1 /*
   2  * Copyright (c) 1999, 2022, 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     __ authenticate_return_address(exception_pc, rscratch1);
 393 
 394     // make sure that the vm_results are cleared (may be unnecessary)
 395     __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
 396     __ str(zr, Address(rthread, JavaThread::vm_result_2_offset()));
 397     break;
 398   case handle_exception_nofpu_id:
 399   case handle_exception_id:
 400     // At this point all registers MAY be live.
 401     oop_map = save_live_registers(sasm, id != handle_exception_nofpu_id);
 402     break;
 403   case handle_exception_from_callee_id: {
 404     // At this point all registers except exception oop (r0) and
 405     // exception pc (lr) are dead.
 406     const int frame_size = 2 /*fp, return address*/;
 407     oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
 408     sasm->set_frame_size(frame_size);
 409     break;
 410   }
 411   default: ShouldNotReachHere();
 412   }
 413 
 414   // verify that only r0 and r3 are valid at this time
 415   __ invalidate_registers(false, true, true, false, true, true);
 416   // verify that r0 contains a valid exception
 417   __ verify_not_null_oop(exception_oop);
 418 
 419 #ifdef ASSERT
 420   // check that fields in JavaThread for exception oop and issuing pc are
 421   // empty before writing to them
 422   Label oop_empty;
 423   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 424   __ cbz(rscratch1, oop_empty);
 425   __ stop("exception oop already set");
 426   __ bind(oop_empty);
 427 
 428   Label pc_empty;
 429   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 430   __ cbz(rscratch1, pc_empty);
 431   __ stop("exception pc already set");
 432   __ bind(pc_empty);
 433 #endif
 434 
 435   // save exception oop and issuing pc into JavaThread
 436   // (exception handler will load it from here)
 437   __ str(exception_oop, Address(rthread, JavaThread::exception_oop_offset()));
 438   __ str(exception_pc, Address(rthread, JavaThread::exception_pc_offset()));
 439 
 440   // patch throwing pc into return address (has bci & oop map)
 441   __ protect_return_address(exception_pc, rscratch1);
 442   __ str(exception_pc, Address(rfp, 1*BytesPerWord));
 443 
 444   // compute the exception handler.
 445   // the exception oop and the throwing pc are read from the fields in JavaThread
 446   int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
 447   oop_maps->add_gc_map(call_offset, oop_map);
 448 
 449   // r0: handler address
 450   //      will be the deopt blob if nmethod was deoptimized while we looked up
 451   //      handler regardless of whether handler existed in the nmethod.
 452 
 453   // only r0 is valid at this time, all other registers have been destroyed by the runtime call
 454   __ invalidate_registers(false, true, true, true, true, true);
 455 
 456   // patch the return address, this stub will directly return to the exception handler
 457   __ protect_return_address(r0, rscratch1);
 458   __ str(r0, Address(rfp, 1*BytesPerWord));
 459 
 460   switch (id) {
 461   case forward_exception_id:
 462   case handle_exception_nofpu_id:
 463   case handle_exception_id:
 464     // Restore the registers that were saved at the beginning.
 465     restore_live_registers(sasm, id != handle_exception_nofpu_id);
 466     break;
 467   case handle_exception_from_callee_id:
 468     break;
 469   default:  ShouldNotReachHere();
 470   }
 471 
 472   return oop_maps;
 473 }
 474 
 475 
 476 void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
 477   // incoming parameters
 478   const Register exception_oop = r0;
 479   // callee-saved copy of exception_oop during runtime call
 480   const Register exception_oop_callee_saved = r19;
 481   // other registers used in this stub
 482   const Register exception_pc = r3;
 483   const Register handler_addr = r1;
 484 
 485   // verify that only r0, is valid at this time
 486   __ invalidate_registers(false, true, true, true, true, true);
 487 
 488 #ifdef ASSERT
 489   // check that fields in JavaThread for exception oop and issuing pc are empty
 490   Label oop_empty;
 491   __ ldr(rscratch1, Address(rthread, JavaThread::exception_oop_offset()));
 492   __ cbz(rscratch1, oop_empty);
 493   __ stop("exception oop must be empty");
 494   __ bind(oop_empty);
 495 
 496   Label pc_empty;
 497   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 498   __ cbz(rscratch1, pc_empty);
 499   __ stop("exception pc must be empty");
 500   __ bind(pc_empty);
 501 #endif
 502 
 503   // Save our return address because
 504   // exception_handler_for_return_address will destroy it.  We also
 505   // save exception_oop
 506   __ mov(r3, lr);
 507   __ protect_return_address();
 508   __ stp(lr, exception_oop, Address(__ pre(sp, -2 * wordSize)));
 509 
 510   // search the exception handler address of the caller (using the return address)
 511   __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rthread, r3);
 512   // r0: exception handler address of the caller
 513 
 514   // Only R0 is valid at this time; all other registers have been
 515   // destroyed by the call.
 516   __ invalidate_registers(false, true, true, true, false, true);
 517 
 518   // move result of call into correct register
 519   __ mov(handler_addr, r0);
 520 
 521   // get throwing pc (= return address).
 522   // lr has been destroyed by the call
 523   __ ldp(lr, exception_oop, Address(__ post(sp, 2 * wordSize)));
 524   __ authenticate_return_address();
 525   __ mov(r3, lr);
 526 
 527   __ verify_not_null_oop(exception_oop);
 528 
 529   // continue at exception handler (return address removed)
 530   // note: do *not* remove arguments when unwinding the
 531   //       activation since the caller assumes having
 532   //       all arguments on the stack when entering the
 533   //       runtime to determine the exception handler
 534   //       (GC happens at call site with arguments!)
 535   // r0: exception oop
 536   // r3: throwing pc
 537   // r1: exception handler
 538   __ br(handler_addr);
 539 }
 540 
 541 
 542 
 543 OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
 544   // use the maximum number of runtime-arguments here because it is difficult to
 545   // distinguish each RT-Call.
 546   // Note: This number affects also the RT-Call in generate_handle_exception because
 547   //       the oop-map is shared for all calls.
 548   DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
 549   assert(deopt_blob != NULL, "deoptimization blob must have been created");
 550 
 551   OopMap* oop_map = save_live_registers(sasm);
 552 
 553   __ mov(c_rarg0, rthread);
 554   Label retaddr;
 555   __ set_last_Java_frame(sp, rfp, retaddr, rscratch1);
 556   // do the call
 557   __ lea(rscratch1, RuntimeAddress(target));
 558   __ blr(rscratch1);
 559   __ bind(retaddr);
 560   OopMapSet* oop_maps = new OopMapSet();
 561   oop_maps->add_gc_map(__ offset(), oop_map);
 562   // verify callee-saved register
 563 #ifdef ASSERT
 564   { Label L;
 565     __ get_thread(rscratch1);
 566     __ cmp(rthread, rscratch1);
 567     __ br(Assembler::EQ, L);
 568     __ stop("StubAssembler::call_RT: rthread not callee saved?");
 569     __ bind(L);
 570   }
 571 #endif
 572 
 573   __ reset_last_Java_frame(true);
 574 
 575 #ifdef ASSERT
 576   // check that fields in JavaThread for exception oop and issuing pc are empty
 577   Label oop_empty;
 578   __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
 579   __ cbz(rscratch1, oop_empty);
 580   __ stop("exception oop must be empty");
 581   __ bind(oop_empty);
 582 
 583   Label pc_empty;
 584   __ ldr(rscratch1, Address(rthread, JavaThread::exception_pc_offset()));
 585   __ cbz(rscratch1, pc_empty);
 586   __ stop("exception pc must be empty");
 587   __ bind(pc_empty);
 588 #endif
 589 
 590   // Runtime will return true if the nmethod has been deoptimized, this is the
 591   // expected scenario and anything else is  an error. Note that we maintain a
 592   // check on the result purely as a defensive measure.
 593   Label no_deopt;
 594   __ cbz(r0, no_deopt);                                // Have we deoptimized?
 595 
 596   // Perform a re-execute. The proper return  address is already on the stack,
 597   // we just need  to restore registers, pop  all of our frame  but the return
 598   // address and jump to the deopt blob.
 599   restore_live_registers(sasm);
 600   __ leave();
 601   __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
 602 
 603   __ bind(no_deopt);
 604   __ stop("deopt not performed");
 605 
 606   return oop_maps;
 607 }
 608 
 609 
 610 OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
 611 
 612   const Register exception_oop = r0;
 613   const Register exception_pc  = r3;
 614 
 615   // for better readability
 616   const bool must_gc_arguments = true;
 617   const bool dont_gc_arguments = false;
 618 
 619   // default value; overwritten for some optimized stubs that are called from methods that do not use the fpu
 620   bool save_fpu_registers = true;
 621 
 622   // stub code & info for the different stubs
 623   OopMapSet* oop_maps = NULL;
 624   OopMap* oop_map = NULL;
 625   switch (id) {
 626     {
 627     case forward_exception_id:
 628       {
 629         oop_maps = generate_handle_exception(id, sasm);
 630         __ leave();
 631         __ ret(lr);
 632       }
 633       break;
 634 
 635     case throw_div0_exception_id:
 636       { StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments, does_not_return);
 637         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
 638       }
 639       break;
 640 
 641     case throw_null_pointer_exception_id:
 642       { StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments, does_not_return);
 643         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
 644       }
 645       break;
 646 
 647     case new_instance_id:
 648     case fast_new_instance_id:
 649     case fast_new_instance_init_check_id:
 650       {
 651         Register klass = r3; // Incoming
 652         Register obj   = r0; // Result
 653 
 654         if (id == new_instance_id) {
 655           __ set_info("new_instance", dont_gc_arguments);
 656         } else if (id == fast_new_instance_id) {
 657           __ set_info("fast new_instance", dont_gc_arguments);
 658         } else {
 659           assert(id == fast_new_instance_init_check_id, "bad StubID");
 660           __ set_info("fast new_instance init check", dont_gc_arguments);
 661         }
 662 
 663         // If TLAB is disabled, see if there is support for inlining contiguous
 664         // allocations.
 665         // Otherwise, just go to the slow path.
 666         if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
 667             !UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 668           Label slow_path;
 669           Register obj_size = r19;
 670           Register t1       = r10;
 671           Register t2       = r11;
 672           assert_different_registers(klass, obj, obj_size, t1, t2);
 673 
 674           __ stp(r19, zr, Address(__ pre(sp, -2 * wordSize)));
 675 
 676           if (id == fast_new_instance_init_check_id) {
 677             // make sure the klass is initialized
 678             __ ldrb(rscratch1, Address(klass, InstanceKlass::init_state_offset()));
 679             __ cmpw(rscratch1, InstanceKlass::fully_initialized);
 680             __ br(Assembler::NE, slow_path);
 681           }
 682 
 683 #ifdef ASSERT
 684           // assert object can be fast path allocated
 685           {
 686             Label ok, not_ok;
 687             __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 688             __ cmp(obj_size, (u1)0);
 689             __ br(Assembler::LE, not_ok);  // make sure it's an instance (LH > 0)
 690             __ tstw(obj_size, Klass::_lh_instance_slow_path_bit);
 691             __ br(Assembler::EQ, ok);
 692             __ bind(not_ok);
 693             __ stop("assert(can be fast path allocated)");
 694             __ should_not_reach_here();
 695             __ bind(ok);
 696           }
 697 #endif // ASSERT
 698 
 699           // get the instance size (size is positive so movl is fine for 64bit)
 700           __ ldrw(obj_size, Address(klass, Klass::layout_helper_offset()));
 701 
 702           __ eden_allocate(obj, obj_size, 0, t1, slow_path);
 703 
 704           __ initialize_object(obj, klass, obj_size, 0, t1, t2, /* is_tlab_allocated */ false);
 705           __ verify_oop(obj);
 706           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 707           __ ret(lr);
 708 
 709           __ bind(slow_path);
 710           __ ldp(r19, zr, Address(__ post(sp, 2 * wordSize)));
 711         }
 712 
 713         __ enter();
 714         OopMap* map = save_live_registers(sasm);
 715         int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
 716         oop_maps = new OopMapSet();
 717         oop_maps->add_gc_map(call_offset, map);
 718         restore_live_registers_except_r0(sasm);
 719         __ verify_oop(obj);
 720         __ leave();
 721         __ ret(lr);
 722 
 723         // r0,: new instance
 724       }
 725 
 726       break;
 727 
 728     case load_klass_id:
 729       {
 730         StubFrame f(sasm, "load_klass", dont_gc_arguments);
 731         save_live_registers_no_oop_map(sasm, true);
 732         f.load_argument(0, r0); // obj
 733         __ call_VM_leaf(CAST_FROM_FN_PTR(address, oopDesc::load_nklass_runtime), r0);
 734         restore_live_registers_except_r0(sasm, true);
 735       }
 736       break;
 737 
 738     case counter_overflow_id:
 739       {
 740         Register bci = r0, method = r1;
 741         __ enter();
 742         OopMap* map = save_live_registers(sasm);
 743         // Retrieve bci
 744         __ ldrw(bci, Address(rfp, 2*BytesPerWord));
 745         // And a pointer to the Method*
 746         __ ldr(method, Address(rfp, 3*BytesPerWord));
 747         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
 748         oop_maps = new OopMapSet();
 749         oop_maps->add_gc_map(call_offset, map);
 750         restore_live_registers(sasm);
 751         __ leave();
 752         __ ret(lr);
 753       }
 754       break;
 755 
 756     case new_type_array_id:
 757     case new_object_array_id:
 758       {
 759         Register length   = r19; // Incoming
 760         Register klass    = r3; // Incoming
 761         Register obj      = r0; // Result
 762 
 763         if (id == new_type_array_id) {
 764           __ set_info("new_type_array", dont_gc_arguments);
 765         } else {
 766           __ set_info("new_object_array", dont_gc_arguments);
 767         }
 768 
 769 #ifdef ASSERT
 770         // assert object type is really an array of the proper kind
 771         {
 772           Label ok;
 773           Register t0 = obj;
 774           __ ldrw(t0, Address(klass, Klass::layout_helper_offset()));
 775           __ asrw(t0, t0, Klass::_lh_array_tag_shift);
 776           int tag = ((id == new_type_array_id)
 777                      ? Klass::_lh_array_tag_type_value
 778                      : Klass::_lh_array_tag_obj_value);
 779           __ mov(rscratch1, tag);
 780           __ cmpw(t0, rscratch1);
 781           __ br(Assembler::EQ, ok);
 782           __ stop("assert(is an array klass)");
 783           __ should_not_reach_here();
 784           __ bind(ok);
 785         }
 786 #endif // ASSERT
 787 
 788         // If TLAB is disabled, see if there is support for inlining contiguous
 789         // allocations.
 790         // Otherwise, just go to the slow path.
 791         if (!UseTLAB && Universe::heap()->supports_inline_contig_alloc()) {
 792           Register arr_size = r5;
 793           Register t1       = r10;
 794           Register t2       = r11;
 795           Label slow_path;
 796           assert_different_registers(length, klass, obj, arr_size, t1, t2);
 797 
 798           // check that array length is small enough for fast path.
 799           __ mov(rscratch1, C1_MacroAssembler::max_array_allocation_length);
 800           __ cmpw(length, rscratch1);
 801           __ br(Assembler::HI, slow_path);
 802 
 803           // get the allocation size: round_up(hdr + length << (layout_helper & 0x1F))
 804           // since size is positive ldrw does right thing on 64bit
 805           __ ldrw(t1, Address(klass, Klass::layout_helper_offset()));
 806           // since size is positive movw does right thing on 64bit
 807           __ movw(arr_size, length);
 808           __ lslvw(arr_size, length, t1);
 809           __ ubfx(t1, t1, Klass::_lh_header_size_shift,
 810                   exact_log2(Klass::_lh_header_size_mask + 1));
 811           __ add(arr_size, arr_size, t1);
 812           __ add(arr_size, arr_size, MinObjAlignmentInBytesMask); // align up
 813           __ andr(arr_size, arr_size, ~MinObjAlignmentInBytesMask);
 814 
 815           __ eden_allocate(obj, arr_size, 0, t1, slow_path);  // preserves arr_size
 816 
 817           __ initialize_header(obj, klass, length, t1, t2);
 818           __ ldrb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
 819           assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
 820           assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
 821           __ andr(t1, t1, Klass::_lh_header_size_mask);
 822           __ sub(arr_size, arr_size, t1);  // body length
 823           __ add(t1, t1, obj);       // body start
 824           __ initialize_body(t1, arr_size, 0, t1, t2);
 825           __ membar(Assembler::StoreStore);
 826           __ verify_oop(obj);
 827 
 828           __ ret(lr);
 829 
 830           __ bind(slow_path);
 831         }
 832 
 833         __ enter();
 834         OopMap* map = save_live_registers(sasm);
 835         int call_offset;
 836         if (id == new_type_array_id) {
 837           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
 838         } else {
 839           call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
 840         }
 841 
 842         oop_maps = new OopMapSet();
 843         oop_maps->add_gc_map(call_offset, map);
 844         restore_live_registers_except_r0(sasm);
 845 
 846         __ verify_oop(obj);
 847         __ leave();
 848         __ ret(lr);
 849 
 850         // r0: new array
 851       }
 852       break;
 853 
 854     case new_multi_array_id:
 855       { StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
 856         // r0,: klass
 857         // r19,: rank
 858         // r2: address of 1st dimension
 859         OopMap* map = save_live_registers(sasm);
 860         __ mov(c_rarg1, r0);
 861         __ mov(c_rarg3, r2);
 862         __ mov(c_rarg2, r19);
 863         int call_offset = __ call_RT(r0, noreg, CAST_FROM_FN_PTR(address, new_multi_array), r1, r2, r3);
 864 
 865         oop_maps = new OopMapSet();
 866         oop_maps->add_gc_map(call_offset, map);
 867         restore_live_registers_except_r0(sasm);
 868 
 869         // r0,: new multi array
 870         __ verify_oop(r0);
 871       }
 872       break;
 873 
 874     case register_finalizer_id:
 875       {
 876         __ set_info("register_finalizer", dont_gc_arguments);
 877 
 878         // This is called via call_runtime so the arguments
 879         // will be place in C abi locations
 880 
 881         __ verify_oop(c_rarg0);
 882 
 883         // load the klass and check the has finalizer flag
 884         Label register_finalizer;
 885         Register t = r5;
 886         __ load_klass(t, r0);
 887         __ ldrw(t, Address(t, Klass::access_flags_offset()));
 888         __ tbnz(t, exact_log2(JVM_ACC_HAS_FINALIZER), register_finalizer);
 889         __ ret(lr);
 890 
 891         __ bind(register_finalizer);
 892         __ enter();
 893         OopMap* oop_map = save_live_registers(sasm);
 894         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), r0);
 895         oop_maps = new OopMapSet();
 896         oop_maps->add_gc_map(call_offset, oop_map);
 897 
 898         // Now restore all the live registers
 899         restore_live_registers(sasm);
 900 
 901         __ leave();
 902         __ ret(lr);
 903       }
 904       break;
 905 
 906     case throw_class_cast_exception_id:
 907       { StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments, does_not_return);
 908         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
 909       }
 910       break;
 911 
 912     case throw_incompatible_class_change_error_id:
 913       { StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments, does_not_return);
 914         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
 915       }
 916       break;
 917 
 918     case slow_subtype_check_id:
 919       {
 920         // Typical calling sequence:
 921         // __ push(klass_RInfo);  // object klass or other subclass
 922         // __ push(sup_k_RInfo);  // array element klass or other superclass
 923         // __ bl(slow_subtype_check);
 924         // Note that the subclass is pushed first, and is therefore deepest.
 925         enum layout {
 926           r0_off, r0_off_hi,
 927           r2_off, r2_off_hi,
 928           r4_off, r4_off_hi,
 929           r5_off, r5_off_hi,
 930           sup_k_off, sup_k_off_hi,
 931           klass_off, klass_off_hi,
 932           framesize,
 933           result_off = sup_k_off
 934         };
 935 
 936         __ set_info("slow_subtype_check", dont_gc_arguments);
 937         __ push(RegSet::of(r0, r2, r4, r5), sp);
 938 
 939         // This is called by pushing args and not with C abi
 940         // __ ldr(r4, Address(sp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
 941         // __ ldr(r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
 942 
 943         __ ldp(r4, r0, Address(sp, (sup_k_off) * VMRegImpl::stack_slot_size));
 944 
 945         Label miss;
 946         __ check_klass_subtype_slow_path(r4, r0, r2, r5, NULL, &miss);
 947 
 948         // fallthrough on success:
 949         __ mov(rscratch1, 1);
 950         __ str(rscratch1, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 951         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 952         __ ret(lr);
 953 
 954         __ bind(miss);
 955         __ str(zr, Address(sp, (result_off) * VMRegImpl::stack_slot_size)); // result
 956         __ pop(RegSet::of(r0, r2, r4, r5), sp);
 957         __ ret(lr);
 958       }
 959       break;
 960 
 961     case monitorenter_nofpu_id:
 962       save_fpu_registers = false;
 963       // fall through
 964     case monitorenter_id:
 965       {
 966         StubFrame f(sasm, "monitorenter", dont_gc_arguments);
 967         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 968 
 969         // Called with store_parameter and not C abi
 970 
 971         f.load_argument(1, r0); // r0,: object
 972         f.load_argument(0, r1); // r1,: lock address
 973 
 974         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), r0, r1);
 975 
 976         oop_maps = new OopMapSet();
 977         oop_maps->add_gc_map(call_offset, map);
 978         restore_live_registers(sasm, save_fpu_registers);
 979       }
 980       break;
 981 
 982     case monitorexit_nofpu_id:
 983       save_fpu_registers = false;
 984       // fall through
 985     case monitorexit_id:
 986       {
 987         StubFrame f(sasm, "monitorexit", dont_gc_arguments);
 988         OopMap* map = save_live_registers(sasm, save_fpu_registers);
 989 
 990         // Called with store_parameter and not C abi
 991 
 992         f.load_argument(0, r0); // r0,: lock address
 993 
 994         // note: really a leaf routine but must setup last java sp
 995         //       => use call_RT for now (speed can be improved by
 996         //       doing last java sp setup manually)
 997         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), r0);
 998 
 999         oop_maps = new OopMapSet();
1000         oop_maps->add_gc_map(call_offset, map);
1001         restore_live_registers(sasm, save_fpu_registers);
1002       }
1003       break;
1004 
1005     case deoptimize_id:
1006       {
1007         StubFrame f(sasm, "deoptimize", dont_gc_arguments, does_not_return);
1008         OopMap* oop_map = save_live_registers(sasm);
1009         f.load_argument(0, c_rarg1);
1010         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize), c_rarg1);
1011 
1012         oop_maps = new OopMapSet();
1013         oop_maps->add_gc_map(call_offset, oop_map);
1014         restore_live_registers(sasm);
1015         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1016         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1017         __ leave();
1018         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1019       }
1020       break;
1021 
1022     case throw_range_check_failed_id:
1023       { StubFrame f(sasm, "range_check_failed", dont_gc_arguments, does_not_return);
1024         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
1025       }
1026       break;
1027 
1028     case unwind_exception_id:
1029       { __ set_info("unwind_exception", dont_gc_arguments);
1030         // note: no stubframe since we are about to leave the current
1031         //       activation and we are calling a leaf VM function only.
1032         generate_unwind_exception(sasm);
1033       }
1034       break;
1035 
1036     case access_field_patching_id:
1037       { StubFrame f(sasm, "access_field_patching", dont_gc_arguments, does_not_return);
1038         // we should set up register map
1039         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
1040       }
1041       break;
1042 
1043     case load_klass_patching_id:
1044       { StubFrame f(sasm, "load_klass_patching", dont_gc_arguments, does_not_return);
1045         // we should set up register map
1046         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
1047       }
1048       break;
1049 
1050     case load_mirror_patching_id:
1051       { StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments, does_not_return);
1052         // we should set up register map
1053         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
1054       }
1055       break;
1056 
1057     case load_appendix_patching_id:
1058       { StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments, does_not_return);
1059         // we should set up register map
1060         oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
1061       }
1062       break;
1063 
1064     case handle_exception_nofpu_id:
1065     case handle_exception_id:
1066       { StubFrame f(sasm, "handle_exception", dont_gc_arguments);
1067         oop_maps = generate_handle_exception(id, sasm);
1068       }
1069       break;
1070 
1071     case handle_exception_from_callee_id:
1072       { StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
1073         oop_maps = generate_handle_exception(id, sasm);
1074       }
1075       break;
1076 
1077     case throw_index_exception_id:
1078       { StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments, does_not_return);
1079         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
1080       }
1081       break;
1082 
1083     case throw_array_store_exception_id:
1084       { StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments, does_not_return);
1085         // tos + 0: link
1086         //     + 1: return address
1087         oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
1088       }
1089       break;
1090 
1091     case predicate_failed_trap_id:
1092       {
1093         StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments, does_not_return);
1094 
1095         OopMap* map = save_live_registers(sasm);
1096 
1097         int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
1098         oop_maps = new OopMapSet();
1099         oop_maps->add_gc_map(call_offset, map);
1100         restore_live_registers(sasm);
1101         __ leave();
1102         DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
1103         assert(deopt_blob != NULL, "deoptimization blob must have been created");
1104 
1105         __ far_jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
1106       }
1107       break;
1108 
1109     case dtrace_object_alloc_id:
1110       { // c_rarg0: object
1111         StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
1112         save_live_registers(sasm);
1113 
1114         __ call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), c_rarg0);
1115 
1116         restore_live_registers(sasm);
1117       }
1118       break;
1119 
1120     default:
1121       { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments, does_not_return);
1122         __ mov(r0, (int)id);
1123         __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), r0);
1124       }
1125       break;
1126     }
1127   }
1128   return oop_maps;
1129 }
1130 
1131 #undef __
1132 
1133 const char *Runtime1::pd_name_for_address(address entry) { Unimplemented(); return 0; }