#include "X86JITInfo.h"
#include "X86Relocations.h"
#include "X86Subtarget.h"
-#include "llvm/CodeGen/MachineCodeEmitter.h"
+#include "llvm/Function.h"
#include "llvm/Config/alloca.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/Support/ErrorHandling.h"
#include <cstdlib>
#include <cstring>
using namespace llvm;
#define GETASMPREFIX(X) GETASMPREFIX2(X)
#define ASMPREFIX GETASMPREFIX(__USER_LABEL_PREFIX__)
+// Check if building with -fPIC
+#if defined(__PIC__) && __PIC__ && defined(__linux__)
+#define ASMCALLSUFFIX "@PLT"
+#else
+#define ASMCALLSUFFIX
+#endif
+
+// For ELF targets, use a .size and .type directive, to let tools
+// know the extent of functions defined in assembler.
+#if defined(__ELF__)
+# define SIZE(sym) ".size " #sym ", . - " #sym "\n"
+# define TYPE_FUNCTION(sym) ".type " #sym ", @function\n"
+#else
+# define SIZE(sym)
+# define TYPE_FUNCTION(sym)
+#endif
+
// Provide a convenient way for disabling usage of CFI directives.
// This is needed for old/broken assemblers (for example, gas on
// Darwin is pretty old and doesn't support these directives)
".text\n"
".align 8\n"
".globl " ASMPREFIX "X86CompilationCallback\n"
+ TYPE_FUNCTION(X86CompilationCallback)
ASMPREFIX "X86CompilationCallback:\n"
CFI(".cfi_startproc\n")
// Save RBP
// JIT callee
"movq %rbp, %rdi\n" // Pass prev frame and return address
"movq 8(%rbp), %rsi\n"
- "call " ASMPREFIX "X86CompilationCallback2\n"
+ "call " ASMPREFIX "X86CompilationCallback2" ASMCALLSUFFIX "\n"
// Restore all XMM arg registers
"movaps 112(%rsp), %xmm7\n"
"movaps 96(%rsp), %xmm6\n"
CFI(".cfi_restore %rbp\n")
"ret\n"
CFI(".cfi_endproc\n")
+ SIZE(X86CompilationCallback)
);
# else
- // No inline assembler support on this platform
- void X86CompilationCallback() {
- assert(0 && "Cannot call X86CompilationCallback() on a non-x86 arch!\n");
- abort();
- }
+ // No inline assembler support on this platform. The routine is in external
+ // file.
+ void X86CompilationCallback();
+
# endif
#elif defined (X86_32_JIT)
# ifndef _MSC_VER
asm(
".text\n"
".align 8\n"
- ".globl " ASMPREFIX "X86CompilationCallback\n"
+ ".globl " ASMPREFIX "X86CompilationCallback\n"
+ TYPE_FUNCTION(X86CompilationCallback)
ASMPREFIX "X86CompilationCallback:\n"
CFI(".cfi_startproc\n")
"pushl %ebp\n"
"movl 4(%ebp), %eax\n" // Pass prev frame and return address
"movl %eax, 4(%esp)\n"
"movl %ebp, (%esp)\n"
- "call " ASMPREFIX "X86CompilationCallback2\n"
+ "call " ASMPREFIX "X86CompilationCallback2" ASMCALLSUFFIX "\n"
"movl %ebp, %esp\n" // Restore ESP
CFI(".cfi_def_cfa_register %esp\n")
"subl $12, %esp\n"
CFI(".cfi_restore %ebp\n")
"ret\n"
CFI(".cfi_endproc\n")
+ SIZE(X86CompilationCallback)
);
// Same as X86CompilationCallback but also saves XMM argument registers.
asm(
".text\n"
".align 8\n"
- ".globl " ASMPREFIX "X86CompilationCallback_SSE\n"
+ ".globl " ASMPREFIX "X86CompilationCallback_SSE\n"
+ TYPE_FUNCTION(X86CompilationCallback_SSE)
ASMPREFIX "X86CompilationCallback_SSE:\n"
CFI(".cfi_startproc\n")
"pushl %ebp\n"
"movl 4(%ebp), %eax\n" // Pass prev frame and return address
"movl %eax, 4(%esp)\n"
"movl %ebp, (%esp)\n"
- "call " ASMPREFIX "X86CompilationCallback2\n"
+ "call " ASMPREFIX "X86CompilationCallback2" ASMCALLSUFFIX "\n"
"addl $16, %esp\n"
"movaps 48(%esp), %xmm3\n"
CFI(".cfi_restore %xmm3\n")
CFI(".cfi_restore %ebp\n")
"ret\n"
CFI(".cfi_endproc\n")
+ SIZE(X86CompilationCallback_SSE)
);
# else
void X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr);
#else // Not an i386 host
void X86CompilationCallback() {
- assert(0 && "Cannot call X86CompilationCallback() on a non-x86 arch!\n");
- abort();
+ LLVM_UNREACHABLE("Cannot call X86CompilationCallback() on a non-x86 arch!");
}
#endif
}
-/// X86CompilationCallback - This is the target-specific function invoked by the
+/// X86CompilationCallback2 - This is the target-specific function invoked by the
/// function stub when we did not know the real target of a call. This function
/// must locate the start of the stub or call site and pass it into the JIT
/// compiler function.
-extern "C" void X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr) {
+extern "C" void ATTRIBUTE_USED
+X86CompilationCallback2(intptr_t *StackPtr, intptr_t RetAddr) {
intptr_t *RetAddrLoc = &StackPtr[1];
assert(*RetAddrLoc == RetAddr &&
"Could not find return address on the stack!");
// Rewrite the call target... so that we don't end up here every time we
// execute the call.
#if defined (X86_64_JIT)
- *(intptr_t *)(RetAddr - 0xa) = NewVal;
+ if (!isStub)
+ *(intptr_t *)(RetAddr - 0xa) = NewVal;
#else
*(intptr_t *)RetAddr = (intptr_t)(NewVal-RetAddr-4);
#endif
// when the requested function finally gets called. This also makes the
// 0xCD byte (interrupt) dead, so the marker doesn't effect anything.
#if defined (X86_64_JIT)
- ((unsigned char*)RetAddr)[0] = (2 | (4 << 3) | (3 << 6));
+ // If the target address is within 32-bit range of the stub, use a
+ // PC-relative branch instead of loading the actual address. (This is
+ // considerably shorter than the 64-bit immediate load already there.)
+ // We assume here intptr_t is 64 bits.
+ intptr_t diff = NewVal-RetAddr+7;
+ if (diff >= -2147483648LL && diff <= 2147483647LL) {
+ *(unsigned char*)(RetAddr-0xc) = 0xE9;
+ *(intptr_t *)(RetAddr-0xb) = diff & 0xffffffff;
+ } else {
+ *(intptr_t *)(RetAddr - 0xa) = NewVal;
+ ((unsigned char*)RetAddr)[0] = (2 | (4 << 3) | (3 << 6));
+ }
#else
((unsigned char*)RetAddr)[-1] = 0xE9;
#endif
return X86CompilationCallback;
}
-void *X86JITInfo::emitGlobalValueLazyPtr(void *GV, MachineCodeEmitter &MCE) {
+void *X86JITInfo::emitGlobalValueIndirectSym(const GlobalValue* GV, void *ptr,
+ JITCodeEmitter &JCE) {
#if defined (X86_64_JIT)
- MCE.startFunctionStub(8, 8);
- MCE.emitWordLE(((unsigned *)&GV)[0]);
- MCE.emitWordLE(((unsigned *)&GV)[1]);
+ JCE.startGVStub(GV, 8, 8);
+ JCE.emitWordLE((unsigned)(intptr_t)ptr);
+ JCE.emitWordLE((unsigned)(((intptr_t)ptr) >> 32));
#else
- MCE.startFunctionStub(4, 4);
- MCE.emitWordLE((intptr_t)GV);
+ JCE.startGVStub(GV, 4, 4);
+ JCE.emitWordLE((intptr_t)ptr);
#endif
- return MCE.finishFunctionStub(0);
+ return JCE.finishGVStub(GV);
}
-void *X86JITInfo::emitFunctionStub(void *Fn, MachineCodeEmitter &MCE) {
+void *X86JITInfo::emitFunctionStub(const Function* F, void *Fn,
+ JITCodeEmitter &JCE) {
// Note, we cast to intptr_t here to silence a -pedantic warning that
// complains about casting a function pointer to a normal pointer.
#if defined (X86_32_JIT) && !defined (_MSC_VER)
#endif
if (NotCC) {
#if defined (X86_64_JIT)
- MCE.startFunctionStub(13, 4);
- MCE.emitByte(0x49); // REX prefix
- MCE.emitByte(0xB8+2); // movabsq r10
- MCE.emitWordLE(((unsigned *)&Fn)[0]);
- MCE.emitWordLE(((unsigned *)&Fn)[1]);
- MCE.emitByte(0x41); // REX prefix
- MCE.emitByte(0xFF); // jmpq *r10
- MCE.emitByte(2 | (4 << 3) | (3 << 6));
+ JCE.startGVStub(F, 13, 4);
+ JCE.emitByte(0x49); // REX prefix
+ JCE.emitByte(0xB8+2); // movabsq r10
+ JCE.emitWordLE((unsigned)(intptr_t)Fn);
+ JCE.emitWordLE((unsigned)(((intptr_t)Fn) >> 32));
+ JCE.emitByte(0x41); // REX prefix
+ JCE.emitByte(0xFF); // jmpq *r10
+ JCE.emitByte(2 | (4 << 3) | (3 << 6));
#else
- MCE.startFunctionStub(5, 4);
- MCE.emitByte(0xE9);
- MCE.emitWordLE((intptr_t)Fn-MCE.getCurrentPCValue()-4);
+ JCE.startGVStub(F, 5, 4);
+ JCE.emitByte(0xE9);
+ JCE.emitWordLE((intptr_t)Fn-JCE.getCurrentPCValue()-4);
#endif
- return MCE.finishFunctionStub(0);
+ return JCE.finishGVStub(F);
}
#if defined (X86_64_JIT)
- MCE.startFunctionStub(14, 4);
- MCE.emitByte(0x49); // REX prefix
- MCE.emitByte(0xB8+2); // movabsq r10
- MCE.emitWordLE(((unsigned *)&Fn)[0]);
- MCE.emitWordLE(((unsigned *)&Fn)[1]);
- MCE.emitByte(0x41); // REX prefix
- MCE.emitByte(0xFF); // callq *r10
- MCE.emitByte(2 | (2 << 3) | (3 << 6));
+ JCE.startGVStub(F, 14, 4);
+ JCE.emitByte(0x49); // REX prefix
+ JCE.emitByte(0xB8+2); // movabsq r10
+ JCE.emitWordLE((unsigned)(intptr_t)Fn);
+ JCE.emitWordLE((unsigned)(((intptr_t)Fn) >> 32));
+ JCE.emitByte(0x41); // REX prefix
+ JCE.emitByte(0xFF); // callq *r10
+ JCE.emitByte(2 | (2 << 3) | (3 << 6));
#else
- MCE.startFunctionStub(6, 4);
- MCE.emitByte(0xE8); // Call with 32 bit pc-rel destination...
+ JCE.startGVStub(F, 6, 4);
+ JCE.emitByte(0xE8); // Call with 32 bit pc-rel destination...
- MCE.emitWordLE((intptr_t)Fn-MCE.getCurrentPCValue()-4);
+ JCE.emitWordLE((intptr_t)Fn-JCE.getCurrentPCValue()-4);
#endif
- MCE.emitByte(0xCD); // Interrupt - Just a marker identifying the stub!
- return MCE.finishFunctionStub(0);
+ JCE.emitByte(0xCD); // Interrupt - Just a marker identifying the stub!
+ return JCE.finishGVStub(F);
+}
+
+void X86JITInfo::emitFunctionStubAtAddr(const Function* F, void *Fn, void *Stub,
+ JITCodeEmitter &JCE) {
+ // Note, we cast to intptr_t here to silence a -pedantic warning that
+ // complains about casting a function pointer to a normal pointer.
+ JCE.startGVStub(F, Stub, 5);
+ JCE.emitByte(0xE9);
+#if defined (X86_64_JIT)
+ assert(((((intptr_t)Fn-JCE.getCurrentPCValue()-5) << 32) >> 32) ==
+ ((intptr_t)Fn-JCE.getCurrentPCValue()-5)
+ && "PIC displacement does not fit in displacement field!");
+#endif
+ JCE.emitWordLE((intptr_t)Fn-JCE.getCurrentPCValue()-4);
+ JCE.finishGVStub(F);
}
/// getPICJumpTableEntry - Returns the value of the jumptable entry for the
/// specific basic block.
-intptr_t X86JITInfo::getPICJumpTableEntry(intptr_t BB, intptr_t Entry) {
+uintptr_t X86JITInfo::getPICJumpTableEntry(uintptr_t BB, uintptr_t Entry) {
+#if defined(X86_64_JIT)
+ return BB - Entry;
+#else
return BB - PICBase;
+#endif
}
/// relocate - Before the JIT can run a block of code that has been emitted,
}
}
}
+
+char* X86JITInfo::allocateThreadLocalMemory(size_t size) {
+#if defined(X86_32_JIT) && !defined(__APPLE__) && !defined(_MSC_VER)
+ TLSOffset -= size;
+ return TLSOffset;
+#else
+ LLVM_UNREACHABLE("Cannot allocate thread local storage on this arch!");
+ return 0;
+#endif
+}