1 /*
   2  * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef CPU_ARM_MACROASSEMBLER_ARM_HPP
  26 #define CPU_ARM_MACROASSEMBLER_ARM_HPP
  27 
  28 #include "code/relocInfo.hpp"
  29 #include "utilities/powerOfTwo.hpp"
  30 
  31 // Introduced AddressLiteral and its subclasses to ease portability from
  32 // x86 and avoid relocation issues
  33 class AddressLiteral {
  34   RelocationHolder _rspec;
  35   // Typically we use AddressLiterals we want to use their rval
  36   // However in some situations we want the lval (effect address) of the item.
  37   // We provide a special factory for making those lvals.
  38   bool _is_lval;
  39 
  40   address          _target;
  41 
  42  private:
  43   static relocInfo::relocType reloc_for_target(address target) {
  44     // Used for ExternalAddress or when the type is not specified
  45     // Sometimes ExternalAddress is used for values which aren't
  46     // exactly addresses, like the card table base.
  47     // external_word_type can't be used for values in the first page
  48     // so just skip the reloc in that case.
  49     return external_word_Relocation::can_be_relocated(target) ? relocInfo::external_word_type : relocInfo::none;
  50   }
  51 
  52   void set_rspec(relocInfo::relocType rtype);
  53 
  54  protected:
  55   // creation
  56   AddressLiteral()
  57     : _is_lval(false),
  58       _target(nullptr)
  59   {}
  60 
  61   public:
  62 
  63   AddressLiteral(address target, relocInfo::relocType rtype) {
  64     _is_lval = false;
  65     _target = target;
  66     set_rspec(rtype);
  67   }
  68 
  69   AddressLiteral(address target, RelocationHolder const& rspec)
  70     : _rspec(rspec),
  71       _is_lval(false),
  72       _target(target)
  73   {}
  74 
  75   AddressLiteral(address target) {
  76     _is_lval = false;
  77     _target = target;
  78     set_rspec(reloc_for_target(target));
  79   }
  80 
  81   AddressLiteral addr() {
  82     AddressLiteral ret = *this;
  83     ret._is_lval = true;
  84     return ret;
  85   }
  86 
  87  private:
  88 
  89   address target() { return _target; }
  90   bool is_lval() { return _is_lval; }
  91 
  92   relocInfo::relocType reloc() const { return _rspec.type(); }
  93   const RelocationHolder& rspec() const { return _rspec; }
  94 
  95   friend class Assembler;
  96   friend class MacroAssembler;
  97   friend class Address;
  98   friend class LIR_Assembler;
  99   friend class InlinedAddress;
 100 };
 101 
 102 class ExternalAddress: public AddressLiteral {
 103 
 104   public:
 105 
 106   ExternalAddress(address target) : AddressLiteral(target) {}
 107 
 108 };
 109 
 110 class InternalAddress: public AddressLiteral {
 111 
 112   public:
 113 
 114   InternalAddress(address target) : AddressLiteral(target, relocInfo::internal_word_type) {}
 115 
 116 };
 117 
 118 // Inlined constants, for use with ldr_literal / bind_literal
 119 // Note: InlinedInteger not supported (use move_slow(Register,int[,cond]))
 120 class InlinedLiteral: StackObj {
 121  public:
 122   Label label; // need to be public for direct access with &
 123   InlinedLiteral() {
 124   }
 125 };
 126 
 127 class InlinedMetadata: public InlinedLiteral {
 128  private:
 129   Metadata *_data;
 130 
 131  public:
 132   InlinedMetadata(Metadata *data): InlinedLiteral() {
 133     _data = data;
 134   }
 135   Metadata *data() { return _data; }
 136 };
 137 
 138 // Currently unused
 139 // class InlinedOop: public InlinedLiteral {
 140 //  private:
 141 //   jobject _jobject;
 142 //
 143 //  public:
 144 //   InlinedOop(jobject target): InlinedLiteral() {
 145 //     _jobject = target;
 146 //   }
 147 //   jobject jobject() { return _jobject; }
 148 // };
 149 
 150 class InlinedAddress: public InlinedLiteral {
 151  private:
 152   AddressLiteral _literal;
 153 
 154  public:
 155 
 156   InlinedAddress(jobject object): InlinedLiteral(), _literal((address)object, relocInfo::oop_type) {
 157     ShouldNotReachHere(); // use mov_oop (or implement InlinedOop)
 158   }
 159 
 160   InlinedAddress(Metadata *data): InlinedLiteral(), _literal((address)data, relocInfo::metadata_type) {
 161     ShouldNotReachHere(); // use InlinedMetadata or mov_metadata
 162   }
 163 
 164   InlinedAddress(address target, const RelocationHolder &rspec): InlinedLiteral(), _literal(target, rspec) {
 165     assert(rspec.type() != relocInfo::oop_type, "Do not use InlinedAddress for oops");
 166     assert(rspec.type() != relocInfo::metadata_type, "Do not use InlinedAddress for metadatas");
 167   }
 168 
 169   InlinedAddress(address target, relocInfo::relocType rtype): InlinedLiteral(), _literal(target, rtype) {
 170     assert(rtype != relocInfo::oop_type, "Do not use InlinedAddress for oops");
 171     assert(rtype != relocInfo::metadata_type, "Do not use InlinedAddress for metadatas");
 172   }
 173 
 174   // Note: default is relocInfo::none for InlinedAddress
 175   InlinedAddress(address target): InlinedLiteral(), _literal(target, relocInfo::none) {
 176   }
 177 
 178   address target() { return _literal.target(); }
 179 
 180   const RelocationHolder& rspec() const { return _literal.rspec(); }
 181 };
 182 
 183 class InlinedString: public InlinedLiteral {
 184  private:
 185   const char* _msg;
 186 
 187  public:
 188   InlinedString(const char* msg): InlinedLiteral() {
 189     _msg = msg;
 190   }
 191   const char* msg() { return _msg; }
 192 };
 193 
 194 class MacroAssembler: public Assembler {
 195 protected:
 196 
 197   // Support for VM calls
 198   //
 199 
 200   // This is the base routine called by the different versions of call_VM_leaf.
 201   void call_VM_leaf_helper(address entry_point, int number_of_arguments);
 202 
 203   // This is the base routine called by the different versions of call_VM. The interpreter
 204   // may customize this version by overriding it for its purposes (e.g., to save/restore
 205   // additional registers when doing a VM call).
 206   virtual void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions);
 207 public:
 208 
 209   MacroAssembler(CodeBuffer* code) : Assembler(code) {}
 210 
 211   // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
 212   // The implementation is only non-empty for the InterpreterMacroAssembler,
 213   // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
 214   virtual void check_and_handle_popframe() {}
 215   virtual void check_and_handle_earlyret() {}
 216 
 217   // By default, we do not need relocation information for non
 218   // patchable absolute addresses. However, when needed by some
 219   // extensions, ignore_non_patchable_relocations can be modified,
 220   // returning false to preserve all relocation information.
 221   inline bool ignore_non_patchable_relocations() { return true; }
 222 
 223   void align(int modulus);
 224   void align(int modulus, int target);
 225 
 226   // Support for VM calls
 227   //
 228   // It is imperative that all calls into the VM are handled via the call_VM methods.
 229   // They make sure that the stack linkage is setup correctly. call_VM's correspond
 230   // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
 231 
 232   void call_VM(Register oop_result, address entry_point, bool check_exceptions = true);
 233   void call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions = true);
 234   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 235   void call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 236 
 237   // The following methods are required by templateTable.cpp,
 238   // but not used on ARM.
 239   void call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
 240   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
 241   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
 242   void call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
 243 
 244   // Note: The super_call_VM calls are not used on ARM
 245 
 246   // Raw call, without saving/restoring registers, exception handling, etc.
 247   // Mainly used from various stubs.
 248   // Note: if 'save_R9_if_scratched' is true, call_VM may on some
 249   // platforms save values on the stack. Set it to false (and handle
 250   // R9 in the callers) if the top of the stack must not be modified
 251   // by call_VM.
 252   void call_VM(address entry_point, bool save_R9_if_scratched);
 253 
 254   void call_VM_leaf(address entry_point);
 255   void call_VM_leaf(address entry_point, Register arg_1);
 256   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
 257   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
 258   void call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
 259 
 260   void get_vm_result_oop(Register oop_result, Register tmp);
 261   void get_vm_result_metadata(Register metadata_result, Register tmp);
 262 
 263   // Always sets/resets sp, which default to SP if (last_sp == noreg)
 264   // Optionally sets/resets fp (use noreg to avoid setting it)
 265   // Optionally sets/resets pc depending on save_last_java_pc flag
 266   // Note: when saving PC, set_last_Java_frame returns PC's offset in the code section
 267   //       (for oop_maps offset computation)
 268   int set_last_Java_frame(Register last_sp, Register last_fp, bool save_last_java_pc, Register tmp);
 269   void reset_last_Java_frame(Register tmp);
 270   // status set in set_last_Java_frame for reset_last_Java_frame
 271   bool _fp_saved;
 272   bool _pc_saved;
 273 
 274 #ifdef PRODUCT
 275 #define BLOCK_COMMENT(str) /* nothing */
 276 #define STOP(error) __ stop(error)
 277 #else
 278 #define BLOCK_COMMENT(str) __ block_comment(str)
 279 #define STOP(error) __ block_comment(error); __ stop(error)
 280 #endif
 281 
 282   void lookup_virtual_method(Register recv_klass,
 283                              Register vtable_index,
 284                              Register method_result);
 285 
 286   // Test sub_klass against super_klass, with fast and slow paths.
 287 
 288   // The fast path produces a tri-state answer: yes / no / maybe-slow.
 289   // One of the three labels can be null, meaning take the fall-through.
 290   // No registers are killed, except temp_regs.
 291   void check_klass_subtype_fast_path(Register sub_klass,
 292                                      Register super_klass,
 293                                      Register temp_reg,
 294                                      Register temp_reg2,
 295                                      Label* L_success,
 296                                      Label* L_failure,
 297                                      Label* L_slow_path);
 298 
 299   // The rest of the type check; must be wired to a corresponding fast path.
 300   // It does not repeat the fast path logic, so don't use it standalone.
 301   // temp_reg3 can be noreg, if no temps are available.
 302   // Updates the sub's secondary super cache as necessary.
 303   // If set_cond_codes:
 304   // - condition codes will be Z on success, NZ on failure.
 305   // - temp_reg will be 0 on success, non-0 on failure
 306   void check_klass_subtype_slow_path(Register sub_klass,
 307                                      Register super_klass,
 308                                      Register temp_reg,
 309                                      Register temp_reg2,
 310                                      Register temp_reg3, // auto assigned if noreg
 311                                      Label* L_success,
 312                                      Label* L_failure,
 313                                      bool set_cond_codes = false);
 314 
 315   // Simplified, combined version, good for typical uses.
 316   // temp_reg3 can be noreg, if no temps are available. It is used only on slow path.
 317   // Falls through on failure.
 318   void check_klass_subtype(Register sub_klass,
 319                            Register super_klass,
 320                            Register temp_reg,
 321                            Register temp_reg2,
 322                            Register temp_reg3, // auto assigned on slow path if noreg
 323                            Label& L_success);
 324 
 325   // Returns address of receiver parameter, using tmp as base register. tmp and params_count can be the same.
 326   Address receiver_argument_address(Register params_base, Register params_count, Register tmp);
 327 
 328   void _verify_oop(Register reg, const char* s, const char* file, int line);
 329   void _verify_oop_addr(Address addr, const char * s, const char* file, int line);
 330 
 331   // TODO: verify method and klass metadata (compare against vptr?)
 332   void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
 333   void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line) {}
 334 
 335 #define verify_oop(reg) _verify_oop(reg, "broken oop " #reg, __FILE__, __LINE__)
 336 #define verify_oop_addr(addr) _verify_oop_addr(addr, "broken oop ", __FILE__, __LINE__)
 337 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
 338 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
 339 
 340   void null_check(Register reg, Register tmp, int offset = -1);
 341   inline void null_check(Register reg) { null_check(reg, noreg, -1); } // for C1 lir_null_check
 342 
 343   // Puts address of allocated object into register `obj` and end of allocated object into register `obj_end`.
 344   void tlab_allocate(Register obj, Register obj_end, Register tmp1,
 345                      RegisterOrConstant size_expression, Label& slow_case);
 346 
 347   void zero_memory(Register start, Register end, Register tmp);
 348 
 349   static bool needs_explicit_null_check(intptr_t offset);
 350   static bool uses_implicit_null_check(void* address);
 351 
 352   void arm_stack_overflow_check(int frame_size_in_bytes, Register tmp);
 353   void arm_stack_overflow_check(Register Rsize, Register tmp);
 354 
 355   void bang_stack_with_offset(int offset) {
 356     ShouldNotReachHere();
 357   }
 358 
 359   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 360   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 361 
 362   void nop() {
 363     mov(R0, R0);
 364   }
 365 
 366   void push(Register rd, AsmCondition cond = al) {
 367     assert(rd != SP, "unpredictable instruction");
 368     str(rd, Address(SP, -wordSize, pre_indexed), cond);
 369   }
 370 
 371   void push(RegisterSet reg_set, AsmCondition cond = al) {
 372     assert(!reg_set.contains(SP), "unpredictable instruction");
 373     stmdb(SP, reg_set, writeback, cond);
 374   }
 375 
 376   void pop(Register rd, AsmCondition cond = al) {
 377     assert(rd != SP, "unpredictable instruction");
 378     ldr(rd, Address(SP, wordSize, post_indexed), cond);
 379   }
 380 
 381   void pop(RegisterSet reg_set, AsmCondition cond = al) {
 382     assert(!reg_set.contains(SP), "unpredictable instruction");
 383     ldmia(SP, reg_set, writeback, cond);
 384   }
 385 
 386   void fpushd(FloatRegister fd, AsmCondition cond = al) {
 387     fstmdbd(SP, FloatRegisterSet(fd), writeback, cond);
 388   }
 389 
 390   void fpushs(FloatRegister fd, AsmCondition cond = al) {
 391     fstmdbs(SP, FloatRegisterSet(fd), writeback, cond);
 392   }
 393 
 394   void fpopd(FloatRegister fd, AsmCondition cond = al) {
 395     fldmiad(SP, FloatRegisterSet(fd), writeback, cond);
 396   }
 397 
 398   void fpops(FloatRegister fd, AsmCondition cond = al) {
 399     fldmias(SP, FloatRegisterSet(fd), writeback, cond);
 400   }
 401 
 402   void fpush(FloatRegisterSet reg_set) {
 403     fstmdbd(SP, reg_set, writeback);
 404   }
 405 
 406   void fpop(FloatRegisterSet reg_set) {
 407     fldmiad(SP, reg_set, writeback);
 408   }
 409 
 410   void fpush_hardfp(FloatRegisterSet reg_set) {
 411 #ifndef __SOFTFP__
 412     fpush(reg_set);
 413 #endif
 414   }
 415 
 416   void fpop_hardfp(FloatRegisterSet reg_set) {
 417 #ifndef __SOFTFP__
 418     fpop(reg_set);
 419 #endif
 420   }
 421 
 422   // Order access primitives
 423   enum Membar_mask_bits {
 424     StoreStore = 1 << 3,
 425     LoadStore  = 1 << 2,
 426     StoreLoad  = 1 << 1,
 427     LoadLoad   = 1 << 0
 428   };
 429 
 430   void membar(Membar_mask_bits mask,
 431               Register tmp,
 432               bool preserve_flags = true,
 433               Register load_tgt = noreg);
 434 
 435   void breakpoint(AsmCondition cond = al);
 436   void stop(const char* msg);
 437   // prints msg and continues
 438   void warn(const char* msg);
 439   void unimplemented(const char* what = "");
 440   void should_not_reach_here()                   { stop("should not reach here"); }
 441   static void debug(const char* msg, const intx* registers);
 442 
 443   // Create a walkable frame to help tracking down who called this code.
 444   // Returns the frame size in words.
 445   int should_not_call_this() {
 446     raw_push(FP, LR);
 447     should_not_reach_here();
 448     flush();
 449     return 2; // frame_size_in_words (FP+LR)
 450   }
 451 
 452   int save_all_registers();
 453   void restore_all_registers();
 454   int save_caller_save_registers();
 455   void restore_caller_save_registers();
 456 
 457   void add_rc(Register dst, Register arg1, RegisterOrConstant arg2);
 458 
 459   // add_slow and mov_slow are used to manipulate offsets larger than 1024,
 460   // these functions are not expected to handle all possible constants,
 461   // only those that can really occur during compilation
 462   void add_slow(Register rd, Register rn, int c);
 463   void sub_slow(Register rd, Register rn, int c);
 464 
 465 
 466   void mov_slow(Register rd, intptr_t c, AsmCondition cond = al);
 467   void mov_slow(Register rd, const char *string);
 468   void mov_slow(Register rd, address addr);
 469 
 470   void patchable_mov_oop(Register rd, jobject o, int oop_index) {
 471     mov_oop(rd, o, oop_index);
 472   }
 473   void mov_oop(Register rd, jobject o, int index = 0, AsmCondition cond = al);
 474 
 475   void patchable_mov_metadata(Register rd, Metadata* o, int index) {
 476     mov_metadata(rd, o, index);
 477   }
 478   void mov_metadata(Register rd, Metadata* o, int index = 0);
 479 
 480   void mov_float(FloatRegister fd, jfloat c, AsmCondition cond = al);
 481   void mov_double(FloatRegister fd, jdouble c, AsmCondition cond = al);
 482 
 483 
 484   // Note: this variant of mov_address assumes the address moves with
 485   // the code. Do *not* implement it with non-relocated instructions,
 486   // unless PC-relative.
 487   void mov_relative_address(Register rd, address addr, AsmCondition cond = al) {
 488     int offset = addr - pc() - 8;
 489     assert((offset & 3) == 0, "bad alignment");
 490     if (offset >= 0) {
 491       assert(AsmOperand::is_rotated_imm(offset), "addr too far");
 492       add(rd, PC, offset, cond);
 493     } else {
 494       assert(AsmOperand::is_rotated_imm(-offset), "addr too far");
 495       sub(rd, PC, -offset, cond);
 496     }
 497   }
 498 
 499   // Runtime address that may vary from one execution to another.
 500   // Warning: do not implement as a PC relative address.
 501   void mov_address(Register rd, address addr) {
 502     mov_address(rd, addr, RelocationHolder::none);
 503   }
 504 
 505   // rspec can be RelocationHolder::none (for ignored symbolic Relocation).
 506   // In that case, the address is absolute and the generated code need
 507   // not be relocable.
 508   void mov_address(Register rd, address addr, RelocationHolder const& rspec) {
 509     assert(rspec.type() != relocInfo::runtime_call_type, "do not use mov_address for runtime calls");
 510     assert(rspec.type() != relocInfo::static_call_type, "do not use mov_address for relocable calls");
 511     if (rspec.type() == relocInfo::none) {
 512       // absolute address, relocation not needed
 513       mov_slow(rd, (intptr_t)addr);
 514       return;
 515     }
 516     if (VM_Version::supports_movw()) {
 517       relocate(rspec);
 518       int c = (int)addr;
 519       movw(rd, c & 0xffff);
 520       if ((unsigned int)c >> 16) {
 521         movt(rd, (unsigned int)c >> 16);
 522       }
 523       return;
 524     }
 525     Label skip_literal;
 526     InlinedAddress addr_literal(addr, rspec);
 527     ldr_literal(rd, addr_literal);
 528     b(skip_literal);
 529     bind_literal(addr_literal);
 530     bind(skip_literal);
 531   }
 532 
 533   // Note: Do not define mov_address for a Label
 534   //
 535   // Load from addresses potentially within the code are now handled
 536   // InlinedLiteral subclasses (to allow more flexibility on how the
 537   // ldr_literal is performed).
 538 
 539   void ldr_literal(Register rd, InlinedAddress& L) {
 540     assert(L.rspec().type() != relocInfo::runtime_call_type, "avoid ldr_literal for calls");
 541     assert(L.rspec().type() != relocInfo::static_call_type, "avoid ldr_literal for calls");
 542     relocate(L.rspec());
 543     ldr(rd, Address(PC, target(L.label) - pc() - 8));
 544   }
 545 
 546   void ldr_literal(Register rd, InlinedString& L) {
 547     const char* msg = L.msg();
 548     if (code()->consts()->contains((address)msg)) {
 549       // string address moves with the code
 550       ldr(rd, Address(PC, ((address)msg) - pc() - 8));
 551       return;
 552     }
 553     // Warning: use external strings with care. They are not relocated
 554     // if the code moves. If needed, use code_string to move them
 555     // to the consts section.
 556     ldr(rd, Address(PC, target(L.label) - pc() - 8));
 557   }
 558 
 559   void ldr_literal(Register rd, InlinedMetadata& L) {
 560     // relocation done in the bind_literal for metadatas
 561     ldr(rd, Address(PC, target(L.label) - pc() - 8));
 562   }
 563 
 564   void bind_literal(InlinedAddress& L) {
 565     bind(L.label);
 566     assert(L.rspec().type() != relocInfo::metadata_type, "Must use InlinedMetadata");
 567     // We currently do not use oop 'bound' literals.
 568     // If the code evolves and the following assert is triggered,
 569     // we need to implement InlinedOop (see InlinedMetadata).
 570     assert(L.rspec().type() != relocInfo::oop_type, "Inlined oops not supported");
 571     // Note: relocation is handled by relocate calls in ldr_literal
 572     AbstractAssembler::emit_address((address)L.target());
 573   }
 574 
 575   void bind_literal(InlinedString& L) {
 576     const char* msg = L.msg();
 577     if (code()->consts()->contains((address)msg)) {
 578       // The Label should not be used; avoid binding it
 579       // to detect errors.
 580       return;
 581     }
 582     bind(L.label);
 583     AbstractAssembler::emit_address((address)L.msg());
 584   }
 585 
 586   void bind_literal(InlinedMetadata& L) {
 587     bind(L.label);
 588     relocate(metadata_Relocation::spec_for_immediate());
 589     AbstractAssembler::emit_address((address)L.data());
 590   }
 591 
 592   void ldr_label(Register rd, Label& L) {
 593     ldr(rd, Address(PC, target(L) - pc() - 8));
 594   }
 595 
 596   void resolve_oop_handle(Register result);
 597   void load_mirror(Register mirror, Register method, Register tmp);
 598 
 599   void enter() {
 600     raw_push(FP, LR);
 601     mov(FP, SP);
 602   }
 603 
 604   void leave() {
 605     mov(SP, FP);
 606     raw_pop(FP, LR);
 607   }
 608 
 609 #define ARM_INSTR_1(common_mnemonic, arm32_mnemonic, arg_type) \
 610   void common_mnemonic(arg_type arg) { \
 611       arm32_mnemonic(arg); \
 612   }
 613 
 614 #define ARM_INSTR_2(common_mnemonic, arm32_mnemonic, arg1_type, arg2_type) \
 615   void common_mnemonic(arg1_type arg1, arg2_type arg2) { \
 616       arm32_mnemonic(arg1, arg2); \
 617   }
 618 
 619 #define ARM_INSTR_3(common_mnemonic, arm32_mnemonic, arg1_type, arg2_type, arg3_type) \
 620   void common_mnemonic(arg1_type arg1, arg2_type arg2, arg3_type arg3) { \
 621       arm32_mnemonic(arg1, arg2, arg3); \
 622   }
 623 
 624   ARM_INSTR_1(jump, bx,  Register)
 625   ARM_INSTR_1(call, blx, Register)
 626 
 627   ARM_INSTR_2(cbz_32,  cbz,  Register, Label&)
 628   ARM_INSTR_2(cbnz_32, cbnz, Register, Label&)
 629 
 630   ARM_INSTR_2(ldr_u32, ldr,  Register, Address)
 631   ARM_INSTR_2(ldr_s32, ldr,  Register, Address)
 632   ARM_INSTR_2(str_32,  str,  Register, Address)
 633 
 634   ARM_INSTR_2(mvn_32,  mvn,  Register, Register)
 635   ARM_INSTR_2(cmp_32,  cmp,  Register, Register)
 636   ARM_INSTR_2(neg_32,  neg,  Register, Register)
 637   ARM_INSTR_2(clz_32,  clz,  Register, Register)
 638   ARM_INSTR_2(rbit_32, rbit, Register, Register)
 639 
 640   ARM_INSTR_2(cmp_32,  cmp,  Register, int)
 641   ARM_INSTR_2(cmn_32,  cmn,  Register, int)
 642 
 643   ARM_INSTR_3(add_32,  add,  Register, Register, Register)
 644   ARM_INSTR_3(sub_32,  sub,  Register, Register, Register)
 645   ARM_INSTR_3(subs_32, subs, Register, Register, Register)
 646   ARM_INSTR_3(mul_32,  mul,  Register, Register, Register)
 647   ARM_INSTR_3(and_32,  andr, Register, Register, Register)
 648   ARM_INSTR_3(orr_32,  orr,  Register, Register, Register)
 649   ARM_INSTR_3(eor_32,  eor,  Register, Register, Register)
 650 
 651   ARM_INSTR_3(add_32,  add,  Register, Register, AsmOperand)
 652   ARM_INSTR_3(sub_32,  sub,  Register, Register, AsmOperand)
 653   ARM_INSTR_3(orr_32,  orr,  Register, Register, AsmOperand)
 654   ARM_INSTR_3(eor_32,  eor,  Register, Register, AsmOperand)
 655   ARM_INSTR_3(and_32,  andr, Register, Register, AsmOperand)
 656 
 657 
 658   ARM_INSTR_3(add_32,  add,  Register, Register, int)
 659   ARM_INSTR_3(adds_32, adds, Register, Register, int)
 660   ARM_INSTR_3(sub_32,  sub,  Register, Register, int)
 661   ARM_INSTR_3(subs_32, subs, Register, Register, int)
 662 
 663   ARM_INSTR_2(tst_32,  tst,  Register, unsigned int)
 664   ARM_INSTR_2(tst_32,  tst,  Register, AsmOperand)
 665 
 666   ARM_INSTR_3(and_32,  andr, Register, Register, uint)
 667   ARM_INSTR_3(orr_32,  orr,  Register, Register, uint)
 668   ARM_INSTR_3(eor_32,  eor,  Register, Register, uint)
 669 
 670   ARM_INSTR_1(cmp_zero_float,  fcmpzs, FloatRegister)
 671   ARM_INSTR_1(cmp_zero_double, fcmpzd, FloatRegister)
 672 
 673   ARM_INSTR_2(ldr_float,   flds,   FloatRegister, Address)
 674   ARM_INSTR_2(str_float,   fsts,   FloatRegister, Address)
 675   ARM_INSTR_2(mov_float,   fcpys,  FloatRegister, FloatRegister)
 676   ARM_INSTR_2(neg_float,   fnegs,  FloatRegister, FloatRegister)
 677   ARM_INSTR_2(abs_float,   fabss,  FloatRegister, FloatRegister)
 678   ARM_INSTR_2(sqrt_float,  fsqrts, FloatRegister, FloatRegister)
 679   ARM_INSTR_2(cmp_float,   fcmps,  FloatRegister, FloatRegister)
 680 
 681   ARM_INSTR_3(add_float,   fadds,  FloatRegister, FloatRegister, FloatRegister)
 682   ARM_INSTR_3(sub_float,   fsubs,  FloatRegister, FloatRegister, FloatRegister)
 683   ARM_INSTR_3(mul_float,   fmuls,  FloatRegister, FloatRegister, FloatRegister)
 684   ARM_INSTR_3(div_float,   fdivs,  FloatRegister, FloatRegister, FloatRegister)
 685 
 686   ARM_INSTR_2(ldr_double,  fldd,   FloatRegister, Address)
 687   ARM_INSTR_2(str_double,  fstd,   FloatRegister, Address)
 688   ARM_INSTR_2(mov_double,  fcpyd,  FloatRegister, FloatRegister)
 689   ARM_INSTR_2(neg_double,  fnegd,  FloatRegister, FloatRegister)
 690   ARM_INSTR_2(cmp_double,  fcmpd,  FloatRegister, FloatRegister)
 691   ARM_INSTR_2(abs_double,  fabsd,  FloatRegister, FloatRegister)
 692   ARM_INSTR_2(sqrt_double, fsqrtd, FloatRegister, FloatRegister)
 693 
 694   ARM_INSTR_3(add_double,  faddd,  FloatRegister, FloatRegister, FloatRegister)
 695   ARM_INSTR_3(sub_double,  fsubd,  FloatRegister, FloatRegister, FloatRegister)
 696   ARM_INSTR_3(mul_double,  fmuld,  FloatRegister, FloatRegister, FloatRegister)
 697   ARM_INSTR_3(div_double,  fdivd,  FloatRegister, FloatRegister, FloatRegister)
 698 
 699   ARM_INSTR_2(convert_f2d, fcvtds, FloatRegister, FloatRegister)
 700   ARM_INSTR_2(convert_d2f, fcvtsd, FloatRegister, FloatRegister)
 701 
 702   ARM_INSTR_2(mov_fpr2gpr_float, fmrs, Register, FloatRegister)
 703 
 704 #undef ARM_INSTR_1
 705 #undef ARM_INSTR_2
 706 #undef ARM_INSTR_3
 707 
 708 
 709 
 710   void tbz(Register rt, int bit, Label& L) {
 711     assert(0 <= bit && bit < BitsPerWord, "bit number is out of range");
 712     tst(rt, 1 << bit);
 713     b(L, eq);
 714   }
 715 
 716   void tbnz(Register rt, int bit, Label& L) {
 717     assert(0 <= bit && bit < BitsPerWord, "bit number is out of range");
 718     tst(rt, 1 << bit);
 719     b(L, ne);
 720   }
 721 
 722   void cbz(Register rt, Label& L) {
 723     cmp(rt, 0);
 724     b(L, eq);
 725   }
 726 
 727   void cbz(Register rt, address target) {
 728     cmp(rt, 0);
 729     b(target, eq);
 730   }
 731 
 732   void cbnz(Register rt, Label& L) {
 733     cmp(rt, 0);
 734     b(L, ne);
 735   }
 736 
 737   void ret(Register dst = LR) {
 738     bx(dst);
 739   }
 740 
 741 
 742   Register zero_register(Register tmp) {
 743     mov(tmp, 0);
 744     return tmp;
 745   }
 746 
 747   void logical_shift_left(Register dst, Register src, int shift) {
 748     mov(dst, AsmOperand(src, lsl, shift));
 749   }
 750 
 751   void logical_shift_left_32(Register dst, Register src, int shift) {
 752     mov(dst, AsmOperand(src, lsl, shift));
 753   }
 754 
 755   void logical_shift_right(Register dst, Register src, int shift) {
 756     mov(dst, AsmOperand(src, lsr, shift));
 757   }
 758 
 759   void arith_shift_right(Register dst, Register src, int shift) {
 760     mov(dst, AsmOperand(src, asr, shift));
 761   }
 762 
 763   void asr_32(Register dst, Register src, int shift) {
 764     mov(dst, AsmOperand(src, asr, shift));
 765   }
 766 
 767   // If <cond> holds, compares r1 and r2. Otherwise, flags are set so that <cond> does not hold.
 768   void cond_cmp(Register r1, Register r2, AsmCondition cond) {
 769     cmp(r1, r2, cond);
 770   }
 771 
 772   // If <cond> holds, compares r and imm. Otherwise, flags are set so that <cond> does not hold.
 773   void cond_cmp(Register r, int imm, AsmCondition cond) {
 774     cmp(r, imm, cond);
 775   }
 776 
 777   void align_reg(Register dst, Register src, int align) {
 778     assert (is_power_of_2(align), "should be");
 779     bic(dst, src, align-1);
 780   }
 781 
 782   void prefetch_read(Address addr) {
 783     pld(addr);
 784   }
 785 
 786   void raw_push(Register r1, Register r2) {
 787     assert(r1->encoding() < r2->encoding(), "should be ordered");
 788     push(RegisterSet(r1) | RegisterSet(r2));
 789   }
 790 
 791   void raw_pop(Register r1, Register r2) {
 792     assert(r1->encoding() < r2->encoding(), "should be ordered");
 793     pop(RegisterSet(r1) | RegisterSet(r2));
 794   }
 795 
 796   void raw_push(Register r1, Register r2, Register r3) {
 797     assert(r1->encoding() < r2->encoding() && r2->encoding() < r3->encoding(), "should be ordered");
 798     push(RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3));
 799   }
 800 
 801   void raw_pop(Register r1, Register r2, Register r3) {
 802     assert(r1->encoding() < r2->encoding() && r2->encoding() < r3->encoding(), "should be ordered");
 803     pop(RegisterSet(r1) | RegisterSet(r2) | RegisterSet(r3));
 804   }
 805 
 806   // Restores registers r1 and r2 previously saved by raw_push(r1, r2, ret_addr) and returns by ret_addr. Clobbers LR.
 807   void raw_pop_and_ret(Register r1, Register r2) {
 808     raw_pop(r1, r2, PC);
 809   }
 810 
 811   void indirect_jump(Address addr, Register scratch) {
 812     ldr(PC, addr);
 813   }
 814 
 815   void indirect_jump(InlinedAddress& literal, Register scratch) {
 816     ldr_literal(PC, literal);
 817   }
 818 
 819   void neg(Register dst, Register src) {
 820     rsb(dst, src, 0);
 821   }
 822 
 823   void branch_if_negative_32(Register r, Label& L) {
 824     // TODO: This function and branch_if_any_negative_32 could possibly
 825     // be revised after the aarch64 removal.
 826     // tbnz is not used instead of tst & b.mi because destination may be out of tbnz range (+-32KB)
 827     // since these methods are used in LIR_Assembler::emit_arraycopy() to jump to stub entry.
 828     tst_32(r, r);
 829     b(L, mi);
 830   }
 831 
 832   void branch_if_any_negative_32(Register r1, Register r2, Register tmp, Label& L) {
 833     orrs(tmp, r1, r2);
 834     b(L, mi);
 835   }
 836 
 837   void branch_if_any_negative_32(Register r1, Register r2, Register r3, Register tmp, Label& L) {
 838     orr_32(tmp, r1, r2);
 839     orrs(tmp, tmp, r3);
 840     b(L, mi);
 841   }
 842 
 843   void add_ptr_scaled_int32(Register dst, Register r1, Register r2, int shift) {
 844       add(dst, r1, AsmOperand(r2, lsl, shift));
 845   }
 846 
 847   void sub_ptr_scaled_int32(Register dst, Register r1, Register r2, int shift) {
 848     sub(dst, r1, AsmOperand(r2, lsl, shift));
 849   }
 850 
 851   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 852   void c2bool(Register x);
 853 
 854     // klass oop manipulations if compressed
 855 
 856   void load_klass(Register dst_klass, Register src_oop, AsmCondition cond = al);
 857 
 858   void store_klass(Register src_klass, Register dst_oop);
 859 
 860 
 861     // oop manipulations
 862 
 863   void load_heap_oop(Register dst, Address src, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
 864   void store_heap_oop(Address obj, Register new_val, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
 865   void store_heap_oop_null(Address obj, Register new_val, Register tmp1 = noreg, Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
 866 
 867   void access_load_at(BasicType type, DecoratorSet decorators, Address src, Register dst, Register tmp1, Register tmp2, Register tmp3);
 868   void access_store_at(BasicType type, DecoratorSet decorators, Address obj, Register new_val, Register tmp1, Register tmp2, Register tmp3, bool is_null);
 869 
 870   void ldr_global_ptr(Register reg, address address_of_global);
 871   void ldr_global_s32(Register reg, address address_of_global);
 872   void ldrb_global(Register reg, address address_of_global);
 873 
 874   // address_placeholder_instruction is invalid instruction and is used
 875   // as placeholder in code for address of label
 876   enum { address_placeholder_instruction = 0xFFFFFFFF };
 877 
 878   void emit_address(Label& L) {
 879     assert(!L.is_bound(), "otherwise address will not be patched");
 880     target(L);       // creates relocation which will be patched later
 881 
 882     assert ((offset() & (wordSize-1)) == 0, "should be aligned by word size");
 883 
 884     AbstractAssembler::emit_address((address)address_placeholder_instruction);
 885   }
 886 
 887   void b(address target, AsmCondition cond = al) {
 888     Assembler::b(target, cond);                 \
 889   }
 890   void b(Label& L, AsmCondition cond = al) {
 891     // internal jumps
 892     Assembler::b(target(L), cond);
 893   }
 894 
 895   void bl(address target, AsmCondition cond = al) {
 896     Assembler::bl(target, cond);
 897   }
 898   void bl(Label& L, AsmCondition cond = al) {
 899     // internal calls
 900     Assembler::bl(target(L), cond);
 901   }
 902 
 903   void adr(Register dest, Label& L, AsmCondition cond = al) {
 904     int delta = target(L) - pc() - 8;
 905     if (delta >= 0) {
 906       add(dest, PC, delta, cond);
 907     } else {
 908       sub(dest, PC, -delta, cond);
 909     }
 910   }
 911 
 912   // Variable-length jump and calls. We now distinguish only the
 913   // patchable case from the other cases. Patchable must be
 914   // distinguised from relocable. Relocable means the generated code
 915   // containing the jump/call may move. Patchable means that the
 916   // targeted address may be changed later.
 917 
 918   // Non patchable versions.
 919   // - used only for relocInfo::runtime_call_type and relocInfo::none
 920   // - may use relative or absolute format (do not use relocInfo::none
 921   //   if the generated code may move)
 922   // - the implementation takes into account switch to THUMB mode if the
 923   //   destination is a THUMB address
 924   // - the implementation supports far targets
 925   //
 926   // To reduce regression risk, scratch still defaults to noreg on
 927   // arm32. This results in patchable instructions. However, if
 928   // patching really matters, the call sites should be modified and
 929   // use patchable_call or patchable_jump. If patching is not required
 930   // and if a register can be cloberred, it should be explicitly
 931   // specified to allow future optimizations.
 932   void jump(address target,
 933             relocInfo::relocType rtype = relocInfo::runtime_call_type,
 934             Register scratch = noreg, AsmCondition cond = al);
 935 
 936   void call(address target,
 937             RelocationHolder rspec, AsmCondition cond = al);
 938 
 939   void call(address target,
 940             relocInfo::relocType rtype = relocInfo::runtime_call_type,
 941             AsmCondition cond = al) {
 942     call(target, Relocation::spec_simple(rtype), cond);
 943   }
 944 
 945   void jump(AddressLiteral dest) {
 946     jump(dest.target(), dest.reloc());
 947   }
 948   void jump(address dest, relocInfo::relocType rtype, AsmCondition cond) {
 949     jump(dest, rtype, Rtemp, cond);
 950   }
 951 
 952   void call(AddressLiteral dest) {
 953     call(dest.target(), dest.reloc());
 954   }
 955 
 956   // Patchable version:
 957   // - set_destination can be used to atomically change the target
 958   //
 959   // The targets for patchable_jump and patchable_call must be in the
 960   // code cache.
 961   // [ including possible extensions of the code cache, like AOT code ]
 962   //
 963   // To reduce regression risk, scratch still defaults to noreg on
 964   // arm32. If a register can be cloberred, it should be explicitly
 965   // specified to allow future optimizations.
 966   void patchable_jump(address target,
 967                       relocInfo::relocType rtype = relocInfo::runtime_call_type,
 968                       Register scratch = noreg, AsmCondition cond = al
 969                       );
 970 
 971   // patchable_call may scratch Rtemp
 972   int patchable_call(address target,
 973                      RelocationHolder const& rspec,
 974                      bool c2 = false);
 975 
 976   int patchable_call(address target,
 977                      relocInfo::relocType rtype,
 978                      bool c2 = false) {
 979     return patchable_call(target, Relocation::spec_simple(rtype), c2);
 980   }
 981 
 982 
 983   static bool _reachable_from_cache(address target);
 984   static bool _cache_fully_reachable();
 985   bool cache_fully_reachable();
 986   bool reachable_from_cache(address target);
 987 
 988   void zero_extend(Register rd, Register rn, int bits);
 989   void sign_extend(Register rd, Register rn, int bits);
 990 
 991   inline void zap_high_non_significant_bits(Register r) {
 992   }
 993 
 994   void cmpoop(Register obj1, Register obj2);
 995 
 996   void long_move(Register rd_lo, Register rd_hi,
 997                  Register rn_lo, Register rn_hi,
 998                  AsmCondition cond = al);
 999   void long_shift(Register rd_lo, Register rd_hi,
1000                   Register rn_lo, Register rn_hi,
1001                   AsmShift shift, Register count);
1002   void long_shift(Register rd_lo, Register rd_hi,
1003                   Register rn_lo, Register rn_hi,
1004                   AsmShift shift, int count);
1005 
1006   void atomic_cas(Register tmpreg1, Register tmpreg2, Register oldval, Register newval, Register base, int offset);
1007   void atomic_cas_bool(Register oldval, Register newval, Register base, int offset, Register tmpreg);
1008   void atomic_cas64(Register temp_lo, Register temp_hi, Register temp_result, Register oldval_lo, Register oldval_hi, Register newval_lo, Register newval_hi, Register base, int offset);
1009 
1010   void cas_for_lock_acquire(Register oldval, Register newval, Register base, Register tmp, Label &slow_case, bool allow_fallthrough_on_failure = false, bool one_shot = false);
1011   void cas_for_lock_release(Register oldval, Register newval, Register base, Register tmp, Label &slow_case, bool allow_fallthrough_on_failure = false, bool one_shot = false);
1012 
1013   // Attempt to fast-lock an object
1014   // Registers:
1015   //  - obj: the object to be locked
1016   //  - t1, t2, t3: temp registers. If corresponding bit in savemask is set, they get saved, otherwise blown.
1017   // Result:
1018   //  - Success: fallthrough
1019   //  - Error:   break to slow, Z cleared.
1020   void fast_lock(Register obj, Register t1, Register t2, Register t3, unsigned savemask, Label& slow);
1021 
1022   // Attempt to fast-unlock an object
1023   // Registers:
1024   //  - obj: the object to be unlocked
1025   //  - t1, t2, t3: temp registers. If corresponding bit in savemask is set, they get saved, otherwise blown.
1026   // Result:
1027   //  - Success: fallthrough
1028   //  - Error:   break to slow, Z cleared.
1029   void fast_unlock(Register obj, Register t1, Register t2, Register t3, unsigned savemask, Label& slow);
1030 
1031 #ifndef PRODUCT
1032   // Preserves flags and all registers.
1033   // On SMP the updated value might not be visible to external observers without a synchronization barrier
1034   void cond_atomic_inc32(AsmCondition cond, int* counter_addr);
1035 #endif // !PRODUCT
1036 
1037   // unconditional non-atomic increment
1038   void inc_counter(address counter_addr, Register tmpreg1, Register tmpreg2);
1039   void inc_counter(uint* counter_addr, Register tmpreg1, Register tmpreg2) {
1040     inc_counter((address) counter_addr, tmpreg1, tmpreg2);
1041   }
1042 
1043   void pd_patch_instruction(address branch, address target, const char* file, int line);
1044 
1045   // Loading and storing values by size and signed-ness;
1046   // size must not exceed wordSize (i.e. 8-byte values are not supported on 32-bit ARM);
1047   // each of these calls generates exactly one load or store instruction,
1048   // so src can be pre- or post-indexed address.
1049   // 32-bit ARM variants also support conditional execution
1050   void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, AsmCondition cond = al);
1051   void store_sized_value(Register src, Address dst, size_t size_in_bytes, AsmCondition cond = al);
1052 
1053   void lookup_interface_method(Register recv_klass,
1054                                Register intf_klass,
1055                                RegisterOrConstant itable_index,
1056                                Register method_result,
1057                                Register temp_reg1,
1058                                Register temp_reg2,
1059                                Label& L_no_such_interface);
1060 
1061 
1062   void floating_cmp(Register dst);
1063 
1064   // improved x86 portability (minimizing source code changes)
1065 
1066   void ldr_literal(Register rd, AddressLiteral addr) {
1067     relocate(addr.rspec());
1068     ldr(rd, Address(PC, addr.target() - pc() - 8));
1069   }
1070 
1071   void lea(Register Rd, AddressLiteral addr) {
1072     // Never dereferenced, as on x86 (lval status ignored)
1073     mov_address(Rd, addr.target(), addr.rspec());
1074   }
1075 
1076   void restore_default_fp_mode();
1077 
1078   void safepoint_poll(Register tmp1, Label& slow_path);
1079   void get_polling_page(Register dest);
1080   void read_polling_page(Register dest, relocInfo::relocType rtype);
1081 
1082   static int ic_check_size();
1083   int ic_check(int end_alignment);
1084 };
1085 
1086 
1087 // The purpose of this class is to build several code fragments of the same size
1088 // in order to allow fast table branch.
1089 
1090 class FixedSizeCodeBlock {
1091 public:
1092   FixedSizeCodeBlock(MacroAssembler* masm, int size_in_instrs, bool enabled);
1093   ~FixedSizeCodeBlock();
1094 
1095 private:
1096   MacroAssembler* _masm;
1097   address _start;
1098   int _size_in_instrs;
1099   bool _enabled;
1100 };
1101 
1102 
1103 #endif // CPU_ARM_MACROASSEMBLER_ARM_HPP