43// 44// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 45// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 46// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 47// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 48// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 49// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 50// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 51// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 52// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 53// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 54// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 55// 56// Authors: Gabe Black 57 58//////////////////////////////////////////////////////////////////// 59// 60// Code to "specialize" a microcode sequence to use a particular 61// variety of operands 62// 63 64let {{ 65 # This code builds up a decode block which decodes based on switchval. 66 # vals is a dict which matches case values with what should be decoded to. 67 # Each element of the dict is a list containing a function and then the 68 # arguments to pass to it. 69 def doSplitDecode(switchVal, vals, default = None): 70 blocks = OutputBlocks() 71 blocks.decode_block = 'switch(%s) {\n' % switchVal 72 for (val, todo) in vals.items(): 73 new_blocks = todo[0](*todo[1:]) 74 new_blocks.decode_block = \ 75 '\tcase %s: %s\n' % (val, new_blocks.decode_block) 76 blocks.append(new_blocks) 77 if default: 78 new_blocks = default[0](*default[1:]) 79 new_blocks.decode_block = \ 80 '\tdefault: %s\n' % new_blocks.decode_block 81 blocks.append(new_blocks) 82 blocks.decode_block += '}\n' 83 return blocks 84}}; 85 86let {{ 87 def doRipRelativeDecode(Name, opTypes, env): 88 # print "RIPing %s with opTypes %s" % (Name, opTypes) 89 env.memoryInst = True 90 normEnv = copy.copy(env) 91 normEnv.addToDisassembly( 92 '''printMem(out, env.seg, env.scale, env.index, env.base, 93 machInst.displacement, env.addressSize, false);''') 94 normBlocks = specializeInst(Name + "_M", copy.copy(opTypes), normEnv) 95 ripEnv = copy.copy(env) 96 ripEnv.addToDisassembly( 97 '''printMem(out, env.seg, 1, 0, 0, 98 machInst.displacement, env.addressSize, true);''') 99 ripBlocks = specializeInst(Name + "_P", copy.copy(opTypes), ripEnv) 100 101 blocks = OutputBlocks() 102 blocks.append(normBlocks) 103 blocks.append(ripBlocks) 104 105 blocks.decode_block = ''' 106 if(machInst.modRM.mod == 0 && 107 machInst.modRM.rm == 5 && 108 machInst.mode.submode == SixtyFourBitMode) 109 { %s } 110 else 111 { %s }''' % \ 112 (ripBlocks.decode_block, normBlocks.decode_block) 113 return blocks 114}}; 115 116let {{ 117 class OpType(object): 118 parser = re.compile(r"(?P<tag>[A-Z]+)(?P<size>[a-z]*)|(r(?P<reg>[A-Z0-9]+)(?P<rsize>[a-z]*))") 119 def __init__(self, opTypeString): 120 match = OpType.parser.search(opTypeString) 121 if match == None: 122 raise Exception, "Problem parsing operand type %s" % opTypeString 123 self.reg = match.group("reg") 124 self.tag = match.group("tag") 125 self.size = match.group("size") 126 if not self.size: 127 self.size = match.group("rsize") 128 129 ModRMRegIndex = "(MODRM_REG | (REX_R << 3))" 130 ModRMRMIndex = "(MODRM_RM | (REX_B << 3))" 131 InstRegIndex = "(OPCODE_OP_BOTTOM3 | (REX_B << 3))" 132 133 # This function specializes the given piece of code to use a particular 134 # set of argument types described by "opTypes". 135 def specializeInst(Name, opTypes, env): 136 # print "Specializing %s with opTypes %s" % (Name, opTypes) 137 while len(opTypes): 138 # Parse the operand type string we're working with 139 opType = OpType(opTypes[0]) 140 opTypes.pop(0) 141 142 if opType.tag not in ("I", "J", "P", "PR", "Q", "V", "VR", "W"): 143 if opType.size: 144 env.setSize(opType.size) 145 146 if opType.reg: 147 #Figure out what to do with fixed register operands 148 #This is the index to use, so we should stick it some place. 149 if opType.reg in ("A", "B", "C", "D"): 150 regString = "INTREG_R%sX" % opType.reg 151 else: 152 regString = "INTREG_R%s" % opType.reg 153 env.addReg(regString) 154 env.addToDisassembly( 155 "printReg(out, %s, regSize);\n" % regString) 156 Name += "_R" 157 elif opType.tag == "B": 158 # This refers to registers whose index is encoded as part of the opcode 159 env.addToDisassembly( 160 "printReg(out, %s, regSize);\n" % InstRegIndex) 161 Name += "_R" 162 env.addReg(InstRegIndex) 163 elif opType.tag == "M": 164 # This refers to memory. The macroop constructor sets up modrm 165 # addressing. Non memory modrm settings should cause an error. 166 env.doModRM = True 167 return doRipRelativeDecode(Name, opTypes, env) 168 elif opType.tag == None or opType.size == None: 169 raise Exception, "Problem parsing operand tag: %s" % opType.tag 170 elif opType.tag == "C": 171 # A control register indexed by the "reg" field 172 env.addReg(ModRMRegIndex) 173 env.addToDisassembly( 174 "ccprintf(out, \"CR%%d\", %s);\n" % ModRMRegIndex) 175 Name += "_C" 176 elif opType.tag == "D": 177 # A debug register indexed by the "reg" field 178 env.addReg(ModRMRegIndex) 179 env.addToDisassembly( 180 "ccprintf(out, \"DR%%d\", %s);\n" % ModRMRegIndex) 181 Name += "_D" 182 elif opType.tag == "S": 183 # A segment selector register indexed by the "reg" field 184 env.addReg(ModRMRegIndex) 185 env.addToDisassembly( 186 "printSegment(out, %s);\n" % ModRMRegIndex) 187 Name += "_S" 188 elif opType.tag in ("G", "P", "T", "V"): 189 # Use the "reg" field of the ModRM byte to select the register 190 env.addReg(ModRMRegIndex) 191 env.addToDisassembly( 192 "printReg(out, %s, regSize);\n" % ModRMRegIndex) 193 if opType.tag == "P": 194 Name += "_MMX" 195 elif opType.tag == "V": 196 Name += "_XMM" 197 else: 198 Name += "_R" 199 elif opType.tag in ("E", "Q", "W"): 200 # This might refer to memory or to a register. We need to 201 # divide it up farther. 202 regEnv = copy.copy(env) 203 regEnv.addReg(ModRMRMIndex) 204 regEnv.addToDisassembly( 205 "printReg(out, %s, regSize);\n" % ModRMRMIndex) 206 # This refers to memory. The macroop constructor should set up 207 # modrm addressing. 208 memEnv = copy.copy(env) 209 memEnv.doModRM = True 210 regSuffix = "_R" 211 if opType.tag == "Q": 212 regSuffix = "_MMX" 213 elif opType.tag == "W": 214 regSuffix = "_XMM" 215 return doSplitDecode("MODRM_MOD", 216 {"3" : (specializeInst, Name + regSuffix, 217 copy.copy(opTypes), regEnv)}, 218 (doRipRelativeDecode, Name, 219 copy.copy(opTypes), memEnv)) 220 elif opType.tag in ("I", "J"): 221 # Immediates 222 env.addToDisassembly( 223 "ccprintf(out, \"%#x\", machInst.immediate);\n") 224 Name += "_I" 225 elif opType.tag == "O": 226 # Immediate containing a memory offset 227 Name += "_MI" 228 elif opType.tag in ("PR", "R", "VR"): 229 # Non register modrm settings should cause an error 230 env.addReg(ModRMRMIndex) 231 env.addToDisassembly( 232 "printReg(out, %s, regSize);\n" % ModRMRMIndex) 233 if opType.tag == "PR": 234 Name += "_MMX" 235 elif opType.tag == "VR": 236 Name += "_XMM" 237 else: 238 Name += "_R" 239 elif opType.tag in ("X", "Y"): 240 # This type of memory addressing is for string instructions. 241 # They'll use the right index and segment internally. 242 if opType.tag == "X": 243 env.addToDisassembly( 244 '''printMem(out, env.seg, 245 1, X86ISA::ZeroReg, X86ISA::INTREG_RSI, 0, 246 env.addressSize, false);''') 247 else: 248 env.addToDisassembly( 249 '''printMem(out, SEGMENT_REG_ES, 250 1, X86ISA::ZeroReg, X86ISA::INTREG_RDI, 0, 251 env.addressSize, false);''') 252 Name += "_M" 253 else: 254 raise Exception, "Unrecognized tag %s." % opType.tag 255 256 # Generate code to return a macroop of the given name which will 257 # operate in the "emulation environment" env 258 return genMacroop(Name, env) 259}};
| 25// 26// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 27// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 28// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 29// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 30// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 31// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 32// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 33// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 34// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 35// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 36// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 37// 38// Authors: Gabe Black 39 40//////////////////////////////////////////////////////////////////// 41// 42// Code to "specialize" a microcode sequence to use a particular 43// variety of operands 44// 45 46let {{ 47 # This code builds up a decode block which decodes based on switchval. 48 # vals is a dict which matches case values with what should be decoded to. 49 # Each element of the dict is a list containing a function and then the 50 # arguments to pass to it. 51 def doSplitDecode(switchVal, vals, default = None): 52 blocks = OutputBlocks() 53 blocks.decode_block = 'switch(%s) {\n' % switchVal 54 for (val, todo) in vals.items(): 55 new_blocks = todo[0](*todo[1:]) 56 new_blocks.decode_block = \ 57 '\tcase %s: %s\n' % (val, new_blocks.decode_block) 58 blocks.append(new_blocks) 59 if default: 60 new_blocks = default[0](*default[1:]) 61 new_blocks.decode_block = \ 62 '\tdefault: %s\n' % new_blocks.decode_block 63 blocks.append(new_blocks) 64 blocks.decode_block += '}\n' 65 return blocks 66}}; 67 68let {{ 69 def doRipRelativeDecode(Name, opTypes, env): 70 # print "RIPing %s with opTypes %s" % (Name, opTypes) 71 env.memoryInst = True 72 normEnv = copy.copy(env) 73 normEnv.addToDisassembly( 74 '''printMem(out, env.seg, env.scale, env.index, env.base, 75 machInst.displacement, env.addressSize, false);''') 76 normBlocks = specializeInst(Name + "_M", copy.copy(opTypes), normEnv) 77 ripEnv = copy.copy(env) 78 ripEnv.addToDisassembly( 79 '''printMem(out, env.seg, 1, 0, 0, 80 machInst.displacement, env.addressSize, true);''') 81 ripBlocks = specializeInst(Name + "_P", copy.copy(opTypes), ripEnv) 82 83 blocks = OutputBlocks() 84 blocks.append(normBlocks) 85 blocks.append(ripBlocks) 86 87 blocks.decode_block = ''' 88 if(machInst.modRM.mod == 0 && 89 machInst.modRM.rm == 5 && 90 machInst.mode.submode == SixtyFourBitMode) 91 { %s } 92 else 93 { %s }''' % \ 94 (ripBlocks.decode_block, normBlocks.decode_block) 95 return blocks 96}}; 97 98let {{ 99 class OpType(object): 100 parser = re.compile(r"(?P<tag>[A-Z]+)(?P<size>[a-z]*)|(r(?P<reg>[A-Z0-9]+)(?P<rsize>[a-z]*))") 101 def __init__(self, opTypeString): 102 match = OpType.parser.search(opTypeString) 103 if match == None: 104 raise Exception, "Problem parsing operand type %s" % opTypeString 105 self.reg = match.group("reg") 106 self.tag = match.group("tag") 107 self.size = match.group("size") 108 if not self.size: 109 self.size = match.group("rsize") 110 111 ModRMRegIndex = "(MODRM_REG | (REX_R << 3))" 112 ModRMRMIndex = "(MODRM_RM | (REX_B << 3))" 113 InstRegIndex = "(OPCODE_OP_BOTTOM3 | (REX_B << 3))" 114 115 # This function specializes the given piece of code to use a particular 116 # set of argument types described by "opTypes". 117 def specializeInst(Name, opTypes, env): 118 # print "Specializing %s with opTypes %s" % (Name, opTypes) 119 while len(opTypes): 120 # Parse the operand type string we're working with 121 opType = OpType(opTypes[0]) 122 opTypes.pop(0) 123 124 if opType.tag not in ("I", "J", "P", "PR", "Q", "V", "VR", "W"): 125 if opType.size: 126 env.setSize(opType.size) 127 128 if opType.reg: 129 #Figure out what to do with fixed register operands 130 #This is the index to use, so we should stick it some place. 131 if opType.reg in ("A", "B", "C", "D"): 132 regString = "INTREG_R%sX" % opType.reg 133 else: 134 regString = "INTREG_R%s" % opType.reg 135 env.addReg(regString) 136 env.addToDisassembly( 137 "printReg(out, %s, regSize);\n" % regString) 138 Name += "_R" 139 elif opType.tag == "B": 140 # This refers to registers whose index is encoded as part of the opcode 141 env.addToDisassembly( 142 "printReg(out, %s, regSize);\n" % InstRegIndex) 143 Name += "_R" 144 env.addReg(InstRegIndex) 145 elif opType.tag == "M": 146 # This refers to memory. The macroop constructor sets up modrm 147 # addressing. Non memory modrm settings should cause an error. 148 env.doModRM = True 149 return doRipRelativeDecode(Name, opTypes, env) 150 elif opType.tag == None or opType.size == None: 151 raise Exception, "Problem parsing operand tag: %s" % opType.tag 152 elif opType.tag == "C": 153 # A control register indexed by the "reg" field 154 env.addReg(ModRMRegIndex) 155 env.addToDisassembly( 156 "ccprintf(out, \"CR%%d\", %s);\n" % ModRMRegIndex) 157 Name += "_C" 158 elif opType.tag == "D": 159 # A debug register indexed by the "reg" field 160 env.addReg(ModRMRegIndex) 161 env.addToDisassembly( 162 "ccprintf(out, \"DR%%d\", %s);\n" % ModRMRegIndex) 163 Name += "_D" 164 elif opType.tag == "S": 165 # A segment selector register indexed by the "reg" field 166 env.addReg(ModRMRegIndex) 167 env.addToDisassembly( 168 "printSegment(out, %s);\n" % ModRMRegIndex) 169 Name += "_S" 170 elif opType.tag in ("G", "P", "T", "V"): 171 # Use the "reg" field of the ModRM byte to select the register 172 env.addReg(ModRMRegIndex) 173 env.addToDisassembly( 174 "printReg(out, %s, regSize);\n" % ModRMRegIndex) 175 if opType.tag == "P": 176 Name += "_MMX" 177 elif opType.tag == "V": 178 Name += "_XMM" 179 else: 180 Name += "_R" 181 elif opType.tag in ("E", "Q", "W"): 182 # This might refer to memory or to a register. We need to 183 # divide it up farther. 184 regEnv = copy.copy(env) 185 regEnv.addReg(ModRMRMIndex) 186 regEnv.addToDisassembly( 187 "printReg(out, %s, regSize);\n" % ModRMRMIndex) 188 # This refers to memory. The macroop constructor should set up 189 # modrm addressing. 190 memEnv = copy.copy(env) 191 memEnv.doModRM = True 192 regSuffix = "_R" 193 if opType.tag == "Q": 194 regSuffix = "_MMX" 195 elif opType.tag == "W": 196 regSuffix = "_XMM" 197 return doSplitDecode("MODRM_MOD", 198 {"3" : (specializeInst, Name + regSuffix, 199 copy.copy(opTypes), regEnv)}, 200 (doRipRelativeDecode, Name, 201 copy.copy(opTypes), memEnv)) 202 elif opType.tag in ("I", "J"): 203 # Immediates 204 env.addToDisassembly( 205 "ccprintf(out, \"%#x\", machInst.immediate);\n") 206 Name += "_I" 207 elif opType.tag == "O": 208 # Immediate containing a memory offset 209 Name += "_MI" 210 elif opType.tag in ("PR", "R", "VR"): 211 # Non register modrm settings should cause an error 212 env.addReg(ModRMRMIndex) 213 env.addToDisassembly( 214 "printReg(out, %s, regSize);\n" % ModRMRMIndex) 215 if opType.tag == "PR": 216 Name += "_MMX" 217 elif opType.tag == "VR": 218 Name += "_XMM" 219 else: 220 Name += "_R" 221 elif opType.tag in ("X", "Y"): 222 # This type of memory addressing is for string instructions. 223 # They'll use the right index and segment internally. 224 if opType.tag == "X": 225 env.addToDisassembly( 226 '''printMem(out, env.seg, 227 1, X86ISA::ZeroReg, X86ISA::INTREG_RSI, 0, 228 env.addressSize, false);''') 229 else: 230 env.addToDisassembly( 231 '''printMem(out, SEGMENT_REG_ES, 232 1, X86ISA::ZeroReg, X86ISA::INTREG_RDI, 0, 233 env.addressSize, false);''') 234 Name += "_M" 235 else: 236 raise Exception, "Unrecognized tag %s." % opType.tag 237 238 # Generate code to return a macroop of the given name which will 239 # operate in the "emulation environment" env 240 return genMacroop(Name, env) 241}};
|