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   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 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   __ tst(O0);
 338   __ brx(Assembler::equal, false, Assembler::pt, no_deopt);
 339   __ delayed()->nop();
 340 
 341   // return to the deoptimization handler entry for unpacking and rexecute
 342   // if we simply returned the we'd deopt as if any call we patched had just
 343   // returned.
 344 
 345   restore_live_registers(sasm);
 346   __ restore();
 347   __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type);
 348   __ delayed()->nop();
 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         // we're handling an exception in the context of a compiled
 370         // frame.  The registers have been saved in the standard
 371         // places.  Perform an exception lookup in the caller and
 372         // dispatch to the handler if found.  Otherwise unwind and
 373         // dispatch to the callers exception handler.
 374 
 375         oop_maps = new OopMapSet();
 376         OopMap* oop_map = generate_oop_map(sasm, true);
 377 
 378         // transfer the pending exception to the exception_oop
 379         __ ld_ptr(G2_thread, in_bytes(JavaThread::pending_exception_offset()), Oexception);
 380         __ ld_ptr(Oexception, 0, G0);
 381         __ st_ptr(G0, G2_thread, in_bytes(JavaThread::pending_exception_offset()));
 382         __ add(I7, frame::pc_return_offset, Oissuing_pc);
 383 
 384         generate_handle_exception(sasm, oop_maps, oop_map);
 385         __ should_not_reach_here();
 386       }
 387       break;
 388 
 389     case new_instance_id:
 390     case fast_new_instance_id:
 391     case fast_new_instance_init_check_id:
 392       {
 393         Register G5_klass = G5; // Incoming
 394         Register O0_obj   = O0; // Outgoing
 395 
 396         if (id == new_instance_id) {
 397           __ set_info("new_instance", dont_gc_arguments);
 398         } else if (id == fast_new_instance_id) {
 399           __ set_info("fast new_instance", dont_gc_arguments);
 400         } else {
 401           assert(id == fast_new_instance_init_check_id, "bad StubID");
 402           __ set_info("fast new_instance init check", dont_gc_arguments);
 403         }
 404 
 405         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 406             UseTLAB && FastTLABRefill) {
 407           Label slow_path;
 408           Register G1_obj_size = G1;
 409           Register G3_t1 = G3;
 410           Register G4_t2 = G4;
 411           assert_different_registers(G5_klass, G1_obj_size, G3_t1, G4_t2);
 412 
 413           // Push a frame since we may do dtrace notification for the
 414           // allocation which requires calling out and we don't want
 415           // to stomp the real return address.
 416           __ save_frame(0);
 417 
 418           if (id == fast_new_instance_init_check_id) {
 419             // make sure the klass is initialized
 420             __ ld(G5_klass, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc), G3_t1);
 421             __ cmp(G3_t1, instanceKlass::fully_initialized);
 422             __ br(Assembler::notEqual, false, Assembler::pn, slow_path);
 423             __ delayed()->nop();
 424           }
 425 #ifdef ASSERT
 426           // assert object can be fast path allocated
 427           {
 428             Label ok, not_ok;
 429           __ ld(G5_klass, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc), G1_obj_size);
 430           __ cmp(G1_obj_size, 0);  // make sure it's an instance (LH > 0)
 431           __ br(Assembler::lessEqual, false, Assembler::pn, not_ok);
 432           __ delayed()->nop();
 433           __ btst(Klass::_lh_instance_slow_path_bit, G1_obj_size);
 434           __ br(Assembler::zero, false, Assembler::pn, ok);
 435           __ delayed()->nop();
 436           __ bind(not_ok);
 437           __ stop("assert(can be fast path allocated)");
 438           __ should_not_reach_here();
 439           __ bind(ok);
 440           }
 441 #endif // ASSERT
 442           // if we got here then the TLAB allocation failed, so try
 443           // refilling the TLAB or allocating directly from eden.
 444           Label retry_tlab, try_eden;
 445           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G5_klass
 446 
 447           __ bind(retry_tlab);
 448 
 449           // get the instance size
 450           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 451 
 452           __ tlab_allocate(O0_obj, G1_obj_size, 0, G3_t1, slow_path);
 453 
 454           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 455           __ verify_oop(O0_obj);
 456           __ mov(O0, I0);
 457           __ ret();
 458           __ delayed()->restore();
 459 
 460           __ bind(try_eden);
 461           // get the instance size
 462           __ ld(G5_klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), G1_obj_size);
 463           __ eden_allocate(O0_obj, G1_obj_size, 0, G3_t1, G4_t2, slow_path);
 464           __ incr_allocated_bytes(G1_obj_size, G3_t1, G4_t2);
 465 
 466           __ initialize_object(O0_obj, G5_klass, G1_obj_size, 0, G3_t1, G4_t2);
 467           __ verify_oop(O0_obj);
 468           __ mov(O0, I0);
 469           __ ret();
 470           __ delayed()->restore();
 471 
 472           __ bind(slow_path);
 473 
 474           // pop this frame so generate_stub_call can push it's own
 475           __ restore();
 476         }
 477 
 478         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_instance), G5_klass);
 479         // I0->O0: new instance
 480       }
 481 
 482       break;
 483 
 484     case counter_overflow_id:
 485         // G4 contains bci, G5 contains method
 486       oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, counter_overflow), G4, G5);
 487       break;
 488 
 489     case new_type_array_id:
 490     case new_object_array_id:
 491       {
 492         Register G5_klass = G5; // Incoming
 493         Register G4_length = G4; // Incoming
 494         Register O0_obj   = O0; // Outgoing
 495 
 496         Address klass_lh(G5_klass, ((klassOopDesc::header_size() * HeapWordSize)
 497                                     + Klass::layout_helper_offset_in_bytes()));
 498         assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 499         assert(Klass::_lh_header_size_mask == 0xFF, "bytewise");
 500         // Use this offset to pick out an individual byte of the layout_helper:
 501         const int klass_lh_header_size_offset = ((BytesPerInt - 1)  // 3 - 2 selects byte {0,1,0,0}
 502                                                  - Klass::_lh_header_size_shift / BitsPerByte);
 503 
 504         if (id == new_type_array_id) {
 505           __ set_info("new_type_array", dont_gc_arguments);
 506         } else {
 507           __ set_info("new_object_array", dont_gc_arguments);
 508         }
 509 
 510 #ifdef ASSERT
 511         // assert object type is really an array of the proper kind
 512         {
 513           Label ok;
 514           Register G3_t1 = G3;
 515           __ ld(klass_lh, G3_t1);
 516           __ sra(G3_t1, Klass::_lh_array_tag_shift, G3_t1);
 517           int tag = ((id == new_type_array_id)
 518                      ? Klass::_lh_array_tag_type_value
 519                      : Klass::_lh_array_tag_obj_value);
 520           __ cmp(G3_t1, tag);
 521           __ brx(Assembler::equal, false, Assembler::pt, ok);
 522           __ delayed()->nop();
 523           __ stop("assert(is an array klass)");
 524           __ should_not_reach_here();
 525           __ bind(ok);
 526         }
 527 #endif // ASSERT
 528 
 529         if (UseTLAB && FastTLABRefill) {
 530           Label slow_path;
 531           Register G1_arr_size = G1;
 532           Register G3_t1 = G3;
 533           Register O1_t2 = O1;
 534           assert_different_registers(G5_klass, G4_length, G1_arr_size, G3_t1, O1_t2);
 535 
 536           // check that array length is small enough for fast path
 537           __ set(C1_MacroAssembler::max_array_allocation_length, G3_t1);
 538           __ cmp(G4_length, G3_t1);
 539           __ br(Assembler::greaterUnsigned, false, Assembler::pn, slow_path);
 540           __ delayed()->nop();
 541 
 542           // if we got here then the TLAB allocation failed, so try
 543           // refilling the TLAB or allocating directly from eden.
 544           Label retry_tlab, try_eden;
 545           __ tlab_refill(retry_tlab, try_eden, slow_path); // preserves G4_length and G5_klass
 546 
 547           __ bind(retry_tlab);
 548 
 549           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 550           __ ld(klass_lh, G3_t1);
 551           __ sll(G4_length, G3_t1, G1_arr_size);
 552           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 553           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 554           __ add(G1_arr_size, G3_t1, G1_arr_size);
 555           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);  // align up
 556           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 557 
 558           __ tlab_allocate(O0_obj, G1_arr_size, 0, G3_t1, slow_path);  // preserves G1_arr_size
 559 
 560           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 561           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 562           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 563           __ add(O0_obj, G3_t1, G3_t1);       // body start
 564           __ initialize_body(G3_t1, O1_t2);
 565           __ verify_oop(O0_obj);
 566           __ retl();
 567           __ delayed()->nop();
 568 
 569           __ bind(try_eden);
 570           // get the allocation size: (length << (layout_helper & 0x1F)) + header_size
 571           __ ld(klass_lh, G3_t1);
 572           __ sll(G4_length, G3_t1, G1_arr_size);
 573           __ srl(G3_t1, Klass::_lh_header_size_shift, G3_t1);
 574           __ and3(G3_t1, Klass::_lh_header_size_mask, G3_t1);
 575           __ add(G1_arr_size, G3_t1, G1_arr_size);
 576           __ add(G1_arr_size, MinObjAlignmentInBytesMask, G1_arr_size);
 577           __ and3(G1_arr_size, ~MinObjAlignmentInBytesMask, G1_arr_size);
 578 
 579           __ eden_allocate(O0_obj, G1_arr_size, 0, G3_t1, O1_t2, slow_path);  // preserves G1_arr_size
 580           __ incr_allocated_bytes(G1_arr_size, G3_t1, O1_t2);
 581 
 582           __ initialize_header(O0_obj, G5_klass, G4_length, G3_t1, O1_t2);
 583           __ ldub(klass_lh, G3_t1, klass_lh_header_size_offset);
 584           __ sub(G1_arr_size, G3_t1, O1_t2);  // body length
 585           __ add(O0_obj, G3_t1, G3_t1);       // body start
 586           __ initialize_body(G3_t1, O1_t2);
 587           __ verify_oop(O0_obj);
 588           __ retl();
 589           __ delayed()->nop();
 590 
 591           __ bind(slow_path);
 592         }
 593 
 594         if (id == new_type_array_id) {
 595           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_type_array), G5_klass, G4_length);
 596         } else {
 597           oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_object_array), G5_klass, G4_length);
 598         }
 599         // I0 -> O0: new array
 600       }
 601       break;
 602 
 603     case new_multi_array_id:
 604       { // O0: klass
 605         // O1: rank
 606         // O2: address of 1st dimension
 607         __ set_info("new_multi_array", dont_gc_arguments);
 608         oop_maps = generate_stub_call(sasm, I0, CAST_FROM_FN_PTR(address, new_multi_array), I0, I1, I2);
 609         // I0 -> O0: new multi array
 610       }
 611       break;
 612 
 613     case register_finalizer_id:
 614       {
 615         __ set_info("register_finalizer", dont_gc_arguments);
 616 
 617         // load the klass and check the has finalizer flag
 618         Label register_finalizer;
 619         Register t = O1;
 620         __ load_klass(O0, t);
 621         __ ld(t, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc), t);
 622         __ set(JVM_ACC_HAS_FINALIZER, G3);
 623         __ andcc(G3, t, G0);
 624         __ br(Assembler::notZero, false, Assembler::pt, register_finalizer);
 625         __ delayed()->nop();
 626 
 627         // do a leaf return
 628         __ retl();
 629         __ delayed()->nop();
 630 
 631         __ bind(register_finalizer);
 632         OopMap* oop_map = save_live_registers(sasm);
 633         int call_offset = __ call_RT(noreg, noreg,
 634                                      CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), I0);
 635         oop_maps = new OopMapSet();
 636         oop_maps->add_gc_map(call_offset, oop_map);
 637 
 638         // Now restore all the live registers
 639         restore_live_registers(sasm);
 640 
 641         __ ret();
 642         __ delayed()->restore();
 643       }
 644       break;
 645 
 646     case throw_range_check_failed_id:
 647       { __ set_info("range_check_failed", dont_gc_arguments); // arguments will be discarded
 648         // G4: index
 649         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
 650       }
 651       break;
 652 
 653     case throw_index_exception_id:
 654       { __ set_info("index_range_check_failed", dont_gc_arguments); // arguments will be discarded
 655         // G4: index
 656         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
 657       }
 658       break;
 659 
 660     case throw_div0_exception_id:
 661       { __ set_info("throw_div0_exception", dont_gc_arguments);
 662         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 663       }
 664       break;
 665 
 666     case throw_null_pointer_exception_id:
 667       { __ set_info("throw_null_pointer_exception", dont_gc_arguments);
 668         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 669       }
 670       break;
 671 
 672     case handle_exception_id:
 673       {
 674         __ set_info("handle_exception", dont_gc_arguments);
 675         // make a frame and preserve the caller's caller-save registers
 676 
 677         oop_maps = new OopMapSet();
 678         OopMap* oop_map = save_live_registers(sasm);
 679         __ mov(Oexception->after_save(),  Oexception);
 680         __ mov(Oissuing_pc->after_save(), Oissuing_pc);
 681         generate_handle_exception(sasm, oop_maps, oop_map);
 682       }
 683       break;
 684 
 685     case unwind_exception_id:
 686       {
 687         // O0: exception
 688         // I7: address of call to this method
 689 
 690         __ set_info("unwind_exception", dont_gc_arguments);
 691         __ mov(Oexception, Oexception->after_save());
 692         __ add(I7, frame::pc_return_offset, Oissuing_pc->after_save());
 693 
 694         __ call_VM_leaf(L7_thread_cache, CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address),
 695                         G2_thread, Oissuing_pc->after_save());
 696         __ verify_not_null_oop(Oexception->after_save());
 697 
 698         // Restore SP from L7 if the exception PC is a MethodHandle call site.
 699         __ mov(O0, G5);  // Save the target address.
 700         __ lduw(Address(G2_thread, JavaThread::is_method_handle_return_offset()), L0);
 701         __ tst(L0);  // Condition codes are preserved over the restore.
 702         __ restore();
 703 
 704         __ jmp(G5, 0);
 705         __ delayed()->movcc(Assembler::notZero, false, Assembler::icc, L7_mh_SP_save, SP);  // Restore SP if required.
 706       }
 707       break;
 708 
 709     case throw_array_store_exception_id:
 710       {
 711         __ set_info("throw_array_store_exception", dont_gc_arguments);
 712         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), false);
 713       }
 714       break;
 715 
 716     case throw_class_cast_exception_id:
 717       {
 718         // G4: object
 719         __ set_info("throw_class_cast_exception", dont_gc_arguments);
 720         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 721       }
 722       break;
 723 
 724     case throw_incompatible_class_change_error_id:
 725       {
 726         __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments);
 727         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 728       }
 729       break;
 730 
 731     case slow_subtype_check_id:
 732       { // Support for uint StubRoutine::partial_subtype_check( Klass sub, Klass super );
 733         // Arguments :
 734         //
 735         //      ret  : G3
 736         //      sub  : G3, argument, destroyed
 737         //      super: G1, argument, not changed
 738         //      raddr: O7, blown by call
 739         Label miss;
 740 
 741         __ save_frame(0);               // Blow no registers!
 742 
 743         __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss);
 744 
 745         __ mov(1, G3);
 746         __ ret();                       // Result in G5 is 'true'
 747         __ delayed()->restore();        // free copy or add can go here
 748 
 749         __ bind(miss);
 750         __ mov(0, G3);
 751         __ ret();                       // Result in G5 is 'false'
 752         __ delayed()->restore();        // free copy or add can go here
 753       }
 754 
 755     case monitorenter_nofpu_id:
 756     case monitorenter_id:
 757       { // G4: object
 758         // G5: lock address
 759         __ set_info("monitorenter", dont_gc_arguments);
 760 
 761         int save_fpu_registers = (id == monitorenter_id);
 762         // make a frame and preserve the caller's caller-save registers
 763         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 764 
 765         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), G4, G5);
 766 
 767         oop_maps = new OopMapSet();
 768         oop_maps->add_gc_map(call_offset, oop_map);
 769         restore_live_registers(sasm, save_fpu_registers);
 770 
 771         __ ret();
 772         __ delayed()->restore();
 773       }
 774       break;
 775 
 776     case monitorexit_nofpu_id:
 777     case monitorexit_id:
 778       { // G4: lock address
 779         // note: really a leaf routine but must setup last java sp
 780         //       => use call_RT for now (speed can be improved by
 781         //       doing last java sp setup manually)
 782         __ set_info("monitorexit", dont_gc_arguments);
 783 
 784         int save_fpu_registers = (id == monitorexit_id);
 785         // make a frame and preserve the caller's caller-save registers
 786         OopMap* oop_map = save_live_registers(sasm, save_fpu_registers);
 787 
 788         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), G4);
 789 
 790         oop_maps = new OopMapSet();
 791         oop_maps->add_gc_map(call_offset, oop_map);
 792         restore_live_registers(sasm, save_fpu_registers);
 793 
 794         __ ret();
 795         __ delayed()->restore();
 796 
 797       }
 798       break;
 799 
 800     case access_field_patching_id:
 801       { __ set_info("access_field_patching", dont_gc_arguments);
 802         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
 803       }
 804       break;
 805 
 806     case load_klass_patching_id:
 807       { __ set_info("load_klass_patching", dont_gc_arguments);
 808         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
 809       }
 810       break;
 811 
 812     case jvmti_exception_throw_id:
 813       { // Oexception : exception
 814         __ set_info("jvmti_exception_throw", dont_gc_arguments);
 815         oop_maps = generate_stub_call(sasm, noreg, CAST_FROM_FN_PTR(address, Runtime1::post_jvmti_exception_throw), I0);
 816       }
 817       break;
 818 
 819     case dtrace_object_alloc_id:
 820       { // O0: object
 821         __ set_info("dtrace_object_alloc", dont_gc_arguments);
 822         // we can't gc here so skip the oopmap but make sure that all
 823         // the live registers get saved.
 824         save_live_registers(sasm);
 825 
 826         __ save_thread(L7_thread_cache);
 827         __ call(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc),
 828                 relocInfo::runtime_call_type);
 829         __ delayed()->mov(I0, O0);
 830         __ restore_thread(L7_thread_cache);
 831 
 832         restore_live_registers(sasm);
 833         __ ret();
 834         __ delayed()->restore();
 835       }
 836       break;
 837 
 838 #ifndef SERIALGC
 839     case g1_pre_barrier_slow_id:
 840       { // G4: previous value of memory
 841         BarrierSet* bs = Universe::heap()->barrier_set();
 842         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 843           __ save_frame(0);
 844           __ set((int)id, O1);
 845           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 846           __ should_not_reach_here();
 847           break;
 848         }
 849 
 850         __ set_info("g1_pre_barrier_slow_id", dont_gc_arguments);
 851 
 852         Register pre_val = G4;
 853         Register tmp  = G1_scratch;
 854         Register tmp2 = G3_scratch;
 855 
 856         Label refill, restart;
 857         bool with_frame = false; // I don't know if we can do with-frame.
 858         int satb_q_index_byte_offset =
 859           in_bytes(JavaThread::satb_mark_queue_offset() +
 860                    PtrQueue::byte_offset_of_index());
 861         int satb_q_buf_byte_offset =
 862           in_bytes(JavaThread::satb_mark_queue_offset() +
 863                    PtrQueue::byte_offset_of_buf());
 864         __ bind(restart);
 865         __ ld_ptr(G2_thread, satb_q_index_byte_offset, tmp);
 866 
 867         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false,
 868                           Assembler::pn, tmp, refill);
 869 
 870         // If the branch is taken, no harm in executing this in the delay slot.
 871         __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, tmp2);
 872         __ sub(tmp, oopSize, tmp);
 873 
 874         __ st_ptr(pre_val, tmp2, tmp);  // [_buf + index] := <address_of_card>
 875         // Use return-from-leaf
 876         __ retl();
 877         __ delayed()->st_ptr(tmp, G2_thread, satb_q_index_byte_offset);
 878 
 879         __ bind(refill);
 880         __ save_frame(0);
 881 
 882         __ mov(pre_val, L0);
 883         __ mov(tmp,     L1);
 884         __ mov(tmp2,    L2);
 885 
 886         __ call_VM_leaf(L7_thread_cache,
 887                         CAST_FROM_FN_PTR(address,
 888                                          SATBMarkQueueSet::handle_zero_index_for_thread),
 889                                          G2_thread);
 890 
 891         __ mov(L0, pre_val);
 892         __ mov(L1, tmp);
 893         __ mov(L2, tmp2);
 894 
 895         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 896         __ delayed()->restore();
 897       }
 898       break;
 899 
 900     case g1_post_barrier_slow_id:
 901       {
 902         BarrierSet* bs = Universe::heap()->barrier_set();
 903         if (bs->kind() != BarrierSet::G1SATBCTLogging) {
 904           __ save_frame(0);
 905           __ set((int)id, O1);
 906           __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), I0);
 907           __ should_not_reach_here();
 908           break;
 909         }
 910 
 911         __ set_info("g1_post_barrier_slow_id", dont_gc_arguments);
 912 
 913         Register addr = G4;
 914         Register cardtable = G5;
 915         Register tmp  = G1_scratch;
 916         Register tmp2 = G3_scratch;
 917         jbyte* byte_map_base = ((CardTableModRefBS*)bs)->byte_map_base;
 918 
 919         Label not_already_dirty, restart, refill;
 920 
 921 #ifdef _LP64
 922         __ srlx(addr, CardTableModRefBS::card_shift, addr);
 923 #else
 924         __ srl(addr, CardTableModRefBS::card_shift, addr);
 925 #endif
 926 
 927         AddressLiteral rs(byte_map_base);
 928         __ set(rs, cardtable);         // cardtable := <card table base>
 929         __ ldub(addr, cardtable, tmp); // tmp := [addr + cardtable]
 930 
 931         __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
 932                           tmp, not_already_dirty);
 933         // Get cardtable + tmp into a reg by itself -- useful in the take-the-branch
 934         // case, harmless if not.
 935         __ delayed()->add(addr, cardtable, tmp2);
 936 
 937         // We didn't take the branch, so we're already dirty: return.
 938         // Use return-from-leaf
 939         __ retl();
 940         __ delayed()->nop();
 941 
 942         // Not dirty.
 943         __ bind(not_already_dirty);
 944         // First, dirty it.
 945         __ stb(G0, tmp2, 0);  // [cardPtr] := 0  (i.e., dirty).
 946 
 947         Register tmp3 = cardtable;
 948         Register tmp4 = tmp;
 949 
 950         // these registers are now dead
 951         addr = cardtable = tmp = noreg;
 952 
 953         int dirty_card_q_index_byte_offset =
 954           in_bytes(JavaThread::dirty_card_queue_offset() +
 955                    PtrQueue::byte_offset_of_index());
 956         int dirty_card_q_buf_byte_offset =
 957           in_bytes(JavaThread::dirty_card_queue_offset() +
 958                    PtrQueue::byte_offset_of_buf());
 959         __ bind(restart);
 960         __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, tmp3);
 961 
 962         __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
 963                           tmp3, refill);
 964         // If the branch is taken, no harm in executing this in the delay slot.
 965         __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, tmp4);
 966         __ sub(tmp3, oopSize, tmp3);
 967 
 968         __ st_ptr(tmp2, tmp4, tmp3);  // [_buf + index] := <address_of_card>
 969         // Use return-from-leaf
 970         __ retl();
 971         __ delayed()->st_ptr(tmp3, G2_thread, dirty_card_q_index_byte_offset);
 972 
 973         __ bind(refill);
 974         __ save_frame(0);
 975 
 976         __ mov(tmp2, L0);
 977         __ mov(tmp3, L1);
 978         __ mov(tmp4, L2);
 979 
 980         __ call_VM_leaf(L7_thread_cache,
 981                         CAST_FROM_FN_PTR(address,
 982                                          DirtyCardQueueSet::handle_zero_index_for_thread),
 983                                          G2_thread);
 984 
 985         __ mov(L0, tmp2);
 986         __ mov(L1, tmp3);
 987         __ mov(L2, tmp4);
 988 
 989         __ br(Assembler::always, /*annul*/false, Assembler::pt, restart);
 990         __ delayed()->restore();
 991       }
 992       break;
 993 #endif // !SERIALGC
 994 
 995     default:
 996       { __ set_info("unimplemented entry", dont_gc_arguments);
 997         __ save_frame(0);
 998         __ set((int)id, O1);
 999         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), O1);
1000         __ should_not_reach_here();
1001       }
1002       break;
1003   }
1004   return oop_maps;
1005 }
1006 
1007 
1008 void Runtime1::generate_handle_exception(StubAssembler* sasm, OopMapSet* oop_maps, OopMap* oop_map, bool) {
1009   Label no_deopt;
1010 
1011   __ verify_not_null_oop(Oexception);
1012 
1013   // save the exception and issuing pc in the thread
1014   __ st_ptr(Oexception, G2_thread, in_bytes(JavaThread::exception_oop_offset()));
1015   __ st_ptr(Oissuing_pc, G2_thread, in_bytes(JavaThread::exception_pc_offset()));
1016 
1017   // save the real return address and use the throwing pc as the return address to lookup (has bci & oop map)
1018   __ mov(I7, L0);
1019   __ mov(Oissuing_pc, I7);
1020   __ sub(I7, frame::pc_return_offset, I7);
1021   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
1022 
1023   // Note: if nmethod has been deoptimized then regardless of
1024   // whether it had a handler or not we will deoptimize
1025   // by entering the deopt blob with a pending exception.
1026 
1027 #ifdef ASSERT
1028   Label done;
1029   __ tst(O0);
1030   __ br(Assembler::notZero, false, Assembler::pn, done);
1031   __ delayed()->nop();
1032   __ stop("should have found address");
1033   __ bind(done);
1034 #endif
1035 
1036   // restore the registers that were saved at the beginning and jump to the exception handler.
1037   restore_live_registers(sasm);
1038 
1039   __ jmp(O0, 0);
1040   __ delayed()->restore();
1041 
1042   oop_maps->add_gc_map(call_offset, oop_map);
1043 }
1044 
1045 
1046 #undef __
1047 
1048 #define __ masm->
1049 
1050 const char *Runtime1::pd_name_for_address(address entry) {
1051   return "<unknown function>";
1052 }