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 void save_live_registers_no_oop_map(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 
 293 static OopMap* save_live_registers(StubAssembler* sasm,
 294                                    bool save_fpu_registers = true) {
 295   save_live_registers_no_oop_map(sasm, save_fpu_registers);
 296   return generate_oop_map(sasm, save_fpu_registers);
 297 }
 298 
 299 static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
 300   if (restore_fpu_registers) {
 301     for (int i = 0; i < 32; i += 4)
 302       __ ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
 303           as_FloatRegister(i+3), __ T1D, Address(__ post(sp, 4 * wordSize)));
 304   } else {
 305     __ add(sp, sp, 32 * wordSize);
 306   }
 307 
 308   __ pop(RegSet::range(r0, r29), sp);
 309 }
 310 
 311 static void restore_live_registers_except_r0(StubAssembler* sasm, bool restore_fpu_registers = true)  {
 312 
 313   if (restore_fpu_registers) {
 314     for (int i = 0; i < 32; i += 4)
 315       __ ld1(as_FloatRegister(i), as_FloatRegister(i+1), as_FloatRegister(i+2),
 316           as_FloatRegister(i+3), __ T1D, Address(__ post(sp, 4 * wordSize)));
 317   } else {
 318     __ add(sp, sp, 32 * wordSize);
 319   }
 320 
 321   __ ldp(zr, r1, Address(__ post(sp, 16)));
 322   __ pop(RegSet::range(r2, r29), sp);
 323 }
 324 
 325 
 326 
 327 void Runtime1::initialize_pd() {
 328   int i;
 329   int sp_offset = 0;
 330 
 331   // all float registers are saved explicitly
 332   assert(FrameMap::nof_fpu_regs == 32, "double registers not handled here");
 333   for (i = 0; i < FrameMap::nof_fpu_regs; i++) {
 334     fpu_reg_save_offsets[i] = sp_offset;
 335     sp_offset += 2;   // SP offsets are in halfwords
 336   }
 337 
 338   for (i = 0; i < FrameMap::nof_cpu_regs; i++) {
 339     Register r = as_Register(i);
 340     cpu_reg_save_offsets[i] = sp_offset;
 341     sp_offset += 2;   // SP offsets are in halfwords
 342   }
 343 }
 344 
 345 
 346 // target: the entry point of the method that creates and posts the exception oop
 347 // has_argument: true if the exception needs arguments (passed in rscratch1 and rscratch2)
 348 
 349 OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
 350   // make a frame and preserve the caller's caller-save registers
 351   OopMap* oop_map = save_live_registers(sasm);
 352   int call_offset;
 353   if (!has_argument) {
 354     call_offset = __ call_RT(noreg, noreg, target);
 355   } else {
 356     __ mov(c_rarg1, rscratch1);
 357     __ mov(c_rarg2, rscratch2);
 358     call_offset = __ call_RT(noreg, noreg, target);
 359   }
 360   OopMapSet* oop_maps = new OopMapSet();
 361   oop_maps->add_gc_map(call_offset, oop_map);
 362   return oop_maps;
 363 }
 364 
 365 
 366 OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
 367   __ block_comment("generate_handle_exception");
 368 
 369   // incoming parameters
 370   const Register exception_oop = r0;
 371   const Register exception_pc  = r3;
 372   // other registers used in this stub
 373 
 374   // Save registers, if required.
 375   OopMapSet* oop_maps = new OopMapSet();
 376   OopMap* oop_map = NULL;
 377   switch (id) {
 378   case forward_exception_id:
 379     // We're handling an exception in the context of a compiled frame.
 380     // The registers have been saved in the standard places.  Perform
 381     // an exception lookup in the caller and dispatch to the handler
 382     // if found.  Otherwise unwind and dispatch to the callers
 383     // exception handler.
 384     oop_map = generate_oop_map(sasm, 1 /*thread*/);
 385 
 386     // load and clear pending exception oop into r0
 387     __ ldr(exception_oop, Address(rthread, Thread::pending_exception_offset()));
 388     __ str(zr, Address(rthread, Thread::pending_exception_offset()));
 389 
 390     // load issuing PC (the return address for this stub) into r3
 391     __ ldr(exception_pc, Address(rfp, 1*BytesPerWord));
 392 
 393     // make sure that the vm_results are cleared (may be unnecessary)
 394     __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
 395     __ str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
 396     break;
 397   case handle_exception_nofpu_id:
 398   case handle_exception_id:
 399     // At this point all registers MAY be live.
 400     oop_map = save_live_registers(sasm, id != handle_exception_nofpu_id);
 401     break;
 402   case handle_exception_from_callee_id: {
 403     // At this point all registers except exception oop (r0) and
 404     // exception pc (lr) are dead.
 405     const int frame_size = 2 /*fp, return address*/;
 406     oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
 407     sasm->set_frame_size(frame_size);
 408     break;
 409   }
 410   default: ShouldNotReachHere();
 411   }
 412 
 413   // verify that only r0 and r3 are valid at this time
 414   __ invalidate_registers(false, true, true, false, true, true);
 415   // verify that r0 contains a valid exception
 416   __ verify_not_null_oop(exception_oop);
 417 
 418 #ifdef ASSERT
 419   // check that fields in JavaThread for exception oop and issuing pc are
 420   // empty before writing to them
 421   Label oop_empty;
 422   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 423   __ cbz(rscratch1, oop_empty);
 424   __ stop("exception oop already set");
 425   __ bind(oop_empty);
 426 
 427   Label pc_empty;
 428   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 429   __ cbz(rscratch1, pc_empty);
 430   __ stop("exception pc already set");
 431   __ bind(pc_empty);
 432 #endif
 433 
 434   // save exception oop and issuing pc into JavaThread
 435   // (exception handler will load it from here)
 436   __ str(exception_oop, Address(rthread, JavaThread::exception_oop_offset()));
 437   __ str(exception_pc, Address(rthread, JavaThread::exception_pc_offset()));
 438 
 439   // patch throwing pc into return address (has bci & oop map)
 440   __ str(exception_pc, Address(rfp, 1*BytesPerWord));
 441 
 442   // compute the exception handler.
 443   // the exception oop and the throwing pc are read from the fields in JavaThread
 444   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
 445   oop_maps->add_gc_map(call_offset, oop_map);
 446 
 447   // r0: handler address
 448   //      will be the deopt blob if nmethod was deoptimized while we looked up
 449   //      handler regardless of whether handler existed in the nmethod.
 450 
 451   // only r0 is valid at this time, all other registers have been destroyed by the runtime call
 452   __ invalidate_registers(false, true, true, true, true, true);
 453 
 454   // patch the return address, this stub will directly return to the exception handler
 455   __ str(r0, Address(rfp, 1*BytesPerWord));
 456 
 457   switch (id) {
 458   case forward_exception_id:
 459   case handle_exception_nofpu_id:
 460   case handle_exception_id:
 461     // Restore the registers that were saved at the beginning.
 462     restore_live_registers(sasm, id != handle_exception_nofpu_id);
 463     break;
 464   case handle_exception_from_callee_id:
 465     break;
 466   default:  ShouldNotReachHere();
 467   }
 468 
 469   return oop_maps;
 470 }
 471 
 472 
 473 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
 474   // incoming parameters
 475   const Register exception_oop = r0;
 476   // callee-saved copy of exception_oop during runtime call
 477   const Register exception_oop_callee_saved = r19;
 478   // other registers used in this stub
 479   const Register exception_pc = r3;
 480   const Register handler_addr = r1;
 481 
 482   // verify that only r0, is valid at this time
 483   __ invalidate_registers(false, true, true, true, true, true);
 484 
 485 #ifdef ASSERT
 486   // check that fields in JavaThread for exception oop and issuing pc are empty
 487   Label oop_empty;
 488   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 489   __ cbz(rscratch1, oop_empty);
 490   __ stop("exception oop must be empty");
 491   __ bind(oop_empty);
 492 
 493   Label pc_empty;
 494   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 495   __ cbz(rscratch1, pc_empty);
 496   __ stop("exception pc must be empty");
 497   __ bind(pc_empty);
 498 #endif
 499 
 500   // Save our return address because
 501   // exception_handler_for_return_address will destroy it.  We also
 502   // save exception_oop
 503   __ stp(lr, exception_oop, Address(__ pre(sp, -2 * wordSize)));
 504 
 505   // search the exception handler address of the caller (using the return address)
 506   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, lr);
 507   // r0: exception handler address of the caller
 508 
 509   // Only R0 is valid at this time; all other registers have been
 510   // destroyed by the call.
 511   __ invalidate_registers(false, true, true, true, false, true);
 512 
 513   // move result of call into correct register
 514   __ mov(handler_addr, r0);
 515 
 516   // get throwing pc (= return address).
 517   // lr has been destroyed by the call
 518   __ ldp(lr, exception_oop, Address(__ post(sp, 2 * wordSize)));
 519   __ mov(r3, lr);
 520 
 521   __ verify_not_null_oop(exception_oop);
 522 
 523   // continue at exception handler (return address removed)
 524   // note: do *not* remove arguments when unwinding the
 525   //       activation since the caller assumes having
 526   //       all arguments on the stack when entering the
 527   //       runtime to determine the exception handler
 528   //       (GC happens at call site with arguments!)
 529   // r0: exception oop
 530   // r3: throwing pc
 531   // r1: exception handler
 532   __ br(handler_addr);
 533 }
 534 
 535 
 536 
 537 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 538   // use the maximum number of runtime-arguments here because it is difficult to
 539   // distinguish each RT-Call.
 540   // Note: This number affects also the RT-Call in generate_handle_exception because
 541   //       the oop-map is shared for all calls.
 542   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 543   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 544 
 545   OopMap* oop_map = save_live_registers(sasm);
 546 
 547   __ mov(c_rarg0, rthread);
 548   Label retaddr;
 549   __ set_last_Java_frame(sp, rfp, retaddr, rscratch1);
 550   // do the call
 551   __ lea(rscratch1, RuntimeAddress(target));
 552   __ blr(rscratch1);
 553   __ bind(retaddr);
 554   OopMapSet* oop_maps = new OopMapSet();
 555   oop_maps->add_gc_map(__ offset(), oop_map);
 556   // verify callee-saved register
 557 #ifdef ASSERT
 558   { Label L;
 559     __ get_thread(rscratch1);
 560     __ cmp(rthread, rscratch1);
 561     __ br(Assembler::EQ, L);
 562     __ stop("StubAssembler::call_RT: rthread not callee saved?");
 563     __ bind(L);
 564   }
 565 #endif
 566 
 567   __ reset_last_Java_frame(true);
 568 
 569 #ifdef ASSERT
 570   // check that fields in JavaThread for exception oop and issuing pc are empty
 571   Label oop_empty;
 572   __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 573   __ cbz(rscratch1, oop_empty);
 574   __ stop("exception oop must be empty");
 575   __ bind(oop_empty);
 576 
 577   Label pc_empty;
 578   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 579   __ cbz(rscratch1, pc_empty);
 580   __ stop("exception pc must be empty");
 581   __ bind(pc_empty);
 582 #endif
 583 
 584   // Runtime will return true if the nmethod has been deoptimized, this is the
 585   // expected scenario and anything else is  an error. Note that we maintain a
 586   // check on the result purely as a defensive measure.
 587   Label no_deopt;
 588   __ cbz(r0, no_deopt);                                // Have we deoptimized?
 589 
 590   // Perform a re-execute. The proper return  address is already on the stack,
 591   // we just need  to restore registers, pop  all of our frame  but the return
 592   // address and jump to the deopt blob.
 593   restore_live_registers(sasm);
 594   __ leave();
 595   __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
 596 
 597   __ bind(no_deopt);
 598   __ stop("deopt not performed");
 599 
 600   return oop_maps;
 601 }
 602 
 603 
 604 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 605 
 606   const Register exception_oop = r0;
 607   const Register exception_pc  = r3;
 608 
 609   // for better readability
 610   const bool must_gc_arguments = true;
 611   const bool dont_gc_arguments = false;
 612 
 613   // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
 614   bool save_fpu_registers = true;
 615 
 616   // stub code & info for the different stubs
 617   OopMapSet* oop_maps = NULL;
 618   OopMap* oop_map = NULL;
 619   switch (id) {
 620     {
 621     case forward_exception_id:
 622       {
 623         oop_maps = generate_handle_exception(id, sasm);
 624         __ leave();
 625         __ ret(lr);
 626       }
 627       break;
 628 
 629     case throw_div0_exception_id:
 630       { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments, does_not_return);
 631         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 632       }
 633       break;
 634 
 635     case throw_null_pointer_exception_id:
 636       { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments, does_not_return);
 637         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 638       }
 639       break;
 640 
 641     case new_instance_id:
 642     case fast_new_instance_id:
 643     case fast_new_instance_init_check_id:
 644       {
 645         Register klass = r3; // Incoming
 646         Register obj   = r0; // Result
 647 
 648         if (id == new_instance_id) {
 649           __ set_info("new_instance", dont_gc_arguments);
 650         } else if (id == fast_new_instance_id) {
 651           __ set_info("fast new_instance", dont_gc_arguments);
 652         } else {
 653           assert(id == fast_new_instance_init_check_id, "bad StubID");
 654           __ set_info("fast new_instance init check", dont_gc_arguments);
 655         }
 656 
 657         // If TLAB is disabled, see if there is support for inlining contiguous
 658         // allocations.
 659         // Otherwise, just go to the slow path.
 660         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 661             !UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 662           Label slow_path;
 663           Register obj_size = r19;
 664           Register t1       = r10;
 665           Register t2       = r11;
 666           assert_different_registers(klass, obj, obj_size, t1, t2);
 667 
 668           __ stp(r19, zr, Address(__ pre(sp, -2 * wordSize)));
 669 
 670           if (id == fast_new_instance_init_check_id) {
 671             // make sure the klass is initialized
 672             __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
 673             __ cmpw(rscratch1, InstanceKlass::fully_initialized);
 674             __ br(Assembler::NE, slow_path);
 675           }
 676 
 677 #ifdef ASSERT
 678           // assert object can be fast path allocated
 679           {
 680             Label ok, not_ok;
 681             __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 682             __ cmp(obj_size, (u1)0);
 683             __ br(Assembler::LE, not_ok);  // make sure it's an instance (LH > 0)
 684             __ tstw(obj_size, Klass::_lh_instance_slow_path_bit);
 685             __ br(Assembler::EQ, ok);
 686             __ bind(not_ok);
 687             __ stop("assert(can be fast path allocated)");
 688             __ should_not_reach_here();
 689             __ bind(ok);
 690           }
 691 #endif // ASSERT
 692 
 693           // get the instance size (size is postive so movl is fine for 64bit)
 694           __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 695 
 696           __ eden_allocate(obj, obj_size, 0, t1, slow_path);
 697 
 698           __ initialize_object(obj, klass, obj_size, 0, t1, t2, /* is_tlab_allocated */ false);
 699           __ verify_oop(obj);
 700           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 701           __ ret(lr);
 702 
 703           __ bind(slow_path);
 704           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 705         }
 706 
 707         __ enter();
 708         OopMap* map = save_live_registers(sasm);
 709         int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
 710         oop_maps = new OopMapSet();
 711         oop_maps->add_gc_map(call_offset, map);
 712         restore_live_registers_except_r0(sasm);
 713         __ verify_oop(obj);
 714         __ leave();
 715         __ ret(lr);
 716 
 717         // r0,: new instance
 718       }
 719 
 720       break;
 721 
 722     case load_klass_id:
 723       {
 724         StubFrame f(sasm, "load_klass", dont_gc_arguments);
 725         save_live_registers_no_oop_map(sasm, true);
 726         f.load_argument(0, r0); // obj
 727         __ call_VM_leaf(CAST_FROM_FN_PTR(address, oopDesc::load_klass_runtime), r0);
 728         restore_live_registers_except_r0(sasm, true);
 729       }
 730       break;
 731 
 732     case counter_overflow_id:
 733       {
 734         Register bci = r0, method = r1;
 735         __ enter();
 736         OopMap* map = save_live_registers(sasm);
 737         // Retrieve bci
 738         __ ldrw(bci, Address(rfp, 2*BytesPerWord));
 739         // And a pointer to the Method*
 740         __ ldr(method, Address(rfp, 3*BytesPerWord));
 741         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
 742         oop_maps = new OopMapSet();
 743         oop_maps->add_gc_map(call_offset, map);
 744         restore_live_registers(sasm);
 745         __ leave();
 746         __ ret(lr);
 747       }
 748       break;
 749 
 750     case new_type_array_id:
 751     case new_object_array_id:
 752       {
 753         Register length   = r19; // Incoming
 754         Register klass    = r3; // Incoming
 755         Register obj      = r0; // Result
 756 
 757         if (id == new_type_array_id) {
 758           __ set_info("new_type_array", dont_gc_arguments);
 759         } else {
 760           __ set_info("new_object_array", dont_gc_arguments);
 761         }
 762 
 763 #ifdef ASSERT
 764         // assert object type is really an array of the proper kind
 765         {
 766           Label ok;
 767           Register t0 = obj;
 768           __ ldrw(t0, Address(klass, Klass::layout_helper_offset()));
 769           __ asrw(t0, t0, Klass::_lh_array_tag_shift);
 770           int tag = ((id == new_type_array_id)
 771                      ? Klass::_lh_array_tag_type_value
 772                      : Klass::_lh_array_tag_obj_value);
 773           __ mov(rscratch1, tag);
 774           __ cmpw(t0, rscratch1);
 775           __ br(Assembler::EQ, ok);
 776           __ stop("assert(is an array klass)");
 777           __ should_not_reach_here();
 778           __ bind(ok);
 779         }
 780 #endif // ASSERT
 781 
 782         // If TLAB is disabled, see if there is support for inlining contiguous
 783         // allocations.
 784         // Otherwise, just go to the slow path.
 785         if (!UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 786           Register arr_size = r5;
 787           Register t1       = r10;
 788           Register t2       = r11;
 789           Label slow_path;
 790           assert_different_registers(length, klass, obj, arr_size, t1, t2);
 791 
 792           // check that array length is small enough for fast path.
 793           __ mov(rscratch1, C1_MacroAssembler::max_array_allocation_length);
 794           __ cmpw(length, rscratch1);
 795           __ br(Assembler::HI, slow_path);
 796 
 797           // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
 798           // since size is positive ldrw does right thing on 64bit
 799           __ ldrw(t1, Address(klass, Klass::layout_helper_offset()));
 800           // since size is positive movw does right thing on 64bit
 801           __ movw(arr_size, length);
 802           __ lslvw(arr_size, length, t1);
 803           __ ubfx(t1, t1, Klass::_lh_header_size_shift,
 804                   exact_log2(Klass::_lh_header_size_mask + 1));
 805           __ add(arr_size, arr_size, t1);
 806           __ add(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
 807           __ andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
 808 
 809           __ eden_allocate(obj, arr_size, 0, t1, slow_path);  // preserves arr_size
 810 
 811           __ initialize_header(obj, klass, length, t1, t2);
 812           __ ldrb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
 813           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 814           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
 815           __ andr(t1, t1, Klass::_lh_header_size_mask);
 816           __ sub(arr_size, arr_size, t1);  // body length
 817           __ add(t1, t1, obj);       // body start
 818           __ initialize_body(t1, arr_size, 0, t1, t2);
 819           __ membar(Assembler::StoreStore);
 820           __ verify_oop(obj);
 821 
 822           __ ret(lr);
 823 
 824           __ bind(slow_path);
 825         }
 826 
 827         __ enter();
 828         OopMap* map = save_live_registers(sasm);
 829         int call_offset;
 830         if (id == new_type_array_id) {
 831           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
 832         } else {
 833           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
 834         }
 835 
 836         oop_maps = new OopMapSet();
 837         oop_maps->add_gc_map(call_offset, map);
 838         restore_live_registers_except_r0(sasm);
 839 
 840         __ verify_oop(obj);
 841         __ leave();
 842         __ ret(lr);
 843 
 844         // r0: new array
 845       }
 846       break;
 847 
 848     case new_multi_array_id:
 849       { StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
 850         // r0,: klass
 851         // r19,: rank
 852         // r2: address of 1st dimension
 853         OopMap* map = save_live_registers(sasm);
 854         __ mov(c_rarg1, r0);
 855         __ mov(c_rarg3, r2);
 856         __ mov(c_rarg2, r19);
 857         int call_offset = __ call_RT(r0, noreg, CAST_FROM_FN_PTR(address, new_multi_array), r1, r2, r3);
 858 
 859         oop_maps = new OopMapSet();
 860         oop_maps->add_gc_map(call_offset, map);
 861         restore_live_registers_except_r0(sasm);
 862 
 863         // r0,: new multi array
 864         __ verify_oop(r0);
 865       }
 866       break;
 867 
 868     case register_finalizer_id:
 869       {
 870         __ set_info("register_finalizer", dont_gc_arguments);
 871 
 872         // This is called via call_runtime so the arguments
 873         // will be place in C abi locations
 874 
 875         __ verify_oop(c_rarg0);
 876 
 877         // load the klass and check the has finalizer flag
 878         Label register_finalizer;
 879         Register t = r5;
 880         __ load_klass(t, r0);
 881         __ ldrw(t, Address(t, Klass::access_flags_offset()));
 882         __ tbnz(t, exact_log2(JVM_ACC_HAS_FINALIZER), register_finalizer);
 883         __ ret(lr);
 884 
 885         __ bind(register_finalizer);
 886         __ enter();
 887         OopMap* oop_map = save_live_registers(sasm);
 888         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), r0);
 889         oop_maps = new OopMapSet();
 890         oop_maps->add_gc_map(call_offset, oop_map);
 891 
 892         // Now restore all the live registers
 893         restore_live_registers(sasm);
 894 
 895         __ leave();
 896         __ ret(lr);
 897       }
 898       break;
 899 
 900     case throw_class_cast_exception_id:
 901       { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments, does_not_return);
 902         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 903       }
 904       break;
 905 
 906     case throw_incompatible_class_change_error_id:
 907       { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments, does_not_return);
 908         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 909       }
 910       break;
 911 
 912     case slow_subtype_check_id:
 913       {
 914         // Typical calling sequence:
 915         // __ push(klass_RInfo);  // object klass or other subclass
 916         // __ push(sup_k_RInfo);  // array element klass or other superclass
 917         // __ bl(slow_subtype_check);
 918         // Note that the subclass is pushed first, and is therefore deepest.
 919         enum layout {
 920           r0_off, r0_off_hi,
 921           r2_off, r2_off_hi,
 922           r4_off, r4_off_hi,
 923           r5_off, r5_off_hi,
 924           sup_k_off, sup_k_off_hi,
 925           klass_off, klass_off_hi,
 926           framesize,
 927           result_off = sup_k_off
 928         };
 929 
 930         __ set_info("slow_subtype_check", dont_gc_arguments);
 931         __ push(RegSet::of(r0, r2, r4, r5), sp);
 932 
 933         // This is called by pushing args and not with C abi
 934         // __ ldr(r4, Address(sp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
 935         // __ ldr(r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
 936 
 937         __ ldp(r4, r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size));
 938 
 939         Label miss;
 940         __ check_klass_subtype_slow_path(r4, r0, r2, r5, NULL, &miss);
 941 
 942         // fallthrough on success:
 943         __ mov(rscratch1, 1);
 944         __ str(rscratch1, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 945         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 946         __ ret(lr);
 947 
 948         __ bind(miss);
 949         __ str(zr, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 950         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 951         __ ret(lr);
 952       }
 953       break;
 954 
 955     case monitorenter_nofpu_id:
 956       save_fpu_registers = false;
 957       // fall through
 958     case monitorenter_id:
 959       {
 960         StubFrame f(sasm, "monitorenter", dont_gc_arguments);
 961         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 962 
 963         // Called with store_parameter and not C abi
 964 
 965         f.load_argument(1, r0); // r0,: object
 966         f.load_argument(0, r1); // r1,: lock address
 967 
 968         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), r0, r1);
 969 
 970         oop_maps = new OopMapSet();
 971         oop_maps->add_gc_map(call_offset, map);
 972         restore_live_registers(sasm, save_fpu_registers);
 973       }
 974       break;
 975 
 976     case monitorexit_nofpu_id:
 977       save_fpu_registers = false;
 978       // fall through
 979     case monitorexit_id:
 980       {
 981         StubFrame f(sasm, "monitorexit", dont_gc_arguments);
 982         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 983 
 984         // Called with store_parameter and not C abi
 985 
 986         f.load_argument(0, r0); // r0,: lock address
 987 
 988         // note: really a leaf routine but must setup last java sp
 989         //       => use call_RT for now (speed can be improved by
 990         //       doing last java sp setup manually)
 991         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), r0);
 992 
 993         oop_maps = new OopMapSet();
 994         oop_maps->add_gc_map(call_offset, map);
 995         restore_live_registers(sasm, save_fpu_registers);
 996       }
 997       break;
 998 
 999     case deoptimize_id:
1000       {
1001         StubFrame f(sasm, "deoptimize", dont_gc_arguments, does_not_return);
1002         OopMap* oop_map = save_live_registers(sasm);
1003         f.load_argument(0, c_rarg1);
1004         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), c_rarg1);
1005 
1006         oop_maps = new OopMapSet();
1007         oop_maps->add_gc_map(call_offset, oop_map);
1008         restore_live_registers(sasm);
1009         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1010         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1011         __ leave();
1012         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1013       }
1014       break;
1015 
1016     case throw_range_check_failed_id:
1017       { StubFrame f(sasm, "range_check_failed", dont_gc_arguments, does_not_return);
1018         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1019       }
1020       break;
1021 
1022     case unwind_exception_id:
1023       { __ set_info("unwind_exception", dont_gc_arguments);
1024         // note: no stubframe since we are about to leave the current
1025         //       activation and we are calling a leaf VM function only.
1026         generate_unwind_exception(sasm);
1027       }
1028       break;
1029 
1030     case access_field_patching_id:
1031       { StubFrame f(sasm, "access_field_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, access_field_patching));
1034       }
1035       break;
1036 
1037     case load_klass_patching_id:
1038       { StubFrame f(sasm, "load_klass_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_klass_patching));
1041       }
1042       break;
1043 
1044     case load_mirror_patching_id:
1045       { StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments, does_not_return);
1046         // we should set up register map
1047         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
1048       }
1049       break;
1050 
1051     case load_appendix_patching_id:
1052       { StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments, does_not_return);
1053         // we should set up register map
1054         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
1055       }
1056       break;
1057 
1058     case handle_exception_nofpu_id:
1059     case handle_exception_id:
1060       { StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1061         oop_maps = generate_handle_exception(id, sasm);
1062       }
1063       break;
1064 
1065     case handle_exception_from_callee_id:
1066       { StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
1067         oop_maps = generate_handle_exception(id, sasm);
1068       }
1069       break;
1070 
1071     case throw_index_exception_id:
1072       { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments, does_not_return);
1073         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1074       }
1075       break;
1076 
1077     case throw_array_store_exception_id:
1078       { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments, does_not_return);
1079         // tos + 0: link
1080         //     + 1: return address
1081         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
1082       }
1083       break;
1084 
1085     case predicate_failed_trap_id:
1086       {
1087         StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments, does_not_return);
1088 
1089         OopMap* map = save_live_registers(sasm);
1090 
1091         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1092         oop_maps = new OopMapSet();
1093         oop_maps->add_gc_map(call_offset, map);
1094         restore_live_registers(sasm);
1095         __ leave();
1096         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1097         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1098 
1099         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1100       }
1101       break;
1102 
1103     case dtrace_object_alloc_id:
1104       { // c_rarg0: object
1105         StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
1106         save_live_registers(sasm);
1107 
1108         __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), c_rarg0);
1109 
1110         restore_live_registers(sasm);
1111       }
1112       break;
1113 
1114     default:
1115       { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments, does_not_return);
1116         __ mov(r0, (int)id);
1117         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), r0);
1118       }
1119       break;
1120     }
1121   }
1122   return oop_maps;
1123 }
1124 
1125 #undef __
1126 
1127 const char *Runtime1::pd_name_for_address(address entry) { Unimplemented(); return 0; }