// Copyright (c) 2007-2008 The Hewlett-Packard Development Company // All rights reserved. // // Redistribution and use of this software in source and binary forms, // with or without modification, are permitted provided that the // following conditions are met: // // The software must be used only for Non-Commercial Use which means any // use which is NOT directed to receiving any direct monetary // compensation for, or commercial advantage from such use. Illustrative // examples of non-commercial use are academic research, personal study, // teaching, education and corporate research & development. // Illustrative examples of commercial use are distributing products for // commercial advantage and providing services using the software for // commercial advantage. // // If you wish to use this software or functionality therein that may be // covered by patents for commercial use, please contact: // Director of Intellectual Property Licensing // Office of Strategy and Technology // Hewlett-Packard Company // 1501 Page Mill Road // Palo Alto, California 94304 // // Redistributions of source code must retain the above copyright notice, // this list of conditions and the following disclaimer. Redistributions // in binary form must reproduce the above copyright notice, this list of // conditions and the following disclaimer in the documentation and/or // other materials provided with the distribution. Neither the name of // the COPYRIGHT HOLDER(s), HEWLETT-PACKARD COMPANY, nor the names of its // contributors may be used to endorse or promote products derived from // this software without specific prior written permission. No right of // sublicense is granted herewith. Derivatives of the software and // output created using the software may be prepared, but only for // Non-Commercial Uses. Derivatives of the software may be shared with // others provided: (i) the others agree to abide by the list of // conditions herein which includes the Non-Commercial Use restrictions; // and (ii) such Derivatives of the software include the above copyright // notice to acknowledge the contribution from this software where // applicable, this list of conditions and the disclaimer below. // // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. // // Authors: Gabe Black ////////////////////////////////////////////////////////////////////////// // // RegOp Microop templates // ////////////////////////////////////////////////////////////////////////// def template MicroRegOpExecute {{ Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, Trace::InstRecord *traceData) const { Fault fault = NoFault; DPRINTF(X86, "The data size is %d\n", dataSize); %(op_decl)s; %(op_rd)s; if(%(cond_check)s) { %(code)s; %(flag_code)s; } else { %(else_code)s; } //Write the resulting state to the execution context if(fault == NoFault) { %(op_wb)s; } return fault; } }}; def template MicroRegOpImmExecute {{ Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, Trace::InstRecord *traceData) const { Fault fault = NoFault; %(op_decl)s; %(op_rd)s; if(%(cond_check)s) { %(code)s; %(flag_code)s; } else { %(else_code)s; } //Write the resulting state to the execution context if(fault == NoFault) { %(op_wb)s; } return fault; } }}; def template MicroRegOpDeclare {{ class %(class_name)s : public %(base_class)s { protected: void buildMe(); public: %(class_name)s(ExtMachInst _machInst, const char * instMnem, bool isMicro, bool isDelayed, bool isFirst, bool isLast, InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext); %(class_name)s(ExtMachInst _machInst, const char * instMnem, InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext); %(BasicExecDeclare)s }; }}; def template MicroRegOpImmDeclare {{ class %(class_name)s : public %(base_class)s { protected: void buildMe(); public: %(class_name)s(ExtMachInst _machInst, const char * instMnem, bool isMicro, bool isDelayed, bool isFirst, bool isLast, InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext); %(class_name)s(ExtMachInst _machInst, const char * instMnem, InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext); %(BasicExecDeclare)s }; }}; def template MicroRegOpConstructor {{ inline void %(class_name)s::buildMe() { %(constructor)s; } inline %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext) : %(base_class)s(machInst, "%(mnemonic)s", instMnem, false, false, false, false, _src1, _src2, _dest, _dataSize, _ext, %(op_class)s) { buildMe(); } inline %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, bool isMicro, bool isDelayed, bool isFirst, bool isLast, InstRegIndex _src1, InstRegIndex _src2, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext) : %(base_class)s(machInst, "%(mnemonic)s", instMnem, isMicro, isDelayed, isFirst, isLast, _src1, _src2, _dest, _dataSize, _ext, %(op_class)s) { buildMe(); } }}; def template MicroRegOpImmConstructor {{ inline void %(class_name)s::buildMe() { %(constructor)s; } inline %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext) : %(base_class)s(machInst, "%(mnemonic)s", instMnem, false, false, false, false, _src1, _imm8, _dest, _dataSize, _ext, %(op_class)s) { buildMe(); } inline %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, bool isMicro, bool isDelayed, bool isFirst, bool isLast, InstRegIndex _src1, uint16_t _imm8, InstRegIndex _dest, uint8_t _dataSize, uint16_t _ext) : %(base_class)s(machInst, "%(mnemonic)s", instMnem, isMicro, isDelayed, isFirst, isLast, _src1, _imm8, _dest, _dataSize, _ext, %(op_class)s) { buildMe(); } }}; output header {{ void divide(uint64_t dividend, uint64_t divisor, uint64_t "ient, uint64_t &remainder); enum SegmentSelectorCheck { SegNoCheck, SegCSCheck, SegCallGateCheck, SegIntGateCheck, SegSoftIntGateCheck, SegSSCheck, SegIretCheck, SegIntCSCheck, SegTRCheck, SegTSSCheck, SegInGDTCheck, SegLDTCheck }; enum LongModeDescriptorType { LDT64 = 2, AvailableTSS64 = 9, BusyTSS64 = 0xb, CallGate64 = 0xc, IntGate64 = 0xe, TrapGate64 = 0xf }; }}; output decoder {{ void divide(uint64_t dividend, uint64_t divisor, uint64_t "ient, uint64_t &remainder) { //Check for divide by zero. if (divisor == 0) panic("Divide by zero!\\n"); //If the divisor is bigger than the dividend, don't do anything. if (divisor <= dividend) { //Shift the divisor so it's msb lines up with the dividend. int dividendMsb = findMsbSet(dividend); int divisorMsb = findMsbSet(divisor); int shift = dividendMsb - divisorMsb; divisor <<= shift; //Compute what we'll add to the quotient if the divisor isn't //now larger than the dividend. uint64_t quotientBit = 1; quotientBit <<= shift; //If we need to step back a bit (no pun intended) because the //divisor got too to large, do that here. This is the "or two" //part of one or two bit division. if (divisor > dividend) { quotientBit >>= 1; divisor >>= 1; } //Decrement the remainder and increment the quotient. quotient += quotientBit; remainder -= divisor; } } }}; let {{ # Make these empty strings so that concatenating onto # them will always work. header_output = "" decoder_output = "" exec_output = "" immTemplates = ( MicroRegOpImmDeclare, MicroRegOpImmConstructor, MicroRegOpImmExecute) regTemplates = ( MicroRegOpDeclare, MicroRegOpConstructor, MicroRegOpExecute) class RegOpMeta(type): def buildCppClasses(self, name, Name, suffix, \ code, flag_code, cond_check, else_code): # Globals to stick the output in global header_output global decoder_output global exec_output # Stick all the code together so it can be searched at once allCode = "|".join((code, flag_code, cond_check, else_code)) # If op2 is used anywhere, make register and immediate versions # of this code. matcher = re.compile("(?s?)op2(?P\\.\\w+)?") match = matcher.search(allCode) if match: typeQual = "" if match.group("typeQual"): typeQual = match.group("typeQual") src2_name = "%spsrc2%s" % (match.group("prefix"), typeQual) self.buildCppClasses(name, Name, suffix, matcher.sub(src2_name, code), matcher.sub(src2_name, flag_code), matcher.sub(src2_name, cond_check), matcher.sub(src2_name, else_code)) self.buildCppClasses(name + "i", Name, suffix + "Imm", matcher.sub("imm8", code), matcher.sub("imm8", flag_code), matcher.sub("imm8", cond_check), matcher.sub("imm8", else_code)) return # If there's something optional to do with flags, generate # a version without it and fix up this version to use it. if flag_code != "" or cond_check != "true": self.buildCppClasses(name, Name, suffix, code, "", "true", else_code) suffix = "Flags" + suffix # If psrc1 or psrc2 is used, we need to actually insert code to # compute it. matcher = re.compile("(?\\.\\w+)?") if matcher.search(allCode): microopClasses[name + 'i'] = cls return cls class RegOp(X86Microop): __metaclass__ = RegOpMeta # This class itself doesn't act as a microop abstract = True # Default template parameter values flag_code = "" cond_check = "true" else_code = ";" def __init__(self, dest, src1, op2, flags = None, dataSize = "env.dataSize"): self.dest = dest self.src1 = src1 self.op2 = op2 self.flags = flags self.dataSize = dataSize if flags is None: self.ext = 0 else: if not isinstance(flags, (list, tuple)): raise Exception, "flags must be a list or tuple of flags" self.ext = " | ".join(flags) self.className += "Flags" def getAllocator(self, *microFlags): className = self.className if self.mnemonic == self.base_mnemonic + 'i': className += "Imm" allocator = '''new %(class_name)s(machInst, macrocodeBlock %(flags)s, %(src1)s, %(op2)s, %(dest)s, %(dataSize)s, %(ext)s)''' % { "class_name" : className, "flags" : self.microFlagsText(microFlags), "src1" : self.src1, "op2" : self.op2, "dest" : self.dest, "dataSize" : self.dataSize, "ext" : self.ext} return allocator class LogicRegOp(RegOp): abstract = True flag_code = ''' //Don't have genFlags handle the OF or CF bits uint64_t mask = CFBit | ECFBit | OFBit; ccFlagBits = genFlags(ccFlagBits, ext & ~mask, DestReg, psrc1, op2); //If a logic microop wants to set these, it wants to set them to 0. ccFlagBits &= ~(CFBit & ext); ccFlagBits &= ~(ECFBit & ext); ccFlagBits &= ~(OFBit & ext); ''' class FlagRegOp(RegOp): abstract = True flag_code = \ "ccFlagBits = genFlags(ccFlagBits, ext, DestReg, psrc1, op2);" class SubRegOp(RegOp): abstract = True flag_code = \ "ccFlagBits = genFlags(ccFlagBits, ext, DestReg, psrc1, ~op2, true);" class CondRegOp(RegOp): abstract = True cond_check = "checkCondition(ccFlagBits, ext)" class RdRegOp(RegOp): abstract = True def __init__(self, dest, src1=None, dataSize="env.dataSize"): if not src1: src1 = dest super(RdRegOp, self).__init__(dest, src1, \ "InstRegIndex(NUM_INTREGS)", None, dataSize) class WrRegOp(RegOp): abstract = True def __init__(self, src1, src2, flags=None, dataSize="env.dataSize"): super(WrRegOp, self).__init__("InstRegIndex(NUM_INTREGS)", \ src1, src2, flags, dataSize) class Add(FlagRegOp): code = 'DestReg = merge(DestReg, psrc1 + op2, dataSize);' class Or(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 | op2, dataSize);' class Adc(FlagRegOp): code = ''' CCFlagBits flags = ccFlagBits; DestReg = merge(DestReg, psrc1 + op2 + flags.cf, dataSize); ''' class Sbb(SubRegOp): code = ''' CCFlagBits flags = ccFlagBits; DestReg = merge(DestReg, psrc1 - op2 - flags.cf, dataSize); ''' class And(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 & op2, dataSize)' class Sub(SubRegOp): code = 'DestReg = merge(DestReg, psrc1 - op2, dataSize)' class Xor(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 ^ op2, dataSize)' class Mul1s(WrRegOp): code = ''' ProdLow = psrc1 * op2; int halfSize = (dataSize * 8) / 2; uint64_t shifter = (1ULL << halfSize); uint64_t hiResult; uint64_t psrc1_h = psrc1 / shifter; uint64_t psrc1_l = psrc1 & mask(halfSize); uint64_t psrc2_h = op2 / shifter; uint64_t psrc2_l = op2 & mask(halfSize); hiResult = ((psrc1_l * psrc2_h + psrc1_h * psrc2_l + ((psrc1_l * psrc2_l) / shifter)) /shifter) + psrc1_h * psrc2_h; if (spsrc1 < 0) hiResult -= op2; int64_t bigSop2 = sop2; if (bigSop2 < 0) hiResult -= psrc1; ProdHi = hiResult; ''' class Mul1u(WrRegOp): code = ''' ProdLow = psrc1 * op2; int halfSize = (dataSize * 8) / 2; uint64_t shifter = (1ULL << halfSize); uint64_t psrc1_h = psrc1 / shifter; uint64_t psrc1_l = psrc1 & mask(halfSize); uint64_t psrc2_h = op2 / shifter; uint64_t psrc2_l = op2 & mask(halfSize); ProdHi = ((psrc1_l * psrc2_h + psrc1_h * psrc2_l + ((psrc1_l * psrc2_l) / shifter)) / shifter) + psrc1_h * psrc2_h; ''' class Mulel(RdRegOp): code = 'DestReg = merge(SrcReg1, ProdLow, dataSize);' class Muleh(RdRegOp): def __init__(self, dest, src1=None, flags=None, dataSize="env.dataSize"): if not src1: src1 = dest super(RdRegOp, self).__init__(dest, src1, \ "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = 'DestReg = merge(SrcReg1, ProdHi, dataSize);' flag_code = ''' if (ProdHi) ccFlagBits = ccFlagBits | (ext & (CFBit | OFBit | ECFBit)); else ccFlagBits = ccFlagBits & ~(ext & (CFBit | OFBit | ECFBit)); ''' # One or two bit divide class Div1(WrRegOp): code = ''' //These are temporaries so that modifying them later won't make //the ISA parser think they're also sources. uint64_t quotient = 0; uint64_t remainder = psrc1; //Similarly, this is a temporary so changing it doesn't make it //a source. uint64_t divisor = op2; //This is a temporary just for consistency and clarity. uint64_t dividend = remainder; //Do the division. divide(dividend, divisor, quotient, remainder); //Record the final results. Remainder = remainder; Quotient = quotient; Divisor = divisor; ''' # Step divide class Div2(RegOp): code = ''' uint64_t dividend = Remainder; uint64_t divisor = Divisor; uint64_t quotient = Quotient; uint64_t remainder = dividend; int remaining = op2; //If we overshot, do nothing. This lets us unrool division loops a //little. if (remaining) { //Shift in bits from the low order portion of the dividend while(dividend < divisor && remaining) { dividend = (dividend << 1) | bits(SrcReg1, remaining - 1); quotient <<= 1; remaining--; } remainder = dividend; //Do the division. divide(dividend, divisor, quotient, remainder); } //Keep track of how many bits there are still to pull in. DestReg = merge(DestReg, remaining, dataSize); //Record the final results Remainder = remainder; Quotient = quotient; ''' flag_code = ''' if (DestReg == 0) ccFlagBits = ccFlagBits | (ext & EZFBit); else ccFlagBits = ccFlagBits & ~(ext & EZFBit); ''' class Divq(RdRegOp): code = 'DestReg = merge(SrcReg1, Quotient, dataSize);' class Divr(RdRegOp): code = 'DestReg = merge(SrcReg1, Remainder, dataSize);' class Mov(CondRegOp): code = 'DestReg = merge(SrcReg1, op2, dataSize)' else_code = 'DestReg=DestReg;' # Shift instructions class Sll(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); DestReg = merge(DestReg, psrc1 << shiftAmt, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); int CFBits = 0; //Figure out if we -would- set the CF bits if requested. if (shiftAmt <= dataSize * 8 && bits(SrcReg1, dataSize * 8 - shiftAmt)) { CFBits = 1; } //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && CFBits) ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); //Figure out what the OF bit should be. if ((ext & OFBit) && (CFBits ^ bits(DestReg, dataSize * 8 - 1))) ccFlagBits = ccFlagBits | OFBit; //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Srl(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); // Because what happens to the bits shift -in- on a right shift // is not defined in the C/C++ standard, we have to mask them out // to be sure they're zero. uint64_t logicalMask = mask(dataSize * 8 - shiftAmt); DestReg = merge(DestReg, (psrc1 >> shiftAmt) & logicalMask, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && shiftAmt <= dataSize * 8 && bits(SrcReg1, shiftAmt - 1)) { ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); } //Figure out what the OF bit should be. if ((ext & OFBit) && bits(SrcReg1, dataSize * 8 - 1)) ccFlagBits = ccFlagBits | OFBit; //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Sra(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); // Because what happens to the bits shift -in- on a right shift // is not defined in the C/C++ standard, we have to sign extend // them manually to be sure. uint64_t arithMask = (shiftAmt == 0) ? 0 : -bits(psrc1, dataSize * 8 - 1) << (dataSize * 8 - shiftAmt); DestReg = merge(DestReg, (psrc1 >> shiftAmt) | arithMask, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); //If some combination of the CF bits need to be set, set them. uint8_t effectiveShift = (shiftAmt <= dataSize * 8) ? shiftAmt : (dataSize * 8); if ((ext & (CFBit | ECFBit)) && bits(SrcReg1, effectiveShift - 1)) { ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); } //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Ror(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8); if(realShiftAmt) { uint64_t top = psrc1 << (dataSize * 8 - realShiftAmt); uint64_t bottom = bits(psrc1, dataSize * 8, realShiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); //Find the most and second most significant bits of the result. int msb = bits(DestReg, dataSize * 8 - 1); int smsb = bits(DestReg, dataSize * 8 - 2); //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && msb) ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); //Figure out what the OF bit should be. if ((ext & OFBit) && (msb ^ smsb)) ccFlagBits = ccFlagBits | OFBit; //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Rcr(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); if(shiftAmt) { CCFlagBits flags = ccFlagBits; uint64_t top = flags.cf << (dataSize * 8 - shiftAmt); if(shiftAmt > 1) top |= psrc1 << (dataSize * 8 - shiftAmt + 1); uint64_t bottom = bits(psrc1, dataSize * 8 - 1, shiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); //Figure out what the OF bit should be. if ((ext & OFBit) && ((ccFlagBits & CFBit) ^ bits(SrcReg1, dataSize * 8 - 1))) ccFlagBits = ccFlagBits | OFBit; //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && bits(SrcReg1, shiftAmt - 1)) ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Rol(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8); if(realShiftAmt) { uint64_t top = psrc1 << realShiftAmt; uint64_t bottom = bits(psrc1, dataSize * 8 - 1, dataSize * 8 - realShiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); //The CF bits, if set, would be set to the lsb of the result. int lsb = DestReg & 0x1; int msb = bits(DestReg, dataSize * 8 - 1); //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && lsb) ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); //Figure out what the OF bit should be. if ((ext & OFBit) && (msb ^ lsb)) ccFlagBits = ccFlagBits | OFBit; //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Rcl(RegOp): code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); if(shiftAmt) { CCFlagBits flags = ccFlagBits; uint64_t top = psrc1 << shiftAmt; uint64_t bottom = flags.cf << (shiftAmt - 1); if(shiftAmt > 1) bottom |= bits(psrc1, dataSize * 8 - 1, dataSize * 8 - shiftAmt + 1); DestReg = merge(DestReg, top | bottom, dataSize); } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { //Zero out any flags we might modify. This way we only have to //worry about setting them. ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | OFBit)); int msb = bits(DestReg, dataSize * 8 - 1); int CFBits = bits(SrcReg1, dataSize * 8 - shiftAmt); //If some combination of the CF bits need to be set, set them. if ((ext & (CFBit | ECFBit)) && CFBits) ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit)); //Figure out what the OF bit should be. if ((ext & OFBit) && (msb ^ CFBits)) ccFlagBits = ccFlagBits | OFBit; //Use the regular mechanisms to calculate the other flags. ccFlagBits = genFlags(ccFlagBits, ext & ~(CFBit | ECFBit | OFBit), DestReg, psrc1, op2); } ''' class Wrip(WrRegOp, CondRegOp): code = 'RIP = psrc1 + sop2 + CSBase' else_code="RIP = RIP;" class Wruflags(WrRegOp): code = 'ccFlagBits = psrc1 ^ op2' class Wrflags(WrRegOp): code = ''' MiscReg newFlags = psrc1 ^ op2; MiscReg userFlagMask = 0xDD5; // Get only the user flags ccFlagBits = newFlags & userFlagMask; // Get everything else nccFlagBits = newFlags & ~userFlagMask; ''' class Rdip(RdRegOp): code = 'DestReg = RIP - CSBase' class Ruflags(RdRegOp): code = 'DestReg = ccFlagBits' class Rflags(RdRegOp): code = 'DestReg = ccFlagBits | nccFlagBits' class Ruflag(RegOp): code = ''' int flag = bits(ccFlagBits, imm8); DestReg = merge(DestReg, flag, dataSize); ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : (ccFlagBits & ~EZFBit); ''' def __init__(self, dest, imm, flags=None, \ dataSize="env.dataSize"): super(Ruflag, self).__init__(dest, \ "InstRegIndex(NUM_INTREGS)", imm, flags, dataSize) class Rflag(RegOp): code = ''' MiscReg flagMask = 0x3F7FDD5; MiscReg flags = (nccFlagBits | ccFlagBits) & flagMask; int flag = bits(flags, imm8); DestReg = merge(DestReg, flag, dataSize); ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : (ccFlagBits & ~EZFBit); ''' def __init__(self, dest, imm, flags=None, \ dataSize="env.dataSize"): super(Rflag, self).__init__(dest, \ "InstRegIndex(NUM_INTREGS)", imm, flags, dataSize) class Sext(RegOp): code = ''' IntReg val = psrc1; // Mask the bit position so that it wraps. int bitPos = op2 & (dataSize * 8 - 1); int sign_bit = bits(val, bitPos, bitPos); uint64_t maskVal = mask(bitPos+1); val = sign_bit ? (val | ~maskVal) : (val & maskVal); DestReg = merge(DestReg, val, dataSize); ''' flag_code = ''' if (!sign_bit) ccFlagBits = ccFlagBits & ~(ext & (CFBit | ECFBit | ZFBit | EZFBit)); else ccFlagBits = ccFlagBits | (ext & (CFBit | ECFBit | ZFBit | EZFBit)); ''' class Zext(RegOp): code = 'DestReg = merge(DestReg, bits(psrc1, op2, 0), dataSize);' class Rddr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Rddr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' CR4 cr4 = CR4Op; DR7 dr7 = DR7Op; if ((cr4.de == 1 && (src1 == 4 || src1 == 5)) || src1 >= 8) { fault = new InvalidOpcode(); } else if (dr7.gd) { fault = new DebugException(); } else { DestReg = merge(DestReg, DebugSrc1, dataSize); } ''' class Wrdr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Wrdr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' CR4 cr4 = CR4Op; DR7 dr7 = DR7Op; if ((cr4.de == 1 && (dest == 4 || dest == 5)) || dest >= 8) { fault = new InvalidOpcode(); } else if ((dest == 6 || dest == 7) && bits(psrc1, 63, 32) && machInst.mode.mode == LongMode) { fault = new GeneralProtection(0); } else if (dr7.gd) { fault = new DebugException(); } else { DebugDest = psrc1; } ''' class Rdcr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Rdcr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' if (src1 == 1 || (src1 > 4 && src1 < 8) || (src1 > 8)) { fault = new InvalidOpcode(); } else { DestReg = merge(DestReg, ControlSrc1, dataSize); } ''' class Wrcr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Wrcr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' if (dest == 1 || (dest > 4 && dest < 8) || (dest > 8)) { fault = new InvalidOpcode(); } else { // There are *s in the line below so it doesn't confuse the // parser. They may be unnecessary. //Mis*cReg old*Val = pick(Cont*rolDest, 0, dat*aSize); MiscReg newVal = psrc1; // Check for any modifications that would cause a fault. switch(dest) { case 0: { Efer efer = EferOp; CR0 cr0 = newVal; CR4 oldCr4 = CR4Op; if (bits(newVal, 63, 32) || (!cr0.pe && cr0.pg) || (!cr0.cd && cr0.nw) || (cr0.pg && efer.lme && !oldCr4.pae)) fault = new GeneralProtection(0); } break; case 2: break; case 3: break; case 4: { CR4 cr4 = newVal; // PAE can't be disabled in long mode. if (bits(newVal, 63, 11) || (machInst.mode.mode == LongMode && !cr4.pae)) fault = new GeneralProtection(0); } break; case 8: { if (bits(newVal, 63, 4)) fault = new GeneralProtection(0); } default: panic("Unrecognized control register %d.\\n", dest); } ControlDest = newVal; } ''' # Microops for manipulating segmentation registers class SegOp(CondRegOp): abstract = True def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(SegOp, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) class Wrbase(SegOp): code = ''' SegBaseDest = psrc1; ''' class Wrlimit(SegOp): code = ''' SegLimitDest = psrc1; ''' class Wrsel(SegOp): code = ''' SegSelDest = psrc1; ''' class WrAttr(SegOp): code = ''' SegAttrDest = psrc1; ''' class Rdbase(SegOp): code = ''' DestReg = merge(DestReg, SegBaseSrc1, dataSize); ''' class Rdlimit(SegOp): code = ''' DestReg = merge(DestReg, SegLimitSrc1, dataSize); ''' class RdAttr(SegOp): code = ''' DestReg = merge(DestReg, SegAttrSrc1, dataSize); ''' class Rdsel(SegOp): code = ''' DestReg = merge(DestReg, SegSelSrc1, dataSize); ''' class Rdval(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Rdval, self).__init__(dest, src1, \ "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' DestReg = MiscRegSrc1; ''' class Wrval(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Wrval, self).__init__(dest, src1, \ "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = ''' MiscRegDest = SrcReg1; ''' class Chks(RegOp): def __init__(self, dest, src1, src2=0, flags=None, dataSize="env.dataSize"): super(Chks, self).__init__(dest, src1, src2, flags, dataSize) code = ''' // The selector is in source 1 and can be at most 16 bits. SegSelector selector = DestReg; SegDescriptor desc = SrcReg1; HandyM5Reg m5reg = M5Reg; switch (imm8) { case SegNoCheck: break; case SegCSCheck: // Make sure it's the right type if (desc.s == 0 || desc.type.codeOrData != 1) { fault = new GeneralProtection(0); } else if (m5reg.cpl != desc.dpl) { fault = new GeneralProtection(0); } break; case SegCallGateCheck: panic("CS checks for far calls/jumps through call gates" "not implemented.\\n"); break; case SegSoftIntGateCheck: // Check permissions. if (desc.dpl < m5reg.cpl) { fault = new GeneralProtection(selector); break; } // Fall through on purpose case SegIntGateCheck: // Make sure the gate's the right type. if ((m5reg.mode == LongMode && (desc.type & 0xe) != 0xe) || ((desc.type & 0x6) != 0x6)) { fault = new GeneralProtection(0); } break; case SegSSCheck: if (selector.si || selector.ti) { if (!desc.p) { fault = new StackFault(selector); } } else { if ((m5reg.submode != SixtyFourBitMode || m5reg.cpl == 3) || !(desc.s == 1 && desc.type.codeOrData == 0 && desc.type.w) || (desc.dpl != m5reg.cpl) || (selector.rpl != m5reg.cpl)) { fault = new GeneralProtection(selector); } } break; case SegIretCheck: { if ((!selector.si && !selector.ti) || (selector.rpl < m5reg.cpl) || !(desc.s == 1 && desc.type.codeOrData == 1) || (!desc.type.c && desc.dpl != selector.rpl) || (desc.type.c && desc.dpl > selector.rpl)) { fault = new GeneralProtection(selector); } else if (!desc.p) { fault = new SegmentNotPresent(selector); } break; } case SegIntCSCheck: if (m5reg.mode == LongMode) { if (desc.l != 1 || desc.d != 0) { fault = new GeneralProtection(selector); } } else { panic("Interrupt CS checks not implemented " "in legacy mode.\\n"); } break; case SegTRCheck: if (!selector.si || selector.ti) { fault = new GeneralProtection(selector); } break; case SegTSSCheck: if (!desc.p) { fault = new SegmentNotPresent(selector); } else if (!(desc.type == 0x9 || (desc.type == 1 && m5reg.mode != LongMode))) { fault = new GeneralProtection(selector); } break; case SegInGDTCheck: if (selector.ti) { fault = new GeneralProtection(selector); } break; case SegLDTCheck: if (!desc.p) { fault = new SegmentNotPresent(selector); } else if (desc.type != 0x2) { fault = new GeneralProtection(selector); } break; default: panic("Undefined segment check type.\\n"); } ''' flag_code = ''' // Check for a NULL selector and set ZF,EZF appropriately. ccFlagBits = ccFlagBits & ~(ext & (ZFBit | EZFBit)); if (!selector.si && !selector.ti) ccFlagBits = ccFlagBits | (ext & (ZFBit | EZFBit)); ''' class Wrdh(RegOp): code = ''' SegDescriptor desc = SrcReg1; uint64_t target = bits(SrcReg2, 31, 0) << 32; switch(desc.type) { case LDT64: case AvailableTSS64: case BusyTSS64: replaceBits(target, 23, 0, desc.baseLow); replaceBits(target, 31, 24, desc.baseHigh); break; case CallGate64: case IntGate64: case TrapGate64: replaceBits(target, 15, 0, bits(desc, 15, 0)); replaceBits(target, 31, 16, bits(desc, 63, 48)); break; default: panic("Wrdh used with wrong descriptor type!\\n"); } DestReg = target; ''' class Wrtsc(WrRegOp): code = ''' TscOp = psrc1; ''' class Rdtsc(RdRegOp): code = ''' DestReg = TscOp; ''' class Rdm5reg(RdRegOp): code = ''' DestReg = M5Reg; ''' class Wrdl(RegOp): code = ''' SegDescriptor desc = SrcReg1; SegSelector selector = SrcReg2; if (selector.si || selector.ti) { if (!desc.p) panic("Segment not present.\\n"); SegAttr attr = 0; attr.dpl = desc.dpl; attr.unusable = 0; attr.defaultSize = desc.d; attr.longMode = desc.l; attr.avl = desc.avl; attr.granularity = desc.g; attr.present = desc.p; attr.system = desc.s; attr.type = desc.type; if (!desc.s) { // The expand down bit happens to be set for gates. if (desc.type.e) { panic("Gate descriptor encountered.\\n"); } attr.readable = 1; attr.writable = 1; attr.expandDown = 0; } else { if (desc.type.codeOrData) { attr.expandDown = 0; attr.readable = desc.type.r; attr.writable = 0; } else { attr.expandDown = desc.type.e; attr.readable = 1; attr.writable = desc.type.w; } } Addr base = desc.baseLow | (desc.baseHigh << 24); Addr limit = desc.limitLow | (desc.limitHigh << 16); if (desc.g) limit = (limit << 12) | mask(12); SegBaseDest = base; SegLimitDest = limit; SegAttrDest = attr; } else { SegBaseDest = SegBaseDest; SegLimitDest = SegLimitDest; SegAttrDest = SegAttrDest; } ''' }};