1 /*
   2  * Copyright (c) 1999, 2021, Oracle and/or its affiliates. All rights reserved.
   3  * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This code is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 only, as
   8  * published by the Free Software Foundation.
   9  *
  10  * This code is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  13  * version 2 for more details (a copy is included in the LICENSE file that
  14  * accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #include "precompiled.hpp"
  27 #include "asm/assembler.hpp"
  28 #include "c1/c1_CodeStubs.hpp"
  29 #include "c1/c1_Defs.hpp"
  30 #include "c1/c1_MacroAssembler.hpp"
  31 #include "c1/c1_Runtime1.hpp"
  32 #include "compiler/disassembler.hpp"
  33 #include "compiler/oopMap.hpp"
  34 #include "gc/shared/cardTable.hpp"
  35 #include "gc/shared/cardTableBarrierSet.hpp"
  36 #include "gc/shared/collectedHeap.hpp"
  37 #include "gc/shared/tlab_globals.hpp"
  38 #include "interpreter/interpreter.hpp"
  39 #include "memory/universe.hpp"
  40 #include "nativeInst_aarch64.hpp"
  41 #include "oops/compiledICHolder.hpp"
  42 #include "oops/oop.inline.hpp"
  43 #include "prims/jvmtiExport.hpp"
  44 #include "register_aarch64.hpp"
  45 #include "runtime/sharedRuntime.hpp"
  46 #include "runtime/signature.hpp"
  47 #include "runtime/stubRoutines.hpp"
  48 #include "runtime/vframe.hpp"
  49 #include "runtime/vframeArray.hpp"
  50 #include "utilities/powerOfTwo.hpp"
  51 #include "vmreg_aarch64.inline.hpp"
  52 
  53 
  54 // Implementation of StubAssembler
  55 
  56 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, int args_size) {
  57   // setup registers
  58   assert(!(oop_result1->is_valid() || metadata_result->is_valid()) || oop_result1 != metadata_result, "registers must be different");
  59   assert(oop_result1 != rthread && metadata_result != rthread, "registers must be different");
  60   assert(args_size >= 0, "illegal args_size");
  61   bool align_stack = false;
  62 
  63   mov(c_rarg0, rthread);
  64   set_num_rt_args(0); // Nothing on stack
  65 
  66   Label retaddr;
  67   set_last_Java_frame(sp, rfp, retaddr, rscratch1);
  68 
  69   // do the call
  70   lea(rscratch1, RuntimeAddress(entry));
  71   blr(rscratch1);
  72   bind(retaddr);
  73   int call_offset = offset();
  74   // verify callee-saved register
  75 #ifdef ASSERT
  76   push(r0, sp);
  77   { Label L;
  78     get_thread(r0);
  79     cmp(rthread, r0);
  80     br(Assembler::EQ, L);
  81     stop("StubAssembler::call_RT: rthread not callee saved?");
  82     bind(L);
  83   }
  84   pop(r0, sp);
  85 #endif
  86   reset_last_Java_frame(true);
  87 
  88   // check for pending exceptions
  89   { Label L;
  90     // check for pending exceptions (java_thread is set upon return)
  91     ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
  92     cbz(rscratch1, L);
  93     // exception pending => remove activation and forward to exception handler
  94     // make sure that the vm_results are cleared
  95     if (oop_result1->is_valid()) {
  96       str(zr, Address(rthread, JavaThread::vm_result_offset()));
  97     }
  98     if (metadata_result->is_valid()) {
  99       str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
 100     }
 101     if (frame_size() == no_frame_size) {
 102       leave();
 103       far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
 104     } else if (_stub_id == Runtime1::forward_exception_id) {
 105       should_not_reach_here();
 106     } else {
 107       far_jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
 108     }
 109     bind(L);
 110   }
 111   // get oop results if there are any and reset the values in the thread
 112   if (oop_result1->is_valid()) {
 113     get_vm_result(oop_result1, rthread);
 114   }
 115   if (metadata_result->is_valid()) {
 116     get_vm_result_2(metadata_result, rthread);
 117   }
 118   return call_offset;
 119 }
 120 
 121 
 122 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
 123   mov(c_rarg1, arg1);
 124   return call_RT(oop_result1, metadata_result, entry, 1);
 125 }
 126 
 127 
 128 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
 129   if (c_rarg1 == arg2) {
 130     if (c_rarg2 == arg1) {
 131       mov(rscratch1, arg1);
 132       mov(arg1, arg2);
 133       mov(arg2, rscratch1);
 134     } else {
 135       mov(c_rarg2, arg2);
 136       mov(c_rarg1, arg1);
 137     }
 138   } else {
 139     mov(c_rarg1, arg1);
 140     mov(c_rarg2, arg2);
 141   }
 142   return call_RT(oop_result1, metadata_result, entry, 2);
 143 }
 144 
 145 
 146 int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
 147   // if there is any conflict use the stack
 148   if (arg1 == c_rarg2 || arg1 == c_rarg3 ||
 149       arg2 == c_rarg1 || arg2 == c_rarg3 ||
 150       arg3 == c_rarg1 || arg3 == c_rarg2) {
 151     stp(arg3, arg2, Address(pre(sp, -2 * wordSize)));
 152     stp(arg1, zr, Address(pre(sp, -2 * wordSize)));
 153     ldp(c_rarg1, zr, Address(post(sp, 2 * wordSize)));
 154     ldp(c_rarg3, c_rarg2, Address(post(sp, 2 * wordSize)));
 155   } else {
 156     mov(c_rarg1, arg1);
 157     mov(c_rarg2, arg2);
 158     mov(c_rarg3, arg3);
 159   }
 160   return call_RT(oop_result1, metadata_result, entry, 3);
 161 }
 162 
 163 enum return_state_t {
 164   does_not_return, requires_return
 165 };
 166 
 167 
 168 // Implementation of StubFrame
 169 
 170 class StubFrame: public StackObj {
 171  private:
 172   StubAssembler* _sasm;
 173   bool _return_state;
 174 
 175  public:
 176   StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, return_state_t return_state=requires_return);
 177   void load_argument(int offset_in_words, Register reg);
 178 
 179   ~StubFrame();
 180 };;
 181 
 182 void StubAssembler::prologue(const char* name, bool must_gc_arguments) {
 183   set_info(name, must_gc_arguments);
 184   enter();
 185 }
 186 
 187 void StubAssembler::epilogue() {
 188   leave();
 189   ret(lr);
 190 }
 191 
 192 #define __ _sasm->
 193 
 194 StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments, return_state_t return_state) {
 195   _sasm = sasm;
 196   _return_state = return_state;
 197   __ prologue(name, must_gc_arguments);
 198 }
 199 
 200 // load parameters that were stored with LIR_Assembler::store_parameter
 201 // Note: offsets for store_parameter and load_argument must match
 202 void StubFrame::load_argument(int offset_in_words, Register reg) {
 203   __ load_parameter(offset_in_words, reg);
 204 }
 205 
 206 StubFrame::~StubFrame() {
 207   if (_return_state == requires_return) {
 208     __ epilogue();
 209   } else {
 210     __ should_not_reach_here();
 211   }
 212 }
 213 
 214 #undef __
 215 
 216 
 217 // Implementation of Runtime1
 218 
 219 #define __ sasm->
 220 
 221 const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2;
 222 
 223 // Stack layout for saving/restoring  all the registers needed during a runtime
 224 // call (this includes deoptimization)
 225 // Note: note that users of this frame may well have arguments to some runtime
 226 // while these values are on the stack. These positions neglect those arguments
 227 // but the code in save_live_registers will take the argument count into
 228 // account.
 229 //
 230 
 231 enum reg_save_layout {
 232   reg_save_frame_size = 32 /* float */ + 32 /* integer */
 233 };
 234 
 235 // Save off registers which might be killed by calls into the runtime.
 236 // Tries to smart of about FP registers.  In particular we separate
 237 // saving and describing the FPU registers for deoptimization since we
 238 // have to save the FPU registers twice if we describe them.  The
 239 // deopt blob is the only thing which needs to describe FPU registers.
 240 // In all other cases it should be sufficient to simply save their
 241 // current value.
 242 
 243 static int cpu_reg_save_offsets[FrameMap::nof_cpu_regs];
 244 static int fpu_reg_save_offsets[FrameMap::nof_fpu_regs];
 245 static int reg_save_size_in_words;
 246 static int frame_size_in_bytes = -1;
 247 
 248 static OopMap* generate_oop_map(StubAssembler* sasm, bool save_fpu_registers) {
 249   int frame_size_in_bytes = reg_save_frame_size * BytesPerWord;
 250   sasm->set_frame_size(frame_size_in_bytes / BytesPerWord);
 251   int frame_size_in_slots = frame_size_in_bytes / sizeof(jint);
 252   OopMap* oop_map = new OopMap(frame_size_in_slots, 0);
 253 
 254   for (int i = 0; i < FrameMap::nof_cpu_regs; i++) {
 255     Register r = as_Register(i);
 256     if (i <= 18 && i != rscratch1->encoding() && i != rscratch2->encoding()) {
 257       int sp_offset = cpu_reg_save_offsets[i];
 258       oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 259                                 r->as_VMReg());
 260     }
 261   }
 262 
 263   if (save_fpu_registers) {
 264     for (int i = 0; i < FrameMap::nof_fpu_regs; i++) {
 265       FloatRegister r = as_FloatRegister(i);
 266       {
 267         int sp_offset = fpu_reg_save_offsets[i];
 268         oop_map->set_callee_saved(VMRegImpl::stack2reg(sp_offset),
 269                                   r->as_VMReg());
 270       }
 271     }
 272   }
 273   return oop_map;
 274 }
 275 
 276 static OopMap* save_live_registers(StubAssembler* sasm,
 277                                    bool save_fpu_registers = true) {
 278   __ block_comment("save_live_registers");
 279 
 280   __ push(RegSet::range(r0, r29), sp);         // integer registers except lr & sp
 281 
 282   if (save_fpu_registers) {
 283     for (int i = 31; i>= 0; i -= 4) {
 284       __ sub(sp, sp, 4 * wordSize); // no pre-increment for st1. Emulate it without modifying other registers
 285       __ st1(as_FloatRegister(i-3), as_FloatRegister(i-2), as_FloatRegister(i-1),
 286           as_FloatRegister(i), __ T1D, Address(sp));
 287     }
 288   } else {
 289     __ add(sp, sp, -32 * wordSize);
 290   }
 291 
 292   return generate_oop_map(sasm, save_fpu_registers);
 293 }
 294 
 295 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 296   if (restore_fpu_registers) {
 297     for (int i = 0; i < 32; i += 4)
 298       __ ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
 299           as_FloatRegister(i+3), __ T1D, Address(__ post(sp, 4 * wordSize)));
 300   } else {
 301     __ add(sp, sp, 32 * wordSize);
 302   }
 303 
 304   __ pop(RegSet::range(r0, r29), sp);
 305 }
 306 
 307 static void restore_live_registers_except_r0(StubAssembler* sasm, bool restore_fpu_registers = true)  {
 308 
 309   if (restore_fpu_registers) {
 310     for (int i = 0; i < 32; i += 4)
 311       __ ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
 312           as_FloatRegister(i+3), __ T1D, Address(__ post(sp, 4 * wordSize)));
 313   } else {
 314     __ add(sp, sp, 32 * wordSize);
 315   }
 316 
 317   __ ldp(zr, r1, Address(__ post(sp, 16)));
 318   __ pop(RegSet::range(r2, r29), sp);
 319 }
 320 
 321 
 322 
 323 void Runtime1::initialize_pd() {
 324   int i;
 325   int sp_offset = 0;
 326 
 327   // all float registers are saved explicitly
 328   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
 329   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 330     fpu_reg_save_offsets[i] = sp_offset;
 331     sp_offset += 2;   // SP offsets are in halfwords
 332   }
 333 
 334   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 335     Register r = as_Register(i);
 336     cpu_reg_save_offsets[i] = sp_offset;
 337     sp_offset += 2;   // SP offsets are in halfwords
 338   }
 339 }
 340 
 341 
 342 // target: the entry point of the method that creates and posts the exception oop
 343 // has_argument: true if the exception needs arguments (passed in rscratch1 and rscratch2)
 344 
 345 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 346   // make a frame and preserve the caller's caller-save registers
 347   OopMap* oop_map = save_live_registers(sasm);
 348   int call_offset;
 349   if (!has_argument) {
 350     call_offset = __ call_RT(noreg, noreg, target);
 351   } else {
 352     __ mov(c_rarg1, rscratch1);
 353     __ mov(c_rarg2, rscratch2);
 354     call_offset = __ call_RT(noreg, noreg, target);
 355   }
 356   OopMapSet* oop_maps = new OopMapSet();
 357   oop_maps->add_gc_map(call_offset, oop_map);
 358   return oop_maps;
 359 }
 360 
 361 
 362 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
 363   __ block_comment("generate_handle_exception");
 364 
 365   // incoming parameters
 366   const Register exception_oop = r0;
 367   const Register exception_pc  = r3;
 368   // other registers used in this stub
 369 
 370   // Save registers, if required.
 371   OopMapSet* oop_maps = new OopMapSet();
 372   OopMap* oop_map = NULL;
 373   switch (id) {
 374   case forward_exception_id:
 375     // We're handling an exception in the context of a compiled frame.
 376     // The registers have been saved in the standard places.  Perform
 377     // an exception lookup in the caller and dispatch to the handler
 378     // if found.  Otherwise unwind and dispatch to the callers
 379     // exception handler.
 380     oop_map = generate_oop_map(sasm, 1 /*thread*/);
 381 
 382     // load and clear pending exception oop into r0
 383     __ ldr(exception_oop, Address(rthread, Thread::pending_exception_offset()));
 384     __ str(zr, Address(rthread, Thread::pending_exception_offset()));
 385 
 386     // load issuing PC (the return address for this stub) into r3
 387     __ ldr(exception_pc, Address(rfp, 1*BytesPerWord));
 388 
 389     // make sure that the vm_results are cleared (may be unnecessary)
 390     __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
 391     __ str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
 392     break;
 393   case handle_exception_nofpu_id:
 394   case handle_exception_id:
 395     // At this point all registers MAY be live.
 396     oop_map = save_live_registers(sasm, id != handle_exception_nofpu_id);
 397     break;
 398   case handle_exception_from_callee_id: {
 399     // At this point all registers except exception oop (r0) and
 400     // exception pc (lr) are dead.
 401     const int frame_size = 2 /*fp, return address*/;
 402     oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
 403     sasm->set_frame_size(frame_size);
 404     break;
 405   }
 406   default: ShouldNotReachHere();
 407   }
 408 
 409   // verify that only r0 and r3 are valid at this time
 410   __ invalidate_registers(false, true, true, false, true, true);
 411   // verify that r0 contains a valid exception
 412   __ verify_not_null_oop(exception_oop);
 413 
 414 #ifdef ASSERT
 415   // check that fields in JavaThread for exception oop and issuing pc are
 416   // empty before writing to them
 417   Label oop_empty;
 418   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 419   __ cbz(rscratch1, oop_empty);
 420   __ stop("exception oop already set");
 421   __ bind(oop_empty);
 422 
 423   Label pc_empty;
 424   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 425   __ cbz(rscratch1, pc_empty);
 426   __ stop("exception pc already set");
 427   __ bind(pc_empty);
 428 #endif
 429 
 430   // save exception oop and issuing pc into JavaThread
 431   // (exception handler will load it from here)
 432   __ str(exception_oop, Address(rthread, JavaThread::exception_oop_offset()));
 433   __ str(exception_pc, Address(rthread, JavaThread::exception_pc_offset()));
 434 
 435   // patch throwing pc into return address (has bci & oop map)
 436   __ str(exception_pc, Address(rfp, 1*BytesPerWord));
 437 
 438   // compute the exception handler.
 439   // the exception oop and the throwing pc are read from the fields in JavaThread
 440   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
 441   oop_maps->add_gc_map(call_offset, oop_map);
 442 
 443   // r0: handler address
 444   //      will be the deopt blob if nmethod was deoptimized while we looked up
 445   //      handler regardless of whether handler existed in the nmethod.
 446 
 447   // only r0 is valid at this time, all other registers have been destroyed by the runtime call
 448   __ invalidate_registers(false, true, true, true, true, true);
 449 
 450   // patch the return address, this stub will directly return to the exception handler
 451   __ str(r0, Address(rfp, 1*BytesPerWord));
 452 
 453   switch (id) {
 454   case forward_exception_id:
 455   case handle_exception_nofpu_id:
 456   case handle_exception_id:
 457     // Restore the registers that were saved at the beginning.
 458     restore_live_registers(sasm, id != handle_exception_nofpu_id);
 459     break;
 460   case handle_exception_from_callee_id:
 461     break;
 462   default:  ShouldNotReachHere();
 463   }
 464 
 465   return oop_maps;
 466 }
 467 
 468 
 469 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
 470   // incoming parameters
 471   const Register exception_oop = r0;
 472   // callee-saved copy of exception_oop during runtime call
 473   const Register exception_oop_callee_saved = r19;
 474   // other registers used in this stub
 475   const Register exception_pc = r3;
 476   const Register handler_addr = r1;
 477 
 478   // verify that only r0, is valid at this time
 479   __ invalidate_registers(false, true, true, true, true, true);
 480 
 481 #ifdef ASSERT
 482   // check that fields in JavaThread for exception oop and issuing pc are empty
 483   Label oop_empty;
 484   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 485   __ cbz(rscratch1, oop_empty);
 486   __ stop("exception oop must be empty");
 487   __ bind(oop_empty);
 488 
 489   Label pc_empty;
 490   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 491   __ cbz(rscratch1, pc_empty);
 492   __ stop("exception pc must be empty");
 493   __ bind(pc_empty);
 494 #endif
 495 
 496   // Save our return address because
 497   // exception_handler_for_return_address will destroy it.  We also
 498   // save exception_oop
 499   __ stp(lr, exception_oop, Address(__ pre(sp, -2 * wordSize)));
 500 
 501   // search the exception handler address of the caller (using the return address)
 502   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, lr);
 503   // r0: exception handler address of the caller
 504 
 505   // Only R0 is valid at this time; all other registers have been
 506   // destroyed by the call.
 507   __ invalidate_registers(false, true, true, true, false, true);
 508 
 509   // move result of call into correct register
 510   __ mov(handler_addr, r0);
 511 
 512   // get throwing pc (= return address).
 513   // lr has been destroyed by the call
 514   __ ldp(lr, exception_oop, Address(__ post(sp, 2 * wordSize)));
 515   __ mov(r3, lr);
 516 
 517   __ verify_not_null_oop(exception_oop);
 518 
 519   // continue at exception handler (return address removed)
 520   // note: do *not* remove arguments when unwinding the
 521   //       activation since the caller assumes having
 522   //       all arguments on the stack when entering the
 523   //       runtime to determine the exception handler
 524   //       (GC happens at call site with arguments!)
 525   // r0: exception oop
 526   // r3: throwing pc
 527   // r1: exception handler
 528   __ br(handler_addr);
 529 }
 530 
 531 
 532 
 533 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 534   // use the maximum number of runtime-arguments here because it is difficult to
 535   // distinguish each RT-Call.
 536   // Note: This number affects also the RT-Call in generate_handle_exception because
 537   //       the oop-map is shared for all calls.
 538   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 539   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 540 
 541   OopMap* oop_map = save_live_registers(sasm);
 542 
 543   __ mov(c_rarg0, rthread);
 544   Label retaddr;
 545   __ set_last_Java_frame(sp, rfp, retaddr, rscratch1);
 546   // do the call
 547   __ lea(rscratch1, RuntimeAddress(target));
 548   __ blr(rscratch1);
 549   __ bind(retaddr);
 550   OopMapSet* oop_maps = new OopMapSet();
 551   oop_maps->add_gc_map(__ offset(), oop_map);
 552   // verify callee-saved register
 553 #ifdef ASSERT
 554   { Label L;
 555     __ get_thread(rscratch1);
 556     __ cmp(rthread, rscratch1);
 557     __ br(Assembler::EQ, L);
 558     __ stop("StubAssembler::call_RT: rthread not callee saved?");
 559     __ bind(L);
 560   }
 561 #endif
 562 
 563   __ reset_last_Java_frame(true);
 564 
 565 #ifdef ASSERT
 566   // check that fields in JavaThread for exception oop and issuing pc are empty
 567   Label oop_empty;
 568   __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 569   __ cbz(rscratch1, oop_empty);
 570   __ stop("exception oop must be empty");
 571   __ bind(oop_empty);
 572 
 573   Label pc_empty;
 574   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 575   __ cbz(rscratch1, pc_empty);
 576   __ stop("exception pc must be empty");
 577   __ bind(pc_empty);
 578 #endif
 579 
 580   // Runtime will return true if the nmethod has been deoptimized, this is the
 581   // expected scenario and anything else is  an error. Note that we maintain a
 582   // check on the result purely as a defensive measure.
 583   Label no_deopt;
 584   __ cbz(r0, no_deopt);                                // Have we deoptimized?
 585 
 586   // Perform a re-execute. The proper return  address is already on the stack,
 587   // we just need  to restore registers, pop  all of our frame  but the return
 588   // address and jump to the deopt blob.
 589   restore_live_registers(sasm);
 590   __ leave();
 591   __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
 592 
 593   __ bind(no_deopt);
 594   __ stop("deopt not performed");
 595 
 596   return oop_maps;
 597 }
 598 
 599 
 600 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 601 
 602   const Register exception_oop = r0;
 603   const Register exception_pc  = r3;
 604 
 605   // for better readability
 606   const bool must_gc_arguments = true;
 607   const bool dont_gc_arguments = false;
 608 
 609   // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
 610   bool save_fpu_registers = true;
 611 
 612   // stub code & info for the different stubs
 613   OopMapSet* oop_maps = NULL;
 614   OopMap* oop_map = NULL;
 615   switch (id) {
 616     {
 617     case forward_exception_id:
 618       {
 619         oop_maps = generate_handle_exception(id, sasm);
 620         __ leave();
 621         __ ret(lr);
 622       }
 623       break;
 624 
 625     case throw_div0_exception_id:
 626       { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments, does_not_return);
 627         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 628       }
 629       break;
 630 
 631     case throw_null_pointer_exception_id:
 632       { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments, does_not_return);
 633         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 634       }
 635       break;
 636 
 637     case new_instance_id:
 638     case fast_new_instance_id:
 639     case fast_new_instance_init_check_id:
 640       {
 641         Register klass = r3; // Incoming
 642         Register obj   = r0; // Result
 643 
 644         if (id == new_instance_id) {
 645           __ set_info("new_instance", dont_gc_arguments);
 646         } else if (id == fast_new_instance_id) {
 647           __ set_info("fast new_instance", dont_gc_arguments);
 648         } else {
 649           assert(id == fast_new_instance_init_check_id, "bad StubID");
 650           __ set_info("fast new_instance init check", dont_gc_arguments);
 651         }
 652 
 653         // If TLAB is disabled, see if there is support for inlining contiguous
 654         // allocations.
 655         // Otherwise, just go to the slow path.
 656         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 657             !UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 658           Label slow_path;
 659           Register obj_size = r19;
 660           Register t1       = r10;
 661           Register t2       = r11;
 662           assert_different_registers(klass, obj, obj_size, t1, t2);
 663 
 664           __ stp(r19, zr, Address(__ pre(sp, -2 * wordSize)));
 665 
 666           if (id == fast_new_instance_init_check_id) {
 667             // make sure the klass is initialized
 668             __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
 669             __ cmpw(rscratch1, InstanceKlass::fully_initialized);
 670             __ br(Assembler::NE, slow_path);
 671           }
 672 
 673 #ifdef ASSERT
 674           // assert object can be fast path allocated
 675           {
 676             Label ok, not_ok;
 677             __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 678             __ cmp(obj_size, (u1)0);
 679             __ br(Assembler::LE, not_ok);  // make sure it's an instance (LH > 0)
 680             __ tstw(obj_size, Klass::_lh_instance_slow_path_bit);
 681             __ br(Assembler::EQ, ok);
 682             __ bind(not_ok);
 683             __ stop("assert(can be fast path allocated)");
 684             __ should_not_reach_here();
 685             __ bind(ok);
 686           }
 687 #endif // ASSERT
 688 
 689           // get the instance size (size is postive so movl is fine for 64bit)
 690           __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 691 
 692           __ eden_allocate(obj, obj_size, 0, t1, slow_path);
 693 
 694           __ initialize_object(obj, klass, obj_size, 0, t1, t2, /* is_tlab_allocated */ false);
 695           __ verify_oop(obj);
 696           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 697           __ ret(lr);
 698 
 699           __ bind(slow_path);
 700           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 701         }
 702 
 703         __ enter();
 704         OopMap* map = save_live_registers(sasm);
 705         int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
 706         oop_maps = new OopMapSet();
 707         oop_maps->add_gc_map(call_offset, map);
 708         restore_live_registers_except_r0(sasm);
 709         __ verify_oop(obj);
 710         __ leave();
 711         __ ret(lr);
 712 
 713         // r0,: new instance
 714       }
 715 
 716       break;
 717 
 718     case counter_overflow_id:
 719       {
 720         Register bci = r0, method = r1;
 721         __ enter();
 722         OopMap* map = save_live_registers(sasm);
 723         // Retrieve bci
 724         __ ldrw(bci, Address(rfp, 2*BytesPerWord));
 725         // And a pointer to the Method*
 726         __ ldr(method, Address(rfp, 3*BytesPerWord));
 727         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
 728         oop_maps = new OopMapSet();
 729         oop_maps->add_gc_map(call_offset, map);
 730         restore_live_registers(sasm);
 731         __ leave();
 732         __ ret(lr);
 733       }
 734       break;
 735 
 736     case new_type_array_id:
 737     case new_object_array_id:
 738       {
 739         Register length   = r19; // Incoming
 740         Register klass    = r3; // Incoming
 741         Register obj      = r0; // Result
 742 
 743         if (id == new_type_array_id) {
 744           __ set_info("new_type_array", dont_gc_arguments);
 745         } else {
 746           __ set_info("new_object_array", dont_gc_arguments);
 747         }
 748 
 749 #ifdef ASSERT
 750         // assert object type is really an array of the proper kind
 751         {
 752           Label ok;
 753           Register t0 = obj;
 754           __ ldrw(t0, Address(klass, Klass::layout_helper_offset()));
 755           __ asrw(t0, t0, Klass::_lh_array_tag_shift);
 756           int tag = ((id == new_type_array_id)
 757                      ? Klass::_lh_array_tag_type_value
 758                      : Klass::_lh_array_tag_obj_value);
 759           __ mov(rscratch1, tag);
 760           __ cmpw(t0, rscratch1);
 761           __ br(Assembler::EQ, ok);
 762           __ stop("assert(is an array klass)");
 763           __ should_not_reach_here();
 764           __ bind(ok);
 765         }
 766 #endif // ASSERT
 767 
 768         // If TLAB is disabled, see if there is support for inlining contiguous
 769         // allocations.
 770         // Otherwise, just go to the slow path.
 771         if (!UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 772           Register arr_size = r5;
 773           Register t1       = r10;
 774           Register t2       = r11;
 775           Label slow_path;
 776           assert_different_registers(length, klass, obj, arr_size, t1, t2);
 777 
 778           // check that array length is small enough for fast path.
 779           __ mov(rscratch1, C1_MacroAssembler::max_array_allocation_length);
 780           __ cmpw(length, rscratch1);
 781           __ br(Assembler::HI, slow_path);
 782 
 783           // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
 784           // since size is positive ldrw does right thing on 64bit
 785           __ ldrw(t1, Address(klass, Klass::layout_helper_offset()));
 786           // since size is positive movw does right thing on 64bit
 787           __ movw(arr_size, length);
 788           __ lslvw(arr_size, length, t1);
 789           __ ubfx(t1, t1, Klass::_lh_header_size_shift,
 790                   exact_log2(Klass::_lh_header_size_mask + 1));
 791           __ add(arr_size, arr_size, t1);
 792           __ add(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
 793           __ andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
 794 
 795           __ eden_allocate(obj, arr_size, 0, t1, slow_path);  // preserves arr_size
 796 
 797           __ initialize_header(obj, klass, length, t1, t2);
 798           __ ldrb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
 799           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 800           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
 801           __ andr(t1, t1, Klass::_lh_header_size_mask);
 802           __ sub(arr_size, arr_size, t1);  // body length
 803           __ add(t1, t1, obj);       // body start
 804           __ initialize_body(t1, arr_size, 0, t1, t2);
 805           __ membar(Assembler::StoreStore);
 806           __ verify_oop(obj);
 807 
 808           __ ret(lr);
 809 
 810           __ bind(slow_path);
 811         }
 812 
 813         __ enter();
 814         OopMap* map = save_live_registers(sasm);
 815         int call_offset;
 816         if (id == new_type_array_id) {
 817           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
 818         } else {
 819           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
 820         }
 821 
 822         oop_maps = new OopMapSet();
 823         oop_maps->add_gc_map(call_offset, map);
 824         restore_live_registers_except_r0(sasm);
 825 
 826         __ verify_oop(obj);
 827         __ leave();
 828         __ ret(lr);
 829 
 830         // r0: new array
 831       }
 832       break;
 833 
 834     case new_multi_array_id:
 835       { StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
 836         // r0,: klass
 837         // r19,: rank
 838         // r2: address of 1st dimension
 839         OopMap* map = save_live_registers(sasm);
 840         __ mov(c_rarg1, r0);
 841         __ mov(c_rarg3, r2);
 842         __ mov(c_rarg2, r19);
 843         int call_offset = __ call_RT(r0, noreg, CAST_FROM_FN_PTR(address, new_multi_array), r1, r2, r3);
 844 
 845         oop_maps = new OopMapSet();
 846         oop_maps->add_gc_map(call_offset, map);
 847         restore_live_registers_except_r0(sasm);
 848 
 849         // r0,: new multi array
 850         __ verify_oop(r0);
 851       }
 852       break;
 853 
 854     case register_finalizer_id:
 855       {
 856         __ set_info("register_finalizer", dont_gc_arguments);
 857 
 858         // This is called via call_runtime so the arguments
 859         // will be place in C abi locations
 860 
 861         __ verify_oop(c_rarg0);
 862 
 863         // load the klass and check the has finalizer flag
 864         Label register_finalizer;
 865         Register t = r5;
 866         __ load_klass(t, r0);
 867         __ ldrw(t, Address(t, Klass::access_flags_offset()));
 868         __ tbnz(t, exact_log2(JVM_ACC_HAS_FINALIZER), register_finalizer);
 869         __ ret(lr);
 870 
 871         __ bind(register_finalizer);
 872         __ enter();
 873         OopMap* oop_map = save_live_registers(sasm);
 874         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), r0);
 875         oop_maps = new OopMapSet();
 876         oop_maps->add_gc_map(call_offset, oop_map);
 877 
 878         // Now restore all the live registers
 879         restore_live_registers(sasm);
 880 
 881         __ leave();
 882         __ ret(lr);
 883       }
 884       break;
 885 
 886     case throw_class_cast_exception_id:
 887       { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments, does_not_return);
 888         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 889       }
 890       break;
 891 
 892     case throw_incompatible_class_change_error_id:
 893       { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments, does_not_return);
 894         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 895       }
 896       break;
 897 
 898     case slow_subtype_check_id:
 899       {
 900         // Typical calling sequence:
 901         // __ push(klass_RInfo);  // object klass or other subclass
 902         // __ push(sup_k_RInfo);  // array element klass or other superclass
 903         // __ bl(slow_subtype_check);
 904         // Note that the subclass is pushed first, and is therefore deepest.
 905         enum layout {
 906           r0_off, r0_off_hi,
 907           r2_off, r2_off_hi,
 908           r4_off, r4_off_hi,
 909           r5_off, r5_off_hi,
 910           sup_k_off, sup_k_off_hi,
 911           klass_off, klass_off_hi,
 912           framesize,
 913           result_off = sup_k_off
 914         };
 915 
 916         __ set_info("slow_subtype_check", dont_gc_arguments);
 917         __ push(RegSet::of(r0, r2, r4, r5), sp);
 918 
 919         // This is called by pushing args and not with C abi
 920         // __ ldr(r4, Address(sp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
 921         // __ ldr(r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
 922 
 923         __ ldp(r4, r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size));
 924 
 925         Label miss;
 926         __ check_klass_subtype_slow_path(r4, r0, r2, r5, NULL, &miss);
 927 
 928         // fallthrough on success:
 929         __ mov(rscratch1, 1);
 930         __ str(rscratch1, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 931         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 932         __ ret(lr);
 933 
 934         __ bind(miss);
 935         __ str(zr, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 936         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 937         __ ret(lr);
 938       }
 939       break;
 940 
 941     case monitorenter_nofpu_id:
 942       save_fpu_registers = false;
 943       // fall through
 944     case monitorenter_id:
 945       {
 946         StubFrame f(sasm, "monitorenter", dont_gc_arguments);
 947         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 948 
 949         // Called with store_parameter and not C abi
 950 
 951         f.load_argument(1, r0); // r0,: object
 952         f.load_argument(0, r1); // r1,: lock address
 953 
 954         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), r0, r1);
 955 
 956         oop_maps = new OopMapSet();
 957         oop_maps->add_gc_map(call_offset, map);
 958         restore_live_registers(sasm, save_fpu_registers);
 959       }
 960       break;
 961 
 962     case monitorexit_nofpu_id:
 963       save_fpu_registers = false;
 964       // fall through
 965     case monitorexit_id:
 966       {
 967         StubFrame f(sasm, "monitorexit", dont_gc_arguments);
 968         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 969 
 970         // Called with store_parameter and not C abi
 971 
 972         f.load_argument(0, r0); // r0,: lock address
 973 
 974         // note: really a leaf routine but must setup last java sp
 975         //       => use call_RT for now (speed can be improved by
 976         //       doing last java sp setup manually)
 977         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), r0);
 978 
 979         oop_maps = new OopMapSet();
 980         oop_maps->add_gc_map(call_offset, map);
 981         restore_live_registers(sasm, save_fpu_registers);
 982       }
 983       break;
 984 
 985     case deoptimize_id:
 986       {
 987         StubFrame f(sasm, "deoptimize", dont_gc_arguments, does_not_return);
 988         OopMap* oop_map = save_live_registers(sasm);
 989         f.load_argument(0, c_rarg1);
 990         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), c_rarg1);
 991 
 992         oop_maps = new OopMapSet();
 993         oop_maps->add_gc_map(call_offset, oop_map);
 994         restore_live_registers(sasm);
 995         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 996         assert(deopt_blob != NULL, "deoptimization blob must have been created");
 997         __ leave();
 998         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
 999       }
1000       break;
1001 
1002     case throw_range_check_failed_id:
1003       { StubFrame f(sasm, "range_check_failed", dont_gc_arguments, does_not_return);
1004         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1005       }
1006       break;
1007 
1008     case unwind_exception_id:
1009       { __ set_info("unwind_exception", dont_gc_arguments);
1010         // note: no stubframe since we are about to leave the current
1011         //       activation and we are calling a leaf VM function only.
1012         generate_unwind_exception(sasm);
1013       }
1014       break;
1015 
1016     case access_field_patching_id:
1017       { StubFrame f(sasm, "access_field_patching", dont_gc_arguments, does_not_return);
1018         // we should set up register map
1019         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1020       }
1021       break;
1022 
1023     case load_klass_patching_id:
1024       { StubFrame f(sasm, "load_klass_patching", dont_gc_arguments, does_not_return);
1025         // we should set up register map
1026         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1027       }
1028       break;
1029 
1030     case load_mirror_patching_id:
1031       { StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments, does_not_return);
1032         // we should set up register map
1033         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
1034       }
1035       break;
1036 
1037     case load_appendix_patching_id:
1038       { StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments, does_not_return);
1039         // we should set up register map
1040         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
1041       }
1042       break;
1043 
1044     case handle_exception_nofpu_id:
1045     case handle_exception_id:
1046       { StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1047         oop_maps = generate_handle_exception(id, sasm);
1048       }
1049       break;
1050 
1051     case handle_exception_from_callee_id:
1052       { StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
1053         oop_maps = generate_handle_exception(id, sasm);
1054       }
1055       break;
1056 
1057     case throw_index_exception_id:
1058       { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments, does_not_return);
1059         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1060       }
1061       break;
1062 
1063     case throw_array_store_exception_id:
1064       { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments, does_not_return);
1065         // tos + 0: link
1066         //     + 1: return address
1067         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
1068       }
1069       break;
1070 
1071     case predicate_failed_trap_id:
1072       {
1073         StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments, does_not_return);
1074 
1075         OopMap* map = save_live_registers(sasm);
1076 
1077         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1078         oop_maps = new OopMapSet();
1079         oop_maps->add_gc_map(call_offset, map);
1080         restore_live_registers(sasm);
1081         __ leave();
1082         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1083         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1084 
1085         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1086       }
1087       break;
1088 
1089     case dtrace_object_alloc_id:
1090       { // c_rarg0: object
1091         StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
1092         save_live_registers(sasm);
1093 
1094         __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), c_rarg0);
1095 
1096         restore_live_registers(sasm);
1097       }
1098       break;
1099 
1100     default:
1101       { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments, does_not_return);
1102         __ mov(r0, (int)id);
1103         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), r0);
1104       }
1105       break;
1106     }
1107   }
1108   return oop_maps;
1109 }
1110 
1111 #undef __
1112 
1113 const char *Runtime1::pd_name_for_address(address entry) { Unimplemented(); return 0; }