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, In1Sel, In2Sel, In3Sel,
  24                                      OutSel, CROutSel)
  25 from soc.decoder.helpers import exts, gtu, ltu, undefined
  26 from soc.consts import PIb, MSRb  # big-endian (PowerISA versions)
  27 from soc.decoder.power_svp64 import SVP64RM, decode_extra
  28
  29 from collections import namedtuple
  30 import math
  31 import sys
  32
  33 instruction_info = namedtuple('instruction_info',
  34                               'func read_regs uninit_regs write_regs ' +
  35                               'special_regs op_fields form asmregs')
  36
  37 special_sprs = {
  38     'LR': 8,
  39     'CTR': 9,
  40     'TAR': 815,
  41     'XER': 1,
  42     'VRSAVE': 256}
  43
  44
  45 def swap_order(x, nbytes):
  46     x = x.to_bytes(nbytes, byteorder='little')
  47     x = int.from_bytes(x, byteorder='big', signed=False)
  48     return x
  49
  50
  51 REG_SORT_ORDER = {
  52     # TODO (lkcl): adjust other registers that should be in a particular order
  53     # probably CA, CA32, and CR
  54     "RT": 0,
  55     "RA": 0,
  56     "RB": 0,
  57     "RS": 0,
  58     "CR": 0,
  59     "LR": 0,
  60     "CTR": 0,
  61     "TAR": 0,
  62     "CA": 0,
  63     "CA32": 0,
  64     "MSR": 0,
  65
  66     "overflow": 1,
  67 }
  68
  69
  70 def create_args(reglist, extra=None):
  71     retval = list(OrderedSet(reglist))
  72     retval.sort(key=lambda reg: REG_SORT_ORDER[reg])
  73     if extra is not None:
  74         return [extra] + retval
  75     return retval
  76
  77
  78 class Mem:
  79
  80     def __init__(self, row_bytes=8, initial_mem=None):
  81         self.mem = {}
  82         self.bytes_per_word = row_bytes
  83         self.word_log2 = math.ceil(math.log2(row_bytes))
  84         print("Sim-Mem", initial_mem, self.bytes_per_word, self.word_log2)
  85         if not initial_mem:
  86             return
  87
  88         # different types of memory data structures recognised (for convenience)
  89         if isinstance(initial_mem, list):
  90             initial_mem = (0, initial_mem)
  91         if isinstance(initial_mem, tuple):
  92             startaddr, mem = initial_mem
  93             initial_mem = {}
  94             for i, val in enumerate(mem):
  95                 initial_mem[startaddr + row_bytes*i] = (val, row_bytes)
  96
  97         for addr, (val, width) in initial_mem.items():
  98             #val = swap_order(val, width)
  99             self.st(addr, val, width, swap=False)
 100
 101     def _get_shifter_mask(self, wid, remainder):
 102         shifter = ((self.bytes_per_word - wid) - remainder) * \
 103             8  # bits per byte
 104         # XXX https://bugs.libre-soc.org/show_bug.cgi?id=377
 105         # BE/LE mode?
 106         shifter = remainder * 8
 107         mask = (1 << (wid * 8)) - 1
 108         print("width,rem,shift,mask", wid, remainder, hex(shifter), hex(mask))
 109         return shifter, mask
 110
 111     # TODO: Implement ld/st of lesser width
 112     def ld(self, address, width=8, swap=True, check_in_mem=False):
 113         print("ld from addr 0x{:x} width {:d}".format(address, width))
 114         remainder = address & (self.bytes_per_word - 1)
 115         address = address >> self.word_log2
 116         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 117         if address in self.mem:
 118             val = self.mem[address]
 119         elif check_in_mem:
 120             return None
 121         else:
 122             val = 0
 123         print("mem @ 0x{:x} rem {:d} : 0x{:x}".format(address, remainder, val))
 124
 125         if width != self.bytes_per_word:
 126             shifter, mask = self._get_shifter_mask(width, remainder)
 127             print("masking", hex(val), hex(mask << shifter), shifter)
 128             val = val & (mask << shifter)
 129             val >>= shifter
 130         if swap:
 131             val = swap_order(val, width)
 132         print("Read 0x{:x} from addr 0x{:x}".format(val, address))
 133         return val
 134
 135     def st(self, addr, v, width=8, swap=True):
 136         staddr = addr
 137         remainder = addr & (self.bytes_per_word - 1)
 138         addr = addr >> self.word_log2
 139         print("Writing 0x{:x} to ST 0x{:x} "
 140               "memaddr 0x{:x}/{:x}".format(v, staddr, addr, remainder, swap))
 141         assert remainder & (width - 1) == 0, "Unaligned access unsupported!"
 142         if swap:
 143             v = swap_order(v, width)
 144         if width != self.bytes_per_word:
 145             if addr in self.mem:
 146                 val = self.mem[addr]
 147             else:
 148                 val = 0
 149             shifter, mask = self._get_shifter_mask(width, remainder)
 150             val &= ~(mask << shifter)
 151             val |= v << shifter
 152             self.mem[addr] = val
 153         else:
 154             self.mem[addr] = v
 155         print("mem @ 0x{:x}: 0x{:x}".format(addr, self.mem[addr]))
 156
 157     def __call__(self, addr, sz):
 158         val = self.ld(addr.value, sz, swap=False)
 159         print("memread", addr, sz, val)
 160         return SelectableInt(val, sz*8)
 161
 162     def memassign(self, addr, sz, val):
 163         print("memassign", addr, sz, val)
 164         self.st(addr.value, val.value, sz, swap=False)
 165
 166
 167 class GPR(dict):
 168     def __init__(self, decoder, isacaller, svstate, regfile):
 169         dict.__init__(self)
 170         self.sd = decoder
 171         self.isacaller = isacaller
 172         self.svstate = svstate
 173         for i in range(32):
 174             self[i] = SelectableInt(regfile[i], 64)
 175
 176     def __call__(self, ridx):
 177         return self[ridx]
 178
 179     def set_form(self, form):
 180         self.form = form
 181
 182     def getz(self, rnum):
 183         # rnum = rnum.value # only SelectableInt allowed
 184         print("GPR getzero", rnum)
 185         if rnum == 0:
 186             return SelectableInt(0, 64)
 187         return self[rnum]
 188
 189     def _get_regnum(self, attr):
 190         getform = self.sd.sigforms[self.form]
 191         rnum = getattr(getform, attr)
 192         return rnum
 193
 194     def ___getitem__(self, attr):
 195         """ XXX currently not used
 196         """
 197         rnum = self._get_regnum(attr)
 198         offs = self.svstate.srcstep
 199         print("GPR getitem", attr, rnum, "srcoffs", offs)
 200         return self.regfile[rnum]
 201
 202     def dump(self):
 203         for i in range(0, len(self), 8):
 204             s = []
 205             for j in range(8):
 206                 s.append("%08x" % self[i+j].value)
 207             s = ' '.join(s)
 208             print("reg", "%2d" % i, s)
 209
 210
 211 class PC:
 212     def __init__(self, pc_init=0):
 213         self.CIA = SelectableInt(pc_init, 64)
 214         self.NIA = self.CIA + SelectableInt(4, 64) # only true for v3.0B!
 215
 216     def update_nia(self, is_svp64):
 217         increment = 8 if is_svp64 else 4
 218         self.NIA = self.CIA + SelectableInt(increment, 64)
 219
 220     def update(self, namespace, is_svp64):
 221         """updates the program counter (PC) by 4 if v3.0B mode or 8 if SVP64
 222         """
 223         self.CIA = namespace['NIA'].narrow(64)
 224         self.update_nia(is_svp64)
 225         namespace['CIA'] = self.CIA
 226         namespace['NIA'] = self.NIA
 227
 228
 229 # Simple-V: see https://libre-soc.org/openpower/sv
 230 class SVP64State:
 231     def __init__(self, init=0):
 232         self.spr = SelectableInt(init, 32)
 233         # fields of SVSTATE, see https://libre-soc.org/openpower/sv/sprs/
 234         self.maxvl = FieldSelectableInt(self.spr, tuple(range(0,7)))
 235         self.vl = FieldSelectableInt(self.spr, tuple(range(7,14)))
 236         self.srcstep = FieldSelectableInt(self.spr, tuple(range(14,21)))
 237         self.dststep = FieldSelectableInt(self.spr, tuple(range(21,28)))
 238         self.subvl = FieldSelectableInt(self.spr, tuple(range(28,30)))
 239         self.svstep = FieldSelectableInt(self.spr, tuple(range(30,32)))
 240
 241
 242 # SVP64 ReMap field
 243 class SVP64RMFields:
 244     def __init__(self, init=0):
 245         self.spr = SelectableInt(init, 24)
 246         # SVP64 RM fields: see https://libre-soc.org/openpower/sv/svp64/
 247         self.mmode = FieldSelectableInt(self.spr, [0])
 248         self.mask = FieldSelectableInt(self.spr, tuple(range(1,4)))
 249         self.elwidth = FieldSelectableInt(self.spr, tuple(range(4,6)))
 250         self.ewsrc = FieldSelectableInt(self.spr, tuple(range(6,8)))
 251         self.subvl = FieldSelectableInt(self.spr, tuple(range(8,10)))
 252         self.extra = FieldSelectableInt(self.spr, tuple(range(10,19)))
 253         self.mode = FieldSelectableInt(self.spr, tuple(range(19,24)))
 254         # these cover the same extra field, split into parts as EXTRA2
 255         self.extra2 = list(range(4))
 256         self.extra2[0] = FieldSelectableInt(self.spr, tuple(range(10,12)))
 257         self.extra2[1] = FieldSelectableInt(self.spr, tuple(range(12,14)))
 258         self.extra2[2] = FieldSelectableInt(self.spr, tuple(range(14,16)))
 259         self.extra2[3] = FieldSelectableInt(self.spr, tuple(range(16,18)))
 260         self.smask = FieldSelectableInt(self.spr, tuple(range(16,19)))
 261         # and here as well, but EXTRA3
 262         self.extra3 = list(range(3))
 263         self.extra3[0] = FieldSelectableInt(self.spr, tuple(range(10,13)))
 264         self.extra3[1] = FieldSelectableInt(self.spr, tuple(range(13,16)))
 265         self.extra3[2] = FieldSelectableInt(self.spr, tuple(range(16,19)))
 266
 267
 268 SVP64RM_MMODE_SIZE = len(SVP64RMFields().mmode.br)
 269 SVP64RM_MASK_SIZE = len(SVP64RMFields().mask.br)
 270 SVP64RM_ELWIDTH_SIZE = len(SVP64RMFields().elwidth.br)
 271 SVP64RM_EWSRC_SIZE = len(SVP64RMFields().ewsrc.br)
 272 SVP64RM_SUBVL_SIZE = len(SVP64RMFields().subvl.br)
 273 SVP64RM_EXTRA2_SPEC_SIZE = len(SVP64RMFields().extra2[0].br)
 274 SVP64RM_EXTRA3_SPEC_SIZE = len(SVP64RMFields().extra3[0].br)
 275 SVP64RM_SMASK_SIZE = len(SVP64RMFields().smask.br)
 276 SVP64RM_MODE_SIZE = len(SVP64RMFields().mode.br)
 277
 278
 279 # SVP64 Prefix fields: see https://libre-soc.org/openpower/sv/svp64/
 280 class SVP64PrefixFields:
 281     def __init__(self):
 282         self.insn = SelectableInt(0, 32)
 283         # 6 bit major opcode EXT001, 2 bits "identifying" (7, 9), 24 SV ReMap
 284         self.major = FieldSelectableInt(self.insn, tuple(range(0,6)))
 285         self.pid = FieldSelectableInt(self.insn, (7, 9)) # must be 0b11
 286         rmfields = [6, 8] + list(range(10,32)) # SVP64 24-bit RM (ReMap)
 287         self.rm = FieldSelectableInt(self.insn, rmfields)
 288
 289
 290 SV64P_MAJOR_SIZE = len(SVP64PrefixFields().major.br)
 291 SV64P_PID_SIZE = len(SVP64PrefixFields().pid.br)
 292 SV64P_RM_SIZE = len(SVP64PrefixFields().rm.br)
 293
 294
 295 class SPR(dict):
 296     def __init__(self, dec2, initial_sprs={}):
 297         self.sd = dec2
 298         dict.__init__(self)
 299         for key, v in initial_sprs.items():
 300             if isinstance(key, SelectableInt):
 301                 key = key.value
 302             key = special_sprs.get(key, key)
 303             if isinstance(key, int):
 304                 info = spr_dict[key]
 305             else:
 306                 info = spr_byname[key]
 307             if not isinstance(v, SelectableInt):
 308                 v = SelectableInt(v, info.length)
 309             self[key] = v
 310
 311     def __getitem__(self, key):
 312         print("get spr", key)
 313         print("dict", self.items())
 314         # if key in special_sprs get the special spr, otherwise return key
 315         if isinstance(key, SelectableInt):
 316             key = key.value
 317         if isinstance(key, int):
 318             key = spr_dict[key].SPR
 319         key = special_sprs.get(key, key)
 320         if key == 'HSRR0':  # HACK!
 321             key = 'SRR0'
 322         if key == 'HSRR1':  # HACK!
 323             key = 'SRR1'
 324         if key in self:
 325             res = dict.__getitem__(self, key)
 326         else:
 327             if isinstance(key, int):
 328                 info = spr_dict[key]
 329             else:
 330                 info = spr_byname[key]
 331             dict.__setitem__(self, key, SelectableInt(0, info.length))
 332             res = dict.__getitem__(self, key)
 333         print("spr returning", key, res)
 334         return res
 335
 336     def __setitem__(self, key, value):
 337         if isinstance(key, SelectableInt):
 338             key = key.value
 339         if isinstance(key, int):
 340             key = spr_dict[key].SPR
 341             print("spr key", key)
 342         key = special_sprs.get(key, key)
 343         if key == 'HSRR0':  # HACK!
 344             self.__setitem__('SRR0', value)
 345         if key == 'HSRR1':  # HACK!
 346             self.__setitem__('SRR1', value)
 347         print("setting spr", key, value)
 348         dict.__setitem__(self, key, value)
 349
 350     def __call__(self, ridx):
 351         return self[ridx]
 352
 353 def get_pdecode_idx_in(dec2, name):
 354     op = dec2.dec.op
 355     in1_sel = yield op.in1_sel
 356     in2_sel = yield op.in2_sel
 357     in3_sel = yield op.in3_sel
 358     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 359     in1 = yield dec2.e.read_reg1.data
 360     in2 = yield dec2.e.read_reg2.data
 361     in3 = yield dec2.e.read_reg3.data
 362     in1_isvec = yield dec2.in1_isvec
 363     in2_isvec = yield dec2.in2_isvec
 364     in3_isvec = yield dec2.in3_isvec
 365     print ("get_pdecode_idx", in1_sel, In1Sel.RA.value, in1, in1_isvec)
 366     # identify which regnames map to in1/2/3
 367     if name == 'RA':
 368         if (in1_sel == In1Sel.RA.value or
 369             (in1_sel == In1Sel.RA_OR_ZERO.value and in1 != 0)):
 370             return in1, in1_isvec
 371         if in1_sel == In1Sel.RA_OR_ZERO.value:
 372             return in1, in1_isvec
 373     elif name == 'RB':
 374         if in2_sel == In2Sel.RB.value:
 375             return in2, in2_isvec
 376         if in3_sel == In3Sel.RB.value:
 377             return in3, in3_isvec
 378     # XXX TODO, RC doesn't exist yet!
 379     elif name == 'RC':
 380         assert False, "RC does not exist yet"
 381     elif name == 'RS':
 382         if in1_sel == In1Sel.RS.value:
 383             return in1, in1_isvec
 384         if in2_sel == In2Sel.RS.value:
 385             return in2, in2_isvec
 386         if in3_sel == In3Sel.RS.value:
 387             return in3, in3_isvec
 388     return None, False
 389
 390
 391 def get_pdecode_cr_out(dec2, name):
 392     op = dec2.dec.op
 393     out_sel = yield op.cr_out
 394     out_bitfield = yield dec2.dec_cr_out.cr_bitfield.data
 395     sv_cr_out = yield op.sv_cr_out
 396     spec = yield dec2.crout_svdec.spec
 397     sv_override = yield dec2.dec_cr_out.sv_override
 398     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 399     out = yield dec2.e.write_cr.data
 400     o_isvec = yield dec2.o_isvec
 401     print ("get_pdecode_cr_out", out_sel, CROutSel.CR0.value, out, o_isvec)
 402     print ("    sv_cr_out", sv_cr_out)
 403     print ("    cr_bf", out_bitfield)
 404     print ("    spec", spec)
 405     print ("    override", sv_override)
 406     # identify which regnames map to out / o2
 407     if name == 'CR0':
 408         if out_sel == CROutSel.CR0.value:
 409             return out, o_isvec
 410     print ("get_pdecode_idx_out not found", name)
 411     return None, False
 412
 413
 414 def get_pdecode_idx_out(dec2, name):
 415     op = dec2.dec.op
 416     out_sel = yield op.out_sel
 417     # get the IN1/2/3 from the decoder (includes SVP64 remap and isvec)
 418     out = yield dec2.e.write_reg.data
 419     o_isvec = yield dec2.o_isvec
 420     print ("get_pdecode_idx_out", out_sel, OutSel.RA.value, out, o_isvec)
 421     # identify which regnames map to out / o2
 422     if name == 'RA':
 423         if out_sel == OutSel.RA.value:
 424             return out, o_isvec
 425     elif name == 'RT':
 426         if out_sel == OutSel.RT.value:
 427             return out, o_isvec
 428     print ("get_pdecode_idx_out not found", name)
 429     return None, False
 430
 431
 432 # XXX TODO
 433 def get_pdecode_idx_out2(dec2, name):
 434     op = dec2.dec.op
 435     print ("TODO: get_pdecode_idx_out2", name)
 436     return None, False
 437
 438
 439 class ISACaller:
 440     # decoder2 - an instance of power_decoder2
 441     # regfile - a list of initial values for the registers
 442     # initial_{etc} - initial values for SPRs, Condition Register, Mem, MSR
 443     # respect_pc - tracks the program counter.  requires initial_insns
 444     def __init__(self, decoder2, regfile, initial_sprs=None, initial_cr=0,
 445                  initial_mem=None, initial_msr=0,
 446                  initial_svstate=0,
 447                  initial_insns=None, respect_pc=False,
 448                  disassembly=None,
 449                  initial_pc=0,
 450                  bigendian=False):
 451
 452         self.bigendian = bigendian
 453         self.halted = False
 454         self.is_svp64_mode = False
 455         self.respect_pc = respect_pc
 456         if initial_sprs is None:
 457             initial_sprs = {}
 458         if initial_mem is None:
 459             initial_mem = {}
 460         if initial_insns is None:
 461             initial_insns = {}
 462             assert self.respect_pc == False, "instructions required to honor pc"
 463
 464         print("ISACaller insns", respect_pc, initial_insns, disassembly)
 465         print("ISACaller initial_msr", initial_msr)
 466
 467         # "fake program counter" mode (for unit testing)
 468         self.fake_pc = 0
 469         disasm_start = 0
 470         if not respect_pc:
 471             if isinstance(initial_mem, tuple):
 472                 self.fake_pc = initial_mem[0]
 473                 disasm_start = self.fake_pc
 474         else:
 475             disasm_start = initial_pc
 476
 477         # disassembly: we need this for now (not given from the decoder)
 478         self.disassembly = {}
 479         if disassembly:
 480             for i, code in enumerate(disassembly):
 481                 self.disassembly[i*4 + disasm_start] = code
 482
 483         # set up registers, instruction memory, data memory, PC, SPRs, MSR
 484         self.svp64rm = SVP64RM()
 485         if isinstance(initial_svstate, int):
 486             initial_svstate = SVP64State(initial_svstate)
 487         self.svstate = initial_svstate
 488         self.gpr = GPR(decoder2, self, self.svstate, regfile)
 489         self.mem = Mem(row_bytes=8, initial_mem=initial_mem)
 490         self.imem = Mem(row_bytes=4, initial_mem=initial_insns)
 491         self.pc = PC()
 492         self.spr = SPR(decoder2, initial_sprs)
 493         self.msr = SelectableInt(initial_msr, 64)  # underlying reg
 494
 495         # TODO, needed here:
 496         # FPR (same as GPR except for FP nums)
 497         # 4.2.2 p124 FPSCR (definitely "separate" - not in SPR)
 498         #            note that mffs, mcrfs, mtfsf "manage" this FPSCR
 499         # 2.3.1 CR (and sub-fields CR0..CR6 - CR0 SO comes from XER.SO)
 500         #         note that mfocrf, mfcr, mtcr, mtocrf, mcrxrx "manage" CRs
 501         #         -- Done
 502         # 2.3.2 LR   (actually SPR #8) -- Done
 503         # 2.3.3 CTR  (actually SPR #9) -- Done
 504         # 2.3.4 TAR  (actually SPR #815)
 505         # 3.2.2 p45 XER  (actually SPR #1) -- Done
 506         # 3.2.3 p46 p232 VRSAVE (actually SPR #256)
 507
 508         # create CR then allow portions of it to be "selectable" (below)
 509         #rev_cr = int('{:016b}'.format(initial_cr)[::-1], 2)
 510         self.cr = SelectableInt(initial_cr, 64)  # underlying reg
 511         #self.cr = FieldSelectableInt(self._cr, list(range(32, 64)))
 512
 513         # "undefined", just set to variable-bit-width int (use exts "max")
 514         #self.undefined = SelectableInt(0, 256)  # TODO, not hard-code 256!
 515
 516         self.namespace = {}
 517         self.namespace.update(self.spr)
 518         self.namespace.update({'GPR': self.gpr,
 519                                'MEM': self.mem,
 520                                'SPR': self.spr,
 521                                'memassign': self.memassign,
 522                                'NIA': self.pc.NIA,
 523                                'CIA': self.pc.CIA,
 524                                'CR': self.cr,
 525                                'MSR': self.msr,
 526                                'undefined': undefined,
 527                                'mode_is_64bit': True,
 528                                'SO': XER_bits['SO']
 529                                })
 530
 531         # update pc to requested start point
 532         self.set_pc(initial_pc)
 533
 534         # field-selectable versions of Condition Register TODO check bitranges?
 535         self.crl = []
 536         for i in range(8):
 537             bits = tuple(range(i*4+32, (i+1)*4+32))  # errr... maybe?
 538             _cr = FieldSelectableInt(self.cr, bits)
 539             self.crl.append(_cr)
 540             self.namespace["CR%d" % i] = _cr
 541
 542         self.decoder = decoder2.dec
 543         self.dec2 = decoder2
 544
 545     def TRAP(self, trap_addr=0x700, trap_bit=PIb.TRAP):
 546         print("TRAP:", hex(trap_addr), hex(self.namespace['MSR'].value))
 547         # store CIA(+4?) in SRR0, set NIA to 0x700
 548         # store MSR in SRR1, set MSR to um errr something, have to check spec
 549         self.spr['SRR0'].value = self.pc.CIA.value
 550         self.spr['SRR1'].value = self.namespace['MSR'].value
 551         self.trap_nia = SelectableInt(trap_addr, 64)
 552         self.spr['SRR1'][trap_bit] = 1  # change *copy* of MSR in SRR1
 553
 554         # set exception bits.  TODO: this should, based on the address
 555         # in figure 66 p1065 V3.0B and the table figure 65 p1063 set these
 556         # bits appropriately.  however it turns out that *for now* in all
 557         # cases (all trap_addrs) the exact same thing is needed.
 558         self.msr[MSRb.IR] = 0
 559         self.msr[MSRb.DR] = 0
 560         self.msr[MSRb.FE0] = 0
 561         self.msr[MSRb.FE1] = 0
 562         self.msr[MSRb.EE] = 0
 563         self.msr[MSRb.RI] = 0
 564         self.msr[MSRb.SF] = 1
 565         self.msr[MSRb.TM] = 0
 566         self.msr[MSRb.VEC] = 0
 567         self.msr[MSRb.VSX] = 0
 568         self.msr[MSRb.PR] = 0
 569         self.msr[MSRb.FP] = 0
 570         self.msr[MSRb.PMM] = 0
 571         self.msr[MSRb.TEs] = 0
 572         self.msr[MSRb.TEe] = 0
 573         self.msr[MSRb.UND] = 0
 574         self.msr[MSRb.LE] = 1
 575
 576     def memassign(self, ea, sz, val):
 577         self.mem.memassign(ea, sz, val)
 578
 579     def prep_namespace(self, formname, op_fields):
 580         # TODO: get field names from form in decoder*1* (not decoder2)
 581         # decoder2 is hand-created, and decoder1.sigform is auto-generated
 582         # from spec
 583         # then "yield" fields only from op_fields rather than hard-coded
 584         # list, here.
 585         fields = self.decoder.sigforms[formname]
 586         for name in op_fields:
 587             if name == 'spr':
 588                 sig = getattr(fields, name.upper())
 589             else:
 590                 sig = getattr(fields, name)
 591             val = yield sig
 592             # these are all opcode fields involved in index-selection of CR,
 593             # and need to do "standard" arithmetic.  CR[BA+32] for example
 594             # would, if using SelectableInt, only be 5-bit.
 595             if name in ['BF', 'BFA', 'BC', 'BA', 'BB', 'BT', 'BI']:
 596                 self.namespace[name] = val
 597             else:
 598                 self.namespace[name] = SelectableInt(val, sig.width)
 599
 600         self.namespace['XER'] = self.spr['XER']
 601         self.namespace['CA'] = self.spr['XER'][XER_bits['CA']].value
 602         self.namespace['CA32'] = self.spr['XER'][XER_bits['CA32']].value
 603
 604     def handle_carry_(self, inputs, outputs, already_done):
 605         inv_a = yield self.dec2.e.do.invert_in
 606         if inv_a:
 607             inputs[0] = ~inputs[0]
 608
 609         imm_ok = yield self.dec2.e.do.imm_data.ok
 610         if imm_ok:
 611             imm = yield self.dec2.e.do.imm_data.data
 612             inputs.append(SelectableInt(imm, 64))
 613         assert len(outputs) >= 1
 614         print("outputs", repr(outputs))
 615         if isinstance(outputs, list) or isinstance(outputs, tuple):
 616             output = outputs[0]
 617         else:
 618             output = outputs
 619         gts = []
 620         for x in inputs:
 621             print("gt input", x, output)
 622             gt = (gtu(x, output))
 623             gts.append(gt)
 624         print(gts)
 625         cy = 1 if any(gts) else 0
 626         print("CA", cy, gts)
 627         if not (1 & already_done):
 628             self.spr['XER'][XER_bits['CA']] = cy
 629
 630         print("inputs", already_done, inputs)
 631         # 32 bit carry
 632         # ARGH... different for OP_ADD... *sigh*...
 633         op = yield self.dec2.e.do.insn_type
 634         if op == MicrOp.OP_ADD.value:
 635             res32 = (output.value & (1 << 32)) != 0
 636             a32 = (inputs[0].value & (1 << 32)) != 0
 637             if len(inputs) >= 2:
 638                 b32 = (inputs[1].value & (1 << 32)) != 0
 639             else:
 640                 b32 = False
 641             cy32 = res32 ^ a32 ^ b32
 642             print("CA32 ADD", cy32)
 643         else:
 644             gts = []
 645             for x in inputs:
 646                 print("input", x, output)
 647                 print("     x[32:64]", x, x[32:64])
 648                 print("     o[32:64]", output, output[32:64])
 649                 gt = (gtu(x[32:64], output[32:64])) == SelectableInt(1, 1)
 650                 gts.append(gt)
 651             cy32 = 1 if any(gts) else 0
 652             print("CA32", cy32, gts)
 653         if not (2 & already_done):
 654             self.spr['XER'][XER_bits['CA32']] = cy32
 655
 656     def handle_overflow(self, inputs, outputs, div_overflow):
 657         if hasattr(self.dec2.e.do, "invert_in"):
 658             inv_a = yield self.dec2.e.do.invert_in
 659             if inv_a:
 660                 inputs[0] = ~inputs[0]
 661
 662         imm_ok = yield self.dec2.e.do.imm_data.ok
 663         if imm_ok:
 664             imm = yield self.dec2.e.do.imm_data.data
 665             inputs.append(SelectableInt(imm, 64))
 666         assert len(outputs) >= 1
 667         print("handle_overflow", inputs, outputs, div_overflow)
 668         if len(inputs) < 2 and div_overflow is None:
 669             return
 670
 671         # div overflow is different: it's returned by the pseudo-code
 672         # because it's more complex than can be done by analysing the output
 673         if div_overflow is not None:
 674             ov, ov32 = div_overflow, div_overflow
 675         # arithmetic overflow can be done by analysing the input and output
 676         elif len(inputs) >= 2:
 677             output = outputs[0]
 678
 679             # OV (64-bit)
 680             input_sgn = [exts(x.value, x.bits) < 0 for x in inputs]
 681             output_sgn = exts(output.value, output.bits) < 0
 682             ov = 1 if input_sgn[0] == input_sgn[1] and \
 683                 output_sgn != input_sgn[0] else 0
 684
 685             # OV (32-bit)
 686             input32_sgn = [exts(x.value, 32) < 0 for x in inputs]
 687             output32_sgn = exts(output.value, 32) < 0
 688             ov32 = 1 if input32_sgn[0] == input32_sgn[1] and \
 689                 output32_sgn != input32_sgn[0] else 0
 690
 691         self.spr['XER'][XER_bits['OV']] = ov
 692         self.spr['XER'][XER_bits['OV32']] = ov32
 693         so = self.spr['XER'][XER_bits['SO']]
 694         so = so | ov
 695         self.spr['XER'][XER_bits['SO']] = so
 696
 697     def handle_comparison(self, outputs, cr_idx=0):
 698         out = outputs[0]
 699         assert isinstance(out, SelectableInt), \
 700             "out zero not a SelectableInt %s" % repr(outputs)
 701         print("handle_comparison", out.bits, hex(out.value))
 702         # TODO - XXX *processor* in 32-bit mode
 703         # https://bugs.libre-soc.org/show_bug.cgi?id=424
 704         # if is_32bit:
 705         #    o32 = exts(out.value, 32)
 706         #    print ("handle_comparison exts 32 bit", hex(o32))
 707         out = exts(out.value, out.bits)
 708         print("handle_comparison exts", hex(out))
 709         zero = SelectableInt(out == 0, 1)
 710         positive = SelectableInt(out > 0, 1)
 711         negative = SelectableInt(out < 0, 1)
 712         SO = self.spr['XER'][XER_bits['SO']]
 713         print("handle_comparison SO", SO)
 714         cr_field = selectconcat(negative, positive, zero, SO)
 715         self.crl[cr_idx].eq(cr_field)
 716
 717     def set_pc(self, pc_val):
 718         self.namespace['NIA'] = SelectableInt(pc_val, 64)
 719         self.pc.update(self.namespace, self.is_svp64_mode)
 720
 721     def setup_one(self):
 722         """set up one instruction
 723         """
 724         if self.respect_pc:
 725             pc = self.pc.CIA.value
 726         else:
 727             pc = self.fake_pc
 728         self._pc = pc
 729         ins = self.imem.ld(pc, 4, False, True)
 730         if ins is None:
 731             raise KeyError("no instruction at 0x%x" % pc)
 732         print("setup: 0x%x 0x%x %s" % (pc, ins & 0xffffffff, bin(ins)))
 733         print("CIA NIA", self.respect_pc, self.pc.CIA.value, self.pc.NIA.value)
 734
 735         yield self.dec2.sv_rm.eq(0)
 736         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff)
 737         yield self.dec2.dec.bigendian.eq(self.bigendian)
 738         yield self.dec2.state.msr.eq(self.msr.value)
 739         yield self.dec2.state.pc.eq(pc)
 740         yield self.dec2.state.svstate.eq(self.svstate.spr.value)
 741
 742         # SVP64.  first, check if the opcode is EXT001, and SVP64 id bits set
 743         yield Settle()
 744         opcode = yield self.dec2.dec.opcode_in
 745         pfx = SVP64PrefixFields() # TODO should probably use SVP64PrefixDecoder
 746         pfx.insn.value = opcode
 747         major = pfx.major.asint(msb0=True) # MSB0 inversion
 748         print ("prefix test: opcode:", major, bin(major),
 749                 pfx.insn[7] == 0b1, pfx.insn[9] == 0b1)
 750         self.is_svp64_mode = ((major == 0b000001) and
 751                               pfx.insn[7].value == 0b1 and
 752                               pfx.insn[9].value == 0b1)
 753         self.pc.update_nia(self.is_svp64_mode)
 754         if not self.is_svp64_mode:
 755             return
 756
 757         # in SVP64 mode.  decode/print out svp64 prefix, get v3.0B instruction
 758         print ("svp64.rm", bin(pfx.rm.asint(msb0=True)))
 759         print ("    svstate.vl", self.svstate.vl.asint(msb0=True))
 760         print ("    svstate.mvl", self.svstate.maxvl.asint(msb0=True))
 761         sv_rm = pfx.rm.asint(msb0=True)
 762         ins = self.imem.ld(pc+4, 4, False, True)
 763         print("     svsetup: 0x%x 0x%x %s" % (pc+4, ins & 0xffffffff, bin(ins)))
 764         yield self.dec2.dec.raw_opcode_in.eq(ins & 0xffffffff) # v3.0B suffix
 765         yield self.dec2.sv_rm.eq(sv_rm)                        # svp64 prefix
 766         yield Settle()
 767
 768     def execute_one(self):
 769         """execute one instruction
 770         """
 771         # get the disassembly code for this instruction
 772         if self.is_svp64_mode:
 773             code = self.disassembly[self._pc+4]
 774             print("    svp64 sim-execute", hex(self._pc), code)
 775         else:
 776             code = self.disassembly[self._pc]
 777             print("sim-execute", hex(self._pc), code)
 778         opname = code.split(' ')[0]
 779         yield from self.call(opname)
 780
 781         # don't use this except in special circumstances
 782         if not self.respect_pc:
 783             self.fake_pc += 4
 784
 785         print("execute one, CIA NIA", self.pc.CIA.value, self.pc.NIA.value)
 786
 787     def get_assembly_name(self):
 788         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 789         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 790         dec_insn = yield self.dec2.e.do.insn
 791         asmcode = yield self.dec2.dec.op.asmcode
 792         print("get assembly name asmcode", asmcode, hex(dec_insn))
 793         asmop = insns.get(asmcode, None)
 794         int_op = yield self.dec2.dec.op.internal_op
 795
 796         # sigh reconstruct the assembly instruction name
 797         if hasattr(self.dec2.e.do, "oe"):
 798             ov_en = yield self.dec2.e.do.oe.oe
 799             ov_ok = yield self.dec2.e.do.oe.ok
 800         else:
 801             ov_en = False
 802             ov_ok = False
 803         if hasattr(self.dec2.e.do, "rc"):
 804             rc_en = yield self.dec2.e.do.rc.rc
 805             rc_ok = yield self.dec2.e.do.rc.ok
 806         else:
 807             rc_en = False
 808             rc_ok = False
 809         # grrrr have to special-case MUL op (see DecodeOE)
 810         print("ov %d en %d rc %d en %d op %d" %
 811               (ov_ok, ov_en, rc_ok, rc_en, int_op))
 812         if int_op in [MicrOp.OP_MUL_H64.value, MicrOp.OP_MUL_H32.value]:
 813             print("mul op")
 814             if rc_en & rc_ok:
 815                 asmop += "."
 816         else:
 817             if not asmop.endswith("."):  # don't add "." to "andis."
 818                 if rc_en & rc_ok:
 819                     asmop += "."
 820         if hasattr(self.dec2.e.do, "lk"):
 821             lk = yield self.dec2.e.do.lk
 822             if lk:
 823                 asmop += "l"
 824         print("int_op", int_op)
 825         if int_op in [MicrOp.OP_B.value, MicrOp.OP_BC.value]:
 826             AA = yield self.dec2.dec.fields.FormI.AA[0:-1]
 827             print("AA", AA)
 828             if AA:
 829                 asmop += "a"
 830         spr_msb = yield from self.get_spr_msb()
 831         if int_op == MicrOp.OP_MFCR.value:
 832             if spr_msb:
 833                 asmop = 'mfocrf'
 834             else:
 835                 asmop = 'mfcr'
 836         # XXX TODO: for whatever weird reason this doesn't work
 837         # https://bugs.libre-soc.org/show_bug.cgi?id=390
 838         if int_op == MicrOp.OP_MTCRF.value:
 839             if spr_msb:
 840                 asmop = 'mtocrf'
 841             else:
 842                 asmop = 'mtcrf'
 843         return asmop
 844
 845     def get_spr_msb(self):
 846         dec_insn = yield self.dec2.e.do.insn
 847         return dec_insn & (1 << 20) != 0  # sigh - XFF.spr[-1]?
 848
 849     def call(self, name):
 850         """call(opcode) - the primary execution point for instructions
 851         """
 852         name = name.strip()  # remove spaces if not already done so
 853         if self.halted:
 854             print("halted - not executing", name)
 855             return
 856
 857         # TODO, asmregs is from the spec, e.g. add RT,RA,RB
 858         # see http://bugs.libre-riscv.org/show_bug.cgi?id=282
 859         asmop = yield from self.get_assembly_name()
 860         print("call", name, asmop)
 861
 862         # check privileged
 863         int_op = yield self.dec2.dec.op.internal_op
 864         spr_msb = yield from self.get_spr_msb()
 865
 866         instr_is_privileged = False
 867         if int_op in [MicrOp.OP_ATTN.value,
 868                       MicrOp.OP_MFMSR.value,
 869                       MicrOp.OP_MTMSR.value,
 870                       MicrOp.OP_MTMSRD.value,
 871                       # TODO: OP_TLBIE
 872                       MicrOp.OP_RFID.value]:
 873             instr_is_privileged = True
 874         if int_op in [MicrOp.OP_MFSPR.value,
 875                       MicrOp.OP_MTSPR.value] and spr_msb:
 876             instr_is_privileged = True
 877
 878         print("is priv", instr_is_privileged, hex(self.msr.value),
 879               self.msr[MSRb.PR])
 880         # check MSR priv bit and whether op is privileged: if so, throw trap
 881         if instr_is_privileged and self.msr[MSRb.PR] == 1:
 882             self.TRAP(0x700, PIb.PRIV)
 883             self.namespace['NIA'] = self.trap_nia
 884             self.pc.update(self.namespace, self.is_svp64_mode)
 885             return
 886
 887         # check halted condition
 888         if name == 'attn':
 889             self.halted = True
 890             return
 891
 892         # check illegal instruction
 893         illegal = False
 894         if name not in ['mtcrf', 'mtocrf']:
 895             illegal = name != asmop
 896
 897         if illegal:
 898             print("illegal", name, asmop)
 899             self.TRAP(0x700, PIb.ILLEG)
 900             self.namespace['NIA'] = self.trap_nia
 901             self.pc.update(self.namespace, self.is_svp64_mode)
 902             print("name %s != %s - calling ILLEGAL trap, PC: %x" %
 903                   (name, asmop, self.pc.CIA.value))
 904             return
 905
 906         info = self.instrs[name]
 907         yield from self.prep_namespace(info.form, info.op_fields)
 908
 909         # preserve order of register names
 910         input_names = create_args(list(info.read_regs) +
 911                                   list(info.uninit_regs))
 912         print(input_names)
 913
 914         # get SVP64 entry for the current instruction
 915         sv_rm = self.svp64rm.instrs.get(name)
 916         if sv_rm is not None:
 917             dest_cr, src_cr, src_byname, dest_byname = decode_extra(sv_rm)
 918         else:
 919             dest_cr, src_cr, src_byname, dest_byname = False, False, {}, {}
 920         print ("sv rm", sv_rm, dest_cr, src_cr, src_byname, dest_byname)
 921
 922         # get SVSTATE srcstep.  TODO: dststep (twin predication)
 923         srcstep = self.svstate.srcstep.asint(msb0=True)
 924         vl = self.svstate.vl.asint(msb0=True)
 925         mvl = self.svstate.maxvl.asint(msb0=True)
 926
 927         # VL=0 in SVP64 mode means "do nothing: skip instruction"
 928         if self.is_svp64_mode and vl == 0:
 929             self.pc.update(self.namespace, self.is_svp64_mode)
 930             print("end of call", self.namespace['CIA'], self.namespace['NIA'])
 931             return
 932
 933         # main input registers (RT, RA ...)
 934         inputs = []
 935         for name in input_names:
 936             # using PowerDecoder2, first, find the decoder index.
 937             # (mapping name RA RB RC RS to in1, in2, in3)
 938             regnum, is_vec = yield from get_pdecode_idx_in(self.dec2, name)
 939             if regnum is None:
 940                 # doing this is not part of svp64, it's because output
 941                 # registers, to be modified, need to be in the namespace.
 942                 regnum, is_vec = yield from get_pdecode_idx_out(self.dec2, name)
 943             # here's where we go "vector".  TODO: zero-testing (RA_IS_ZERO)
 944             # XXX already done by PowerDecoder2, now
 945             #if is_vec:
 946             #   regnum += srcstep # TODO, elwidth overrides
 947
 948             # in case getting the register number is needed, _RA, _RB
 949             regname = "_" + name
 950             self.namespace[regname] = regnum
 951             print('reading reg %s %d' % (name, regnum), is_vec)
 952             reg_val = self.gpr(regnum)
 953             inputs.append(reg_val)
 954
 955         # "special" registers
 956         for special in info.special_regs:
 957             if special in special_sprs:
 958                 inputs.append(self.spr[special])
 959             else:
 960                 inputs.append(self.namespace[special])
 961
 962         # clear trap (trap) NIA
 963         self.trap_nia = None
 964
 965         print("inputs", inputs)
 966         results = info.func(self, *inputs)
 967         print("results", results)
 968
 969         # "inject" decorator takes namespace from function locals: we need to
 970         # overwrite NIA being overwritten (sigh)
 971         if self.trap_nia is not None:
 972             self.namespace['NIA'] = self.trap_nia
 973
 974         print("after func", self.namespace['CIA'], self.namespace['NIA'])
 975
 976         # detect if CA/CA32 already in outputs (sra*, basically)
 977         already_done = 0
 978         if info.write_regs:
 979             output_names = create_args(info.write_regs)
 980             for name in output_names:
 981                 if name == 'CA':
 982                     already_done |= 1
 983                 if name == 'CA32':
 984                     already_done |= 2
 985
 986         print("carry already done?", bin(already_done))
 987         if hasattr(self.dec2.e.do, "output_carry"):
 988             carry_en = yield self.dec2.e.do.output_carry
 989         else:
 990             carry_en = False
 991         if carry_en:
 992             yield from self.handle_carry_(inputs, results, already_done)
 993
 994         # detect if overflow was in return result
 995         overflow = None
 996         if info.write_regs:
 997             for name, output in zip(output_names, results):
 998                 if name == 'overflow':
 999                     overflow = output
1000
1001         if hasattr(self.dec2.e.do, "oe"):
1002             ov_en = yield self.dec2.e.do.oe.oe
1003             ov_ok = yield self.dec2.e.do.oe.ok
1004         else:
1005             ov_en = False
1006             ov_ok = False
1007         print("internal overflow", overflow, ov_en, ov_ok)
1008         if ov_en & ov_ok:
1009             yield from self.handle_overflow(inputs, results, overflow)
1010
1011         if hasattr(self.dec2.e.do, "rc"):
1012             rc_en = yield self.dec2.e.do.rc.rc
1013         else:
1014             rc_en = False
1015         if rc_en:
1016             regnum, is_vec = yield from get_pdecode_cr_out(self.dec2, "CR0")
1017             self.handle_comparison(results, regnum)
1018
1019         # any modified return results?
1020         if info.write_regs:
1021             for name, output in zip(output_names, results):
1022                 if name == 'overflow':  # ignore, done already (above)
1023                     continue
1024                 if isinstance(output, int):
1025                     output = SelectableInt(output, 256)
1026                 if name in ['CA', 'CA32']:
1027                     if carry_en:
1028                         print("writing %s to XER" % name, output)
1029                         self.spr['XER'][XER_bits[name]] = output.value
1030                     else:
1031                         print("NOT writing %s to XER" % name, output)
1032                 elif name in info.special_regs:
1033                     print('writing special %s' % name, output, special_sprs)
1034                     if name in special_sprs:
1035                         self.spr[name] = output
1036                     else:
1037                         self.namespace[name].eq(output)
1038                     if name == 'MSR':
1039                         print('msr written', hex(self.msr.value))
1040                 else:
1041                     regnum, is_vec = yield from get_pdecode_idx_out(self.dec2,
1042                                                 name)
1043                     if regnum is None:
1044                         # temporary hack for not having 2nd output
1045                         regnum = yield getattr(self.decoder, name)
1046                         is_vec = False
1047                     print('writing reg %d %s' % (regnum, str(output)), is_vec)
1048                     if output.bits > 64:
1049                         output = SelectableInt(output.value, 64)
1050                     self.gpr[regnum] = output
1051
1052         # check if it is the SVSTATE.src/dest step that needs incrementing
1053         # this is our Sub-Program-Counter loop from 0 to VL-1
1054         if self.is_svp64_mode:
1055             # XXX twin predication TODO
1056             vl = self.svstate.vl.asint(msb0=True)
1057             mvl = self.svstate.maxvl.asint(msb0=True)
1058             srcstep = self.svstate.srcstep.asint(msb0=True)
1059             print ("    svstate.vl", vl)
1060             print ("    svstate.mvl", mvl)
1061             print ("    svstate.srcstep", srcstep)
1062             # check if srcstep needs incrementing by one, stop PC advancing
1063             # svp64 loop can end early if the dest is scalar
1064             svp64_dest_vector = not (yield self.dec2.no_out_vec)
1065             if svp64_dest_vector and srcstep != vl-1:
1066                 self.svstate.srcstep += SelectableInt(1, 7)
1067                 self.pc.NIA.value = self.pc.CIA.value
1068                 self.namespace['NIA'] = self.pc.NIA
1069                 print("end of sub-pc call", self.namespace['CIA'],
1070                                      self.namespace['NIA'])
1071                 return # DO NOT allow PC to update whilst Sub-PC loop running
1072             # reset to zero
1073             self.svstate.srcstep[0:7] = 0
1074             print ("    svstate.srcstep loop end (PC to update)")
1075             self.pc.update_nia(self.is_svp64_mode)
1076             self.namespace['NIA'] = self.pc.NIA
1077
1078         # UPDATE program counter
1079         self.pc.update(self.namespace, self.is_svp64_mode)
1080         print("end of call", self.namespace['CIA'], self.namespace['NIA'])
1081
1082
1083 def inject():
1084     """Decorator factory.
1085
1086     this decorator will "inject" variables into the function's namespace,
1087     from the *dictionary* in self.namespace.  it therefore becomes possible
1088     to make it look like a whole stack of variables which would otherwise
1089     need "self." inserted in front of them (*and* for those variables to be
1090     added to the instance) "appear" in the function.
1091
1092     "self.namespace['SI']" for example becomes accessible as just "SI" but
1093     *only* inside the function, when decorated.
1094     """
1095     def variable_injector(func):
1096         @wraps(func)
1097         def decorator(*args, **kwargs):
1098             try:
1099                 func_globals = func.__globals__  # Python 2.6+
1100             except AttributeError:
1101                 func_globals = func.func_globals  # Earlier versions.
1102
1103             context = args[0].namespace  # variables to be injected
1104             saved_values = func_globals.copy()  # Shallow copy of dict.
1105             func_globals.update(context)
1106             result = func(*args, **kwargs)
1107             print("globals after", func_globals['CIA'], func_globals['NIA'])
1108             print("args[0]", args[0].namespace['CIA'],
1109                   args[0].namespace['NIA'])
1110             args[0].namespace = func_globals
1111             #exec (func.__code__, func_globals)
1112
1113             # finally:
1114             #    func_globals = saved_values  # Undo changes.
1115
1116             return result
1117
1118         return decorator
1119
1120     return variable_injector