit_id:
 763       { // G4: lock address
 764         // note: really a leaf routine but must setup last java sp
 765         //       => use call_RT for now (speed can be improved by
 766         //       doing last java sp setup manually)
 767         __ set_info("monitorexit", dont_gc_arguments);
 768 
 769         int save_fpu_registers = (id == monitorexit_id);
 770         // make a frame and preserve the caller's caller-save registers
 771         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 772 
 773         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
 774 
 775         oop_maps = new OopMapSet();
 776         oop_maps->add_gc_map(call_offset, oop_map);
 777         restore_live_registers(sasm, save_fpu_registers);
 778 
 779         __ ret();
 780         __ delayed()->restore();
 781 
 782       }
 783       break;
 784 
 785     case access_field_patching_id:
 786       { __ set_info("access_field_patching", dont_gc_arguments);
 787         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
 788       }
 789       break;
 790 
 791     case load_klass_patching_id:
 792       { __ set_info("load_klass_patching", dont_gc_arguments);
 793         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
 794       }
 795       break;
 796 
 797     case jvmti_exception_throw_id:
 798       { // Oexception : exception
 799         __ set_info("jvmti_exception_throw", dont_gc_arguments);
 800         oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0);
 801       }
 802       break;
 803 
 804     case dtrace_object_alloc_id:
 805       { // O0: object
 806         __ set_info("dtrace_object_alloc", dont_gc_arguments);
 807         // we can't gc here so skip the oopmap but make sure that all
 808         // the live registers get saved.
 809         save_live_registers(sasm);
 810 
 811         __ save_thread(L7_thread_cache);
 812         __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
 813                 relocInfo::runtime_call_type);
 814         __ delayed()->mov(I0, O0);
 815         __ restore_thread(L7_thread_cache);
 816 
 817         restore_live_registers(sasm);
 818         __ ret();
 819         __ delayed()->restore();
 820       }
 821       break;
 822 
 823 #ifndef SERIALGC
 824     case g1_pre_barrier_slow_id:
 825       { // G4: previous value of memory
 826         BarrierSet* bs = Universe::heap()->barrier_set();
 827         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 828           __ save_frame(0);
 829           __ set((int)id, O1);
 830           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 831           __ should_not_reach_here();
 832           break;
 833         }
 834 
 835         __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
 836 
 837         Register pre_val = G4;
 838         Register tmp  = G1_scratch;
 839         Register tmp2 = G3_scratch;
 840 
 841         Label refill, restart;
 842         bool with_frame = false; // I don't know if we can do with-frame.
 843         int satb_q_index_byte_offset =
 844           in_bytes(JavaThread::satb_mark_queue_offset() +
 845                    PtrQueue::byte_offset_of_index());
 846         int satb_q_buf_byte_offset =
 847           in_bytes(JavaThread::satb_mark_queue_offset() +
 848                    PtrQueue::byte_offset_of_buf());
 849         __ bind(restart);
 850         __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
 851 
 852         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false,
 853                           Assembler::pn, tmp, refill);
 854 
 855         // If the branch is taken, no harm in executing this in the delay slot.
 856         __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
 857         __ sub(tmp, oopSize, tmp);
 858 
 859         __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
 860         // Use return-from-leaf
 861         __ retl();
 862         __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
 863 
 864         __ bind(refill);
 865         __ save_frame(0);
 866 
 867         __ mov(pre_val, L0);
 868         __ mov(tmp,     L1);
 869         __ mov(tmp2,    L2);
 870 
 871         __ call_VM_leaf(L7_thread_cache,
 872                         CAST_FROM_FN_PTR(address,
 873                                          SATBMarkQueueSet::handle_zero_index_for_thread),
 874                                          G2_thread);
 875 
 876         __ mov(L0, pre_val);
 877         __ mov(L1, tmp);
 878         __ mov(L2, tmp2);
 879 
 880         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 881         __ delayed()->restore();
 882       }
 883       break;
 884 
 885     case g1_post_barrier_slow_id:
 886       {
 887         BarrierSet* bs = Universe::heap()->barrier_set();
 888         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 889           __ save_frame(0);
 890           __ set((int)id, O1);
 891           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 892           __ should_not_reach_here();
 893           break;
 894         }
 895 
 896         __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
 897 
 898         Register addr = G4;
 899         Register cardtable = G5;
 900         Register tmp  = G1_scratch;
 901         Register tmp2 = G3_scratch;
 902         jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
 903 
 904         Label not_already_dirty, restart, refill;
 905 
 906 #ifdef _LP64
 907         __ srlx(addr, CardTableModRefBS::card_shift, addr);
 908 #else
 909         __ srl(addr, CardTableModRefBS::card_shift, addr);
 910 #endif
 911 
 912         AddressLiteral rs(byte_map_base);
 913         __ set(rs, cardtable);         // cardtable := <card table base>
 914         __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
 915 
 916         __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
 917                           tmp, not_already_dirty);
 918         // Get cardtable + tmp into a reg by itself -- useful in the take-the-branch
 919         // case, harmless if not.
 920         __ delayed()->add(addr, cardtable, tmp2);
 921 
 922         // We didn't take the branch, so we're already dirty: return.
 923         // Use return-from-leaf
 924         __ retl();
 925         __ delayed()->nop();
 926 
 927         // Not dirty.
 928         __ bind(not_already_dirty);
 929         // First, dirty it.
 930         __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
 931 
 932         Register tmp3 = cardtable;
 933         Register tmp4 = tmp;
 934 
 935         // these registers are now dead
 936         addr = cardtable = tmp = noreg;
 937 
 938         int dirty_card_q_index_byte_offset =
 939           in_bytes(JavaThread::dirty_card_queue_offset() +
 940                    PtrQueue::byte_offset_of_index());
 941         int dirty_card_q_buf_byte_offset =
 942           in_bytes(JavaThread::dirty_card_queue_offset() +
 943                    PtrQueue::byte_offset_of_buf());
 944         __ bind(restart);
 945         __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
 946 
 947         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
 948                           tmp3, refill);
 949         // If the branch is taken, no harm in executing this in the delay slot.
 950         __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
 951         __ sub(tmp3, oopSize, tmp3);
 952 
 953         __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
 954         // Use return-from-leaf
 955         __ retl();
 956         __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
 957 
 958         __ bind(refill);
 959         __ save_frame(0);
 960 
 961         __ mov(tmp2, L0);
 962         __ mov(tmp3, L1);
 963         __ mov(tmp4, L2);
 964 
 965         __ call_VM_leaf(L7_thread_cache,
 966                         CAST_FROM_FN_PTR(address,
 967                                          DirtyCardQueueSet::handle_zero_index_for_thread),
 968                                          G2_thread);
 969 
 970         __ mov(L0, tmp2);
 971         __ mov(L1, tmp3);
 972         __ mov(L2, tmp4);
 973 
 974         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 975         __ delayed()->restore();
 976       }
 977       break;
 978 #endif // !SERIALGC
 979 
 980     default:
 981       { __ set_info("unimplemented entry", dont_gc_arguments);
 982         __ save_frame(0);
 983         __ set((int)id, O1);
 984         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
 985         __ should_not_reach_here();
 986       }
 987       break;
 988   }
 989   return oop_maps;
 990 }
 991 
 992 
 993 void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) {
 994   Label no_deopt;
 995 
 996   __ verify_not_null_oop(Oexception);
 997 
 998   // save the exception and issuing pc in the thread
 999   __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1000   __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1001 
1002   // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map)
1003   __ mov(I7, L0);
1004   __ mov(Oissuing_pc, I7);
1005   __ sub(I7, frame::pc_return_offset, I7);
1006   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1007 
1008   // Note: if nmethod has been deoptimized then regardless of
1009   // whether it had a handler or not we will deoptimize
1010   // by entering the deopt blob with a pending exception.
1011 
1012 #ifdef ASSERT
1013   Label done;
1014   __ tst(O0);
1015   __ br(Assembler::notZero, false, Assembler::pn, done);
1016   __ delayed()->nop();
1017   __ stop("should have found address");
1018   __ bind(done);
1019 #endif
1020 
1021   // restore the registers that were saved at the beginning and jump to the exception handler.
1022   restore_live_registers(sasm);
1023 
1024   __ jmp(O0, 0);
1025   __ delayed()->restore();
1026 
1027   oop_maps->add_gc_map(call_offset, oop_map);
1028 }
1029 
1030 
1031 #undef __
1032 
1033 #define __ masm->