summaryrefslogtreecommitdiffstats
path: root/contrib/llvm/lib/Target/X86/Disassembler
diff options
context:
space:
mode:
Diffstat (limited to 'contrib/llvm/lib/Target/X86/Disassembler')
-rw-r--r--contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp77
-rw-r--r--contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.h26
-rw-r--r--contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c165
-rw-r--r--contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h8
4 files changed, 173 insertions, 103 deletions
diff --git a/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp b/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp
index 4a0d2ec..3aacb20 100644
--- a/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp
+++ b/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.cpp
@@ -21,13 +21,16 @@
#include "llvm/MC/MCDisassembler.h"
#include "llvm/MC/MCDisassembler.h"
#include "llvm/MC/MCInst.h"
-#include "llvm/Target/TargetRegistry.h"
+#include "llvm/MC/MCSubtargetInfo.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MemoryObject.h"
+#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/raw_ostream.h"
#define GET_REGINFO_ENUM
#include "X86GenRegisterInfo.inc"
+#define GET_INSTRINFO_ENUM
+#include "X86GenInstrInfo.inc"
#include "X86GenEDInfo.inc"
using namespace llvm;
@@ -64,8 +67,8 @@ extern Target TheX86_32Target, TheX86_64Target;
static bool translateInstruction(MCInst &target,
InternalInstruction &source);
-X86GenericDisassembler::X86GenericDisassembler(DisassemblerMode mode) :
- MCDisassembler(),
+X86GenericDisassembler::X86GenericDisassembler(const MCSubtargetInfo &STI, DisassemblerMode mode) :
+ MCDisassembler(STI),
fMode(mode) {
}
@@ -106,28 +109,34 @@ static void logger(void* arg, const char* log) {
// Public interface for the disassembler
//
-bool X86GenericDisassembler::getInstruction(MCInst &instr,
- uint64_t &size,
- const MemoryObject &region,
- uint64_t address,
- raw_ostream &vStream) const {
+MCDisassembler::DecodeStatus
+X86GenericDisassembler::getInstruction(MCInst &instr,
+ uint64_t &size,
+ const MemoryObject &region,
+ uint64_t address,
+ raw_ostream &vStream,
+ raw_ostream &cStream) const {
InternalInstruction internalInstr;
+
+ dlog_t loggerFn = logger;
+ if (&vStream == &nulls())
+ loggerFn = 0; // Disable logging completely if it's going to nulls().
int ret = decodeInstruction(&internalInstr,
regionReader,
(void*)&region,
- logger,
+ loggerFn,
(void*)&vStream,
address,
fMode);
if (ret) {
size = internalInstr.readerCursor - address;
- return false;
+ return Fail;
}
else {
size = internalInstr.length;
- return !translateInstruction(instr, internalInstr);
+ return (!translateInstruction(instr, internalInstr)) ? Success : Fail;
}
}
@@ -183,8 +192,46 @@ static void translateImmediate(MCInst &mcInst, uint64_t immediate,
break;
}
}
+ // By default sign-extend all X86 immediates based on their encoding.
+ else if (type == TYPE_IMM8 || type == TYPE_IMM16 || type == TYPE_IMM32 ||
+ type == TYPE_IMM64) {
+ uint32_t Opcode = mcInst.getOpcode();
+ switch (operand.encoding) {
+ default:
+ break;
+ case ENCODING_IB:
+ // Special case those X86 instructions that use the imm8 as a set of
+ // bits, bit count, etc. and are not sign-extend.
+ if (Opcode != X86::BLENDPSrri && Opcode != X86::BLENDPDrri &&
+ Opcode != X86::PBLENDWrri && Opcode != X86::MPSADBWrri &&
+ Opcode != X86::DPPSrri && Opcode != X86::DPPDrri &&
+ Opcode != X86::INSERTPSrr && Opcode != X86::VBLENDPSYrri &&
+ Opcode != X86::VBLENDPSYrmi && Opcode != X86::VBLENDPDYrri &&
+ Opcode != X86::VBLENDPDYrmi && Opcode != X86::VPBLENDWrri &&
+ Opcode != X86::VMPSADBWrri && Opcode != X86::VDPPSYrri &&
+ Opcode != X86::VDPPSYrmi && Opcode != X86::VDPPDrri &&
+ Opcode != X86::VINSERTPSrr)
+ type = TYPE_MOFFS8;
+ break;
+ case ENCODING_IW:
+ type = TYPE_MOFFS16;
+ break;
+ case ENCODING_ID:
+ type = TYPE_MOFFS32;
+ break;
+ case ENCODING_IO:
+ type = TYPE_MOFFS64;
+ break;
+ }
+ }
switch (type) {
+ case TYPE_XMM128:
+ mcInst.addOperand(MCOperand::CreateReg(X86::XMM0 + (immediate >> 4)));
+ return;
+ case TYPE_XMM256:
+ mcInst.addOperand(MCOperand::CreateReg(X86::YMM0 + (immediate >> 4)));
+ return;
case TYPE_MOFFS8:
case TYPE_REL8:
if(immediate & 0x80)
@@ -543,12 +590,12 @@ static bool translateInstruction(MCInst &mcInst,
return false;
}
-static MCDisassembler *createX86_32Disassembler(const Target &T) {
- return new X86Disassembler::X86_32Disassembler;
+static MCDisassembler *createX86_32Disassembler(const Target &T, const MCSubtargetInfo &STI) {
+ return new X86Disassembler::X86_32Disassembler(STI);
}
-static MCDisassembler *createX86_64Disassembler(const Target &T) {
- return new X86Disassembler::X86_64Disassembler;
+static MCDisassembler *createX86_64Disassembler(const Target &T, const MCSubtargetInfo &STI) {
+ return new X86Disassembler::X86_64Disassembler(STI);
}
extern "C" void LLVMInitializeX86Disassembler() {
diff --git a/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.h b/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.h
index 550cf9d..6ac9a0f 100644
--- a/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.h
+++ b/contrib/llvm/lib/Target/X86/Disassembler/X86Disassembler.h
@@ -92,6 +92,7 @@ struct InternalInstruction;
namespace llvm {
class MCInst;
+class MCSubtargetInfo;
class MemoryObject;
class raw_ostream;
@@ -107,16 +108,17 @@ protected:
/// Constructor - Initializes the disassembler.
///
/// @param mode - The X86 architecture mode to decode for.
- X86GenericDisassembler(DisassemblerMode mode);
+ X86GenericDisassembler(const MCSubtargetInfo &STI, DisassemblerMode mode);
public:
~X86GenericDisassembler();
/// getInstruction - See MCDisassembler.
- bool getInstruction(MCInst &instr,
- uint64_t &size,
- const MemoryObject &region,
- uint64_t address,
- raw_ostream &vStream) const;
+ DecodeStatus getInstruction(MCInst &instr,
+ uint64_t &size,
+ const MemoryObject &region,
+ uint64_t address,
+ raw_ostream &vStream,
+ raw_ostream &cStream) const;
/// getEDInfo - See MCDisassembler.
EDInstInfo *getEDInfo() const;
@@ -127,24 +129,24 @@ private:
/// X86_16Disassembler - 16-bit X86 disassembler.
class X86_16Disassembler : public X86GenericDisassembler {
public:
- X86_16Disassembler() :
- X86GenericDisassembler(MODE_16BIT) {
+ X86_16Disassembler(const MCSubtargetInfo &STI) :
+ X86GenericDisassembler(STI, MODE_16BIT) {
}
};
/// X86_16Disassembler - 32-bit X86 disassembler.
class X86_32Disassembler : public X86GenericDisassembler {
public:
- X86_32Disassembler() :
- X86GenericDisassembler(MODE_32BIT) {
+ X86_32Disassembler(const MCSubtargetInfo &STI) :
+ X86GenericDisassembler(STI, MODE_32BIT) {
}
};
/// X86_16Disassembler - 64-bit X86 disassembler.
class X86_64Disassembler : public X86GenericDisassembler {
public:
- X86_64Disassembler() :
- X86GenericDisassembler(MODE_64BIT) {
+ X86_64Disassembler(const MCSubtargetInfo &STI) :
+ X86GenericDisassembler(STI, MODE_64BIT) {
}
};
diff --git a/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c b/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
index de1610b..f9b0fe5 100644
--- a/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
+++ b/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoder.c
@@ -58,8 +58,8 @@ static InstructionContext contextForAttrs(uint8_t attrMask) {
* @return - TRUE if the ModR/M byte is required, FALSE otherwise.
*/
static int modRMRequired(OpcodeType type,
- InstructionContext insnContext,
- uint8_t opcode) {
+ InstructionContext insnContext,
+ uint8_t opcode) {
const struct ContextDecision* decision = 0;
switch (type) {
@@ -391,7 +391,7 @@ static int readPrefixes(struct InternalInstruction* insn) {
return -1;
}
- if (insn->mode == MODE_64BIT || byte1 & 0x8) {
+ if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
insn->vexSize = 3;
insn->necessaryPrefixLocation = insn->readerCursor - 1;
}
@@ -406,12 +406,14 @@ static int readPrefixes(struct InternalInstruction* insn) {
consumeByte(insn, &insn->vexPrefix[2]);
/* We simulate the REX prefix for simplicity's sake */
-
- insn->rexPrefix = 0x40
- | (wFromVEX3of3(insn->vexPrefix[2]) << 3)
- | (rFromVEX2of3(insn->vexPrefix[1]) << 2)
- | (xFromVEX2of3(insn->vexPrefix[1]) << 1)
- | (bFromVEX2of3(insn->vexPrefix[1]) << 0);
+
+ if (insn->mode == MODE_64BIT) {
+ insn->rexPrefix = 0x40
+ | (wFromVEX3of3(insn->vexPrefix[2]) << 3)
+ | (rFromVEX2of3(insn->vexPrefix[1]) << 2)
+ | (xFromVEX2of3(insn->vexPrefix[1]) << 1)
+ | (bFromVEX2of3(insn->vexPrefix[1]) << 0);
+ }
switch (ppFromVEX3of3(insn->vexPrefix[2]))
{
@@ -433,7 +435,7 @@ static int readPrefixes(struct InternalInstruction* insn) {
return -1;
}
- if (insn->mode == MODE_64BIT || byte1 & 0x8) {
+ if (insn->mode == MODE_64BIT || (byte1 & 0xc0) == 0xc0) {
insn->vexSize = 2;
}
else {
@@ -444,8 +446,10 @@ static int readPrefixes(struct InternalInstruction* insn) {
insn->vexPrefix[0] = byte;
consumeByte(insn, &insn->vexPrefix[1]);
- insn->rexPrefix = 0x40
- | (rFromVEX2of2(insn->vexPrefix[1]) << 2);
+ if (insn->mode == MODE_64BIT) {
+ insn->rexPrefix = 0x40
+ | (rFromVEX2of2(insn->vexPrefix[1]) << 2);
+ }
switch (ppFromVEX2of2(insn->vexPrefix[1]))
{
@@ -700,34 +704,6 @@ static BOOL is16BitEquvalent(const char* orig, const char* equiv) {
}
/*
- * is64BitEquivalent - Determines whether two instruction names refer to
- * equivalent instructions but one is 64-bit whereas the other is not.
- *
- * @param orig - The instruction that is not 64-bit
- * @param equiv - The instruction that is 64-bit
- */
-static BOOL is64BitEquivalent(const char* orig, const char* equiv) {
- off_t i;
-
- for (i = 0;; i++) {
- if (orig[i] == '\0' && equiv[i] == '\0')
- return TRUE;
- if (orig[i] == '\0' || equiv[i] == '\0')
- return FALSE;
- if (orig[i] != equiv[i]) {
- if ((orig[i] == 'W' || orig[i] == 'L') && equiv[i] == 'Q')
- continue;
- if ((orig[i] == '1' || orig[i] == '3') && equiv[i] == '6')
- continue;
- if ((orig[i] == '6' || orig[i] == '2') && equiv[i] == '4')
- continue;
- return FALSE;
- }
- }
-}
-
-
-/*
* getID - Determines the ID of an instruction, consuming the ModR/M byte as
* appropriate for extended and escape opcodes. Determines the attributes and
* context for the instruction before doing so.
@@ -763,8 +739,6 @@ static int getID(struct InternalInstruction* insn) {
break;
}
- if (wFromVEX3of3(insn->vexPrefix[2]))
- attrMask |= ATTR_REXW;
if (lFromVEX3of3(insn->vexPrefix[2]))
attrMask |= ATTR_VEXL;
}
@@ -789,63 +763,55 @@ static int getID(struct InternalInstruction* insn) {
}
}
else {
- if (insn->rexPrefix & 0x08)
- attrMask |= ATTR_REXW;
-
if (isPrefixAtLocation(insn, 0x66, insn->necessaryPrefixLocation))
attrMask |= ATTR_OPSIZE;
else if (isPrefixAtLocation(insn, 0xf3, insn->necessaryPrefixLocation))
attrMask |= ATTR_XS;
else if (isPrefixAtLocation(insn, 0xf2, insn->necessaryPrefixLocation))
attrMask |= ATTR_XD;
-
}
+ if (insn->rexPrefix & 0x08)
+ attrMask |= ATTR_REXW;
+
if (getIDWithAttrMask(&instructionID, insn, attrMask))
return -1;
/* The following clauses compensate for limitations of the tables. */
- if ((attrMask & ATTR_XD) && (attrMask & ATTR_REXW)) {
+ if ((attrMask & ATTR_VEXL) && (attrMask & ATTR_REXW)) {
/*
- * Although for SSE instructions it is usually necessary to treat REX.W+F2
- * as F2 for decode (in the absence of a 64BIT_REXW_XD category) there is
- * an occasional instruction where F2 is incidental and REX.W is the more
- * significant. If the decoded instruction is 32-bit and adding REX.W
- * instead of F2 changes a 32 to a 64, we adopt the new encoding.
+ * Some VEX instructions ignore the L-bit, but use the W-bit. Normally L-bit
+ * has precedence since there are no L-bit with W-bit entries in the tables.
+ * So if the L-bit isn't significant we should use the W-bit instead.
*/
-
+
const struct InstructionSpecifier *spec;
- uint16_t instructionIDWithREXw;
- const struct InstructionSpecifier *specWithREXw;
-
+ uint16_t instructionIDWithWBit;
+ const struct InstructionSpecifier *specWithWBit;
+
spec = specifierForUID(instructionID);
-
- if (getIDWithAttrMask(&instructionIDWithREXw,
+
+ if (getIDWithAttrMask(&instructionIDWithWBit,
insn,
- attrMask & (~ATTR_XD))) {
- /*
- * Decoding with REX.w would yield nothing; give up and return original
- * decode.
- */
-
+ (attrMask & (~ATTR_VEXL)) | ATTR_REXW)) {
insn->instructionID = instructionID;
insn->spec = spec;
return 0;
}
-
- specWithREXw = specifierForUID(instructionIDWithREXw);
-
- if (is64BitEquivalent(spec->name, specWithREXw->name)) {
- insn->instructionID = instructionIDWithREXw;
- insn->spec = specWithREXw;
+
+ specWithWBit = specifierForUID(instructionIDWithWBit);
+
+ if (instructionID != instructionIDWithWBit) {
+ insn->instructionID = instructionIDWithWBit;
+ insn->spec = specWithWBit;
} else {
insn->instructionID = instructionID;
insn->spec = spec;
}
return 0;
}
-
+
if (insn->prefixPresent[0x66] && !(attrMask & ATTR_OPSIZE)) {
/*
* The instruction tables make no distinction between instructions that
@@ -885,6 +851,43 @@ static int getID(struct InternalInstruction* insn) {
}
return 0;
}
+
+ if (insn->opcodeType == ONEBYTE && insn->opcode == 0x90 &&
+ insn->rexPrefix & 0x01) {
+ /*
+ * NOOP shouldn't decode as NOOP if REX.b is set. Instead
+ * it should decode as XCHG %r8, %eax.
+ */
+
+ const struct InstructionSpecifier *spec;
+ uint16_t instructionIDWithNewOpcode;
+ const struct InstructionSpecifier *specWithNewOpcode;
+
+ spec = specifierForUID(instructionID);
+
+ /* Borrow opcode from one of the other XCHGar opcodes */
+ insn->opcode = 0x91;
+
+ if (getIDWithAttrMask(&instructionIDWithNewOpcode,
+ insn,
+ attrMask)) {
+ insn->opcode = 0x90;
+
+ insn->instructionID = instructionID;
+ insn->spec = spec;
+ return 0;
+ }
+
+ specWithNewOpcode = specifierForUID(instructionIDWithNewOpcode);
+
+ /* Change back */
+ insn->opcode = 0x90;
+
+ insn->instructionID = instructionIDWithNewOpcode;
+ insn->spec = specWithNewOpcode;
+
+ return 0;
+ }
insn->instructionID = instructionID;
insn->spec = specifierForUID(insn->instructionID);
@@ -1434,11 +1437,10 @@ static int readImmediate(struct InternalInstruction* insn, uint8_t size) {
}
/*
- * readVVVV - Consumes an immediate operand from an instruction, given the
- * desired operand size.
+ * readVVVV - Consumes vvvv from an instruction if it has a VEX prefix.
*
* @param insn - The instruction whose operand is to be read.
- * @return - 0 if the immediate was successfully consumed; nonzero
+ * @return - 0 if the vvvv was successfully consumed; nonzero
* otherwise.
*/
static int readVVVV(struct InternalInstruction* insn) {
@@ -1451,6 +1453,9 @@ static int readVVVV(struct InternalInstruction* insn) {
else
return -1;
+ if (insn->mode != MODE_64BIT)
+ insn->vvvv &= 0x7;
+
return 0;
}
@@ -1463,8 +1468,14 @@ static int readVVVV(struct InternalInstruction* insn) {
*/
static int readOperands(struct InternalInstruction* insn) {
int index;
+ int hasVVVV, needVVVV;
dbgprintf(insn, "readOperands()");
+
+ /* If non-zero vvvv specified, need to make sure one of the operands
+ uses it. */
+ hasVVVV = !readVVVV(insn);
+ needVVVV = hasVVVV && (insn->vvvv != 0);
for (index = 0; index < X86_MAX_OPERANDS; ++index) {
switch (insn->spec->operands[index].encoding) {
@@ -1537,7 +1548,8 @@ static int readOperands(struct InternalInstruction* insn) {
return -1;
break;
case ENCODING_VVVV:
- if (readVVVV(insn))
+ needVVVV = 0; /* Mark that we have found a VVVV operand. */
+ if (!hasVVVV)
return -1;
if (fixupReg(insn, &insn->spec->operands[index]))
return -1;
@@ -1549,6 +1561,9 @@ static int readOperands(struct InternalInstruction* insn) {
return -1;
}
}
+
+ /* If we didn't find ENCODING_VVVV operand, but non-zero vvvv present, fail */
+ if (needVVVV) return -1;
return 0;
}
diff --git a/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h b/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
index 70315ed..8b79335 100644
--- a/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
+++ b/contrib/llvm/lib/Target/X86/Disassembler/X86DisassemblerDecoderCommon.h
@@ -81,12 +81,18 @@ enum attributeBits {
"but not the operands") \
ENUM_ENTRY(IC_XS, 2, "may say something about the opcode " \
"but not the operands") \
+ ENUM_ENTRY(IC_XD_OPSIZE, 3, "requires an OPSIZE prefix, so " \
+ "operands change width") \
+ ENUM_ENTRY(IC_XS_OPSIZE, 3, "requires an OPSIZE prefix, so " \
+ "operands change width") \
ENUM_ENTRY(IC_64BIT_REXW, 4, "requires a REX.W prefix, so operands "\
"change width; overrides IC_OPSIZE") \
ENUM_ENTRY(IC_64BIT_OPSIZE, 3, "Just as meaningful as IC_OPSIZE") \
ENUM_ENTRY(IC_64BIT_XD, 5, "XD instructions are SSE; REX.W is " \
"secondary") \
ENUM_ENTRY(IC_64BIT_XS, 5, "Just as meaningful as IC_64BIT_XD") \
+ ENUM_ENTRY(IC_64BIT_XD_OPSIZE, 3, "Just as meaningful as IC_XD_OPSIZE") \
+ ENUM_ENTRY(IC_64BIT_XS_OPSIZE, 3, "Just as meaningful as IC_XS_OPSIZE") \
ENUM_ENTRY(IC_64BIT_REXW_XS, 6, "OPSIZE could mean a different " \
"opcode") \
ENUM_ENTRY(IC_64BIT_REXW_XD, 6, "Just as meaningful as " \
@@ -104,7 +110,7 @@ enum attributeBits {
ENUM_ENTRY(IC_VEX_W_OPSIZE, 4, "requires VEX, W, and OpSize") \
ENUM_ENTRY(IC_VEX_L, 3, "requires VEX and the L prefix") \
ENUM_ENTRY(IC_VEX_L_XS, 4, "requires VEX and the L and XS prefix")\
- ENUM_ENTRY(IC_VEX_L_XD, 4, "requires VEX and the L and XS prefix")\
+ ENUM_ENTRY(IC_VEX_L_XD, 4, "requires VEX and the L and XD prefix")\
ENUM_ENTRY(IC_VEX_L_OPSIZE, 4, "requires VEX, L, and OpSize")
OpenPOWER on IntegriCloud