lib/Target/X86/X86JITInfo.cpp

   1 //===-- X86JITInfo.cpp - Implement the JIT interfaces for the X86 target --===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file was developed by the LLVM research group and is distributed under
   6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This file implements the JIT interfaces for the X86 target.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #define DEBUG_TYPE "jit"
  15 #include "X86JITInfo.h"
  16 #include "X86Relocations.h"
  17 #include "X86Subtarget.h"
  18 #include "llvm/CodeGen/MachineCodeEmitter.h"
  19 #include "llvm/Config/alloca.h"
  20 #include <cstdlib>
  21 using namespace llvm;
  22
  23 #ifdef _MSC_VER
  24   extern "C" void *_AddressOfReturnAddress(void);
  25   #pragma intrinsic(_AddressOfReturnAddress)
  26 #endif
  27
  28 void X86JITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
  29   unsigned char *OldByte = (unsigned char *)Old;
  30   *OldByte++ = 0xE9;                // Emit JMP opcode.
  31   unsigned *OldWord = (unsigned *)OldByte;
  32   unsigned NewAddr = (intptr_t)New;
  33   unsigned OldAddr = (intptr_t)OldWord;
  34   *OldWord = NewAddr - OldAddr - 4; // Emit PC-relative addr of New code.
  35 }
  36
  37
  38 /// JITCompilerFunction - This contains the address of the JIT function used to
  39 /// compile a function lazily.
  40 static TargetJITInfo::JITCompilerFn JITCompilerFunction;
  41
  42 // Get the ASMPREFIX for the current host.  This is often '_'.
  43 #ifndef __USER_LABEL_PREFIX__
  44 #define __USER_LABEL_PREFIX__
  45 #endif
  46 #define GETASMPREFIX2(X) #X
  47 #define GETASMPREFIX(X) GETASMPREFIX2(X)
  48 #define ASMPREFIX GETASMPREFIX(__USER_LABEL_PREFIX__)
  49
  50 // Provide a wrapper for X86CompilationCallback2 that saves non-traditional
  51 // callee saved registers, for the fastcc calling convention.
  52 extern "C" {
  53 #if defined(__x86_64__)
  54   // No need to save EAX/EDX for X86-64.
  55   void X86CompilationCallback(void);
  56   asm(
  57     ".text\n"
  58     ".align 8\n"
  59     ".globl " ASMPREFIX "X86CompilationCallback\n"
  60   ASMPREFIX "X86CompilationCallback:\n"
  61     // Save RBP
  62     "pushq   %rbp\n"
  63     // Save RSP
  64     "movq    %rsp, %rbp\n"
  65     // Save all int arg registers
  66     "pushq   %rdi\n"
  67     "pushq   %rsi\n"
  68     "pushq   %rdx\n"
  69     "pushq   %rcx\n"
  70     "pushq   %r8\n"
  71     "pushq   %r9\n"
  72     // Align stack on 16-byte boundary. ESP might not be properly aligned
  73     // (8 byte) if this is called from an indirect stub.
  74     "andq    $-16, %rsp\n"
  75     // Save all XMM arg registers
  76     "subq    $128, %rsp\n"
  77     "movaps  %xmm0, (%rsp)\n"
  78     "movaps  %xmm1, 16(%rsp)\n"
  79     "movaps  %xmm2, 32(%rsp)\n"
  80     "movaps  %xmm3, 48(%rsp)\n"
  81     "movaps  %xmm4, 64(%rsp)\n"
  82     "movaps  %xmm5, 80(%rsp)\n"
  83     "movaps  %xmm6, 96(%rsp)\n"
  84     "movaps  %xmm7, 112(%rsp)\n"
  85     // JIT callee
  86     "movq    %rbp, %rdi\n"    // Pass prev frame and return address
  87     "movq    8(%rbp), %rsi\n"
  88     "call    " ASMPREFIX "X86CompilationCallback2\n"
  89     // Restore all XMM arg registers
  90     "movaps  112(%rsp), %xmm7\n"
  91     "movaps  96(%rsp), %xmm6\n"
  92     "movaps  80(%rsp), %xmm5\n"
  93     "movaps  64(%rsp), %xmm4\n"
  94     "movaps  48(%rsp), %xmm3\n"
  95     "movaps  32(%rsp), %xmm2\n"
  96     "movaps  16(%rsp), %xmm1\n"
  97     "movaps  (%rsp), %xmm0\n"
  98     // Restore RSP
  99     "movq    %rbp, %rsp\n"
 100     // Restore all int arg registers
 101     "subq    $48, %rsp\n"
 102     "popq    %r9\n"
 103     "popq    %r8\n"
 104     "popq    %rcx\n"
 105     "popq    %rdx\n"
 106     "popq    %rsi\n"
 107     "popq    %rdi\n"
 108     // Restore RBP
 109     "popq    %rbp\n"
 110     "ret\n");
 111 #elif defined(__i386__) || defined(i386) || defined(_M_IX86)
 112 #ifndef _MSC_VER
 113   void X86CompilationCallback(void);
 114   asm(
 115     ".text\n"
 116     ".align 8\n"
 117     ".globl " ASMPREFIX  "X86CompilationCallback\n"
 118   ASMPREFIX "X86CompilationCallback:\n"
 119     "pushl   %ebp\n"
 120     "movl    %esp, %ebp\n"    // Standard prologue
 121     "pushl   %eax\n"
 122     "pushl   %edx\n"          // Save EAX/EDX/ECX
 123     "pushl   %ecx\n"
 124 #if defined(__APPLE__)
 125     "andl    $-16, %esp\n"    // Align ESP on 16-byte boundary
 126 #endif
 127     "subl    $16, %esp\n"
 128     "movl    4(%ebp), %eax\n" // Pass prev frame and return address
 129     "movl    %eax, 4(%esp)\n"
 130     "movl    %ebp, (%esp)\n"
 131     "call    " ASMPREFIX "X86CompilationCallback2\n"
 132     "movl    %ebp, %esp\n"    // Restore ESP
 133     "subl    $12, %esp\n"
 134     "popl    %ecx\n"
 135     "popl    %edx\n"
 136     "popl    %eax\n"
 137     "popl    %ebp\n"
 138     "ret\n");
 139
 140   // Same as X86CompilationCallback but also saves XMM argument registers.
 141   void X86CompilationCallback_SSE(void);
 142   asm(
 143     ".text\n"
 144     ".align 8\n"
 145     ".globl " ASMPREFIX  "X86CompilationCallback_SSE\n"
 146   ASMPREFIX "X86CompilationCallback_SSE:\n"
 147     "pushl   %ebp\n"
 148     "movl    %esp, %ebp\n"    // Standard prologue
 149     "pushl   %eax\n"
 150     "pushl   %edx\n"          // Save EAX/EDX/ECX
 151     "pushl   %ecx\n"
 152     "andl    $-16, %esp\n"    // Align ESP on 16-byte boundary
 153     // Save all XMM arg registers
 154     "subl    $64, %esp\n"
 155     "movaps  %xmm0, (%esp)\n"
 156     "movaps  %xmm1, 16(%esp)\n"
 157     "movaps  %xmm2, 32(%esp)\n"
 158     "movaps  %xmm3, 48(%esp)\n"
 159     "subl    $16, %esp\n"
 160     "movl    4(%ebp), %eax\n" // Pass prev frame and return address
 161     "movl    %eax, 4(%esp)\n"
 162     "movl    %ebp, (%esp)\n"
 163     "call    " ASMPREFIX "X86CompilationCallback2\n"
 164     "addl    $16, %esp\n"
 165     "movaps  48(%esp), %xmm3\n"
 166     "movaps  32(%esp), %xmm2\n"
 167     "movaps  16(%esp), %xmm1\n"
 168     "movaps  (%esp), %xmm0\n"
 169     "movl    %ebp, %esp\n"    // Restore ESP
 170     "subl    $12, %esp\n"
 171     "popl    %ecx\n"
 172     "popl    %edx\n"
 173     "popl    %eax\n"
 174     "popl    %ebp\n"
 175     "ret\n");
 176 #else
 177   void X86CompilationCallback2(void);
 178
 179   _declspec(naked) void X86CompilationCallback(void) {
 180     __asm {
 181       push  eax
 182       push  edx
 183       push  ecx
 184       call  X86CompilationCallback2
 185       pop   ecx
 186       pop   edx
 187       pop   eax
 188       ret
 189     }
 190   }
 191 #endif // _MSC_VER
 192
 193 #else // Not an i386 host
 194   void X86CompilationCallback() {
 195     assert(0 && "Cannot call X86CompilationCallback() on a non-x86 arch!\n");
 196     abort();
 197   }
 198 #endif
 199 }
 200
 201 /// X86CompilationCallback - This is the target-specific function invoked by the
 202 /// function stub when we did not know the real target of a call.  This function
 203 /// must locate the start of the stub or call site and pass it into the JIT
 204 /// compiler function.
 205 #ifdef _MSC_VER
 206 extern "C" void X86CompilationCallback2() {
 207   assert(sizeof(size_t) == 4); // FIXME: handle Win64
 208   intptr_t *RetAddrLoc = (intptr_t *)_AddressOfReturnAddress();
 209   RetAddrLoc += 4;  // skip over ret addr, edx, eax, ecx
 210   intptr_t RetAddr = *RetAddrLoc;
 211 #else
 212 extern "C" void X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr) {
 213   intptr_t *RetAddrLoc = &StackPtr[1];
 214 #endif
 215   assert(*RetAddrLoc == RetAddr &&
 216          "Could not find return address on the stack!");
 217
 218   // It's a stub if there is an interrupt marker after the call.
 219   bool isStub = ((unsigned char*)RetAddr)[0] == 0xCD;
 220
 221   // The call instruction should have pushed the return value onto the stack...
 222 #ifdef __x86_64__
 223   RetAddr--;     // Backtrack to the reference itself...
 224 #else
 225   RetAddr -= 4;  // Backtrack to the reference itself...
 226 #endif
 227
 228 #if 0
 229   DOUT << "In callback! Addr=" << (void*)RetAddr
 230        << " ESP=" << (void*)StackPtr
 231        << ": Resolving call to function: "
 232        << TheVM->getFunctionReferencedName((void*)RetAddr) << "\n";
 233 #endif
 234
 235   // Sanity check to make sure this really is a call instruction.
 236 #ifdef __x86_64__
 237   assert(((unsigned char*)RetAddr)[-2] == 0x41 &&"Not a call instr!");
 238   assert(((unsigned char*)RetAddr)[-1] == 0xFF &&"Not a call instr!");
 239 #else
 240   assert(((unsigned char*)RetAddr)[-1] == 0xE8 &&"Not a call instr!");
 241 #endif
 242
 243   intptr_t NewVal = (intptr_t)JITCompilerFunction((void*)RetAddr);
 244
 245   // Rewrite the call target... so that we don't end up here every time we
 246   // execute the call.
 247 #ifdef __x86_64__
 248   *(intptr_t *)(RetAddr - 0xa) = NewVal;
 249 #else
 250   *(intptr_t *)RetAddr = (intptr_t)(NewVal-RetAddr-4);
 251 #endif
 252
 253   if (isStub) {
 254     // If this is a stub, rewrite the call into an unconditional branch
 255     // instruction so that two return addresses are not pushed onto the stack
 256     // when the requested function finally gets called.  This also makes the
 257     // 0xCD byte (interrupt) dead, so the marker doesn't effect anything.
 258 #ifdef __x86_64__
 259     ((unsigned char*)RetAddr)[0] = (2 | (4 << 3) | (3 << 6));
 260 #else
 261     ((unsigned char*)RetAddr)[-1] = 0xE9;
 262 #endif
 263   }
 264
 265   // Change the return address to reexecute the call instruction...
 266 #ifdef __x86_64__
 267   *RetAddrLoc -= 0xd;
 268 #else
 269   *RetAddrLoc -= 5;
 270 #endif
 271 }
 272
 273 TargetJITInfo::LazyResolverFn
 274 X86JITInfo::getLazyResolverFunction(JITCompilerFn F) {
 275   JITCompilerFunction = F;
 276
 277 #if (defined(__i386__) || defined(i386) || defined(_M_IX86)) && \
 278   !defined(_MSC_VER) && !defined(__x86_64__)
 279   unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
 280   union {
 281     unsigned u[3];
 282     char     c[12];
 283   } text;
 284
 285   if (!X86::GetCpuIDAndInfo(0, &EAX, text.u+0, text.u+2, text.u+1)) {
 286     // FIXME: support for AMD family of processors.
 287     if (memcmp(text.c, "GenuineIntel", 12) == 0) {
 288       X86::GetCpuIDAndInfo(0x1, &EAX, &EBX, &ECX, &EDX);
 289       if ((EDX >> 25) & 0x1)
 290         return X86CompilationCallback_SSE;
 291     }
 292   }
 293 #endif
 294
 295   return X86CompilationCallback;
 296 }
 297
 298 void *X86JITInfo::emitFunctionStub(void *Fn, MachineCodeEmitter &MCE) {
 299   // Note, we cast to intptr_t here to silence a -pedantic warning that
 300   // complains about casting a function pointer to a normal pointer.
 301 #if (defined(__i386__) || defined(i386) || defined(_M_IX86)) && \
 302   !defined(_MSC_VER) && !defined(__x86_64__)
 303   bool NotCC = (Fn != (void*)(intptr_t)X86CompilationCallback &&
 304                 Fn != (void*)(intptr_t)X86CompilationCallback_SSE);
 305 #else
 306   bool NotCC = Fn != (void*)(intptr_t)X86CompilationCallback;
 307 #endif
 308   if (NotCC) {
 309 #ifdef __x86_64__
 310     MCE.startFunctionStub(13, 4);
 311     MCE.emitByte(0x49);          // REX prefix
 312     MCE.emitByte(0xB8+2);        // movabsq r10
 313     MCE.emitWordLE(((unsigned *)&Fn)[0]);
 314     MCE.emitWordLE(((unsigned *)&Fn)[1]);
 315     MCE.emitByte(0x41);          // REX prefix
 316     MCE.emitByte(0xFF);          // jmpq *r10
 317     MCE.emitByte(2 | (4 << 3) | (3 << 6));
 318 #else
 319     MCE.startFunctionStub(5, 4);
 320     MCE.emitByte(0xE9);
 321     MCE.emitWordLE((intptr_t)Fn-MCE.getCurrentPCValue()-4);
 322 #endif
 323     return MCE.finishFunctionStub(0);
 324   }
 325
 326 #ifdef __x86_64__
 327   MCE.startFunctionStub(14, 4);
 328   MCE.emitByte(0x49);          // REX prefix
 329   MCE.emitByte(0xB8+2);        // movabsq r10
 330   MCE.emitWordLE(((unsigned *)&Fn)[0]);
 331   MCE.emitWordLE(((unsigned *)&Fn)[1]);
 332   MCE.emitByte(0x41);          // REX prefix
 333   MCE.emitByte(0xFF);          // callq *r10
 334   MCE.emitByte(2 | (2 << 3) | (3 << 6));
 335 #else
 336   MCE.startFunctionStub(6, 4);
 337   MCE.emitByte(0xE8);   // Call with 32 bit pc-rel destination...
 338
 339   MCE.emitWordLE((intptr_t)Fn-MCE.getCurrentPCValue()-4);
 340 #endif
 341
 342   MCE.emitByte(0xCD);   // Interrupt - Just a marker identifying the stub!
 343   return MCE.finishFunctionStub(0);
 344 }
 345
 346 /// relocate - Before the JIT can run a block of code that has been emitted,
 347 /// it must rewrite the code to contain the actual addresses of any
 348 /// referenced global symbols.
 349 void X86JITInfo::relocate(void *Function, MachineRelocation *MR,
 350                           unsigned NumRelocs, unsigned char* GOTBase) {
 351   for (unsigned i = 0; i != NumRelocs; ++i, ++MR) {
 352     void *RelocPos = (char*)Function + MR->getMachineCodeOffset();
 353     intptr_t ResultPtr = (intptr_t)MR->getResultPointer();
 354     switch ((X86::RelocationType)MR->getRelocationType()) {
 355     case X86::reloc_pcrel_word: {
 356       // PC relative relocation, add the relocated value to the value already in
 357       // memory, after we adjust it for where the PC is.
 358       ResultPtr = ResultPtr-(intptr_t)RelocPos-4-MR->getConstantVal();
 359       *((unsigned*)RelocPos) += (unsigned)ResultPtr;
 360       break;
 361     }
 362     case X86::reloc_absolute_word:
 363       // Absolute relocation, just add the relocated value to the value already
 364       // in memory.
 365       *((unsigned*)RelocPos) += (unsigned)ResultPtr;
 366       break;
 367     case X86::reloc_absolute_dword:
 368       *((intptr_t*)RelocPos) += ResultPtr;
 369       break;
 370     }
 371   }
 372 }