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_code_lea                = 0x8d,
 291 
 292     instruction_VEX_prefix_2bytes       = Assembler::VEX_2bytes,
 293     instruction_VEX_prefix_3bytes       = Assembler::VEX_3bytes,
 294 
 295     instruction_size                    = 4,
 296     instruction_offset                  = 0,
 297     data_offset                         = 2,
 298     next_instruction_offset             = 4
 299   };
 300 
 301   // helper
 302   int instruction_start() const;
 303 
 304   address instruction_address() const;
 305 
 306   address next_instruction_address() const;
 307 
 308   int   offset() const;
 309 
 310   void  set_offset(int x);
 311 
 312   void  add_offset_in_bytes(int add_offset)     { set_offset ( ( offset() + add_offset ) ); }
 313 
 314   void verify();
 315   void print ();
 316 
 317   // unit test stuff
 318   static void test() {}
 319 
 320  private:
 321   inline friend NativeMovRegMem* nativeMovRegMem_at (address address);
 322 };
 323 
 324 inline NativeMovRegMem* nativeMovRegMem_at (address address) {
 325   NativeMovRegMem* test = (NativeMovRegMem*)(address - NativeMovRegMem::instruction_offset);
 326 #ifdef ASSERT
 327   test->verify();
 328 #endif
 329   return test;
 330 }
 331 
 332 class NativeMovRegMemPatching: public NativeMovRegMem {
 333  private:
 334   friend NativeMovRegMemPatching* nativeMovRegMemPatching_at (address address) {
 335     NativeMovRegMemPatching* test = (NativeMovRegMemPatching*)(address - instruction_offset);
 336     #ifdef ASSERT
 337       test->verify();
 338     #endif
 339     return test;
 340   }
 341 };
 342 
 343 
 344 
 345 // An interface for accessing/manipulating native leal instruction of form:
 346 //        leal reg, [reg + offset]
 347 
 348 class NativeLoadAddress: public NativeMovRegMem {
 349 #ifdef AMD64
 350   static const bool has_rex = true;
 351   static const int rex_size = 1;
 352 #else
 353   static const bool has_rex = false;
 354   static const int rex_size = 0;
 355 #endif // AMD64
 356  public:
 357   enum Intel_specific_constants {
 358     instruction_prefix_wide             = Assembler::REX_W,
 359     instruction_prefix_wide_extended    = Assembler::REX_WB,
 360     lea_instruction_code                = 0x8D,
 361     mov64_instruction_code              = 0xB8
 362   };
 363 
 364   void verify();
 365   void print ();
 366 
 367   // unit test stuff
 368   static void test() {}
 369 
 370  private:
 371   friend NativeLoadAddress* nativeLoadAddress_at (address address) {
 372     NativeLoadAddress* test = (NativeLoadAddress*)(address - instruction_offset);
 373     #ifdef ASSERT
 374       test->verify();
 375     #endif
 376     return test;
 377   }
 378 };
 379 
 380 // jump rel32off
 381 
 382 class NativeJump: public NativeInstruction {
 383  public:
 384   enum Intel_specific_constants {
 385     instruction_code            = 0xe9,
 386     instruction_size            =    5,
 387     instruction_offset          =    0,
 388     data_offset                 =    1,
 389     next_instruction_offset     =    5
 390   };
 391 
 392   address instruction_address() const       { return addr_at(instruction_offset); }
 393   address next_instruction_address() const  { return addr_at(next_instruction_offset); }
 394   address jump_destination() const          {
 395      address dest = (int_at(data_offset)+next_instruction_address());
 396      // 32bit used to encode unresolved jmp as jmp -1
 397      // 64bit can't produce this so it used jump to self.
 398      // Now 32bit and 64bit use jump to self as the unresolved address
 399      // which the inline cache code (and relocs) know about
 400 
 401      // return -1 if jump to self
 402     dest = (dest == (address) this) ? (address) -1 : dest;
 403     return dest;
 404   }
 405 
 406   void  set_jump_destination(address dest)  {
 407     intptr_t val = dest - next_instruction_address();
 408     if (dest == (address) -1) {
 409       val = -5; // jump to self
 410     }
 411 #ifdef AMD64
 412     assert((labs(val)  & 0xFFFFFFFF00000000) == 0 || dest == (address)-1, "must be 32bit offset or -1");
 413 #endif // AMD64
 414     set_int_at(data_offset, (jint)val);
 415   }
 416 
 417   // Creation
 418   inline friend NativeJump* nativeJump_at(address address);
 419 
 420   void verify();
 421 
 422   // Unit testing stuff
 423   static void test() {}
 424 
 425   // Insertion of native jump instruction
 426   static void insert(address code_pos, address entry);
 427   // MT-safe insertion of native jump at verified method entry
 428   static void check_verified_entry_alignment(address entry, address verified_entry);
 429   static void patch_verified_entry(address entry, address verified_entry, address dest);
 430 };
 431 
 432 inline NativeJump* nativeJump_at(address address) {
 433   NativeJump* jump = (NativeJump*)(address - NativeJump::instruction_offset);
 434 #ifdef ASSERT
 435   jump->verify();
 436 #endif
 437   return jump;
 438 }
 439 
 440 // Handles all kinds of jump on Intel. Long/far, conditional/unconditional
 441 class NativeGeneralJump: public NativeInstruction {
 442  public:
 443   enum Intel_specific_constants {
 444     // Constants does not apply, since the lengths and offsets depends on the actual jump
 445     // used
 446     // Instruction codes:
 447     //   Unconditional jumps: 0xE9    (rel32off), 0xEB (rel8off)
 448     //   Conditional jumps:   0x0F8x  (rel32off), 0x7x (rel8off)
 449     unconditional_long_jump  = 0xe9,
 450     unconditional_short_jump = 0xeb,
 451     instruction_size = 5
 452   };
 453 
 454   address instruction_address() const       { return addr_at(0); }
 455   address jump_destination()    const;
 456 
 457   // Creation
 458   inline friend NativeGeneralJump* nativeGeneralJump_at(address address);
 459 
 460   // Insertion of native general jump instruction
 461   static void insert_unconditional(address code_pos, address entry);
 462   static void replace_mt_safe(address instr_addr, address code_buffer);
 463 
 464   void verify();
 465 };
 466 
 467 inline NativeGeneralJump* nativeGeneralJump_at(address address) {
 468   NativeGeneralJump* jump = (NativeGeneralJump*)(address);
 469   debug_only(jump->verify();)
 470   return jump;
 471 }
 472 
 473 class NativePopReg : public NativeInstruction {
 474  public:
 475   enum Intel_specific_constants {
 476     instruction_code            = 0x58,
 477     instruction_size            =    1,
 478     instruction_offset          =    0,
 479     data_offset                 =    1,
 480     next_instruction_offset     =    1
 481   };
 482 
 483   // Insert a pop instruction
 484   static void insert(address code_pos, Register reg);
 485 };
 486 
 487 
 488 class NativeIllegalInstruction: public NativeInstruction {
 489  public:
 490   enum Intel_specific_constants {
 491     instruction_code            = 0x0B0F,    // Real byte order is: 0x0F, 0x0B
 492     instruction_size            =    2,
 493     instruction_offset          =    0,
 494     next_instruction_offset     =    2
 495   };
 496 
 497   // Insert illegal opcode as specific address
 498   static void insert(address code_pos);
 499 };
 500 
 501 // return instruction that does not pop values of the stack
 502 class NativeReturn: public NativeInstruction {
 503  public:
 504   enum Intel_specific_constants {
 505     instruction_code            = 0xC3,
 506     instruction_size            =    1,
 507     instruction_offset          =    0,
 508     next_instruction_offset     =    1
 509   };
 510 };
 511 
 512 // return instruction that does pop values of the stack
 513 class NativeReturnX: public NativeInstruction {
 514  public:
 515   enum Intel_specific_constants {
 516     instruction_code            = 0xC2,
 517     instruction_size            =    2,
 518     instruction_offset          =    0,
 519     next_instruction_offset     =    2
 520   };
 521 };
 522 
 523 // Simple test vs memory
 524 class NativeTstRegMem: public NativeInstruction {
 525  public:
 526   enum Intel_specific_constants {
 527     instruction_rex_prefix_mask = 0xF0,
 528     instruction_rex_prefix      = Assembler::REX,
 529     instruction_code_memXregl   = 0x85,
 530     modrm_mask                  = 0x38, // select reg from the ModRM byte
 531     modrm_reg                   = 0x00  // rax
 532   };
 533 };
 534 
 535 inline bool NativeInstruction::is_illegal()      { return (short)int_at(0) == (short)NativeIllegalInstruction::instruction_code; }
 536 inline bool NativeInstruction::is_call()         { return ubyte_at(0) == NativeCall::instruction_code; }
 537 inline bool NativeInstruction::is_return()       { return ubyte_at(0) == NativeReturn::instruction_code ||
 538                                                           ubyte_at(0) == NativeReturnX::instruction_code; }
 539 inline bool NativeInstruction::is_jump()         { return ubyte_at(0) == NativeJump::instruction_code ||
 540                                                           ubyte_at(0) == 0xEB; /* short jump */ }
 541 inline bool NativeInstruction::is_cond_jump()    { return (int_at(0) & 0xF0FF) == 0x800F /* long jump */ ||
 542                                                           (ubyte_at(0) & 0xF0) == 0x70;  /* short jump */ }
 543 inline bool NativeInstruction::is_safepoint_poll() {
 544 #ifdef AMD64
 545   if (Assembler::is_polling_page_far()) {
 546     // two cases, depending on the choice of the base register in the address.
 547     if (((ubyte_at(0) & NativeTstRegMem::instruction_rex_prefix_mask) == NativeTstRegMem::instruction_rex_prefix &&
 548          ubyte_at(1) == NativeTstRegMem::instruction_code_memXregl &&
 549          (ubyte_at(2) & NativeTstRegMem::modrm_mask) == NativeTstRegMem::modrm_reg) ||
 550         ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl &&
 551         (ubyte_at(1) & NativeTstRegMem::modrm_mask) == NativeTstRegMem::modrm_reg) {
 552       return true;
 553     } else {
 554       return false;
 555     }
 556   } else {
 557     if (ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl &&
 558         ubyte_at(1) == 0x05) { // 00 rax 101
 559       address fault = addr_at(6) + int_at(2);
 560       return os::is_poll_address(fault);
 561     } else {
 562       return false;
 563     }
 564   }
 565 #else
 566   return ( ubyte_at(0) == NativeMovRegMem::instruction_code_mem2reg ||
 567            ubyte_at(0) == NativeTstRegMem::instruction_code_memXregl ) &&
 568            (ubyte_at(1)&0xC7) == 0x05 && /* Mod R/M == disp32 */
 569            (os::is_poll_address((address)int_at(2)));
 570 #endif // AMD64
 571 }
 572 
 573 inline bool NativeInstruction::is_mov_literal64() {
 574 #ifdef AMD64
 575   return ((ubyte_at(0) == Assembler::REX_W || ubyte_at(0) == Assembler::REX_WB) &&
 576           (ubyte_at(1) & (0xff ^ NativeMovConstReg::register_mask)) == 0xB8);
 577 #else
 578   return false;
 579 #endif // AMD64
 580 }
 581 
 582 #endif // CPU_X86_VM_NATIVEINST_X86_HPP