1 /*
   2  * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef CPU_X86_VM_NATIVEINST_X86_HPP
  26 #define CPU_X86_VM_NATIVEINST_X86_HPP
  27 
  28 #include "asm/assembler.hpp"
  29 #include "memory/allocation.hpp"
  30 #include "runtime/icache.hpp"
  31 #include "runtime/os.hpp"
  32 
  33 // We have interfaces for the following instructions:
  34 // - NativeInstruction
  35 // - - NativeCall
  36 // - - NativeMovConstReg
  37 // - - NativeMovConstRegPatching
  38 // - - NativeMovRegMem
  39 // - - NativeMovRegMemPatching
  40 // - - NativeJump
  41 // - - NativeIllegalOpCode
  42 // - - NativeGeneralJump
  43 // - - NativeReturn
  44 // - - NativeReturnX (return with argument)
  45 // - - NativePushConst
  46 // - - NativeTstRegMem
  47 
  48 // The base class for different kinds of native instruction abstractions.
  49 // Provides the primitive operations to manipulate code relative to this.
  50 
  51 class NativeInstruction VALUE_OBJ_CLASS_SPEC {
  52   friend class Relocation;
  53 
  54  public:
  55   enum Intel_specific_constants {
  56     nop_instruction_code        = 0x90,
  57     nop_instruction_size        =    1
  58   };
  59 
  60   bool is_nop()                        { return ubyte_at(0) == nop_instruction_code; }
  61   inline bool is_call();
  62   inline bool is_call_reg();
  63   inline bool is_illegal();
  64   inline bool is_return();
  65   inline bool is_jump();
  66   inline bool is_cond_jump();
  67   inline bool is_safepoint_poll();
  68   inline bool is_mov_literal64();
  69 
  70  protected:
  71   address addr_at(int offset) const    { return address(this) + offset; }
  72 
  73   s_char sbyte_at(int offset) const    { return *(s_char*) addr_at(offset); }
  74   u_char ubyte_at(int offset) const    { return *(u_char*) addr_at(offset); }
  75 
  76   jint int_at(int offset) const         { return *(jint*) addr_at(offset); }
  77 
  78   intptr_t ptr_at(int offset) const    { return *(intptr_t*) addr_at(offset); }
  79 
  80   oop  oop_at (int offset) const       { return *(oop*) addr_at(offset); }
  81 
  82 
  83   void set_char_at(int offset, char c)        { *addr_at(offset) = (u_char)c; wrote(offset); }
  84   void set_int_at(int offset, jint  i)        { *(jint*)addr_at(offset) = i;  wrote(offset); }
  85   void set_ptr_at (int offset, intptr_t  ptr) { *(intptr_t*) addr_at(offset) = ptr;  wrote(offset); }
  86   void set_oop_at (int offset, oop  o)        { *(oop*) addr_at(offset) = o;  wrote(offset); }
  87 
  88   // This doesn't really do anything on Intel, but it is the place where
  89   // cache invalidation belongs, generically:
  90   void wrote(int offset);
  91 
  92  public:
  93 
  94   // unit test stuff
  95   static void test() {}                 // override for testing
  96 
  97   inline friend NativeInstruction* nativeInstruction_at(address address);
  98 };
  99 
 100 inline NativeInstruction* nativeInstruction_at(address address) {
 101   NativeInstruction* inst = (NativeInstruction*)address;
 102 #ifdef ASSERT
 103   //inst->verify();
 104 #endif
 105   return inst;
 106 }
 107 
 108 inline NativeCall* nativeCall_at(address address);
 109 // The NativeCall is an abstraction for accessing/manipulating native call imm32/rel32off
 110 // instructions (used to manipulate inline caches, primitive & dll calls, etc.).
 111 
 112 class NativeCall: public NativeInstruction {
 113  public:
 114   enum Intel_specific_constants {
 115     instruction_code            = 0xE8,
 116     instruction_size            =    5,
 117     instruction_offset          =    0,
 118     displacement_offset         =    1,
 119     return_address_offset       =    5
 120   };
 121 
 122   enum { cache_line_size = BytesPerWord };  // conservative estimate!
 123 
 124   address instruction_address() const       { return addr_at(instruction_offset); }
 125   address next_instruction_address() const  { return addr_at(return_address_offset); }
 126   int   displacement() const                { return (jint) int_at(displacement_offset); }
 127   address displacement_address() const      { return addr_at(displacement_offset); }
 128   address return_address() const            { return addr_at(return_address_offset); }
 129   address destination() const;
 130   void  set_destination(address dest)       {
 131 #ifdef AMD64
 132     assert((labs((intptr_t) dest - (intptr_t) return_address())  &
 133             0xFFFFFFFF00000000) == 0,
 134            "must be 32bit offset");
 135 #endif // AMD64
 136     set_int_at(displacement_offset, dest - return_address());
 137   }
 138   void  set_destination_mt_safe(address dest);
 139 
 140   void  verify_alignment() { assert((intptr_t)addr_at(displacement_offset) % BytesPerInt == 0, "must be aligned"); }
 141   void  verify();
 142   void  print();
 143 
 144   // Creation
 145   inline friend NativeCall* nativeCall_at(address address);
 146   inline friend NativeCall* nativeCall_before(address return_address);
 147 
 148   static bool is_call_at(address instr) {
 149     return ((*instr) & 0xFF) == NativeCall::instruction_code;
 150   }
 151 
 152   static bool is_call_before(address return_address) {
 153     return is_call_at(return_address - NativeCall::return_address_offset);
 154   }
 155 
 156   static bool is_call_to(address instr, address target) {
 157     return nativeInstruction_at(instr)->is_call() &&
 158       nativeCall_at(instr)->destination() == target;
 159   }
 160 
 161   // MT-safe patching of a call instruction.
 162   static void insert(address code_pos, address entry);
 163 
 164   static void replace_mt_safe(address instr_addr, address code_buffer);
 165 };
 166 
 167 inline NativeCall* nativeCall_at(address address) {
 168   NativeCall* call = (NativeCall*)(address - NativeCall::instruction_offset);
 169 #ifdef ASSERT
 170   call->verify();
 171 #endif
 172   return call;
 173 }
 174 
 175 inline NativeCall* nativeCall_before(address return_address) {
 176   NativeCall* call = (NativeCall*)(return_address - NativeCall::return_address_offset);
 177 #ifdef ASSERT
 178   call->verify();
 179 #endif
 180   return call;
 181 }
 182 
 183 class NativeCallReg: public NativeInstruction {
 184  public:
 185   enum Intel_specific_constants {
 186     instruction_code            = 0xFF,
 187     instruction_offset          =    0,
 188     return_address_offset_norex =    2,
 189     return_address_offset_rex   =    3
 190   };
 191 
 192   int next_instruction_offset() const  {
 193     if (ubyte_at(0) == NativeCallReg::instruction_code) {
 194       return return_address_offset_norex;
 195     } else {
 196       return return_address_offset_rex;
 197     }
 198   }
 199 };
 200 
 201 // An interface for accessing/manipulating native mov reg, imm32 instructions.
 202 // (used to manipulate inlined 32bit data dll calls, etc.)
 203 class NativeMovConstReg: public NativeInstruction {
 204 #ifdef AMD64
 205   static const bool has_rex = true;
 206   static const int rex_size = 1;
 207 #else
 208   static const bool has_rex = false;
 209   static const int rex_size = 0;
 210 #endif // AMD64
 211  public:
 212   enum Intel_specific_constants {
 213     instruction_code            = 0xB8,
 214     instruction_size            =    1 + rex_size + wordSize,
 215     instruction_offset          =    0,
 216     data_offset                 =    1 + rex_size,
 217     next_instruction_offset     =    instruction_size,
 218     register_mask               = 0x07
 219   };
 220 
 221   address instruction_address() const       { return addr_at(instruction_offset); }
 222   address next_instruction_address() const  { return addr_at(next_instruction_offset); }
 223   intptr_t data() const                     { return ptr_at(data_offset); }
 224   void  set_data(intptr_t x)                { set_ptr_at(data_offset, x); }
 225 
 226   void  verify();
 227   void  print();
 228 
 229   // unit test stuff
 230   static void test() {}
 231 
 232   // Creation
 233   inline friend NativeMovConstReg* nativeMovConstReg_at(address address);
 234   inline friend NativeMovConstReg* nativeMovConstReg_before(address address);
 235 };
 236 
 237 inline NativeMovConstReg* nativeMovConstReg_at(address address) {
 238   NativeMovConstReg* test = (NativeMovConstReg*)(address - NativeMovConstReg::instruction_offset);
 239 #ifdef ASSERT
 240   test->verify();
 241 #endif
 242   return test;
 243 }
 244 
 245 inline NativeMovConstReg* nativeMovConstReg_before(address address) {
 246   NativeMovConstReg* test = (NativeMovConstReg*)(address - NativeMovConstReg::instruction_size - NativeMovConstReg::instruction_offset);
 247 #ifdef ASSERT
 248   test->verify();
 249 #endif
 250   return test;
 251 }
 252 
 253 class NativeMovConstRegPatching: public NativeMovConstReg {
 254  private:
 255     friend NativeMovConstRegPatching* nativeMovConstRegPatching_at(address address) {
 256     NativeMovConstRegPatching* test = (NativeMovConstRegPatching*)(address - instruction_offset);
 257     #ifdef ASSERT
 258       test->verify();
 259     #endif
 260     return test;
 261   }
 262 };
 263 
 264 // An interface for accessing/manipulating native moves of the form:
 265 //      mov[b/w/l/q] [reg + offset], reg   (instruction_code_reg2mem)
 266 //      mov[b/w/l/q] reg, [reg+offset]     (instruction_code_mem2reg
 267 //      mov[s/z]x[w/b/q] [reg + offset], reg
 268 //      fld_s  [reg+offset]
 269 //      fld_d  [reg+offset]
 270 //      fstp_s [reg + offset]
 271 //      fstp_d [reg + offset]
 272 //      mov_literal64  scratch,<pointer> ; mov[b/w/l/q] 0(scratch),reg | mov[b/w/l/q] reg,0(scratch)
 273 //
 274 // Warning: These routines must be able to handle any instruction sequences
 275 // that are generated as a result of the load/store byte,word,long
 276 // macros.  For example: The load_unsigned_byte instruction generates
 277 // an xor reg,reg inst prior to generating the movb instruction.  This
 278 // class must skip the xor instruction.
 279 
 280 class NativeMovRegMem: public NativeInstruction {
 281  public:
 282   enum Intel_specific_constants {
 283     instruction_prefix_wide_lo          = Assembler::REX,
 284     instruction_prefix_wide_hi          = Assembler::REX_WRXB,
 285     instruction_code_xor                = 0x33,
 286     instruction_extended_prefix         = 0x0F,
 287     instruction_code_mem2reg_movslq     = 0x63,
 288     instruction_code_mem2reg_movzxb     = 0xB6,
 289     instruction_code_mem2reg_movsxb     = 0xBE,
 290     instruction_code_mem2reg_movzxw     = 0xB7,
 291     instruction_code_mem2reg_movsxw     = 0xBF,
 292     instruction_operandsize_prefix      = 0x66,
 293     instruction_code_reg2mem            = 0x89,
 294     instruction_code_mem2reg            = 0x8b,
 295     instruction_code_reg2memb           = 0x88,
 296     instruction_code_mem2regb           = 0x8a,
 297     instruction_code_float_s            = 0xd9,
 298     instruction_code_float_d            = 0xdd,
 299     instruction_code_long_volatile      = 0xdf,
 300     instruction_code_xmm_ss_prefix      = 0xf3,
 301     instruction_code_xmm_sd_prefix      = 0xf2,
 302     instruction_code_xmm_code           = 0x0f,
 303     instruction_code_xmm_load           = 0x10,
 304     instruction_code_xmm_store          = 0x11,
 305     instruction_code_xmm_lpd            = 0x12,
 306 
 307     instruction_VEX_prefix_2bytes       = Assembler::VEX_2bytes,
 308     instruction_VEX_prefix_3bytes       = Assembler::VEX_3bytes,
 309 
 310     instruction_size                    = 4,
 311     instruction_offset                  = 0,
 312     data_offset                         = 2,
 313     next_instruction_offset             = 4
 314   };
 315 
 316   // helper
 317   int instruction_start() const;
 318 
 319   address instruction_address() const;
 320 
 321   address next_instruction_address() const;
 322 
 323   int   offset() const;
 324 
 325   void  set_offset(int x);
 326 
 327   void  add_offset_in_bytes(int add_offset)     { set_offset ( ( offset() + add_offset ) ); }
 328 
 329   void verify();
 330   void print ();
 331 
 332   // unit test stuff
 333   static void test() {}
 334 
 335  private:
 336   inline friend NativeMovRegMem* nativeMovRegMem_at (address address);
 337 };
 338 
 339 inline NativeMovRegMem* nativeMovRegMem_at (address address) {
 340   NativeMovRegMem* test = (NativeMovRegMem*)(address - NativeMovRegMem::instruction_offset);
 341 #ifdef ASSERT
 342   test->verify();
 343 #endif
 344   return test;
 345 }
 346 
 347 
 348 // An interface for accessing/manipulating native leal instruction of form:
 349 //        leal reg, [reg + offset]
 350 
 351 class NativeLoadAddress: public NativeMovRegMem {
 352 #ifdef AMD64
 353   static const bool has_rex = true;
 354   static const int rex_size = 1;
 355 #else
 356   static const bool has_rex = false;
 357   static const int rex_size = 0;
 358 #endif // AMD64
 359  public:
 360   enum Intel_specific_constants {
 361     instruction_prefix_wide             = Assembler::REX_W,
 362     instruction_prefix_wide_extended    = Assembler::REX_WB,
 363     lea_instruction_code                = 0x8D,
 364     mov64_instruction_code              = 0xB8
 365   };
 366 
 367   void verify();
 368   void print ();
 369 
 370   // unit test stuff
 371   static void test() {}
 372 
 373  private:
 374   friend NativeLoadAddress* nativeLoadAddress_at (address address) {
 375     NativeLoadAddress* test = (NativeLoadAddress*)(address - instruction_offset);
 376     #ifdef ASSERT
 377       test->verify();
 378     #endif
 379     return test;
 380   }
 381 };
 382 
 383 // jump rel32off
 384 
 385 class NativeJump: public NativeInstruction {
 386  public:
 387   enum Intel_specific_constants {
 388     instruction_code            = 0xe9,
 389     instruction_size            =    5,
 390     instruction_offset          =    0,
 391     data_offset                 =    1,
 392     next_instruction_offset     =    5
 393   };
 394 
 395   address instruction_address() const       { return addr_at(instruction_offset); }
 396   address next_instruction_address() const  { return addr_at(next_instruction_offset); }
 397   address jump_destination() const          {
 398      address dest = (int_at(data_offset)+next_instruction_address());
 399      // 32bit used to encode unresolved jmp as jmp -1
 400      // 64bit can't produce this so it used jump to self.
 401      // Now 32bit and 64bit use jump to self as the unresolved address
 402      // which the inline cache code (and relocs) know about
 403 
 404      // return -1 if jump to self
 405     dest = (dest == (address) this) ? (address) -1 : dest;
 406     return dest;
 407   }
 408 
 409   void  set_jump_destination(address dest)  {
 410     intptr_t val = dest - next_instruction_address();
 411     if (dest == (address) -1) {
 412       val = -5; // jump to self
 413     }
 414 #ifdef AMD64
 415     assert((labs(val)  & 0xFFFFFFFF00000000) == 0 || dest == (address)-1, "must be 32bit offset or -1");
 416 #endif // AMD64
 417     set_int_at(data_offset, (jint)val);
 418   }
 419 
 420   // Creation
 421   inline friend NativeJump* nativeJump_at(address address);
 422 
 423   void verify();
 424 
 425   // Unit testing stuff
 426   static void test() {}
 427 
 428   // Insertion of native jump instruction
 429   static void insert(address code_pos, address entry);
 430   // MT-safe insertion of native jump at verified method entry
 431   static void check_verified_entry_alignment(address entry, address verified_entry);
 432   static void patch_verified_entry(address entry, address verified_entry, address dest);
 433 };
 434 
 435 inline NativeJump* nativeJump_at(address address) {
 436   NativeJump* jump = (NativeJump*)(address - NativeJump::instruction_offset);
 437 #ifdef ASSERT
 438   jump->verify();
 439 #endif
 440   return jump;
 441 }
 442 
 443 // Handles all kinds of jump on Intel. Long/far, conditional/unconditional
 444 class NativeGeneralJump: public NativeInstruction {
 445  public:
 446   enum Intel_specific_constants {
 447     // Constants does not apply, since the lengths and offsets depends on the actual jump
 448     // used
 449     // Instruction codes:
 450     //   Unconditional jumps: 0xE9    (rel32off), 0xEB (rel8off)
 451     //   Conditional jumps:   0x0F8x  (rel32off), 0x7x (rel8off)
 452     unconditional_long_jump  = 0xe9,
 453     unconditional_short_jump = 0xeb,
 454     instruction_size = 5
 455   };
 456 
 457   address instruction_address() const       { return addr_at(0); }
 458   address jump_destination()    const;
 459 
 460   // Creation
 461   inline friend NativeGeneralJump* nativeGeneralJump_at(address address);
 462 
 463   // Insertion of native general jump instruction
 464   static void insert_unconditional(address code_pos, address entry);
 465   static void replace_mt_safe(address instr_addr, address code_buffer);
 466 
 467   void verify();
 468 };
 469 
 470 inline NativeGeneralJump* nativeGeneralJump_at(address address) {
 471   NativeGeneralJump* jump = (NativeGeneralJump*)(address);
 472   debug_only(jump->verify();)
 473   return jump;
 474 }
 475 
 476 class NativePopReg : public NativeInstruction {
 477  public:
 478   enum Intel_specific_constants {
 479     instruction_code            = 0x58,
 480     instruction_size            =    1,
 481     instruction_offset          =    0,
 482     data_offset                 =    1,
 483     next_instruction_offset     =    1
 484   };
 485 
 486   // Insert a pop instruction
 487   static void insert(address code_pos, Register reg);
 488 };
 489 
 490 
 491 class NativeIllegalInstruction: public NativeInstruction {
 492  public:
 493   enum Intel_specific_constants {
 494     instruction_code            = 0x0B0F,    // Real byte order is: 0x0F, 0x0B
 495     instruction_size            =    2,
 496     instruction_offset          =    0,
 497     next_instruction_offset     =    2
 498   };
 499 
 500   // Insert illegal opcode as specific address
 501   static void insert(address code_pos);
 502 };
 503 
 504 // return instruction that does not pop values of the stack
 505 class NativeReturn: public NativeInstruction {
 506  public:
 507   enum Intel_specific_constants {
 508     instruction_code            = 0xC3,
 509     instruction_size            =    1,
 510     instruction_offset          =    0,
 511     next_instruction_offset     =    1
 512   };
 513 };
 514 
 515 // return instruction that does pop values of the stack
 516 class NativeReturnX: public NativeInstruction {
 517  public:
 518   enum Intel_specific_constants {
 519     instruction_code            = 0xC2,
 520     instruction_size            =    2,
 521     instruction_offset          =    0,
 522     next_instruction_offset     =    2
 523   };
 524 };
 525 
 526 // Simple test vs memory
 527 class NativeTstRegMem: public NativeInstruction {
 528  public:
 529   enum Intel_specific_constants {
 530     instruction_rex_prefix_mask = 0xF0,
 531     instruction_rex_prefix      = Assembler::REX,
 532     instruction_code_memXregl   = 0x85,
 533     modrm_mask                  = 0x38, // select reg from the ModRM byte
 534     modrm_reg                   = 0x00  // rax
 535   };
 536 };
 537 
 538 inline bool NativeInstruction::is_illegal()      { return (short)int_at(0) == (short)NativeIllegalInstruction::instruction_code; }
 539 inline bool NativeInstruction::is_call()         { return ubyte_at(0) == NativeCall::instruction_code; }
 540 inline bool NativeInstruction::is_call_reg()     { return ubyte_at(0) == NativeCallReg::instruction_code ||
 541                                                           (ubyte_at(1) == NativeCallReg::instruction_code &&
 542                                                            (ubyte_at(0) == Assembler::REX || ubyte_at(0) == Assembler::REX_B)); }
 543 inline bool NativeInstruction::is_return()       { return ubyte_at(0) == NativeReturn::instruction_code ||
 544                                                           ubyte_at(0) == NativeReturnX::instruction_code; }
 545 inline bool NativeInstruction::is_jump()         { return ubyte_at(0) == NativeJump::instruction_code ||
 546                                                           ubyte_at(0) == 0xEB; /* short jump */ }
 547 inline bool NativeInstruction::is_cond_jump()    { return (int_at(0) & 0xF0FF) == 0x800F /* long jump */ ||
 548                                                           (ubyte_at(0) & 0xF0) == 0x70;  /* short jump */ }
 549 inline bool NativeInstruction::is_safepoint_poll() {
 550 #ifdef AMD64
 551   // Try decoding a near safepoint first:
 552   if (ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl &&
 553       ubyte_at(1) == 0x05) { // 00 rax 101
 554     address fault = addr_at(6) + int_at(2);
 555     NOT_JVMCI(assert(!Assembler::is_polling_page_far(), "unexpected poll encoding");)
 556     return os::is_poll_address(fault);
 557   }
 558   // Now try decoding a far safepoint:
 559   // two cases, depending on the choice of the base register in the address.
 560   if (((ubyte_at(0) & NativeTstRegMem::instruction_rex_prefix_mask) == NativeTstRegMem::instruction_rex_prefix &&
 561        ubyte_at(1) == NativeTstRegMem::instruction_code_memXregl &&
 562        (ubyte_at(2) & NativeTstRegMem::modrm_mask) == NativeTstRegMem::modrm_reg) ||
 563       ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl &&
 564       (ubyte_at(1) & NativeTstRegMem::modrm_mask) == NativeTstRegMem::modrm_reg) {
 565     NOT_JVMCI(assert(Assembler::is_polling_page_far(), "unexpected poll encoding");)
 566     return true;
 567   }
 568   return false;
 569 #else
 570   return ( ubyte_at(0) == NativeMovRegMem::instruction_code_mem2reg ||
 571            ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl ) &&
 572            (ubyte_at(1)&0xC7) == 0x05 && /* Mod R/M == disp32 */
 573            (os::is_poll_address((address)int_at(2)));
 574 #endif // AMD64
 575 }
 576 
 577 inline bool NativeInstruction::is_mov_literal64() {
 578 #ifdef AMD64
 579   return ((ubyte_at(0) == Assembler::REX_W || ubyte_at(0) == Assembler::REX_WB) &&
 580           (ubyte_at(1) & (0xff ^ NativeMovConstReg::register_mask)) == 0xB8);
 581 #else
 582   return false;
 583 #endif // AMD64
 584 }
 585 
 586 #endif // CPU_X86_VM_NATIVEINST_X86_HPP