1 /*
   2  * Copyright (c) 1999, 2012, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  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/compiledICHolder.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 "utilities/macros.hpp"
  39 #include "vmreg_sparc.inline.hpp"
  40 
  41 // Implementation of StubAssembler
  42 
  43 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry_point, int number_of_arguments) {
  44   // for sparc changing the number of arguments doesn't change
  45   // anything about the frame size so we'll always lie and claim that
  46   // we are only passing 1 argument.
  47   set_num_rt_args(1);
  48 
  49   assert_not_delayed();
  50   // bang stack before going to runtime
  51   set(-os::vm_page_size() + STACK_BIAS, G3_scratch);
  52   st(G0, SP, G3_scratch);
  53 
  54   // debugging support
  55   assert(number_of_arguments >= 0   , "cannot have negative number of arguments");
  56 
  57   set_last_Java_frame(SP, noreg);
  58   if (VerifyThread)  mov(G2_thread, O0); // about to be smashed; pass early
  59   save_thread(L7_thread_cache);
  60   // do the call
  61   call(entry_point, relocInfo::runtime_call_type);
  62   if (!VerifyThread) {
  63     delayed()->mov(G2_thread, O0);  // pass thread as first argument
  64   } else {
  65     delayed()->nop();             // (thread already passed)
  66   }
  67   int call_offset = offset();  // offset of return address
  68   restore_thread(L7_thread_cache);
  69   reset_last_Java_frame();
  70 
  71   // check for pending exceptions
  72   { Label L;
  73     Address exception_addr(G2_thread, Thread::pending_exception_offset());
  74     ld_ptr(exception_addr, Gtemp);
  75     br_null_short(Gtemp, pt, L);
  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   // get second result if there is one and reset the value in the thread
 105   if (metadata_result->is_valid()) {
 106     get_vm_result_2  (metadata_result);
 107   } else {
 108     // be a little paranoid and clear the result
 109     Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset());
 110     st_ptr(G0, vm_result_addr_2);
 111   }
 112 
 113   return call_offset;
 114 }
 115 
 116 
 117 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
 118   // O0 is reserved for the thread
 119   mov(arg1, O1);
 120   return call_RT(oop_result1, metadata_result, entry, 1);
 121 }
 122 
 123 
 124 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
 125   // O0 is reserved for the thread
 126   mov(arg1, O1);
 127   mov(arg2, O2); assert(arg2 != O1, "smashed argument");
 128   return call_RT(oop_result1, metadata_result, entry, 2);
 129 }
 130 
 131 
 132 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
 133   // O0 is reserved for the thread
 134   mov(arg1, O1);
 135   mov(arg2, O2); assert(arg2 != O1,               "smashed argument");
 136   mov(arg3, O3); assert(arg3 != O1 && arg3 != O2, "smashed argument");
 137   return call_RT(oop_result1, metadata_result, entry, 3);
 138 }
 139 
 140 
 141 // Implementation of Runtime1
 142 
 143 #define __ sasm->
 144 
 145 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
 146 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
 147 static int reg_save_size_in_words;
 148 static int frame_size_in_bytes = -1;
 149 
 150 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
 151   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 152          "mismatch in calculation");
 153   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 154   int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
 155   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
 156 
 157   int i;
 158   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 159     Register r = as_Register(i);
 160     if (r == G1 || r == G3 || r == G4 || r == G5) {
 161       int sp_offset = cpu_reg_save_offsets[i];
 162       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 163                                 r->as_VMReg());
 164     }
 165   }
 166 
 167   if (save_fpu_registers) {
 168     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 169       FloatRegister r = as_FloatRegister(i);
 170       int sp_offset = fpu_reg_save_offsets[i];
 171       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 172                                 r->as_VMReg());
 173     }
 174   }
 175   return oop_map;
 176 }
 177 
 178 static OopMap* save_live_registers(StubAssembler* sasm, bool save_fpu_registers = true) {
 179   assert(frame_size_in_bytes == __ total_frame_size_in_bytes(reg_save_size_in_words),
 180          "mismatch in calculation");
 181   __ save_frame_c1(frame_size_in_bytes);
 182 
 183   // Record volatile registers as callee-save values in an OopMap so their save locations will be
 184   // propagated to the caller frame's RegisterMap during StackFrameStream construction (needed for
 185   // deoptimization; see compiledVFrame::create_stack_value).  The caller's I, L and O registers
 186   // are saved in register windows - I's and L's in the caller's frame and O's in the stub frame
 187   // (as the stub's I's) when the runtime routine called by the stub creates its frame.
 188   // OopMap frame sizes are in c2 stack slot sizes (sizeof(jint))
 189 
 190   int i;
 191   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 192     Register r = as_Register(i);
 193     if (r == G1 || r == G3 || r == G4 || r == G5) {
 194       int sp_offset = cpu_reg_save_offsets[i];
 195       __ st_ptr(r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 196     }
 197   }
 198 
 199   if (save_fpu_registers) {
 200     for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 201       FloatRegister r = as_FloatRegister(i);
 202       int sp_offset = fpu_reg_save_offsets[i];
 203       __ stf(FloatRegisterImpl::S, r, SP, (sp_offset * BytesPerWord) + STACK_BIAS);
 204     }
 205   }
 206 
 207   return generate_oop_map(sasm, save_fpu_registers);
 208 }
 209 
 210 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 211   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
 212     Register r = as_Register(i);
 213     if (r == G1 || r == G3 || r == G4 || r == G5) {
 214       __ ld_ptr(SP, (cpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 215     }
 216   }
 217 
 218   if (restore_fpu_registers) {
 219     for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
 220       FloatRegister r = as_FloatRegister(i);
 221       __ ldf(FloatRegisterImpl::S, SP, (fpu_reg_save_offsets[i] * BytesPerWord) + STACK_BIAS, r);
 222     }
 223   }
 224 }
 225 
 226 
 227 void Runtime1::initialize_pd() {
 228   // compute word offsets from SP at which live (non-windowed) registers are captured by stub routines
 229   //
 230   // A stub routine will have a frame that is at least large enough to hold
 231   // a register window save area (obviously) and the volatile g registers
 232   // and floating registers. A user of save_live_registers can have a frame
 233   // that has more scratch area in it (although typically they will use L-regs).
 234   // in that case the frame will look like this (stack growing down)
 235   //
 236   // FP -> |             |
 237   //       | scratch mem |
 238   //       |   "      "  |
 239   //       --------------
 240   //       | float regs  |
 241   //       |   "    "    |
 242   //       ---------------
 243   //       | G regs      |
 244   //       | "  "        |
 245   //       ---------------
 246   //       | abi reg.    |
 247   //       | window save |
 248   //       | area        |
 249   // SP -> ---------------
 250   //
 251   int i;
 252   int sp_offset = round_to(frame::register_save_words, 2); //  start doubleword aligned
 253 
 254   // only G int registers are saved explicitly; others are found in register windows
 255   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 256     Register r = as_Register(i);
 257     if (r == G1 || r == G3 || r == G4 || r == G5) {
 258       cpu_reg_save_offsets[i] = sp_offset;
 259       sp_offset++;
 260     }
 261   }
 262 
 263   // all float registers are saved explicitly
 264   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
 265   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 266     fpu_reg_save_offsets[i] = sp_offset;
 267     sp_offset++;
 268   }
 269   reg_save_size_in_words = sp_offset - frame::memory_parameter_word_sp_offset;
 270   // this should match assembler::total_frame_size_in_bytes, which
 271   // isn't callable from this context.  It's checked by an assert when
 272   // it's used though.
 273   frame_size_in_bytes = align_size_up(sp_offset * wordSize, 8);
 274 }
 275 
 276 
 277 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 278   // make a frame and preserve the caller's caller-save registers
 279   OopMap* oop_map = save_live_registers(sasm);
 280   int call_offset;
 281   if (!has_argument) {
 282     call_offset = __ call_RT(noreg, noreg, target);
 283   } else {
 284     call_offset = __ call_RT(noreg, noreg, target, G4);
 285   }
 286   OopMapSet* oop_maps = new OopMapSet();
 287   oop_maps->add_gc_map(call_offset, oop_map);
 288 
 289   __ should_not_reach_here();
 290   return oop_maps;
 291 }
 292 
 293 
 294 OopMapSet* Runtime1::generate_stub_call(StubAssembler* sasm, Register result, address target,
 295                                         Register arg1, Register arg2, Register arg3) {
 296   // make a frame and preserve the caller's caller-save registers
 297   OopMap* oop_map = save_live_registers(sasm);
 298 
 299   int call_offset;
 300   if (arg1 == noreg) {
 301     call_offset = __ call_RT(result, noreg, target);
 302   } else if (arg2 == noreg) {
 303     call_offset = __ call_RT(result, noreg, target, arg1);
 304   } else if (arg3 == noreg) {
 305     call_offset = __ call_RT(result, noreg, target, arg1, arg2);
 306   } else {
 307     call_offset = __ call_RT(result, noreg, target, arg1, arg2, arg3);
 308   }
 309   OopMapSet* oop_maps = NULL;
 310 
 311   oop_maps = new OopMapSet();
 312   oop_maps->add_gc_map(call_offset, oop_map);
 313   restore_live_registers(sasm);
 314 
 315   __ ret();
 316   __ delayed()->restore();
 317 
 318   return oop_maps;
 319 }
 320 
 321 
 322 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 323   // make a frame and preserve the caller's caller-save registers
 324   OopMap* oop_map = save_live_registers(sasm);
 325 
 326   // call the runtime patching routine, returns non-zero if nmethod got deopted.
 327   int call_offset = __ call_RT(noreg, noreg, target);
 328   OopMapSet* oop_maps = new OopMapSet();
 329   oop_maps->add_gc_map(call_offset, oop_map);
 330 
 331   // re-execute the patched instruction or, if the nmethod was deoptmized, return to the
 332   // deoptimization handler entry that will cause re-execution of the current bytecode
 333   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 334   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 335 
 336   Label no_deopt;
 337   __ br_null_short(O0, Assembler::pt, no_deopt);
 338 
 339   // return to the deoptimization handler entry for unpacking and rexecute
 340   // if we simply returned the we'd deopt as if any call we patched had just
 341   // returned.
 342 
 343   restore_live_registers(sasm);
 344 
 345   AddressLiteral dest(deopt_blob->unpack_with_reexecution());
 346   __ jump_to(dest, O0);
 347   __ delayed()->restore();
 348 
 349   __ bind(no_deopt);
 350   restore_live_registers(sasm);
 351   __ ret();
 352   __ delayed()->restore();
 353 
 354   return oop_maps;
 355 }
 356 
 357 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 358 
 359   OopMapSet* oop_maps = NULL;
 360   // for better readability
 361   const bool must_gc_arguments = true;
 362   const bool dont_gc_arguments = false;
 363 
 364   // stub code & info for the different stubs
 365   switch (id) {
 366     case forward_exception_id:
 367       {
 368         oop_maps = generate_handle_exception(id, sasm);
 369       }
 370       break;
 371 
 372     case new_instance_id:
 373     case fast_new_instance_id:
 374     case fast_new_instance_init_check_id:
 375       {
 376         Register G5_klass = G5; // Incoming
 377         Register O0_obj   = O0; // Outgoing
 378 
 379         if (id == new_instance_id) {
 380           __ set_info("new_instance", dont_gc_arguments);
 381         } else if (id == fast_new_instance_id) {
 382           __ set_info("fast new_instance", dont_gc_arguments);
 383         } else {
 384           assert(id == fast_new_instance_init_check_id, "bad StubID");
 385           __ set_info("fast new_instance init check", dont_gc_arguments);
 386         }
 387 
 388         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 389             UseTLAB && FastTLABRefill) {
 390           Label slow_path;
 391           Register G1_obj_size = G1;
 392           Register G3_t1 = G3;
 393           Register G4_t2 = G4;
 394           assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
 395 
 396           // Push a frame since we may do dtrace notification for the
 397           // allocation which requires calling out and we don't want
 398           // to stomp the real return address.
 399           __ save_frame(0);
 400 
 401           if (id == fast_new_instance_init_check_id) {
 402             // make sure the klass is initialized
 403             __ ldub(G5_klass, in_bytes(InstanceKlass::init_state_offset()), G3_t1);
 404             __ cmp_and_br_short(G3_t1, InstanceKlass::fully_initialized, Assembler::notEqual, Assembler::pn, slow_path);
 405           }
 406 #ifdef ASSERT
 407           // assert object can be fast path allocated
 408           {
 409             Label ok, not_ok;
 410           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
 411           // make sure it's an instance (LH > 0)
 412           __ cmp_and_br_short(G1_obj_size, 0, Assembler::lessEqual, Assembler::pn, not_ok);
 413           __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
 414           __ br(Assembler::zero, false, Assembler::pn, ok);
 415           __ delayed()->nop();
 416           __ bind(not_ok);
 417           __ stop("assert(can be fast path allocated)");
 418           __ should_not_reach_here();
 419           __ bind(ok);
 420           }
 421 #endif // ASSERT
 422           // if we got here then the TLAB allocation failed, so try
 423           // refilling the TLAB or allocating directly from eden.
 424           Label retry_tlab, try_eden;
 425           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
 426 
 427           __ bind(retry_tlab);
 428 
 429           // get the instance size
 430           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
 431 
 432           __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
 433 
 434           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 435           __ verify_oop(O0_obj);
 436           __ mov(O0, I0);
 437           __ ret();
 438           __ delayed()->restore();
 439 
 440           __ bind(try_eden);
 441           // get the instance size
 442           __ ld(G5_klass, in_bytes(Klass::layout_helper_offset()), G1_obj_size);
 443           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
 444           __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
 445 
 446           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 447           __ verify_oop(O0_obj);
 448           __ mov(O0, I0);
 449           __ ret();
 450           __ delayed()->restore();
 451 
 452           __ bind(slow_path);
 453 
 454           // pop this frame so generate_stub_call can push it's own
 455           __ restore();
 456         }
 457 
 458         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
 459         // I0->O0: new instance
 460       }
 461 
 462       break;
 463 
 464     case counter_overflow_id:
 465         // G4 contains bci, G5 contains method
 466       oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
 467       break;
 468 
 469     case new_type_array_id:
 470     case new_object_array_id:
 471       {
 472         Register G5_klass = G5; // Incoming
 473         Register G4_length = G4; // Incoming
 474         Register O0_obj   = O0; // Outgoing
 475 
 476         Address klass_lh(G5_klass, Klass::layout_helper_offset());
 477         assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 478         assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
 479         // Use this offset to pick out an individual byte of the layout_helper:
 480         const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
 481                                                  - Klass::_lh_header_size_shift / BitsPerByte);
 482 
 483         if (id == new_type_array_id) {
 484           __ set_info("new_type_array", dont_gc_arguments);
 485         } else {
 486           __ set_info("new_object_array", dont_gc_arguments);
 487         }
 488 
 489 #ifdef ASSERT
 490         // assert object type is really an array of the proper kind
 491         {
 492           Label ok;
 493           Register G3_t1 = G3;
 494           __ ld(klass_lh, G3_t1);
 495           __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
 496           int tag = ((id == new_type_array_id)
 497                      ? Klass::_lh_array_tag_type_value
 498                      : Klass::_lh_array_tag_obj_value);
 499           __ cmp_and_brx_short(G3_t1, tag, Assembler::equal, Assembler::pt, ok);
 500           __ stop("assert(is an array klass)");
 501           __ should_not_reach_here();
 502           __ bind(ok);
 503         }
 504 #endif // ASSERT
 505 
 506         if (UseTLAB && FastTLABRefill) {
 507           Label slow_path;
 508           Register G1_arr_size = G1;
 509           Register G3_t1 = G3;
 510           Register O1_t2 = O1;
 511           assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
 512 
 513           // check that array length is small enough for fast path
 514           __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
 515           __ cmp_and_br_short(G4_length, G3_t1, Assembler::greaterUnsigned, Assembler::pn, slow_path);
 516 
 517           // if we got here then the TLAB allocation failed, so try
 518           // refilling the TLAB or allocating directly from eden.
 519           Label retry_tlab, try_eden;
 520           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
 521 
 522           __ bind(retry_tlab);
 523 
 524           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 525           __ ld(klass_lh, G3_t1);
 526           __ sll(G4_length, G3_t1, G1_arr_size);
 527           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 528           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 529           __ add(G1_arr_size, G3_t1, G1_arr_size);
 530           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
 531           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 532 
 533           __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
 534 
 535           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 536           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 537           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 538           __ add(O0_obj, G3_t1, G3_t1);       // body start
 539           __ initialize_body(G3_t1, O1_t2);
 540           __ verify_oop(O0_obj);
 541           __ retl();
 542           __ delayed()->nop();
 543 
 544           __ bind(try_eden);
 545           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 546           __ ld(klass_lh, G3_t1);
 547           __ sll(G4_length, G3_t1, G1_arr_size);
 548           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 549           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 550           __ add(G1_arr_size, G3_t1, G1_arr_size);
 551           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
 552           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 553 
 554           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
 555           __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
 556 
 557           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 558           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 559           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 560           __ add(O0_obj, G3_t1, G3_t1);       // body start
 561           __ initialize_body(G3_t1, O1_t2);
 562           __ verify_oop(O0_obj);
 563           __ retl();
 564           __ delayed()->nop();
 565 
 566           __ bind(slow_path);
 567         }
 568 
 569         if (id == new_type_array_id) {
 570           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
 571         } else {
 572           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
 573         }
 574         // I0 -> O0: new array
 575       }
 576       break;
 577 
 578     case new_multi_array_id:
 579       { // O0: klass
 580         // O1: rank
 581         // O2: address of 1st dimension
 582         __ set_info("new_multi_array", dont_gc_arguments);
 583         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
 584         // I0 -> O0: new multi array
 585       }
 586       break;
 587 
 588     case register_finalizer_id:
 589       {
 590         __ set_info("register_finalizer", dont_gc_arguments);
 591 
 592         // load the klass and check the has finalizer flag
 593         Label register_finalizer;
 594         Register t = O1;
 595         __ load_klass(O0, t);
 596         __ ld(t, in_bytes(Klass::access_flags_offset()), t);
 597         __ set(JVM_ACC_HAS_FINALIZER, G3);
 598         __ andcc(G3, t, G0);
 599         __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
 600         __ delayed()->nop();
 601 
 602         // do a leaf return
 603         __ retl();
 604         __ delayed()->nop();
 605 
 606         __ bind(register_finalizer);
 607         OopMap* oop_map = save_live_registers(sasm);
 608         int call_offset = __ call_RT(noreg, noreg,
 609                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
 610         oop_maps = new OopMapSet();
 611         oop_maps->add_gc_map(call_offset, oop_map);
 612 
 613         // Now restore all the live registers
 614         restore_live_registers(sasm);
 615 
 616         __ ret();
 617         __ delayed()->restore();
 618       }
 619       break;
 620 
 621     case throw_range_check_failed_id:
 622       { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
 623         // G4: index
 624         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
 625       }
 626       break;
 627 
 628     case throw_index_exception_id:
 629       { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
 630         // G4: index
 631         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
 632       }
 633       break;
 634 
 635     case throw_div0_exception_id:
 636       { __ set_info("throw_div0_exception", dont_gc_arguments);
 637         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 638       }
 639       break;
 640 
 641     case throw_null_pointer_exception_id:
 642       { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
 643         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 644       }
 645       break;
 646 
 647     case handle_exception_id:
 648       { __ set_info("handle_exception", dont_gc_arguments);
 649         oop_maps = generate_handle_exception(id, sasm);
 650       }
 651       break;
 652 
 653     case handle_exception_from_callee_id:
 654       { __ set_info("handle_exception_from_callee", dont_gc_arguments);
 655         oop_maps = generate_handle_exception(id, sasm);
 656       }
 657       break;
 658 
 659     case unwind_exception_id:
 660       {
 661         // O0: exception
 662         // I7: address of call to this method
 663 
 664         __ set_info("unwind_exception", dont_gc_arguments);
 665         __ mov(Oexception, Oexception->after_save());
 666         __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
 667 
 668         __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
 669                         G2_thread, Oissuing_pc->after_save());
 670         __ verify_not_null_oop(Oexception->after_save());
 671 
 672         // Restore SP from L7 if the exception PC is a method handle call site.
 673         __ mov(O0, G5);  // Save the target address.
 674         __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
 675         __ tst(L0);  // Condition codes are preserved over the restore.
 676         __ restore();
 677 
 678         __ jmp(G5, 0);
 679         __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
 680       }
 681       break;
 682 
 683     case throw_array_store_exception_id:
 684       {
 685         __ set_info("throw_array_store_exception", dont_gc_arguments);
 686         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
 687       }
 688       break;
 689 
 690     case throw_class_cast_exception_id:
 691       {
 692         // G4: object
 693         __ set_info("throw_class_cast_exception", dont_gc_arguments);
 694         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 695       }
 696       break;
 697 
 698     case throw_incompatible_class_change_error_id:
 699       {
 700         __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
 701         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 702       }
 703       break;
 704 
 705     case slow_subtype_check_id:
 706       { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
 707         // Arguments :
 708         //
 709         //      ret  : G3
 710         //      sub  : G3, argument, destroyed
 711         //      super: G1, argument, not changed
 712         //      raddr: O7, blown by call
 713         Label miss;
 714 
 715         __ save_frame(0);               // Blow no registers!
 716 
 717         __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
 718 
 719         __ mov(1, G3);
 720         __ ret();                       // Result in G5 is 'true'
 721         __ delayed()->restore();        // free copy or add can go here
 722 
 723         __ bind(miss);
 724         __ mov(0, G3);
 725         __ ret();                       // Result in G5 is 'false'
 726         __ delayed()->restore();        // free copy or add can go here
 727       }
 728 
 729     case monitorenter_nofpu_id:
 730     case monitorenter_id:
 731       { // G4: object
 732         // G5: lock address
 733         __ set_info("monitorenter", dont_gc_arguments);
 734 
 735         int save_fpu_registers = (id == monitorenter_id);
 736         // make a frame and preserve the caller's caller-save registers
 737         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 738 
 739         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
 740 
 741         oop_maps = new OopMapSet();
 742         oop_maps->add_gc_map(call_offset, oop_map);
 743         restore_live_registers(sasm, save_fpu_registers);
 744 
 745         __ ret();
 746         __ delayed()->restore();
 747       }
 748       break;
 749 
 750     case monitorexit_nofpu_id:
 751     case monitorexit_id:
 752       { // G4: lock address
 753         // note: really a leaf routine but must setup last java sp
 754         //       => use call_RT for now (speed can be improved by
 755         //       doing last java sp setup manually)
 756         __ set_info("monitorexit", dont_gc_arguments);
 757 
 758         int save_fpu_registers = (id == monitorexit_id);
 759         // make a frame and preserve the caller's caller-save registers
 760         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 761 
 762         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
 763 
 764         oop_maps = new OopMapSet();
 765         oop_maps->add_gc_map(call_offset, oop_map);
 766         restore_live_registers(sasm, save_fpu_registers);
 767 
 768         __ ret();
 769         __ delayed()->restore();
 770       }
 771       break;
 772 
 773     case deoptimize_id:
 774       {
 775         __ set_info("deoptimize", dont_gc_arguments);
 776         OopMap* oop_map = save_live_registers(sasm);
 777         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
 778         oop_maps = new OopMapSet();
 779         oop_maps->add_gc_map(call_offset, oop_map);
 780         restore_live_registers(sasm);
 781         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 782         assert(deopt_blob != NULL, "deoptimization blob must have been created");
 783         AddressLiteral dest(deopt_blob->unpack_with_reexecution());
 784         __ jump_to(dest, O0);
 785         __ delayed()->restore();
 786       }
 787       break;
 788 
 789     case access_field_patching_id:
 790       { __ set_info("access_field_patching", dont_gc_arguments);
 791         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
 792       }
 793       break;
 794 
 795     case load_klass_patching_id:
 796       { __ set_info("load_klass_patching", dont_gc_arguments);
 797         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
 798       }
 799       break;
 800 
 801     case load_mirror_patching_id:
 802       { __ set_info("load_mirror_patching", dont_gc_arguments);
 803         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
 804       }
 805       break;
 806 
 807     case load_appendix_patching_id:
 808       { __ set_info("load_appendix_patching", dont_gc_arguments);
 809         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
 810       }
 811       break;
 812 
 813     case dtrace_object_alloc_id:
 814       { // O0: object
 815         __ set_info("dtrace_object_alloc", dont_gc_arguments);
 816         // we can't gc here so skip the oopmap but make sure that all
 817         // the live registers get saved.
 818         save_live_registers(sasm);
 819 
 820         __ save_thread(L7_thread_cache);
 821         __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
 822                 relocInfo::runtime_call_type);
 823         __ delayed()->mov(I0, O0);
 824         __ restore_thread(L7_thread_cache);
 825 
 826         restore_live_registers(sasm);
 827         __ ret();
 828         __ delayed()->restore();
 829       }
 830       break;
 831 
 832 #if INCLUDE_ALL_GCS
 833     case g1_pre_barrier_slow_id:
 834       { // G4: previous value of memory
 835         BarrierSet* bs = Universe::heap()->barrier_set();
 836         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 837           __ save_frame(0);
 838           __ set((int)id, O1);
 839           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 840           __ should_not_reach_here();
 841           break;
 842         }
 843 
 844         __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
 845 
 846         Register pre_val = G4;
 847         Register tmp  = G1_scratch;
 848         Register tmp2 = G3_scratch;
 849 
 850         Label refill, restart;
 851         bool with_frame = false; // I don't know if we can do with-frame.
 852         int satb_q_index_byte_offset =
 853           in_bytes(JavaThread::satb_mark_queue_offset() +
 854                    PtrQueue::byte_offset_of_index());
 855         int satb_q_buf_byte_offset =
 856           in_bytes(JavaThread::satb_mark_queue_offset() +
 857                    PtrQueue::byte_offset_of_buf());
 858 
 859         __ bind(restart);
 860         // Load the index into the SATB buffer. PtrQueue::_index is a
 861         // size_t so ld_ptr is appropriate
 862         __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
 863 
 864         // index == 0?
 865         __ cmp_and_brx_short(tmp, G0, Assembler::equal, Assembler::pn, refill);
 866 
 867         __ ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
 868         __ sub(tmp, oopSize, tmp);
 869 
 870         __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
 871         // Use return-from-leaf
 872         __ retl();
 873         __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
 874 
 875         __ bind(refill);
 876         __ save_frame(0);
 877 
 878         __ mov(pre_val, L0);
 879         __ mov(tmp,     L1);
 880         __ mov(tmp2,    L2);
 881 
 882         __ call_VM_leaf(L7_thread_cache,
 883                         CAST_FROM_FN_PTR(address,
 884                                          SATBMarkQueueSet::handle_zero_index_for_thread),
 885                                          G2_thread);
 886 
 887         __ mov(L0, pre_val);
 888         __ mov(L1, tmp);
 889         __ mov(L2, tmp2);
 890 
 891         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 892         __ delayed()->restore();
 893       }
 894       break;
 895 
 896     case g1_post_barrier_slow_id:
 897       {
 898         BarrierSet* bs = Universe::heap()->barrier_set();
 899         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 900           __ save_frame(0);
 901           __ set((int)id, O1);
 902           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 903           __ should_not_reach_here();
 904           break;
 905         }
 906 
 907         __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
 908 
 909         Register addr = G4;
 910         Register cardtable = G5;
 911         Register tmp  = G1_scratch;
 912         Register tmp2 = G3_scratch;
 913         jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
 914 
 915         Label not_already_dirty, restart, refill;
 916 
 917 #ifdef _LP64
 918         __ srlx(addr, CardTableModRefBS::card_shift, addr);
 919 #else
 920         __ srl(addr, CardTableModRefBS::card_shift, addr);
 921 #endif
 922 
 923         AddressLiteral rs(byte_map_base);
 924         __ set(rs, cardtable);         // cardtable := <card table base>
 925         __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
 926 
 927         assert(CardTableModRefBS::dirty_card_val() == 0, "otherwise check this code");
 928         __ cmp_and_br_short(tmp, G0, Assembler::notEqual, Assembler::pt, not_already_dirty);
 929 
 930         // We didn't take the branch, so we're already dirty: return.
 931         // Use return-from-leaf
 932         __ retl();
 933         __ delayed()->nop();
 934 
 935         // Not dirty.
 936         __ bind(not_already_dirty);
 937 
 938         // Get cardtable + tmp into a reg by itself
 939         __ add(addr, cardtable, tmp2);
 940 
 941         // First, dirty it.
 942         __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
 943 
 944         Register tmp3 = cardtable;
 945         Register tmp4 = tmp;
 946 
 947         // these registers are now dead
 948         addr = cardtable = tmp = noreg;
 949 
 950         int dirty_card_q_index_byte_offset =
 951           in_bytes(JavaThread::dirty_card_queue_offset() +
 952                    PtrQueue::byte_offset_of_index());
 953         int dirty_card_q_buf_byte_offset =
 954           in_bytes(JavaThread::dirty_card_queue_offset() +
 955                    PtrQueue::byte_offset_of_buf());
 956 
 957         __ bind(restart);
 958 
 959         // Get the index into the update buffer. PtrQueue::_index is
 960         // a size_t so ld_ptr is appropriate here.
 961         __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
 962 
 963         // index == 0?
 964         __ cmp_and_brx_short(tmp3, G0, Assembler::equal,  Assembler::pn, refill);
 965 
 966         __ ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
 967         __ sub(tmp3, oopSize, tmp3);
 968 
 969         __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
 970         // Use return-from-leaf
 971         __ retl();
 972         __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
 973 
 974         __ bind(refill);
 975         __ save_frame(0);
 976 
 977         __ mov(tmp2, L0);
 978         __ mov(tmp3, L1);
 979         __ mov(tmp4, L2);
 980 
 981         __ call_VM_leaf(L7_thread_cache,
 982                         CAST_FROM_FN_PTR(address,
 983                                          DirtyCardQueueSet::handle_zero_index_for_thread),
 984                                          G2_thread);
 985 
 986         __ mov(L0, tmp2);
 987         __ mov(L1, tmp3);
 988         __ mov(L2, tmp4);
 989 
 990         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 991         __ delayed()->restore();
 992       }
 993       break;
 994 #endif // INCLUDE_ALL_GCS
 995 
 996     case predicate_failed_trap_id:
 997       {
 998         __ set_info("predicate_failed_trap", dont_gc_arguments);
 999         OopMap* oop_map = save_live_registers(sasm);
1000 
1001         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1002 
1003         oop_maps = new OopMapSet();
1004         oop_maps->add_gc_map(call_offset, oop_map);
1005 
1006         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1007         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1008         restore_live_registers(sasm);
1009 
1010         AddressLiteral dest(deopt_blob->unpack_with_reexecution());
1011         __ jump_to(dest, O0);
1012         __ delayed()->restore();
1013       }
1014       break;
1015 
1016     default:
1017       { __ set_info("unimplemented entry", dont_gc_arguments);
1018         __ save_frame(0);
1019         __ set((int)id, O1);
1020         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
1021         __ should_not_reach_here();
1022       }
1023       break;
1024   }
1025   return oop_maps;
1026 }
1027 
1028 
1029 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler* sasm) {
1030   __ block_comment("generate_handle_exception");
1031 
1032   // Save registers, if required.
1033   OopMapSet* oop_maps = new OopMapSet();
1034   OopMap* oop_map = NULL;
1035   switch (id) {
1036   case forward_exception_id:
1037     // We're handling an exception in the context of a compiled frame.
1038     // The registers have been saved in the standard places.  Perform
1039     // an exception lookup in the caller and dispatch to the handler
1040     // if found.  Otherwise unwind and dispatch to the callers
1041     // exception handler.
1042      oop_map = generate_oop_map(sasm, true);
1043 
1044      // transfer the pending exception to the exception_oop
1045      __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
1046      __ ld_ptr(Oexception, 0, G0);
1047      __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
1048      __ add(I7, frame::pc_return_offset, Oissuing_pc);
1049     break;
1050   case handle_exception_id:
1051     // At this point all registers MAY be live.
1052     oop_map = save_live_registers(sasm);
1053     __ mov(Oexception->after_save(),  Oexception);
1054     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
1055     break;
1056   case handle_exception_from_callee_id:
1057     // At this point all registers except exception oop (Oexception)
1058     // and exception pc (Oissuing_pc) are dead.
1059     oop_map = new OopMap(frame_size_in_bytes / sizeof(jint), 0);
1060     sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
1061     __ save_frame_c1(frame_size_in_bytes);
1062     __ mov(Oexception->after_save(),  Oexception);
1063     __ mov(Oissuing_pc->after_save(), Oissuing_pc);
1064     break;
1065   default:  ShouldNotReachHere();
1066   }
1067 
1068   __ verify_not_null_oop(Oexception);
1069 
1070   // save the exception and issuing pc in the thread
1071   __ st_ptr(Oexception,  G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1072   __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1073 
1074   // use the throwing pc as the return address to lookup (has bci & oop map)
1075   __ mov(Oissuing_pc, I7);
1076   __ sub(I7, frame::pc_return_offset, I7);
1077   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1078   oop_maps->add_gc_map(call_offset, oop_map);
1079 
1080   // Note: if nmethod has been deoptimized then regardless of
1081   // whether it had a handler or not we will deoptimize
1082   // by entering the deopt blob with a pending exception.
1083 
1084   // Restore the registers that were saved at the beginning, remove
1085   // the frame and jump to the exception handler.
1086   switch (id) {
1087   case forward_exception_id:
1088   case handle_exception_id:
1089     restore_live_registers(sasm);
1090     __ jmp(O0, 0);
1091     __ delayed()->restore();
1092     break;
1093   case handle_exception_from_callee_id:
1094     // Restore SP from L7 if the exception PC is a method handle call site.
1095     __ mov(O0, G5);  // Save the target address.
1096     __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
1097     __ tst(L0);  // Condition codes are preserved over the restore.
1098     __ restore();
1099 
1100     __ jmp(G5, 0);  // jump to the exception handler
1101     __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
1102     break;
1103   default:  ShouldNotReachHere();
1104   }
1105 
1106   return oop_maps;
1107 }
1108 
1109 
1110 #undef __
1111 
1112 const char *Runtime1::pd_name_for_address(address entry) {
1113   return "<unknown function>";
1114 }