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