rdos/host/asm.awk

#!/usr/bin/awk -f
BEGIN {
  if (!of) of="lst"
  pos=0
  errors=0
  # lookup table for ord()
  for (i = 32; i <= 128; i++) {
    t=sprintf("%c", i)
    _ord_[t]=i
  }
  # 8-bit general purpose registers
  r8["al"]=0
  r8["cl"]=1
  r8["dl"]=2
  r8["bl"]=3
  r8["ah"]=4
  r8["ch"]=5
  r8["dh"]=6
  r8["bh"]=7
  # 16-bit general purpose registers
  r16["ax"]=0
  r16["cx"]=1
  r16["dx"]=2
  r16["bx"]=3
  r16["sp"]=4
  r16["bp"]=5
  r16["si"]=6
  r16["di"]=7
  # segment registers
  sreg["es"]=0
  sreg["cs"]=1
  sreg["ss"]=2
  sreg["ds"]=3
  # indirect access register combinations
  modstr["bxsi"]=0
  modstr["sibx"]=0
  modstr["bxdi"]=1
  modstr["dibx"]=1
  modstr["bpsi"]=2
  modstr["sibp"]=2
  modstr["bpdi"]=3
  modstr["dibp"]=3
  modstr["si"]=4
  modstr["di"]=5
  modstr["bp"]=6
  modstr["bx"]=7
  # ALU operations
  alu["add"]=0
  alu["or"]=1
  alu["adc"]=2
  alu["sbb"]=3
  alu["and"]=4
  alu["sub"]=5
  alu["xor"]=6
  alu["cmp"]=7
  # near conditional jumps
  ops_rel8["jo"]=112+0
  ops_rel8["jno"]=112+1
  ops_rel8["jb"]=112+2
  ops_rel8["jc"]=112+2
  ops_rel8["jnae"]=112+2
  ops_rel8["jae"]=112+3
  ops_rel8["jnb"]=112+3
  ops_rel8["jnc"]=112+3
  ops_rel8["je"]=112+4
  ops_rel8["jz"]=112+4
  ops_rel8["jne"]=112+5
  ops_rel8["jnz"]=112+5
  ops_rel8["jbe"]=112+6
  ops_rel8["jna"]=112+6
  ops_rel8["ja"]=112+7
  ops_rel8["jnbe"]=112+7
  ops_rel8["js"]=112+8
  ops_rel8["jns"]=112+9
  ops_rel8["jp"]=112+10
  ops_rel8["jpe"]=112+10
  ops_rel8["jnp"]=112+11
  ops_rel8["jpo"]=112+11
  ops_rel8["jl"]=112+12
  ops_rel8["jnge"]=112+12
  ops_rel8["jge"]=112+13
  ops_rel8["jbl"]=112+13
  ops_rel8["jle"]=112+14
  ops_rel8["jng"]=112+14
  ops_rel8["jg"]=112+15
  ops_rel8["jnle"]=112+15
  ops_rel8["loopne"]=224
  ops_rel8["loopnz"]=224
  ops_rel8["loope"]=225
  ops_rel8["loopz"]=225
  ops_rel8["loop"]=226
  ops_rel8["jmp"]=235
  # single-byte opcodes without operands
  ops_sb["nop"]=128+16
  ops_sb["movsb"]=164
  ops_sb["movsw"]=165
  ops_sb["stosb"]=170
  ops_sb["stosw"]=171
  ops_sb["lodsb"]=172
  ops_sb["lodsw"]=173
  ops_sb["ret"]=195
  ops_sb["retf"]=203
  ops_sb["hlt"]=244
  # prefix instructions
  prefix["es"]=38
  prefix["cs"]=46
  prefix["ss"]=54
  prefix["ds"]=64
  prefix["lock"]=240
  prefix["rep"]=243
  prefix["repe"]=242
  prefix["repz"]=242
  prefix["repne"]=243
  prefix["repnz"]=243
}
# error string to insert into listing
function err(str) {
  errors++
  if (of=="lst") printf("**** %s\n", str)
}
# submit a assembling result to output
# set label of current line to off
# this outputs a listing line
function submit(off) {
  if (of=="lst") printf("%04X %-18s %s\n", off, hex, $0)
  if (of=="hex" && hex) printf("%s", hex)
  if (label) {
    if (of=="sym") printf("%s\tsym\t%d\n", label, off)
    if (label in prevsym && prevsym[label]!=off) {
      err("label " label " different during second pass, was " prevsym[label] ", now " off)
    }
    sym[label]=off
  }
  pos=pos+length(hex)/2
  hex=""
}
# evaluate an expression
# globals set:
#   ecrit: value known after first pass, 1=yes, 0=no
#   eregs: concatenated list of registers that add to this expr (for modrm)
function expr(str) {
  val=0
  sign=1
  ecrit=1
  eregs=""
  gsub(/-/, "+-", str)
  split(str, ep,"+")
  for (k in ep) {
    if (substr(ep[k],1,1)=="-") {
      gsub(/^-/, "", ep[k])
      sign = -1
    } else {
      sign = 1
    }
    if (substr(ep[k],1,1)==".") ep[k] = plabel ep[k]
    if (ep[k] in r8 || ep[k] in r16 || ep[k] in sreg) {
      if (sign > 0) {
        eregs=eregs ep[k]
      } else {
        err("Registers cannot be subtractive in expressions")
      }
    } else if (match(ep[k], /^[0-9]/)) {
      if (match(ep[k], /h$/)) {
        ep[i]="0x" ep[k]
      }
      val = val + sign*int(ep[k])
    } else if (ep[k] in sym) {
      val = val + sign*sym[ep[k]]
    } else if (ep[k] in prevsym) {
      val = val + sign*prevsym[ep[k]]
      ecrit=0
    } else {
      err("Undefined label " ep[k])
      ecrit=0
    }
  }
  return val
}

function imm(str) {
  val = expr(str)
  if (eregs) err("Registers not allowed here")
  return val
}

function crit(str) {
  val = imm(str)
  if (!ecrit) err("Labels from below not allowed here")
  return val
}

function push_byte(val) {
  #print("; pushb " val)
  if (val<0 || val>=256) err("Value " val " does not fit in byte")
  hex=hex sprintf("%02X",val)
}
function push_signed_byte(val) {
  while (val < 0) val = val + 256
  push_byte(val)
}
function push_word(val) {
  while (val<0) val=val+65536
  t=sprintf("%04X",val)
  hex=hex substr(t,3) substr(t,1,2)
}
# rs is the register set (r8, r16) that can show up in str
function push_modrm(str, spare, rs) {
  mod=0
  rm=0
  if (str in rs) {
    mod=3
    rm=rs[str]
  } else if (substr(str,1,1)=="[") {
    gsub(/^\[|\]$/, "", str)
    disp=expr(str)
    if (!ecrit || disp) {
      mod=2
    }
    if (!eregs) {
      mod=0
      rm=6
    } else if (eregs in modstr) {
      rm=modstr[eregs]
      # [BP] is unencodable, this combination is read as [0000]
      # so we upgrade [BP] to [BP+00]
      if (mod==0 && rm==6) {
        mod=1
      }
    } else {
      err("Bad modR/M")
    }
  }
  #print("; modR/M:", mod, spare, rm)
  push_byte(mod*64+spare*8+rm)
  if (mod == 1) {
    push_byte(disp)
  } else if (mod == 2 || (mod == 0 && rm == 6)) {
    push_word(disp)
  }
}
# common encoding: two operands, one is modrm, other is register via spare field
# last two bits of opcode specify width (byte/word) and whether modrm operand is first or second
function push_op_modrm(opcode) {
  if (rm1 && byteop && op2 in r8) {
    push_byte(opcode)
    push_modrm(op1, r8[op2], r8)
  } else if (rm1 && wordop && op2 in r16) {
    push_byte(opcode+1)
    push_modrm(op1, r16[op2], r16)
  } else if (rm2 && byteop && op1 in r8) {
    push_byte(opcode+2)
    push_modrm(op2, r8[op1], r8)
  } else if (rm2 && wordop && op1 in r16) {
    push_byte(opcode+2+1)
    push_modrm(op2, r16[op1], r16)
  }
}
# common encoding: one operand encoded as modrm with fixed spare field
# operand can be byte or word, encoded in last bit of opcode
function push_op_fixed_spare(opcode, spare) {
  if (byteop) {
    push_byte(opcode)
    push_modrm(op1, spare, r8)
    return 1
  } else if (wordop) {
    push_byte(opcode+1)
    push_modrm(op1, spare, r16)
    return 2
  }
  return 0
}

# dont process empty lines or comment lines
/^( |\t)*(;|%)/ || /^( |\t)*$/ {
  if (of=="lst") printf("%24s%s\n","",$0)
  next
}
# load symbols from previous pass
$2=="sym" {
  prevsym[$1]=int($3)
  #printf("; %s (%s=%X)\n", $0,$1,prevsym[$1])
  next
}
# Start parsing the line
# and set up per-line vars
{
  label=$1
  gsub(/:$/, "",label)
  opn=2
  split("", b, ":")
  byteop=1
  wordop=1
  dwordop=1
}
# no label on line! fixup
/^ / || /^\t/ {
  label=""
  opn=1
}
{
  # implement local labels
  if (substr(label,1,1)==".") {
    label=plabel label
  } else if (label) {
    plabel=label
  }
  # take note if we got a instruction size specifier
  op=$(opn)
  if (op in prefix) {
    push_byte(prefix[op])
    opn=opn+1
    op=$(opn)
  }

  if($(opn+1)=="byte" || $(opn+1)=="short") {
    wordop=0
    dwordop=0
    opn++
  } else if ($(opn+1)=="word" || $(opn+1)=="near") {
    byteop=0
    dwordop=0
    opn++
  } else if ($(opn+1)=="dword" || $(opn+1)=="far") {
    byteop=0
    wordop=0
    opn++
  }
  split("", a, ":")
  c=0
  for (i=opn+1;i<=NF;i++) {
    if (substr($(i),1,1)==";") break
    a[++c]=$(i)
    if (substr($(i),1,1)=="\"") {
      do {
        i++
        j=index($(i), "\"")
        if (j) {
          a[c]=a[c] " " substr($(i), 1, j)
          break
        } else {
          a[c]=a[c] " " $(i)
        }
      } while($(i))
    } else {
      gsub(/,$/, "", a[c])
    }
  }
  op1=a[1]
  op2=a[2]
  # pre-estimate if operand could be encoded as modrm
  rm1=(op1 in r16) || (op1 in r8) || substr(op1,1,1)=="["
  rm2=(op2 in r16) || (op2 in r8) || substr(op2,1,1)=="["
  # if byte register in operands, it cant be a word or dword operation
  if (op1 in r8 || op2 in r8) {
    wordop=0
    dwordop=0
  }
  # if word register in operands, it cant be a byte or dword operation
  if (op1 in r16 || op2 in r16 || op2 in sreg) {
    byteop=0
    dwordop=0
  }
}
# the source line is parsed by here:
#   - op: opcode name
#   - a: array of operands, starting with 1
#   - c: number of operands
#   - byteop, wordop, dwordop: test before encoding, all 1 per default
# pseudo-opcodes
op=="cpu" {
  next
}
op=="org" {
  pos=crit(a[1])
  submit(pos);next
}
op=="equ" {
  val=crit(a[1])
  submit(val);next
}
op=="db" {
  for(i=1;i<=c;i++) {
    if (substr(a[i],1,1)=="\"") {
      for(j=2;j<length(a[i]);j++) {
        push_byte(_ord_[substr(a[i],j,1)])
      }
    } else push_byte(imm(a[i]))
  }
}
op=="dw" {
  for(i=1;i<=c;i++) push_word(imm(a[i]))
}
# arithmetics: ADD, SUB, XOR etc
op in alu {
  push_op_modrm(alu[op]*8)
  if (!hex) {
    size=push_op_fixed_spare(0x80, alu[op])
    if (size==1) {
      push_byte(imm(op2))
    } else if (size==2) {
      push_word(imm(op2))
    }
  }
}
# no idea why this made this extra, this is a AND without storing the result
op=="test" {
  push_op_modrm(132) # 84
  if (!hex) {
    size=push_op_fixed_spare(246, 0)
    if (size==1) {
      push_byte(imm(op2))
    } else if (size==2) {
      push_word(imm(op2))
    }
  }
}
op=="sar" && op2=="1" {
  push_op_fixed_spare(208, 0) # D0 /0
}
op=="inc" && rm1 {
  push_op_fixed_spare(254, 0) # FE /0
}
# MOV variants
op=="mov" {
  # modrm <-> reg
  if (rm1 && op2 in r8) {
    push_byte(136) # 88
    push_modrm(op1, r8[op2], r8)
  } else if (rm1 && op2 in r16) {
    push_byte(137) # 89
    push_modrm(op1, r16[op2], r16)
  } else if (op1 in r8 && rm2) {
    push_byte(138)
    push_modrm(op2, r8[op1], r8)
  } else if (op1 in r16 && rm2) {
    push_byte(139)
    push_modrm(op2, r16[op1], r16)
  # modrm <-> sreg
  } else if (rm1 && op2 in sreg) {
    push_byte(140)
    push_modrm(op1, sreg[op2], r16)
  } else if (rm2 && op1 in sreg) {
    push_byte(142)
    push_modrm(op2, sreg[op1], r16)
  # reg <- imm
  } else if (op1 in r8) {
    push_byte(176+r8[op1])
    push_byte(imm(op2))
  } else if (op1 in r16) {
    push_byte(184+r16[op1])
    push_word(imm(op2))
  # modrm <- imm
  } else if (byteop && rm1) {
    push_byte(198)
    push_modrm(rm1, 0, r16)
  } else if (wordop && rm1) {
    push_byte(199)
    push_modrm(rm1, 0, r16)
  }
}
op=="push" && op1 in r16 {
  push_byte(80+r16[op1])
}
op=="pop" && op1 in r16 {
  push_byte(88+r16[op1])
}
op=="int" && op1=="3" { # CC breakpoint
  push_byte(204)
}
op=="int" { # CD
  push_byte(205)
  push_byte(imm(op1))
}
op=="jmp" {
  val=imm(op1)-(pos+2)
  if (val>-127 && val<128 && ecrit) {
    push_byte(235)
    push_signed_byte(val)
  } else {
    push_byte(233)
    push_word(val-1)
  }
  submit(pos);next
}
op=="call" && wordop {
  push_byte(232)
  push_word(imm(op1)-(pos+3))
}
op=="neg" {
  push_op_fixed_spare(246, 3)
}
# opcodes with rel8 encoding
op in ops_rel8 && byteop && c==1 {
  push_byte(ops_rel8[op])
  push_signed_byte(imm(op1)-(pos+2))
}
# opcodes without arguments
op in ops_sb { push_byte(ops_sb[op]) }
{
  if (!hex) err("no encoding found")
  submit(pos)
}
END{
  if (of=="hex") printf("\n")
  if (errors) exit(1)
}