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  */
  24 
  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"
  39 
  40 // Implementation of StubAssembler
  41 
  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);
  47 
  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);
  52 
  53   // debugging support
  54   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
  55 
  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();
  69 
  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(Gtemp, false, pt, L);
  75     delayed()->nop();
  76     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
  77     st_ptr(G0, vm_result_addr);
  78     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
  79     st_ptr(G0, vm_result_addr_2);
  80 
  81     if (frame_size() == no_frame_size) {
  82       // we use O7 linkage so that forward_exception_entry has the issuing PC
  83       call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type);
  84       delayed()->restore();
  85     } else if (_stub_id == Runtime1::forward_exception_id) {
  86       should_not_reach_here();
  87     } else {
  88       AddressLiteral exc(Runtime1::entry_for(Runtime1::forward_exception_id));
  89       jump_to(exc, G4);
  90       delayed()->nop();
  91     }
  92     bind(L);
  93   }
  94 
  95   // get oop result if there is one and reset the value in the thread
  96   if (oop_result1->is_valid()) {                    // get oop result if there is one and reset it in the thread
  97     get_vm_result  (oop_result1);
  98   } else {
  99     // be a little paranoid and clear the result
 100     Address vm_result_addr(G2_thread, JavaThread::vm_result_offset());
 101     st_ptr(G0, vm_result_addr);
 102   }
 103 
 104   if (oop_result2->is_valid()) {
 105     get_vm_result_2(oop_result2);
 106   } else {
 107     // be a little paranoid and clear the result
 108     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
 109     st_ptr(G0, vm_result_addr_2);
 110   }
 111 
 112   return call_offset;
 113 }
 114 
 115 
 116 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1) {
 117   // O0 is reserved for the thread
 118   mov(arg1, O1);
 119   return call_RT(oop_result1, oop_result2, entry, 1);
 120 }
 121 
 122 
 123 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2) {
 124   // O0 is reserved for the thread
 125   mov(arg1, O1);
 126   mov(arg2, O2); assert(arg2 != O1, "smashed argument");
 127   return call_RT(oop_result1, oop_result2, entry, 2);
 128 }
 129 
 130 
 131 int StubAssembler::call_RT(Register oop_result1, Register oop_result2, address entry, Register arg1, Register arg2, Register arg3) {
 132   // O0 is reserved for the thread
 133   mov(arg1, O1);
 134   mov(arg2, O2); assert(arg2 != O1,               "smashed argument");
 135   mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
 136   return call_RT(oop_result1, oop_result2, entry, 3);
 137 }
 138 
 139 
 140 // Implementation of Runtime1
 141 
 142 #define __ sasm->
 143 
 144 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
 145 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
 146 static int reg_save_size_in_words;
 147 static int frame_size_in_bytes = -1;
 148 
 149 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
 150   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 151          "mismatch in calculation");
 152   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 153   int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
 154   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
 155 
 156   int i;
 157   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 158     Register r = as_Register(i);
 159     if (r == G1 || r == G3 || r == G4 || r == G5) {
 160       int sp_offset = cpu_reg_save_offsets[i];
 161       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 162                                 r->as_VMReg());
 163     }
 164   }
 165 
 166   if (save_fpu_registers) {
 167     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 168       FloatRegister r = as_FloatRegister(i);
 169       int sp_offset = fpu_reg_save_offsets[i];
 170       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 171                                 r->as_VMReg());
 172     }
 173   }
 174   return oop_map;
 175 }
 176 
 177 static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
 178   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 179          "mismatch in calculation");
 180   __ save_frame_c1(frame_size_in_bytes);
 181 
 182   // Record volatile registers as callee-save values in an OopMap so their save locations will be
 183   // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
 184   // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers
 185   // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
 186   // (as the stub's I's) when the runtime routine called by the stub creates its frame.
 187   // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
 188 
 189   int i;
 190   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 191     Register r = as_Register(i);
 192     if (r == G1 || r == G3 || r == G4 || r == G5) {
 193       int sp_offset = cpu_reg_save_offsets[i];
 194       __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 195     }
 196   }
 197 
 198   if (save_fpu_registers) {
 199     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 200       FloatRegister r = as_FloatRegister(i);
 201       int sp_offset = fpu_reg_save_offsets[i];
 202       __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 203     }
 204   }
 205 
 206   return generate_oop_map(sasm, save_fpu_registers);
 207 }
 208 
 209 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 210   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
 211     Register r = as_Register(i);
 212     if (r == G1 || r == G3 || r == G4 || r == G5) {
 213       __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 214     }
 215   }
 216 
 217   if (restore_fpu_registers) {
 218     for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
 219       FloatRegister r = as_FloatRegister(i);
 220       __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 221     }
 222   }
 223 }
 224 
 225 
 226 void Runtime1::initialize_pd() {
 227   // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
 228   //
 229   // A stub routine will have a frame that is at least large enough to hold
 230   // a register window save area (obviously) and the volatile g registers
 231   // and floating registers. A user of save_live_registers can have a frame
 232   // that has more scratch area in it (although typically they will use L-regs).
 233   // in that case the frame will look like this (stack growing down)
 234   //
 235   // FP -> |             |
 236   //       | scratch mem |
 237   //       |   "      "  |
 238   //       --------------
 239   //       | float regs  |
 240   //       |   "    "    |
 241   //       ---------------
 242   //       | G regs      |
 243   //       | "  "        |
 244   //       ---------------
 245   //       | abi reg.    |
 246   //       | window save |
 247   //       | area        |
 248   // SP -> ---------------
 249   //
 250   int i;
 251   int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned
 252 
 253   // only G int registers are saved explicitly; others are found in register windows
 254   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 255     Register r = as_Register(i);
 256     if (r == G1 || r == G3 || r == G4 || r == G5) {
 257       cpu_reg_save_offsets[i] = sp_offset;
 258       sp_offset++;
 259     }
 260   }
 261 
 262   // all float registers are saved explicitly
 263   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
 264   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 265     fpu_reg_save_offsets[i] = sp_offset;
 266     sp_offset++;
 267   }
 268   reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
 269   // this should match assembler::total_frame_size_in_bytes, which
 270   // isn't callable from this context.  It's checked by an assert when
 271   // it's used though.
 272   frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
 273 }
 274 
 275 
 276 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 277   // make a frame and preserve the caller's caller-save registers
 278   OopMap* oop_map = save_live_registers(sasm);
 279   int call_offset;
 280   if (!has_argument) {
 281     call_offset = __ call_RT(noreg, noreg, target);
 282   } else {
 283     call_offset = __ call_RT(noreg, noreg, target, G4);
 284   }
 285   OopMapSet* oop_maps = new OopMapSet();
 286   oop_maps->add_gc_map(call_offset, oop_map);
 287 
 288   __ should_not_reach_here();
 289   return oop_maps;
 290 }
 291 
 292 
 293 OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
 294                                         Register arg1, Register arg2, Register arg3) {
 295   // make a frame and preserve the caller's caller-save registers
 296   OopMap* oop_map = save_live_registers(sasm);
 297 
 298   int call_offset;
 299   if (arg1 == noreg) {
 300     call_offset = __ call_RT(result, noreg, target);
 301   } else if (arg2 == noreg) {
 302     call_offset = __ call_RT(result, noreg, target, arg1);
 303   } else if (arg3 == noreg) {
 304     call_offset = __ call_RT(result, noreg, target, arg1, arg2);
 305   } else {
 306     call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
 307   }
 308   OopMapSet* oop_maps = NULL;
 309 
 310   oop_maps = new OopMapSet();
 311   oop_maps->add_gc_map(call_offset, oop_map);
 312   restore_live_registers(sasm);
 313 
 314   __ ret();
 315   __ delayed()->restore();
 316 
 317   return oop_maps;
 318 }
 319 
 320 
 321 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 322   // make a frame and preserve the caller's caller-save registers
 323   OopMap* oop_map = save_live_registers(sasm);
 324 
 325   // call the runtime patching routine, returns non-zero if nmethod got deopted.
 326   int call_offset = __ call_RT(noreg, noreg, target);
 327   OopMapSet* oop_maps = new OopMapSet();
 328   oop_maps->add_gc_map(call_offset, oop_map);
 329 
 330   // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
 331   // deoptimization handler entry that will cause re-execution of the current bytecode
 332   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 333   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 334 
 335   Label no_deopt;
 336   __ tst(O0);
 337   __ brx(Assembler::equal, false, Assembler::pt, no_deopt);
 338   __ delayed()->nop();
 339 
 340   // return to the deoptimization handler entry for unpacking and rexecute
 341   // if we simply returned the we'd deopt as if any call we patched had just
 342   // returned.
 343 
 344   restore_live_registers(sasm);
 345 
 346   AddressLiteral dest(deopt_blob->unpack_with_reexecution());
 347   __ jump_to(dest, O0);
 348   __ delayed()->restore();
 349 
 350   __ bind(no_deopt);
 351   restore_live_registers(sasm);
 352   __ ret();
 353   __ delayed()->restore();
 354 
 355   return oop_maps;
 356 }
 357 
 358 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 359 
 360   OopMapSet* oop_maps = NULL;
 361   // for better readability
 362   const bool must_gc_arguments = true;
 363   const bool dont_gc_arguments = false;
 364 
 365   // stub code & info for the different stubs
 366   switch (id) {
 367     case forward_exception_id:
 368       {
 369         oop_maps = generate_handle_exception(id, sasm);
 370       }
 371       break;
 372 
 373     case new_instance_id:
 374     case fast_new_instance_id:
 375     case fast_new_instance_init_check_id:
 376       {
 377         Register G5_klass = G5; // Incoming
 378         Register O0_obj   = O0; // Outgoing
 379 
 380         if (id == new_instance_id) {
 381           __ set_info("new_instance", dont_gc_arguments);
 382         } else if (id == fast_new_instance_id) {
 383           __ set_info("fast new_instance", dont_gc_arguments);
 384         } else {
 385           assert(id == fast_new_instance_init_check_id, "bad StubID");
 386           __ set_info("fast new_instance init check", dont_gc_arguments);
 387         }
 388 
 389         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 390             UseTLAB && FastTLABRefill) {
 391           Label slow_path;
 392           Register G1_obj_size = G1;
 393           Register G3_t1 = G3;
 394           Register G4_t2 = G4;
 395           assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
 396 
 397           // Push a frame since we may do dtrace notification for the
 398           // allocation which requires calling out and we don't want
 399           // to stomp the real return address.
 400           __ save_frame(0);
 401 
 402           if (id == fast_new_instance_init_check_id) {
 403             // make sure the klass is initialized
 404             __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);
 405             __ cmp(G3_t1, instanceKlass::fully_initialized);
 406             __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
 407             __ delayed()->nop();
 408           }
 409 #ifdef ASSERT
 410           // assert object can be fast path allocated
 411           {
 412             Label ok, not_ok;
 413           __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);
 414           __ cmp(G1_obj_size, 0);  // make sure it's an instance (LH > 0)
 415           __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);
 416           __ delayed()->nop();
 417           __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
 418           __ br(Assembler::zero, false, Assembler::pn, ok);
 419           __ delayed()->nop();
 420           __ bind(not_ok);
 421           __ stop("assert(can be fast path allocated)");
 422           __ should_not_reach_here();
 423           __ bind(ok);
 424           }
 425 #endif // ASSERT
 426           // if we got here then the TLAB allocation failed, so try
 427           // refilling the TLAB or allocating directly from eden.
 428           Label retry_tlab, try_eden;
 429           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
 430 
 431           __ bind(retry_tlab);
 432 
 433           // get the instance size
 434           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 435 
 436           __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
 437 
 438           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 439           __ verify_oop(O0_obj);
 440           __ mov(O0, I0);
 441           __ ret();
 442           __ delayed()->restore();
 443 
 444           __ bind(try_eden);
 445           // get the instance size
 446           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 447           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
 448           __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
 449 
 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     case counter_overflow_id:
 469         // G4 contains bci, G5 contains method
 470       oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
 471       break;
 472 
 473     case new_type_array_id:
 474     case new_object_array_id:
 475       {
 476         Register G5_klass = G5; // Incoming
 477         Register G4_length = G4; // Incoming
 478         Register O0_obj   = O0; // Outgoing
 479 
 480         Address klass_lh(G5_klass, ((klassOopDesc::header_size() * HeapWordSize)
 481                                     + Klass::layout_helper_offset_in_bytes()));
 482         assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 483         assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
 484         // Use this offset to pick out an individual byte of the layout_helper:
 485         const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
 486                                                  - Klass::_lh_header_size_shift / BitsPerByte);
 487 
 488         if (id == new_type_array_id) {
 489           __ set_info("new_type_array", dont_gc_arguments);
 490         } else {
 491           __ set_info("new_object_array", dont_gc_arguments);
 492         }
 493 
 494 #ifdef ASSERT
 495         // assert object type is really an array of the proper kind
 496         {
 497           Label ok;
 498           Register G3_t1 = G3;
 499           __ ld(klass_lh, G3_t1);
 500           __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
 501           int tag = ((id == new_type_array_id)
 502                      ? Klass::_lh_array_tag_type_value
 503                      : Klass::_lh_array_tag_obj_value);
 504           __ cmp(G3_t1, tag);
 505           __ brx(Assembler::equal, false, Assembler::pt, ok);
 506           __ delayed()->nop();
 507           __ stop("assert(is an array klass)");
 508           __ should_not_reach_here();
 509           __ bind(ok);
 510         }
 511 #endif // ASSERT
 512 
 513         if (UseTLAB && FastTLABRefill) {
 514           Label slow_path;
 515           Register G1_arr_size = G1;
 516           Register G3_t1 = G3;
 517           Register O1_t2 = O1;
 518           assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
 519 
 520           // check that array length is small enough for fast path
 521           __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
 522           __ cmp(G4_length, G3_t1);
 523           __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
 524           __ delayed()->nop();
 525 
 526           // if we got here then the TLAB allocation failed, so try
 527           // refilling the TLAB or allocating directly from eden.
 528           Label retry_tlab, try_eden;
 529           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
 530 
 531           __ bind(retry_tlab);
 532 
 533           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 534           __ ld(klass_lh, G3_t1);
 535           __ sll(G4_length, G3_t1, G1_arr_size);
 536           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 537           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 538           __ add(G1_arr_size, G3_t1, G1_arr_size);
 539           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
 540           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 541 
 542           __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
 543 
 544           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 545           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 546           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 547           __ add(O0_obj, G3_t1, G3_t1);       // body start
 548           __ initialize_body(G3_t1, O1_t2);
 549           __ verify_oop(O0_obj);
 550           __ retl();
 551           __ delayed()->nop();
 552 
 553           __ bind(try_eden);
 554           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 555           __ ld(klass_lh, G3_t1);
 556           __ sll(G4_length, G3_t1, G1_arr_size);
 557           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 558           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 559           __ add(G1_arr_size, G3_t1, G1_arr_size);
 560           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
 561           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 562 
 563           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
 564           __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
 565 
 566           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 567           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 568           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 569           __ add(O0_obj, G3_t1, G3_t1);       // body start
 570           __ initialize_body(G3_t1, O1_t2);
 571           __ verify_oop(O0_obj);
 572           __ retl();
 573           __ delayed()->nop();
 574 
 575           __ bind(slow_path);
 576         }
 577 
 578         if (id == new_type_array_id) {
 579           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
 580         } else {
 581           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
 582         }
 583         // I0 -> O0: new array
 584       }
 585       break;
 586 
 587     case new_multi_array_id:
 588       { // O0: klass
 589         // O1: rank
 590         // O2: address of 1st dimension
 591         __ set_info("new_multi_array", dont_gc_arguments);
 592         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
 593         // I0 -> O0: new multi array
 594       }
 595       break;
 596 
 597     case register_finalizer_id:
 598       {
 599         __ set_info("register_finalizer", dont_gc_arguments);
 600 
 601         // load the klass and check the has finalizer flag
 602         Label register_finalizer;
 603         Register t = O1;
 604         __ load_klass(O0, t);
 605         __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t);
 606         __ set(JVM_ACC_HAS_FINALIZER, G3);
 607         __ andcc(G3, t, G0);
 608         __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
 609         __ delayed()->nop();
 610 
 611         // do a leaf return
 612         __ retl();
 613         __ delayed()->nop();
 614 
 615         __ bind(register_finalizer);
 616         OopMap* oop_map = save_live_registers(sasm);
 617         int call_offset = __ call_RT(noreg, noreg,
 618                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
 619         oop_maps = new OopMapSet();
 620         oop_maps->add_gc_map(call_offset, oop_map);
 621 
 622         // Now restore all the live registers
 623         restore_live_registers(sasm);
 624 
 625         __ ret();
 626         __ delayed()->restore();
 627       }
 628       break;
 629 
 630     case throw_range_check_failed_id:
 631       { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
 632         // G4: index
 633         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
 634       }
 635       break;
 636 
 637     case throw_index_exception_id:
 638       { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
 639         // G4: index
 640         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
 641       }
 642       break;
 643 
 644     case throw_div0_exception_id:
 645       { __ set_info("throw_div0_exception", dont_gc_arguments);
 646         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 647       }
 648       break;
 649 
 650     case throw_null_pointer_exception_id:
 651       { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
 652         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 653       }
 654       break;
 655 
 656     case handle_exception_id:
 657       { __ set_info("handle_exception", dont_gc_arguments);
 658         oop_maps = generate_handle_exception(id, sasm);
 659       }
 660       break;
 661 
 662     case handle_exception_from_callee_id:
 663       { __ set_info("handle_exception_from_callee", dont_gc_arguments);
 664         oop_maps = generate_handle_exception(id, sasm);
 665       }
 666       break;
 667 
 668     case unwind_exception_id:
 669       {
 670         // O0: exception
 671         // I7: address of call to this method
 672 
 673         __ set_info("unwind_exception", dont_gc_arguments);
 674         __ mov(Oexception, Oexception->after_save());
 675         __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
 676 
 677         __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
 678                         G2_thread, Oissuing_pc->after_save());
 679         __ verify_not_null_oop(Oexception->after_save());
 680 
 681         // Restore SP from L7 if the exception PC is a method handle call site.
 682         __ mov(O0, G5);  // Save the target address.
 683         __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
 684         __ tst(L0);  // Condition codes are preserved over the restore.
 685         __ restore();
 686 
 687         __ jmp(G5, 0);
 688         __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
 689       }
 690       break;
 691 
 692     case throw_array_store_exception_id:
 693       {
 694         __ set_info("throw_array_store_exception", dont_gc_arguments);
 695         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
 696       }
 697       break;
 698 
 699     case throw_class_cast_exception_id:
 700       {
 701         // G4: object
 702         __ set_info("throw_class_cast_exception", dont_gc_arguments);
 703         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 704       }
 705       break;
 706 
 707     case throw_incompatible_class_change_error_id:
 708       {
 709         __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
 710         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 711       }
 712       break;
 713 
 714     case slow_subtype_check_id:
 715       { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
 716         // Arguments :
 717         //
 718         //      ret  : G3
 719         //      sub  : G3, argument, destroyed
 720         //      super: G1, argument, not changed
 721         //      raddr: O7, blown by call
 722         Label miss;
 723 
 724         __ save_frame(0);               // Blow no registers!
 725 
 726         __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
 727 
 728         __ mov(1, G3);
 729         __ ret();                       // Result in G5 is 'true'
 730         __ delayed()->restore();        // free copy or add can go here
 731 
 732         __ bind(miss);
 733         __ mov(0, G3);
 734         __ ret();                       // Result in G5 is 'false'
 735         __ delayed()->restore();        // free copy or add can go here
 736       }
 737 
 738     case monitorenter_nofpu_id:
 739     case monitorenter_id:
 740       { // G4: object
 741         // G5: lock address
 742         __ set_info("monitorenter", dont_gc_arguments);
 743 
 744         int save_fpu_registers = (id == monitorenter_id);
 745         // make a frame and preserve the caller's caller-save registers
 746         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 747 
 748         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
 749 
 750         oop_maps = new OopMapSet();
 751         oop_maps->add_gc_map(call_offset, oop_map);
 752         restore_live_registers(sasm, save_fpu_registers);
 753 
 754         __ ret();
 755         __ delayed()->restore();
 756       }
 757       break;
 758 
 759     case monitorexit_nofpu_id:
 760     case monitorexit_id:
 761       { // G4: lock address
 762         // note: really a leaf routine but must setup last java sp
 763         //       => use call_RT for now (speed can be improved by
 764         //       doing last java sp setup manually)
 765         __ set_info("monitorexit", dont_gc_arguments);
 766 
 767         int save_fpu_registers = (id == monitorexit_id);
 768         // make a frame and preserve the caller's caller-save registers
 769         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 770 
 771         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
 772 
 773         oop_maps = new OopMapSet();
 774         oop_maps->add_gc_map(call_offset, oop_map);
 775         restore_live_registers(sasm, save_fpu_registers);
 776 
 777         __ ret();
 778         __ delayed()->restore();
 779 
 780       }
 781       break;
 782 
 783     case access_field_patching_id:
 784       { __ set_info("access_field_patching", dont_gc_arguments);
 785         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
 786       }
 787       break;
 788 
 789     case load_klass_patching_id:
 790       { __ set_info("load_klass_patching", dont_gc_arguments);
 791         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
 792       }
 793       break;
 794 
 795     case jvmti_exception_throw_id:
 796       { // Oexception : exception
 797         __ set_info("jvmti_exception_throw", dont_gc_arguments);
 798         oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0);
 799       }
 800       break;
 801 
 802     case dtrace_object_alloc_id:
 803       { // O0: object
 804         __ set_info("dtrace_object_alloc", dont_gc_arguments);
 805         // we can't gc here so skip the oopmap but make sure that all
 806         // the live registers get saved.
 807         save_live_registers(sasm);
 808 
 809         __ save_thread(L7_thread_cache);
 810         __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
 811                 relocInfo::runtime_call_type);
 812         __ delayed()->mov(I0, O0);
 813         __ restore_thread(L7_thread_cache);
 814 
 815         restore_live_registers(sasm);
 816         __ ret();
 817         __ delayed()->restore();
 818       }
 819       break;
 820 
 821 #ifndef SERIALGC
 822     case g1_pre_barrier_slow_id:
 823       { // G4: previous value of memory
 824         BarrierSet* bs = Universe::heap()->barrier_set();
 825         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 826           __ save_frame(0);
 827           __ set((int)id, O1);
 828           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 829           __ should_not_reach_here();
 830           break;
 831         }
 832 
 833         __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
 834 
 835         Register pre_val = G4;
 836         Register tmp  = G1_scratch;
 837         Register tmp2 = G3_scratch;
 838 
 839         Label refill, restart;
 840         bool with_frame = false; // I don't know if we can do with-frame.
 841         int satb_q_index_byte_offset =
 842           in_bytes(JavaThread::satb_mark_queue_offset() +
 843                    PtrQueue::byte_offset_of_index());
 844         int satb_q_buf_byte_offset =
 845           in_bytes(JavaThread::satb_mark_queue_offset() +
 846                    PtrQueue::byte_offset_of_buf());
 847         __ bind(restart);
 848         __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
 849 
 850         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false,
 851                           Assembler::pn, tmp, refill);
 852 
 853         // If the branch is taken, no harm in executing this in the delay slot.
 854         __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
 855         __ sub(tmp, oopSize, tmp);
 856 
 857         __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
 858         // Use return-from-leaf
 859         __ retl();
 860         __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
 861 
 862         __ bind(refill);
 863         __ save_frame(0);
 864 
 865         __ mov(pre_val, L0);
 866         __ mov(tmp,     L1);
 867         __ mov(tmp2,    L2);
 868 
 869         __ call_VM_leaf(L7_thread_cache,
 870                         CAST_FROM_FN_PTR(address,
 871                                          SATBMarkQueueSet::handle_zero_index_for_thread),
 872                                          G2_thread);
 873 
 874         __ mov(L0, pre_val);
 875         __ mov(L1, tmp);
 876         __ mov(L2, tmp2);
 877 
 878         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 879         __ delayed()->restore();
 880       }
 881       break;
 882 
 883     case g1_post_barrier_slow_id:
 884       {
 885         BarrierSet* bs = Universe::heap()->barrier_set();
 886         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 887           __ save_frame(0);
 888           __ set((int)id, O1);
 889           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 890           __ should_not_reach_here();
 891           break;
 892         }
 893 
 894         __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
 895 
 896         Register addr = G4;
 897         Register cardtable = G5;
 898         Register tmp  = G1_scratch;
 899         Register tmp2 = G3_scratch;
 900         jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
 901 
 902         Label not_already_dirty, restart, refill;
 903 
 904 #ifdef _LP64
 905         __ srlx(addr, CardTableModRefBS::card_shift, addr);
 906 #else
 907         __ srl(addr, CardTableModRefBS::card_shift, addr);
 908 #endif
 909 
 910         AddressLiteral rs(byte_map_base);
 911         __ set(rs, cardtable);         // cardtable := <card table base>
 912         __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
 913 
 914         __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
 915                           tmp, not_already_dirty);
 916         // Get cardtable + tmp into a reg by itself -- useful in the take-the-branch
 917         // case, harmless if not.
 918         __ delayed()->add(addr, cardtable, tmp2);
 919 
 920         // We didn't take the branch, so we're already dirty: return.
 921         // Use return-from-leaf
 922         __ retl();
 923         __ delayed()->nop();
 924 
 925         // Not dirty.
 926         __ bind(not_already_dirty);
 927         // First, dirty it.
 928         __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
 929 
 930         Register tmp3 = cardtable;
 931         Register tmp4 = tmp;
 932 
 933         // these registers are now dead
 934         addr = cardtable = tmp = noreg;
 935 
 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());
 942         __ bind(restart);
 943         __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
 944 
 945         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
 946                           tmp3, refill);
 947         // If the branch is taken, no harm in executing this in the delay slot.
 948         __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
 949         __ sub(tmp3, oopSize, tmp3);
 950 
 951         __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
 952         // Use return-from-leaf
 953         __ retl();
 954         __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
 955 
 956         __ bind(refill);
 957         __ save_frame(0);
 958 
 959         __ mov(tmp2, L0);
 960         __ mov(tmp3, L1);
 961         __ mov(tmp4, L2);
 962 
 963         __ call_VM_leaf(L7_thread_cache,
 964                         CAST_FROM_FN_PTR(address,
 965                                          DirtyCardQueueSet::handle_zero_index_for_thread),
 966                                          G2_thread);
 967 
 968         __ mov(L0, tmp2);
 969         __ mov(L1, tmp3);
 970         __ mov(L2, tmp4);
 971 
 972         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 973         __ delayed()->restore();
 974       }
 975       break;
 976 #endif // !SERIALGC
 977 
 978     default:
 979       { __ set_info("unimplemented entry", dont_gc_arguments);
 980         __ save_frame(0);
 981         __ set((int)id, O1);
 982         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
 983         __ should_not_reach_here();
 984       }
 985       break;
 986   }
 987   return oop_maps;
 988 }
 989 
 990 
 991 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler* sasm) {
 992   __ block_comment("generate_handle_exception");
 993 
 994   // Save registers, if required.
 995   OopMapSet* oop_maps = new OopMapSet();
 996   OopMap* oop_map = NULL;
 997   switch (id) {
 998   case forward_exception_id:
 999     // We're handling an exception in the context of a compiled frame.
1000     // The registers have been saved in the standard places.  Perform
1001     // an exception lookup in the caller and dispatch to the handler
1002     // if found.  Otherwise unwind and dispatch to the callers
1003     // exception handler.
1004      oop_map = generate_oop_map(sasm, true);
1005 
1006      // transfer the pending exception to the exception_oop
1007      __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
1008      __ ld_ptr(Oexception, 0, G0);
1009      __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
1010      __ add(I7, frame::pc_return_offset, Oissuing_pc);
1011     break;
1012   case handle_exception_id:
1013     // At this point all registers MAY be live.
1014     oop_map = save_live_registers(sasm);
1015     __ mov(Oexception->after_save(),  Oexception);
1016     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
1017     break;
1018   case handle_exception_from_callee_id:
1019     // At this point all registers except exception oop (Oexception)
1020     // and exception pc (Oissuing_pc) are dead.
1021     oop_map = new OopMap(frame_size_in_bytes / sizeof(jint), 0);
1022     sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
1023     __ save_frame_c1(frame_size_in_bytes);
1024     __ mov(Oexception->after_save(),  Oexception);
1025     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
1026     break;
1027   default:  ShouldNotReachHere();
1028   }
1029 
1030   __ verify_not_null_oop(Oexception);
1031 
1032   // save the exception and issuing pc in the thread
1033   __ st_ptr(Oexception,  G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1034   __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1035 
1036   // use the throwing pc as the return address to lookup (has bci & oop map)
1037   __ mov(Oissuing_pc, I7);
1038   __ sub(I7, frame::pc_return_offset, I7);
1039   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1040   oop_maps->add_gc_map(call_offset, oop_map);
1041 
1042   // Note: if nmethod has been deoptimized then regardless of
1043   // whether it had a handler or not we will deoptimize
1044   // by entering the deopt blob with a pending exception.
1045 
1046   // Restore the registers that were saved at the beginning, remove
1047   // the frame and jump to the exception handler.
1048   switch (id) {
1049   case forward_exception_id:
1050   case handle_exception_id:
1051     restore_live_registers(sasm);
1052     __ jmp(O0, 0);
1053     __ delayed()->restore();
1054     break;
1055   case handle_exception_from_callee_id:
1056     // Restore SP from L7 if the exception PC is a method handle call site.
1057     __ mov(O0, G5);  // Save the target address.
1058     __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
1059     __ tst(L0);  // Condition codes are preserved over the restore.
1060     __ restore();
1061 
1062     __ jmp(G5, 0);  // jump to the exception handler
1063     __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
1064     break;
1065   default:  ShouldNotReachHere();
1066   }
1067 
1068   return oop_maps;
1069 }
1070 
1071 
1072 #undef __
1073 
1074 const char *Runtime1::pd_name_for_address(address entry) {
1075   return "<unknown function>";
1076 }