src/soc/decoder/isa/caller.py

   1 # SPDX-License-Identifier: LGPLv3+
   2 # Copyright (C) 2020, 2021 Luke Kenneth Casson Leighton <lkcl@lkcl.net>
   3 # Copyright (C) 2020 Michael Nolan
   4 # Funded by NLnet http://nlnet.nl
   5 """core of the python-based POWER9 simulator
   6
   7 this is part of a cycle-accurate POWER9 simulator.  its primary purpose is
   8 not speed, it is for both learning and educational purposes, as well as
   9 a method of verifying the HDL.
  10
  11 related bugs:
  12
  13 * https://bugs.libre-soc.org/show_bug.cgi?id=424
  14 """
  15
  16 from functools import wraps
  17 from copy import copy
  18 from soc.decoder.orderedset import OrderedSet
  19 from soc.decoder.selectable_int import (FieldSelectableInt, SelectableInt,
  20                                         selectconcat)
  21 from soc.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
  22                                      insns, MicrOp)
  23 from soc.decoder.helpers import exts, gtu, ltu, undefined
  24 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  25
  26 from collections import namedtuple
  27 import math
  28 import sys
  29
  30 instruction_info = namedtuple('instruction_info',
  31                               'func read_regs uninit_regs write_regs ' +
  32                               'special_regs op_fields form asmregs')
  33
  34 special_sprs = {
  35     'LR': 8,
  36     'CTR': 9,
  37     'TAR': 815,
  38     'XER': 1,
  39     'VRSAVE': 256}
  40
  41
  42 def swap_order(x, nbytes):
  43     x = x.to_bytes(nbytes, byteorder='little')
  44     x = int.from_bytes(x, byteorder='big', signed=False)
  45     return x
  46
  47
  48 REG_SORT_ORDER = {
  49     # TODO (lkcl): adjust other registers that should be in a particular order
  50     # probably CA, CA32, and CR
  51     "RT": 0,
  52     "RA": 0,
  53     "RB": 0,
  54     "RS": 0,
  55     "CR": 0,
  56     "LR": 0,
  57     "CTR": 0,
  58     "TAR": 0,
  59     "CA": 0,
  60     "CA32": 0,
  61     "MSR": 0,
  62
  63     "overflow": 1,
  64 }
  65
  66
  67 def create_args(reglist, extra=None):
  68     retval = list(OrderedSet(reglist))
  69     retval.sort(key=lambda reg: REG_SORT_ORDER[reg])
  70     if extra is not None:
  71         return [extra] + retval
  72     return retval
  73
  74
  75 class Mem:
  76
  77     def __init__(self, row_bytes=8, initial_mem=None):
  78         self.mem = {}
  79         self.bytes_per_word = row_bytes
  80         self.word_log2 = math.ceil(math.log2(row_bytes))
  81         print("Sim-Mem", initial_mem, self.bytes_per_word, self.word_log2)
  82         if not initial_mem:
  83             return
  84
  85         # different types of memory data structures recognised (for convenience)
  86         if isinstance(initial_mem, list):
  87             initial_mem = (0, initial_mem)
  88         if isinstance(initial_mem, tuple):
  89             startaddr, mem = initial_mem
  90             initial_mem = {}
  91             for i, val in enumerate(mem):
  92                 initial_mem[startaddr + row_bytes*i] = (val, row_bytes)
  93
  94         for addr, (val, width) in initial_mem.items():
  95             #val = swap_order(val, width)
  96             self.st(addr, val, width, swap=False)
  97
  98     def _get_shifter_mask(self, wid, remainder):
  99         shifter = ((self.bytes_per_word - wid) - remainder) * \
 100             8  # bits per byte
 101         # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
 102         # BE/LE mode?
 103         shifter = remainder * 8
 104         mask = (1 << (wid * 8)) - 1
 105         print("width,rem,shift,mask", wid, remainder, hex(shifter), hex(mask))
 106         return shifter, mask
 107
 108     # TODO: Implement ld/st of lesser width
 109     def ld(self, address, width=8, swap=True, check_in_mem=False):
 110         print("ld from addr 0x{:x} width {:d}".format(address, width))
 111         remainder = address & (self.bytes_per_word - 1)
 112         address = address >> self.word_log2
 113         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 114         if address in self.mem:
 115             val = self.mem[address]
 116         elif check_in_mem:
 117             return None
 118         else:
 119             val = 0
 120         print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address, remainder, val))
 121
 122         if width != self.bytes_per_word:
 123             shifter, mask = self._get_shifter_mask(width, remainder)
 124             print("masking", hex(val), hex(mask << shifter), shifter)
 125             val = val & (mask << shifter)
 126             val >>= shifter
 127         if swap:
 128             val = swap_order(val, width)
 129         print("Read 0x{:x} from addr 0x{:x}".format(val, address))
 130         return val
 131
 132     def st(self, addr, v, width=8, swap=True):
 133         staddr = addr
 134         remainder = addr & (self.bytes_per_word - 1)
 135         addr = addr >> self.word_log2
 136         print("Writing 0x{:x} to ST 0x{:x} "
 137               "memaddr 0x{:x}/{:x}".format(v, staddr, addr, remainder, swap))
 138         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 139         if swap:
 140             v = swap_order(v, width)
 141         if width != self.bytes_per_word:
 142             if addr in self.mem:
 143                 val = self.mem[addr]
 144             else:
 145                 val = 0
 146             shifter, mask = self._get_shifter_mask(width, remainder)
 147             val &= ~(mask << shifter)
 148             val |= v << shifter
 149             self.mem[addr] = val
 150         else:
 151             self.mem[addr] = v
 152         print("mem @ 0x{:x}: 0x{:x}".format(addr, self.mem[addr]))
 153
 154     def __call__(self, addr, sz):
 155         val = self.ld(addr.value, sz, swap=False)
 156         print("memread", addr, sz, val)
 157         return SelectableInt(val, sz*8)
 158
 159     def memassign(self, addr, sz, val):
 160         print("memassign", addr, sz, val)
 161         self.st(addr.value, val.value, sz, swap=False)
 162
 163
 164 class GPR(dict):
 165     def __init__(self, decoder, regfile):
 166         dict.__init__(self)
 167         self.sd = decoder
 168         for i in range(32):
 169             self[i] = SelectableInt(regfile[i], 64)
 170
 171     def __call__(self, ridx):
 172         return self[ridx]
 173
 174     def set_form(self, form):
 175         self.form = form
 176
 177     def getz(self, rnum):
 178         # rnum = rnum.value # only SelectableInt allowed
 179         print("GPR getzero", rnum)
 180         if rnum == 0:
 181             return SelectableInt(0, 64)
 182         return self[rnum]
 183
 184     def _get_regnum(self, attr):
 185         getform = self.sd.sigforms[self.form]
 186         rnum = getattr(getform, attr)
 187         return rnum
 188
 189     def ___getitem__(self, attr):
 190         print("GPR getitem", attr)
 191         rnum = self._get_regnum(attr)
 192         return self.regfile[rnum]
 193
 194     def dump(self):
 195         for i in range(0, len(self), 8):
 196             s = []
 197             for j in range(8):
 198                 s.append("%08x" % self[i+j].value)
 199             s = ' '.join(s)
 200             print("reg", "%2d" % i, s)
 201
 202
 203 class PC:
 204     def __init__(self, pc_init=0):
 205         self.CIA = SelectableInt(pc_init, 64)
 206         self.NIA = self.CIA + SelectableInt(4, 64)
 207
 208     def update(self, namespace):
 209         self.CIA = namespace['NIA'].narrow(64)
 210         self.NIA = self.CIA + SelectableInt(4, 64)
 211         namespace['CIA'] = self.CIA
 212         namespace['NIA'] = self.NIA
 213
 214
 215 # Simple-V: see https://libre-soc.org/openpower/sv
 216 # also soc.sv.svstate SVSTATEREC
 217 class SVP64State:
 218     def __init__(self, init=0):
 219         self.spr = SelectableInt(init, 32)
 220         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 221         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7))
 222         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 223         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 224         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 225         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 226         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 227
 228
 229 # SVP64 ReMap field
 230 class SVP64RMFields:
 231     def __init__(self, init=0):
 232         self.spr = SelectableInt(init, 24)
 233         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 234         self.mmode = FieldSelectableInt(self.spr, [0])
 235         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 236         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 237         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 238         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 239         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 240         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 241
 242
 243 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 244 class SPP64PrefixFields:
 245     def __init__(self):
 246         self.insn = SelectableInt(0, 32)
 247         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 248         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 249         self.pid = FieldSelectableInt(self.insn, (7, 9) # must be 0b11
 250         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM
 251         self.rm = FieldSelectableInt(self.spr, rmfields)
 252
 253
 254 class SPR(dict):
 255     def __init__(self, dec2, initial_sprs={}):
 256         self.sd = dec2
 257         dict.__init__(self)
 258         for key, v in initial_sprs.items():
 259             if isinstance(key, SelectableInt):
 260                 key = key.value
 261             key = special_sprs.get(key, key)
 262             if isinstance(key, int):
 263                 info = spr_dict[key]
 264             else:
 265                 info = spr_byname[key]
 266             if not isinstance(v, SelectableInt):
 267                 v = SelectableInt(v, info.length)
 268             self[key] = v
 269
 270     def __getitem__(self, key):
 271         print("get spr", key)
 272         print("dict", self.items())
 273         # if key in special_sprs get the special spr, otherwise return key
 274         if isinstance(key, SelectableInt):
 275             key = key.value
 276         if isinstance(key, int):
 277             key = spr_dict[key].SPR
 278         key = special_sprs.get(key, key)
 279         if key == 'HSRR0':  # HACK!
 280             key = 'SRR0'
 281         if key == 'HSRR1':  # HACK!
 282             key = 'SRR1'
 283         if key in self:
 284             res = dict.__getitem__(self, key)
 285         else:
 286             if isinstance(key, int):
 287                 info = spr_dict[key]
 288             else:
 289                 info = spr_byname[key]
 290             dict.__setitem__(self, key, SelectableInt(0, info.length))
 291             res = dict.__getitem__(self, key)
 292         print("spr returning", key, res)
 293         return res
 294
 295     def __setitem__(self, key, value):
 296         if isinstance(key, SelectableInt):
 297             key = key.value
 298         if isinstance(key, int):
 299             key = spr_dict[key].SPR
 300             print("spr key", key)
 301         key = special_sprs.get(key, key)
 302         if key == 'HSRR0':  # HACK!
 303             self.__setitem__('SRR0', value)
 304         if key == 'HSRR1':  # HACK!
 305             self.__setitem__('SRR1', value)
 306         print("setting spr", key, value)
 307         dict.__setitem__(self, key, value)
 308
 309     def __call__(self, ridx):
 310         return self[ridx]
 311
 312
 313 class ISACaller:
 314     # decoder2 - an instance of power_decoder2
 315     # regfile - a list of initial values for the registers
 316     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 317     # respect_pc - tracks the program counter.  requires initial_insns
 318     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 319                  initial_mem=None, initial_msr=0,
 320                  initial_svstate=0,
 321                  initial_insns=None, respect_pc=False,
 322                  disassembly=None,
 323                  initial_pc=0,
 324                  bigendian=False):
 325
 326         self.bigendian = bigendian
 327         self.halted = False
 328         self.respect_pc = respect_pc
 329         if initial_sprs is None:
 330             initial_sprs = {}
 331         if initial_mem is None:
 332             initial_mem = {}
 333         if initial_insns is None:
 334             initial_insns = {}
 335             assert self.respect_pc == False, "instructions required to honor pc"
 336
 337         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 338         print("ISACaller initial_msr", initial_msr)
 339
 340         # "fake program counter" mode (for unit testing)
 341         self.fake_pc = 0
 342         disasm_start = 0
 343         if not respect_pc:
 344             if isinstance(initial_mem, tuple):
 345                 self.fake_pc = initial_mem[0]
 346                 disasm_start = self.fake_pc
 347         else:
 348             disasm_start = initial_pc
 349
 350         # disassembly: we need this for now (not given from the decoder)
 351         self.disassembly = {}
 352         if disassembly:
 353             for i, code in enumerate(disassembly):
 354                 self.disassembly[i*4 + disasm_start] = code
 355
 356         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 357         self.gpr = GPR(decoder2, regfile)
 358         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 359         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 360         self.pc = PC()
 361         self.svstate = SVSTATE(initial_svstate)
 362         self.spr = SPR(decoder2, initial_sprs)
 363         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 364
 365         # TODO, needed here:
 366         # FPR (same as GPR except for FP nums)
 367         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 368         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 369         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 370         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 371         #         -- Done
 372         # 2.3.2 LR   (actually SPR #8) -- Done
 373         # 2.3.3 CTR  (actually SPR #9) -- Done
 374         # 2.3.4 TAR  (actually SPR #815)
 375         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 376         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 377
 378         # create CR then allow portions of it to be "selectable" (below)
 379         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 380         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 381         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 382
 383         # "undefined", just set to variable-bit-width int (use exts "max")
 384         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 385
 386         self.namespace = {}
 387         self.namespace.update(self.spr)
 388         self.namespace.update({'GPR': self.gpr,
 389                                'MEM': self.mem,
 390                                'SPR': self.spr,
 391                                'memassign': self.memassign,
 392                                'NIA': self.pc.NIA,
 393                                'CIA': self.pc.CIA,
 394                                'CR': self.cr,
 395                                'MSR': self.msr,
 396                                'undefined': undefined,
 397                                'mode_is_64bit': True,
 398                                'SO': XER_bits['SO']
 399                                })
 400
 401         # update pc to requested start point
 402         self.set_pc(initial_pc)
 403
 404         # field-selectable versions of Condition Register TODO check bitranges?
 405         self.crl = []
 406         for i in range(8):
 407             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 408             _cr = FieldSelectableInt(self.cr, bits)
 409             self.crl.append(_cr)
 410             self.namespace["CR%d" % i] = _cr
 411
 412         self.decoder = decoder2.dec
 413         self.dec2 = decoder2
 414
 415     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 416         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 417         # store CIA(+4?) in SRR0, set NIA to 0x700
 418         # store MSR in SRR1, set MSR to um errr something, have to check spec
 419         self.spr['SRR0'].value = self.pc.CIA.value
 420         self.spr['SRR1'].value = self.namespace['MSR'].value
 421         self.trap_nia = SelectableInt(trap_addr, 64)
 422         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 423
 424         # set exception bits.  TODO: this should, based on the address
 425         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 426         # bits appropriately.  however it turns out that *for now* in all
 427         # cases (all trap_addrs) the exact same thing is needed.
 428         self.msr[MSRb.IR] = 0
 429         self.msr[MSRb.DR] = 0
 430         self.msr[MSRb.FE0] = 0
 431         self.msr[MSRb.FE1] = 0
 432         self.msr[MSRb.EE] = 0
 433         self.msr[MSRb.RI] = 0
 434         self.msr[MSRb.SF] = 1
 435         self.msr[MSRb.TM] = 0
 436         self.msr[MSRb.VEC] = 0
 437         self.msr[MSRb.VSX] = 0
 438         self.msr[MSRb.PR] = 0
 439         self.msr[MSRb.FP] = 0
 440         self.msr[MSRb.PMM] = 0
 441         self.msr[MSRb.TEs] = 0
 442         self.msr[MSRb.TEe] = 0
 443         self.msr[MSRb.UND] = 0
 444         self.msr[MSRb.LE] = 1
 445
 446     def memassign(self, ea, sz, val):
 447         self.mem.memassign(ea, sz, val)
 448
 449     def prep_namespace(self, formname, op_fields):
 450         # TODO: get field names from form in decoder*1* (not decoder2)
 451         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 452         # from spec
 453         # then "yield" fields only from op_fields rather than hard-coded
 454         # list, here.
 455         fields = self.decoder.sigforms[formname]
 456         for name in op_fields:
 457             if name == 'spr':
 458                 sig = getattr(fields, name.upper())
 459             else:
 460                 sig = getattr(fields, name)
 461             val = yield sig
 462             # these are all opcode fields involved in index-selection of CR,
 463             # and need to do "standard" arithmetic.  CR[BA+32] for example
 464             # would, if using SelectableInt, only be 5-bit.
 465             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 466                 self.namespace[name] = val
 467             else:
 468                 self.namespace[name] = SelectableInt(val, sig.width)
 469
 470         self.namespace['XER'] = self.spr['XER']
 471         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 472         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 473
 474     def handle_carry_(self, inputs, outputs, already_done):
 475         inv_a = yield self.dec2.e.do.invert_in
 476         if inv_a:
 477             inputs[0] = ~inputs[0]
 478
 479         imm_ok = yield self.dec2.e.do.imm_data.ok
 480         if imm_ok:
 481             imm = yield self.dec2.e.do.imm_data.data
 482             inputs.append(SelectableInt(imm, 64))
 483         assert len(outputs) >= 1
 484         print("outputs", repr(outputs))
 485         if isinstance(outputs, list) or isinstance(outputs, tuple):
 486             output = outputs[0]
 487         else:
 488             output = outputs
 489         gts = []
 490         for x in inputs:
 491             print("gt input", x, output)
 492             gt = (gtu(x, output))
 493             gts.append(gt)
 494         print(gts)
 495         cy = 1 if any(gts) else 0
 496         print("CA", cy, gts)
 497         if not (1 & already_done):
 498             self.spr['XER'][XER_bits['CA']] = cy
 499
 500         print("inputs", already_done, inputs)
 501         # 32 bit carry
 502         # ARGH... different for OP_ADD... *sigh*...
 503         op = yield self.dec2.e.do.insn_type
 504         if op == MicrOp.OP_ADD.value:
 505             res32 = (output.value & (1 << 32)) != 0
 506             a32 = (inputs[0].value & (1 << 32)) != 0
 507             if len(inputs) >= 2:
 508                 b32 = (inputs[1].value & (1 << 32)) != 0
 509             else:
 510                 b32 = False
 511             cy32 = res32 ^ a32 ^ b32
 512             print("CA32 ADD", cy32)
 513         else:
 514             gts = []
 515             for x in inputs:
 516                 print("input", x, output)
 517                 print("     x[32:64]", x, x[32:64])
 518                 print("     o[32:64]", output, output[32:64])
 519                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 520                 gts.append(gt)
 521             cy32 = 1 if any(gts) else 0
 522             print("CA32", cy32, gts)
 523         if not (2 & already_done):
 524             self.spr['XER'][XER_bits['CA32']] = cy32
 525
 526     def handle_overflow(self, inputs, outputs, div_overflow):
 527         if hasattr(self.dec2.e.do, "invert_in"):
 528             inv_a = yield self.dec2.e.do.invert_in
 529             if inv_a:
 530                 inputs[0] = ~inputs[0]
 531
 532         imm_ok = yield self.dec2.e.do.imm_data.ok
 533         if imm_ok:
 534             imm = yield self.dec2.e.do.imm_data.data
 535             inputs.append(SelectableInt(imm, 64))
 536         assert len(outputs) >= 1
 537         print("handle_overflow", inputs, outputs, div_overflow)
 538         if len(inputs) < 2 and div_overflow is None:
 539             return
 540
 541         # div overflow is different: it's returned by the pseudo-code
 542         # because it's more complex than can be done by analysing the output
 543         if div_overflow is not None:
 544             ov, ov32 = div_overflow, div_overflow
 545         # arithmetic overflow can be done by analysing the input and output
 546         elif len(inputs) >= 2:
 547             output = outputs[0]
 548
 549             # OV (64-bit)
 550             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 551             output_sgn = exts(output.value, output.bits) < 0
 552             ov = 1 if input_sgn[0] == input_sgn[1] and \
 553                 output_sgn != input_sgn[0] else 0
 554
 555             # OV (32-bit)
 556             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 557             output32_sgn = exts(output.value, 32) < 0
 558             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 559                 output32_sgn != input32_sgn[0] else 0
 560
 561         self.spr['XER'][XER_bits['OV']] = ov
 562         self.spr['XER'][XER_bits['OV32']] = ov32
 563         so = self.spr['XER'][XER_bits['SO']]
 564         so = so | ov
 565         self.spr['XER'][XER_bits['SO']] = so
 566
 567     def handle_comparison(self, outputs):
 568         out = outputs[0]
 569         assert isinstance(out, SelectableInt), \
 570             "out zero not a SelectableInt %s" % repr(outputs)
 571         print("handle_comparison", out.bits, hex(out.value))
 572         # TODO - XXX *processor* in 32-bit mode
 573         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 574         # if is_32bit:
 575         #    o32 = exts(out.value, 32)
 576         #    print ("handle_comparison exts 32 bit", hex(o32))
 577         out = exts(out.value, out.bits)
 578         print("handle_comparison exts", hex(out))
 579         zero = SelectableInt(out == 0, 1)
 580         positive = SelectableInt(out > 0, 1)
 581         negative = SelectableInt(out < 0, 1)
 582         SO = self.spr['XER'][XER_bits['SO']]
 583         print("handle_comparison SO", SO)
 584         cr_field = selectconcat(negative, positive, zero, SO)
 585         self.crl[0].eq(cr_field)
 586
 587     def set_pc(self, pc_val):
 588         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 589         self.pc.update(self.namespace)
 590
 591     def setup_one(self):
 592         """set up one instruction
 593         """
 594         if self.respect_pc:
 595             pc = self.pc.CIA.value
 596         else:
 597             pc = self.fake_pc
 598         self._pc = pc
 599         ins = self.imem.ld(pc, 4, False, True)
 600         if ins is None:
 601             raise KeyError("no instruction at 0x%x" % pc)
 602         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 603         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 604
 605         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 606         yield self.dec2.dec.bigendian.eq(self.bigendian)
 607         yield self.dec2.state.msr.eq(self.msr.value)
 608         yield self.dec2.state.pc.eq(pc)
 609
 610     def execute_one(self):
 611         """execute one instruction
 612         """
 613         # get the disassembly code for this instruction
 614         code = self.disassembly[self._pc]
 615         print("sim-execute", hex(self._pc), code)
 616         opname = code.split(' ')[0]
 617         yield from self.call(opname)
 618
 619         if not self.respect_pc:
 620             self.fake_pc += 4
 621         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 622
 623     def get_assembly_name(self):
 624         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 625         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 626         dec_insn = yield self.dec2.e.do.insn
 627         asmcode = yield self.dec2.dec.op.asmcode
 628         print("get assembly name asmcode", asmcode, hex(dec_insn))
 629         asmop = insns.get(asmcode, None)
 630         int_op = yield self.dec2.dec.op.internal_op
 631
 632         # sigh reconstruct the assembly instruction name
 633         if hasattr(self.dec2.e.do, "oe"):
 634             ov_en = yield self.dec2.e.do.oe.oe
 635             ov_ok = yield self.dec2.e.do.oe.ok
 636         else:
 637             ov_en = False
 638             ov_ok = False
 639         if hasattr(self.dec2.e.do, "rc"):
 640             rc_en = yield self.dec2.e.do.rc.rc
 641             rc_ok = yield self.dec2.e.do.rc.ok
 642         else:
 643             rc_en = False
 644             rc_ok = False
 645         # grrrr have to special-case MUL op (see DecodeOE)
 646         print("ov %d en %d rc %d en %d op %d" %
 647               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 648         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 649             print("mul op")
 650             if rc_en & rc_ok:
 651                 asmop += "."
 652         else:
 653             if not asmop.endswith("."):  # don't add "." to "andis."
 654                 if rc_en & rc_ok:
 655                     asmop += "."
 656         if hasattr(self.dec2.e.do, "lk"):
 657             lk = yield self.dec2.e.do.lk
 658             if lk:
 659                 asmop += "l"
 660         print("int_op", int_op)
 661         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 662             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 663             print("AA", AA)
 664             if AA:
 665                 asmop += "a"
 666         spr_msb = yield from self.get_spr_msb()
 667         if int_op == MicrOp.OP_MFCR.value:
 668             if spr_msb:
 669                 asmop = 'mfocrf'
 670             else:
 671                 asmop = 'mfcr'
 672         # XXX TODO: for whatever weird reason this doesn't work
 673         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 674         if int_op == MicrOp.OP_MTCRF.value:
 675             if spr_msb:
 676                 asmop = 'mtocrf'
 677             else:
 678                 asmop = 'mtcrf'
 679         return asmop
 680
 681     def get_spr_msb(self):
 682         dec_insn = yield self.dec2.e.do.insn
 683         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 684
 685     def call(self, name):
 686         name = name.strip()  # remove spaces if not already done so
 687         if self.halted:
 688             print("halted - not executing", name)
 689             return
 690
 691         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 692         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 693         asmop = yield from self.get_assembly_name()
 694         print("call", name, asmop)
 695
 696         # check privileged
 697         int_op = yield self.dec2.dec.op.internal_op
 698         spr_msb = yield from self.get_spr_msb()
 699
 700         instr_is_privileged = False
 701         if int_op in [MicrOp.OP_ATTN.value,
 702                       MicrOp.OP_MFMSR.value,
 703                       MicrOp.OP_MTMSR.value,
 704                       MicrOp.OP_MTMSRD.value,
 705                       # TODO: OP_TLBIE
 706                       MicrOp.OP_RFID.value]:
 707             instr_is_privileged = True
 708         if int_op in [MicrOp.OP_MFSPR.value,
 709                       MicrOp.OP_MTSPR.value] and spr_msb:
 710             instr_is_privileged = True
 711
 712         print("is priv", instr_is_privileged, hex(self.msr.value),
 713               self.msr[MSRb.PR])
 714         # check MSR priv bit and whether op is privileged: if so, throw trap
 715         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 716             self.TRAP(0x700, PIb.PRIV)
 717             self.namespace['NIA'] = self.trap_nia
 718             self.pc.update(self.namespace)
 719             return
 720
 721         # check halted condition
 722         if name == 'attn':
 723             self.halted = True
 724             return
 725
 726         # check illegal instruction
 727         illegal = False
 728         if name not in ['mtcrf', 'mtocrf']:
 729             illegal = name != asmop
 730
 731         if illegal:
 732             print("illegal", name, asmop)
 733             self.TRAP(0x700, PIb.ILLEG)
 734             self.namespace['NIA'] = self.trap_nia
 735             self.pc.update(self.namespace)
 736             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 737                   (name, asmop, self.pc.CIA.value))
 738             return
 739
 740         info = self.instrs[name]
 741         yield from self.prep_namespace(info.form, info.op_fields)
 742
 743         # preserve order of register names
 744         input_names = create_args(list(info.read_regs) +
 745                                   list(info.uninit_regs))
 746         print(input_names)
 747
 748         # main registers (RT, RA ...)
 749         inputs = []
 750         for name in input_names:
 751             regnum = yield getattr(self.decoder, name)
 752             regname = "_" + name
 753             self.namespace[regname] = regnum
 754             print('reading reg %d' % regnum)
 755             inputs.append(self.gpr(regnum))
 756
 757         # "special" registers
 758         for special in info.special_regs:
 759             if special in special_sprs:
 760                 inputs.append(self.spr[special])
 761             else:
 762                 inputs.append(self.namespace[special])
 763
 764         # clear trap (trap) NIA
 765         self.trap_nia = None
 766
 767         print(inputs)
 768         results = info.func(self, *inputs)
 769         print(results)
 770
 771         # "inject" decorator takes namespace from function locals: we need to
 772         # overwrite NIA being overwritten (sigh)
 773         if self.trap_nia is not None:
 774             self.namespace['NIA'] = self.trap_nia
 775
 776         print("after func", self.namespace['CIA'], self.namespace['NIA'])
 777
 778         # detect if CA/CA32 already in outputs (sra*, basically)
 779         already_done = 0
 780         if info.write_regs:
 781             output_names = create_args(info.write_regs)
 782             for name in output_names:
 783                 if name == 'CA':
 784                     already_done |= 1
 785                 if name == 'CA32':
 786                     already_done |= 2
 787
 788         print("carry already done?", bin(already_done))
 789         if hasattr(self.dec2.e.do, "output_carry"):
 790             carry_en = yield self.dec2.e.do.output_carry
 791         else:
 792             carry_en = False
 793         if carry_en:
 794             yield from self.handle_carry_(inputs, results, already_done)
 795
 796         # detect if overflow was in return result
 797         overflow = None
 798         if info.write_regs:
 799             for name, output in zip(output_names, results):
 800                 if name == 'overflow':
 801                     overflow = output
 802
 803         if hasattr(self.dec2.e.do, "oe"):
 804             ov_en = yield self.dec2.e.do.oe.oe
 805             ov_ok = yield self.dec2.e.do.oe.ok
 806         else:
 807             ov_en = False
 808             ov_ok = False
 809         print("internal overflow", overflow, ov_en, ov_ok)
 810         if ov_en & ov_ok:
 811             yield from self.handle_overflow(inputs, results, overflow)
 812
 813         if hasattr(self.dec2.e.do, "rc"):
 814             rc_en = yield self.dec2.e.do.rc.rc
 815         else:
 816             rc_en = False
 817         if rc_en:
 818             self.handle_comparison(results)
 819
 820         # any modified return results?
 821         if info.write_regs:
 822             for name, output in zip(output_names, results):
 823                 if name == 'overflow':  # ignore, done already (above)
 824                     continue
 825                 if isinstance(output, int):
 826                     output = SelectableInt(output, 256)
 827                 if name in ['CA', 'CA32']:
 828                     if carry_en:
 829                         print("writing %s to XER" % name, output)
 830                         self.spr['XER'][XER_bits[name]] = output.value
 831                     else:
 832                         print("NOT writing %s to XER" % name, output)
 833                 elif name in info.special_regs:
 834                     print('writing special %s' % name, output, special_sprs)
 835                     if name in special_sprs:
 836                         self.spr[name] = output
 837                     else:
 838                         self.namespace[name].eq(output)
 839                     if name == 'MSR':
 840                         print('msr written', hex(self.msr.value))
 841                 else:
 842                     regnum = yield getattr(self.decoder, name)
 843                     print('writing reg %d %s' % (regnum, str(output)))
 844                     if output.bits > 64:
 845                         output = SelectableInt(output.value, 64)
 846                     self.gpr[regnum] = output
 847
 848         print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 849         # UPDATE program counter
 850         self.pc.update(self.namespace)
 851
 852
 853 def inject():
 854     """Decorator factory.
 855
 856     this decorator will "inject" variables into the function's namespace,
 857     from the *dictionary* in self.namespace.  it therefore becomes possible
 858     to make it look like a whole stack of variables which would otherwise
 859     need "self." inserted in front of them (*and* for those variables to be
 860     added to the instance) "appear" in the function.
 861
 862     "self.namespace['SI']" for example becomes accessible as just "SI" but
 863     *only* inside the function, when decorated.
 864     """
 865     def variable_injector(func):
 866         @wraps(func)
 867         def decorator(*args, **kwargs):
 868             try:
 869                 func_globals = func.__globals__  # Python 2.6+
 870             except AttributeError:
 871                 func_globals = func.func_globals  # Earlier versions.
 872
 873             context = args[0].namespace  # variables to be injected
 874             saved_values = func_globals.copy()  # Shallow copy of dict.
 875             func_globals.update(context)
 876             result = func(*args, **kwargs)
 877             print("globals after", func_globals['CIA'], func_globals['NIA'])
 878             print("args[0]", args[0].namespace['CIA'],
 879                   args[0].namespace['NIA'])
 880             args[0].namespace = func_globals
 881             #exec (func.__code__, func_globals)
 882
 883             # finally:
 884             #    func_globals = saved_values  # Undo changes.
 885
 886             return result
 887
 888         return decorator
 889
 890     return variable_injector