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 nmigen.back.pysim import Settle
  17 from functools import wraps
  18 from copy import copy
  19 from soc.decoder.orderedset import OrderedSet
  20 from soc.decoder.selectable_int import (FieldSelectableInt, SelectableInt,
  21                                         selectconcat)
  22 from soc.decoder.power_enums import (spr_dict, spr_byname, XER_bits,
  23                                      insns, MicrOp)
  24 from soc.decoder.helpers import exts, gtu, ltu, undefined
  25 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  26
  27 from collections import namedtuple
  28 import math
  29 import sys
  30
  31 instruction_info = namedtuple('instruction_info',
  32                               'func read_regs uninit_regs write_regs ' +
  33                               'special_regs op_fields form asmregs')
  34
  35 special_sprs = {
  36     'LR': 8,
  37     'CTR': 9,
  38     'TAR': 815,
  39     'XER': 1,
  40     'VRSAVE': 256}
  41
  42
  43 def swap_order(x, nbytes):
  44     x = x.to_bytes(nbytes, byteorder='little')
  45     x = int.from_bytes(x, byteorder='big', signed=False)
  46     return x
  47
  48
  49 REG_SORT_ORDER = {
  50     # TODO (lkcl): adjust other registers that should be in a particular order
  51     # probably CA, CA32, and CR
  52     "RT": 0,
  53     "RA": 0,
  54     "RB": 0,
  55     "RS": 0,
  56     "CR": 0,
  57     "LR": 0,
  58     "CTR": 0,
  59     "TAR": 0,
  60     "CA": 0,
  61     "CA32": 0,
  62     "MSR": 0,
  63
  64     "overflow": 1,
  65 }
  66
  67
  68 def create_args(reglist, extra=None):
  69     retval = list(OrderedSet(reglist))
  70     retval.sort(key=lambda reg: REG_SORT_ORDER[reg])
  71     if extra is not None:
  72         return [extra] + retval
  73     return retval
  74
  75
  76 class Mem:
  77
  78     def __init__(self, row_bytes=8, initial_mem=None):
  79         self.mem = {}
  80         self.bytes_per_word = row_bytes
  81         self.word_log2 = math.ceil(math.log2(row_bytes))
  82         print("Sim-Mem", initial_mem, self.bytes_per_word, self.word_log2)
  83         if not initial_mem:
  84             return
  85
  86         # different types of memory data structures recognised (for convenience)
  87         if isinstance(initial_mem, list):
  88             initial_mem = (0, initial_mem)
  89         if isinstance(initial_mem, tuple):
  90             startaddr, mem = initial_mem
  91             initial_mem = {}
  92             for i, val in enumerate(mem):
  93                 initial_mem[startaddr + row_bytes*i] = (val, row_bytes)
  94
  95         for addr, (val, width) in initial_mem.items():
  96             #val = swap_order(val, width)
  97             self.st(addr, val, width, swap=False)
  98
  99     def _get_shifter_mask(self, wid, remainder):
 100         shifter = ((self.bytes_per_word - wid) - remainder) * \
 101             8  # bits per byte
 102         # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
 103         # BE/LE mode?
 104         shifter = remainder * 8
 105         mask = (1 << (wid * 8)) - 1
 106         print("width,rem,shift,mask", wid, remainder, hex(shifter), hex(mask))
 107         return shifter, mask
 108
 109     # TODO: Implement ld/st of lesser width
 110     def ld(self, address, width=8, swap=True, check_in_mem=False):
 111         print("ld from addr 0x{:x} width {:d}".format(address, width))
 112         remainder = address & (self.bytes_per_word - 1)
 113         address = address >> self.word_log2
 114         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 115         if address in self.mem:
 116             val = self.mem[address]
 117         elif check_in_mem:
 118             return None
 119         else:
 120             val = 0
 121         print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address, remainder, val))
 122
 123         if width != self.bytes_per_word:
 124             shifter, mask = self._get_shifter_mask(width, remainder)
 125             print("masking", hex(val), hex(mask << shifter), shifter)
 126             val = val & (mask << shifter)
 127             val >>= shifter
 128         if swap:
 129             val = swap_order(val, width)
 130         print("Read 0x{:x} from addr 0x{:x}".format(val, address))
 131         return val
 132
 133     def st(self, addr, v, width=8, swap=True):
 134         staddr = addr
 135         remainder = addr & (self.bytes_per_word - 1)
 136         addr = addr >> self.word_log2
 137         print("Writing 0x{:x} to ST 0x{:x} "
 138               "memaddr 0x{:x}/{:x}".format(v, staddr, addr, remainder, swap))
 139         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 140         if swap:
 141             v = swap_order(v, width)
 142         if width != self.bytes_per_word:
 143             if addr in self.mem:
 144                 val = self.mem[addr]
 145             else:
 146                 val = 0
 147             shifter, mask = self._get_shifter_mask(width, remainder)
 148             val &= ~(mask << shifter)
 149             val |= v << shifter
 150             self.mem[addr] = val
 151         else:
 152             self.mem[addr] = v
 153         print("mem @ 0x{:x}: 0x{:x}".format(addr, self.mem[addr]))
 154
 155     def __call__(self, addr, sz):
 156         val = self.ld(addr.value, sz, swap=False)
 157         print("memread", addr, sz, val)
 158         return SelectableInt(val, sz*8)
 159
 160     def memassign(self, addr, sz, val):
 161         print("memassign", addr, sz, val)
 162         self.st(addr.value, val.value, sz, swap=False)
 163
 164
 165 class GPR(dict):
 166     def __init__(self, decoder, regfile):
 167         dict.__init__(self)
 168         self.sd = decoder
 169         for i in range(32):
 170             self[i] = SelectableInt(regfile[i], 64)
 171
 172     def __call__(self, ridx):
 173         return self[ridx]
 174
 175     def set_form(self, form):
 176         self.form = form
 177
 178     def getz(self, rnum):
 179         # rnum = rnum.value # only SelectableInt allowed
 180         print("GPR getzero", rnum)
 181         if rnum == 0:
 182             return SelectableInt(0, 64)
 183         return self[rnum]
 184
 185     def _get_regnum(self, attr):
 186         getform = self.sd.sigforms[self.form]
 187         rnum = getattr(getform, attr)
 188         return rnum
 189
 190     def ___getitem__(self, attr):
 191         print("GPR getitem", attr)
 192         rnum = self._get_regnum(attr)
 193         return self.regfile[rnum]
 194
 195     def dump(self):
 196         for i in range(0, len(self), 8):
 197             s = []
 198             for j in range(8):
 199                 s.append("%08x" % self[i+j].value)
 200             s = ' '.join(s)
 201             print("reg", "%2d" % i, s)
 202
 203
 204 class PC:
 205     def __init__(self, pc_init=0):
 206         self.CIA = SelectableInt(pc_init, 64)
 207         self.NIA = self.CIA + SelectableInt(4, 64)
 208
 209     def update(self, namespace):
 210         self.CIA = namespace['NIA'].narrow(64)
 211         self.NIA = self.CIA + SelectableInt(4, 64)
 212         namespace['CIA'] = self.CIA
 213         namespace['NIA'] = self.NIA
 214
 215
 216 # Simple-V: see https://libre-soc.org/openpower/sv
 217 # also soc.sv.svstate SVSTATEREC
 218 class SVP64State:
 219     def __init__(self, init=0):
 220         self.spr = SelectableInt(init, 32)
 221         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 222         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
 223         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 224         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 225         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 226         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 227         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 228
 229
 230 # SVP64 ReMap field
 231 class SVP64RMFields:
 232     def __init__(self, init=0):
 233         self.spr = SelectableInt(init, 24)
 234         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 235         self.mmode = FieldSelectableInt(self.spr, [0])
 236         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 237         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 238         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 239         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 240         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 241         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 242
 243
 244 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 245 class SVP64PrefixFields:
 246     def __init__(self):
 247         self.insn = SelectableInt(0, 32)
 248         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 249         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 250         self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
 251         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM
 252         self.rm = FieldSelectableInt(self.insn, rmfields)
 253
 254
 255 class SPR(dict):
 256     def __init__(self, dec2, initial_sprs={}):
 257         self.sd = dec2
 258         dict.__init__(self)
 259         for key, v in initial_sprs.items():
 260             if isinstance(key, SelectableInt):
 261                 key = key.value
 262             key = special_sprs.get(key, key)
 263             if isinstance(key, int):
 264                 info = spr_dict[key]
 265             else:
 266                 info = spr_byname[key]
 267             if not isinstance(v, SelectableInt):
 268                 v = SelectableInt(v, info.length)
 269             self[key] = v
 270
 271     def __getitem__(self, key):
 272         print("get spr", key)
 273         print("dict", self.items())
 274         # if key in special_sprs get the special spr, otherwise return key
 275         if isinstance(key, SelectableInt):
 276             key = key.value
 277         if isinstance(key, int):
 278             key = spr_dict[key].SPR
 279         key = special_sprs.get(key, key)
 280         if key == 'HSRR0':  # HACK!
 281             key = 'SRR0'
 282         if key == 'HSRR1':  # HACK!
 283             key = 'SRR1'
 284         if key in self:
 285             res = dict.__getitem__(self, key)
 286         else:
 287             if isinstance(key, int):
 288                 info = spr_dict[key]
 289             else:
 290                 info = spr_byname[key]
 291             dict.__setitem__(self, key, SelectableInt(0, info.length))
 292             res = dict.__getitem__(self, key)
 293         print("spr returning", key, res)
 294         return res
 295
 296     def __setitem__(self, key, value):
 297         if isinstance(key, SelectableInt):
 298             key = key.value
 299         if isinstance(key, int):
 300             key = spr_dict[key].SPR
 301             print("spr key", key)
 302         key = special_sprs.get(key, key)
 303         if key == 'HSRR0':  # HACK!
 304             self.__setitem__('SRR0', value)
 305         if key == 'HSRR1':  # HACK!
 306             self.__setitem__('SRR1', value)
 307         print("setting spr", key, value)
 308         dict.__setitem__(self, key, value)
 309
 310     def __call__(self, ridx):
 311         return self[ridx]
 312
 313
 314 class ISACaller:
 315     # decoder2 - an instance of power_decoder2
 316     # regfile - a list of initial values for the registers
 317     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 318     # respect_pc - tracks the program counter.  requires initial_insns
 319     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 320                  initial_mem=None, initial_msr=0,
 321                  initial_svstate=0,
 322                  initial_insns=None, respect_pc=False,
 323                  disassembly=None,
 324                  initial_pc=0,
 325                  bigendian=False):
 326
 327         self.bigendian = bigendian
 328         self.halted = False
 329         self.respect_pc = respect_pc
 330         if initial_sprs is None:
 331             initial_sprs = {}
 332         if initial_mem is None:
 333             initial_mem = {}
 334         if initial_insns is None:
 335             initial_insns = {}
 336             assert self.respect_pc == False, "instructions required to honor pc"
 337
 338         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 339         print("ISACaller initial_msr", initial_msr)
 340
 341         # "fake program counter" mode (for unit testing)
 342         self.fake_pc = 0
 343         disasm_start = 0
 344         if not respect_pc:
 345             if isinstance(initial_mem, tuple):
 346                 self.fake_pc = initial_mem[0]
 347                 disasm_start = self.fake_pc
 348         else:
 349             disasm_start = initial_pc
 350
 351         # disassembly: we need this for now (not given from the decoder)
 352         self.disassembly = {}
 353         if disassembly:
 354             for i, code in enumerate(disassembly):
 355                 self.disassembly[i*4 + disasm_start] = code
 356
 357         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 358         self.gpr = GPR(decoder2, regfile)
 359         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 360         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 361         self.pc = PC()
 362         self.svstate = SVP64State(initial_svstate)
 363         self.spr = SPR(decoder2, initial_sprs)
 364         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 365
 366         # TODO, needed here:
 367         # FPR (same as GPR except for FP nums)
 368         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 369         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 370         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 371         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 372         #         -- Done
 373         # 2.3.2 LR   (actually SPR #8) -- Done
 374         # 2.3.3 CTR  (actually SPR #9) -- Done
 375         # 2.3.4 TAR  (actually SPR #815)
 376         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 377         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 378
 379         # create CR then allow portions of it to be "selectable" (below)
 380         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 381         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 382         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 383
 384         # "undefined", just set to variable-bit-width int (use exts "max")
 385         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 386
 387         self.namespace = {}
 388         self.namespace.update(self.spr)
 389         self.namespace.update({'GPR': self.gpr,
 390                                'MEM': self.mem,
 391                                'SPR': self.spr,
 392                                'memassign': self.memassign,
 393                                'NIA': self.pc.NIA,
 394                                'CIA': self.pc.CIA,
 395                                'CR': self.cr,
 396                                'MSR': self.msr,
 397                                'undefined': undefined,
 398                                'mode_is_64bit': True,
 399                                'SO': XER_bits['SO']
 400                                })
 401
 402         # update pc to requested start point
 403         self.set_pc(initial_pc)
 404
 405         # field-selectable versions of Condition Register TODO check bitranges?
 406         self.crl = []
 407         for i in range(8):
 408             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 409             _cr = FieldSelectableInt(self.cr, bits)
 410             self.crl.append(_cr)
 411             self.namespace["CR%d" % i] = _cr
 412
 413         self.decoder = decoder2.dec
 414         self.dec2 = decoder2
 415
 416     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 417         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 418         # store CIA(+4?) in SRR0, set NIA to 0x700
 419         # store MSR in SRR1, set MSR to um errr something, have to check spec
 420         self.spr['SRR0'].value = self.pc.CIA.value
 421         self.spr['SRR1'].value = self.namespace['MSR'].value
 422         self.trap_nia = SelectableInt(trap_addr, 64)
 423         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 424
 425         # set exception bits.  TODO: this should, based on the address
 426         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 427         # bits appropriately.  however it turns out that *for now* in all
 428         # cases (all trap_addrs) the exact same thing is needed.
 429         self.msr[MSRb.IR] = 0
 430         self.msr[MSRb.DR] = 0
 431         self.msr[MSRb.FE0] = 0
 432         self.msr[MSRb.FE1] = 0
 433         self.msr[MSRb.EE] = 0
 434         self.msr[MSRb.RI] = 0
 435         self.msr[MSRb.SF] = 1
 436         self.msr[MSRb.TM] = 0
 437         self.msr[MSRb.VEC] = 0
 438         self.msr[MSRb.VSX] = 0
 439         self.msr[MSRb.PR] = 0
 440         self.msr[MSRb.FP] = 0
 441         self.msr[MSRb.PMM] = 0
 442         self.msr[MSRb.TEs] = 0
 443         self.msr[MSRb.TEe] = 0
 444         self.msr[MSRb.UND] = 0
 445         self.msr[MSRb.LE] = 1
 446
 447     def memassign(self, ea, sz, val):
 448         self.mem.memassign(ea, sz, val)
 449
 450     def prep_namespace(self, formname, op_fields):
 451         # TODO: get field names from form in decoder*1* (not decoder2)
 452         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 453         # from spec
 454         # then "yield" fields only from op_fields rather than hard-coded
 455         # list, here.
 456         fields = self.decoder.sigforms[formname]
 457         for name in op_fields:
 458             if name == 'spr':
 459                 sig = getattr(fields, name.upper())
 460             else:
 461                 sig = getattr(fields, name)
 462             val = yield sig
 463             # these are all opcode fields involved in index-selection of CR,
 464             # and need to do "standard" arithmetic.  CR[BA+32] for example
 465             # would, if using SelectableInt, only be 5-bit.
 466             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 467                 self.namespace[name] = val
 468             else:
 469                 self.namespace[name] = SelectableInt(val, sig.width)
 470
 471         self.namespace['XER'] = self.spr['XER']
 472         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 473         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 474
 475     def handle_carry_(self, inputs, outputs, already_done):
 476         inv_a = yield self.dec2.e.do.invert_in
 477         if inv_a:
 478             inputs[0] = ~inputs[0]
 479
 480         imm_ok = yield self.dec2.e.do.imm_data.ok
 481         if imm_ok:
 482             imm = yield self.dec2.e.do.imm_data.data
 483             inputs.append(SelectableInt(imm, 64))
 484         assert len(outputs) >= 1
 485         print("outputs", repr(outputs))
 486         if isinstance(outputs, list) or isinstance(outputs, tuple):
 487             output = outputs[0]
 488         else:
 489             output = outputs
 490         gts = []
 491         for x in inputs:
 492             print("gt input", x, output)
 493             gt = (gtu(x, output))
 494             gts.append(gt)
 495         print(gts)
 496         cy = 1 if any(gts) else 0
 497         print("CA", cy, gts)
 498         if not (1 & already_done):
 499             self.spr['XER'][XER_bits['CA']] = cy
 500
 501         print("inputs", already_done, inputs)
 502         # 32 bit carry
 503         # ARGH... different for OP_ADD... *sigh*...
 504         op = yield self.dec2.e.do.insn_type
 505         if op == MicrOp.OP_ADD.value:
 506             res32 = (output.value & (1 << 32)) != 0
 507             a32 = (inputs[0].value & (1 << 32)) != 0
 508             if len(inputs) >= 2:
 509                 b32 = (inputs[1].value & (1 << 32)) != 0
 510             else:
 511                 b32 = False
 512             cy32 = res32 ^ a32 ^ b32
 513             print("CA32 ADD", cy32)
 514         else:
 515             gts = []
 516             for x in inputs:
 517                 print("input", x, output)
 518                 print("     x[32:64]", x, x[32:64])
 519                 print("     o[32:64]", output, output[32:64])
 520                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 521                 gts.append(gt)
 522             cy32 = 1 if any(gts) else 0
 523             print("CA32", cy32, gts)
 524         if not (2 & already_done):
 525             self.spr['XER'][XER_bits['CA32']] = cy32
 526
 527     def handle_overflow(self, inputs, outputs, div_overflow):
 528         if hasattr(self.dec2.e.do, "invert_in"):
 529             inv_a = yield self.dec2.e.do.invert_in
 530             if inv_a:
 531                 inputs[0] = ~inputs[0]
 532
 533         imm_ok = yield self.dec2.e.do.imm_data.ok
 534         if imm_ok:
 535             imm = yield self.dec2.e.do.imm_data.data
 536             inputs.append(SelectableInt(imm, 64))
 537         assert len(outputs) >= 1
 538         print("handle_overflow", inputs, outputs, div_overflow)
 539         if len(inputs) < 2 and div_overflow is None:
 540             return
 541
 542         # div overflow is different: it's returned by the pseudo-code
 543         # because it's more complex than can be done by analysing the output
 544         if div_overflow is not None:
 545             ov, ov32 = div_overflow, div_overflow
 546         # arithmetic overflow can be done by analysing the input and output
 547         elif len(inputs) >= 2:
 548             output = outputs[0]
 549
 550             # OV (64-bit)
 551             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 552             output_sgn = exts(output.value, output.bits) < 0
 553             ov = 1 if input_sgn[0] == input_sgn[1] and \
 554                 output_sgn != input_sgn[0] else 0
 555
 556             # OV (32-bit)
 557             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 558             output32_sgn = exts(output.value, 32) < 0
 559             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 560                 output32_sgn != input32_sgn[0] else 0
 561
 562         self.spr['XER'][XER_bits['OV']] = ov
 563         self.spr['XER'][XER_bits['OV32']] = ov32
 564         so = self.spr['XER'][XER_bits['SO']]
 565         so = so | ov
 566         self.spr['XER'][XER_bits['SO']] = so
 567
 568     def handle_comparison(self, outputs):
 569         out = outputs[0]
 570         assert isinstance(out, SelectableInt), \
 571             "out zero not a SelectableInt %s" % repr(outputs)
 572         print("handle_comparison", out.bits, hex(out.value))
 573         # TODO - XXX *processor* in 32-bit mode
 574         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 575         # if is_32bit:
 576         #    o32 = exts(out.value, 32)
 577         #    print ("handle_comparison exts 32 bit", hex(o32))
 578         out = exts(out.value, out.bits)
 579         print("handle_comparison exts", hex(out))
 580         zero = SelectableInt(out == 0, 1)
 581         positive = SelectableInt(out > 0, 1)
 582         negative = SelectableInt(out < 0, 1)
 583         SO = self.spr['XER'][XER_bits['SO']]
 584         print("handle_comparison SO", SO)
 585         cr_field = selectconcat(negative, positive, zero, SO)
 586         self.crl[0].eq(cr_field)
 587
 588     def set_pc(self, pc_val):
 589         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 590         self.pc.update(self.namespace)
 591
 592     def setup_one(self):
 593         """set up one instruction
 594         """
 595         if self.respect_pc:
 596             pc = self.pc.CIA.value
 597         else:
 598             pc = self.fake_pc
 599         self._pc = pc
 600         ins = self.imem.ld(pc, 4, False, True)
 601         if ins is None:
 602             raise KeyError("no instruction at 0x%x" % pc)
 603         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 604         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 605
 606         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 607         yield self.dec2.dec.bigendian.eq(self.bigendian)
 608         yield self.dec2.state.msr.eq(self.msr.value)
 609         yield self.dec2.state.pc.eq(pc)
 610
 611         # SVP64.  first, check if the opcode is EXT001
 612         yield Settle()
 613         opcode = yield self.dec2.dec.opcode_in
 614         pfx = SVP64PrefixFields()
 615         pfx.insn.value = opcode
 616         major = pfx.major.asint(msb0=True) # MSB0 inversion
 617         print ("prefix test: opcode:", major, bin(major))
 618
 619     def execute_one(self):
 620         """execute one instruction
 621         """
 622         # get the disassembly code for this instruction
 623         code = self.disassembly[self._pc]
 624         print("sim-execute", hex(self._pc), code)
 625         opname = code.split(' ')[0]
 626         yield from self.call(opname)
 627
 628         if not self.respect_pc:
 629             self.fake_pc += 4
 630         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 631
 632     def get_assembly_name(self):
 633         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 634         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 635         dec_insn = yield self.dec2.e.do.insn
 636         asmcode = yield self.dec2.dec.op.asmcode
 637         print("get assembly name asmcode", asmcode, hex(dec_insn))
 638         asmop = insns.get(asmcode, None)
 639         int_op = yield self.dec2.dec.op.internal_op
 640
 641         # sigh reconstruct the assembly instruction name
 642         if hasattr(self.dec2.e.do, "oe"):
 643             ov_en = yield self.dec2.e.do.oe.oe
 644             ov_ok = yield self.dec2.e.do.oe.ok
 645         else:
 646             ov_en = False
 647             ov_ok = False
 648         if hasattr(self.dec2.e.do, "rc"):
 649             rc_en = yield self.dec2.e.do.rc.rc
 650             rc_ok = yield self.dec2.e.do.rc.ok
 651         else:
 652             rc_en = False
 653             rc_ok = False
 654         # grrrr have to special-case MUL op (see DecodeOE)
 655         print("ov %d en %d rc %d en %d op %d" %
 656               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 657         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 658             print("mul op")
 659             if rc_en & rc_ok:
 660                 asmop += "."
 661         else:
 662             if not asmop.endswith("."):  # don't add "." to "andis."
 663                 if rc_en & rc_ok:
 664                     asmop += "."
 665         if hasattr(self.dec2.e.do, "lk"):
 666             lk = yield self.dec2.e.do.lk
 667             if lk:
 668                 asmop += "l"
 669         print("int_op", int_op)
 670         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 671             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 672             print("AA", AA)
 673             if AA:
 674                 asmop += "a"
 675         spr_msb = yield from self.get_spr_msb()
 676         if int_op == MicrOp.OP_MFCR.value:
 677             if spr_msb:
 678                 asmop = 'mfocrf'
 679             else:
 680                 asmop = 'mfcr'
 681         # XXX TODO: for whatever weird reason this doesn't work
 682         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 683         if int_op == MicrOp.OP_MTCRF.value:
 684             if spr_msb:
 685                 asmop = 'mtocrf'
 686             else:
 687                 asmop = 'mtcrf'
 688         return asmop
 689
 690     def get_spr_msb(self):
 691         dec_insn = yield self.dec2.e.do.insn
 692         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 693
 694     def call(self, name):
 695         name = name.strip()  # remove spaces if not already done so
 696         if self.halted:
 697             print("halted - not executing", name)
 698             return
 699
 700         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 701         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 702         asmop = yield from self.get_assembly_name()
 703         print("call", name, asmop)
 704
 705         # check privileged
 706         int_op = yield self.dec2.dec.op.internal_op
 707         spr_msb = yield from self.get_spr_msb()
 708
 709         instr_is_privileged = False
 710         if int_op in [MicrOp.OP_ATTN.value,
 711                       MicrOp.OP_MFMSR.value,
 712                       MicrOp.OP_MTMSR.value,
 713                       MicrOp.OP_MTMSRD.value,
 714                       # TODO: OP_TLBIE
 715                       MicrOp.OP_RFID.value]:
 716             instr_is_privileged = True
 717         if int_op in [MicrOp.OP_MFSPR.value,
 718                       MicrOp.OP_MTSPR.value] and spr_msb:
 719             instr_is_privileged = True
 720
 721         print("is priv", instr_is_privileged, hex(self.msr.value),
 722               self.msr[MSRb.PR])
 723         # check MSR priv bit and whether op is privileged: if so, throw trap
 724         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 725             self.TRAP(0x700, PIb.PRIV)
 726             self.namespace['NIA'] = self.trap_nia
 727             self.pc.update(self.namespace)
 728             return
 729
 730         # check halted condition
 731         if name == 'attn':
 732             self.halted = True
 733             return
 734
 735         # check illegal instruction
 736         illegal = False
 737         if name not in ['mtcrf', 'mtocrf']:
 738             illegal = name != asmop
 739
 740         if illegal:
 741             print("illegal", name, asmop)
 742             self.TRAP(0x700, PIb.ILLEG)
 743             self.namespace['NIA'] = self.trap_nia
 744             self.pc.update(self.namespace)
 745             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 746                   (name, asmop, self.pc.CIA.value))
 747             return
 748
 749         info = self.instrs[name]
 750         yield from self.prep_namespace(info.form, info.op_fields)
 751
 752         # preserve order of register names
 753         input_names = create_args(list(info.read_regs) +
 754                                   list(info.uninit_regs))
 755         print(input_names)
 756
 757         # main registers (RT, RA ...)
 758         inputs = []
 759         for name in input_names:
 760             regnum = yield getattr(self.decoder, name)
 761             regname = "_" + name
 762             self.namespace[regname] = regnum
 763             print('reading reg %d' % regnum)
 764             inputs.append(self.gpr(regnum))
 765
 766         # "special" registers
 767         for special in info.special_regs:
 768             if special in special_sprs:
 769                 inputs.append(self.spr[special])
 770             else:
 771                 inputs.append(self.namespace[special])
 772
 773         # clear trap (trap) NIA
 774         self.trap_nia = None
 775
 776         print(inputs)
 777         results = info.func(self, *inputs)
 778         print(results)
 779
 780         # "inject" decorator takes namespace from function locals: we need to
 781         # overwrite NIA being overwritten (sigh)
 782         if self.trap_nia is not None:
 783             self.namespace['NIA'] = self.trap_nia
 784
 785         print("after func", self.namespace['CIA'], self.namespace['NIA'])
 786
 787         # detect if CA/CA32 already in outputs (sra*, basically)
 788         already_done = 0
 789         if info.write_regs:
 790             output_names = create_args(info.write_regs)
 791             for name in output_names:
 792                 if name == 'CA':
 793                     already_done |= 1
 794                 if name == 'CA32':
 795                     already_done |= 2
 796
 797         print("carry already done?", bin(already_done))
 798         if hasattr(self.dec2.e.do, "output_carry"):
 799             carry_en = yield self.dec2.e.do.output_carry
 800         else:
 801             carry_en = False
 802         if carry_en:
 803             yield from self.handle_carry_(inputs, results, already_done)
 804
 805         # detect if overflow was in return result
 806         overflow = None
 807         if info.write_regs:
 808             for name, output in zip(output_names, results):
 809                 if name == 'overflow':
 810                     overflow = output
 811
 812         if hasattr(self.dec2.e.do, "oe"):
 813             ov_en = yield self.dec2.e.do.oe.oe
 814             ov_ok = yield self.dec2.e.do.oe.ok
 815         else:
 816             ov_en = False
 817             ov_ok = False
 818         print("internal overflow", overflow, ov_en, ov_ok)
 819         if ov_en & ov_ok:
 820             yield from self.handle_overflow(inputs, results, overflow)
 821
 822         if hasattr(self.dec2.e.do, "rc"):
 823             rc_en = yield self.dec2.e.do.rc.rc
 824         else:
 825             rc_en = False
 826         if rc_en:
 827             self.handle_comparison(results)
 828
 829         # any modified return results?
 830         if info.write_regs:
 831             for name, output in zip(output_names, results):
 832                 if name == 'overflow':  # ignore, done already (above)
 833                     continue
 834                 if isinstance(output, int):
 835                     output = SelectableInt(output, 256)
 836                 if name in ['CA', 'CA32']:
 837                     if carry_en:
 838                         print("writing %s to XER" % name, output)
 839                         self.spr['XER'][XER_bits[name]] = output.value
 840                     else:
 841                         print("NOT writing %s to XER" % name, output)
 842                 elif name in info.special_regs:
 843                     print('writing special %s' % name, output, special_sprs)
 844                     if name in special_sprs:
 845                         self.spr[name] = output
 846                     else:
 847                         self.namespace[name].eq(output)
 848                     if name == 'MSR':
 849                         print('msr written', hex(self.msr.value))
 850                 else:
 851                     regnum = yield getattr(self.decoder, name)
 852                     print('writing reg %d %s' % (regnum, str(output)))
 853                     if output.bits > 64:
 854                         output = SelectableInt(output.value, 64)
 855                     self.gpr[regnum] = output
 856
 857         print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 858         # UPDATE program counter
 859         self.pc.update(self.namespace)
 860
 861
 862 def inject():
 863     """Decorator factory.
 864
 865     this decorator will "inject" variables into the function's namespace,
 866     from the *dictionary* in self.namespace.  it therefore becomes possible
 867     to make it look like a whole stack of variables which would otherwise
 868     need "self." inserted in front of them (*and* for those variables to be
 869     added to the instance) "appear" in the function.
 870
 871     "self.namespace['SI']" for example becomes accessible as just "SI" but
 872     *only* inside the function, when decorated.
 873     """
 874     def variable_injector(func):
 875         @wraps(func)
 876         def decorator(*args, **kwargs):
 877             try:
 878                 func_globals = func.__globals__  # Python 2.6+
 879             except AttributeError:
 880                 func_globals = func.func_globals  # Earlier versions.
 881
 882             context = args[0].namespace  # variables to be injected
 883             saved_values = func_globals.copy()  # Shallow copy of dict.
 884             func_globals.update(context)
 885             result = func(*args, **kwargs)
 886             print("globals after", func_globals['CIA'], func_globals['NIA'])
 887             print("args[0]", args[0].namespace['CIA'],
 888                   args[0].namespace['NIA'])
 889             args[0].namespace = func_globals
 890             #exec (func.__code__, func_globals)
 891
 892             # finally:
 893             #    func_globals = saved_values  # Undo changes.
 894
 895             return result
 896
 897         return decorator
 898
 899     return variable_injector