Help with 68k assembly - jump tables?

Question

I'm working on reverse engineering a large Amiga program in IDA, and I've made a ton of progress. However, there is some stuff I can't quite figure out. Namely, I have found several subroutines which use what I believe to be "jump tables" - but I can't quite figure out how they work. Does anyone have any advice?

        moveq   #0,d0
        move.b  d7,d0       ; set D0 to a byte from CTRL
        subq.w  #1,d0       ; subtract 1 from it
        blt.w   finish_29ABA    ; if D0 is less than 1, branch
        cmpi.w  #$16,d0
        bge.w   finish_29ABA    ; if D0 is greater than or equal to 16, branch
        add.w   d0,d0       ; otherwise, double D0
        move.w  dword_29918(pc,d0.w),d0
        jmp dword_29918+2(pc,d0.w)
; ---------------------------------------------------------------------------
dword_29918:    dc.l $400036        ; CODE XREF: serialCtrlCmd+E0j
        dc.l $360036
        dc.l $3601A0
        dc.l $3601A0
        dc.l $1A001A0
        dc.l $360040
        dc.l $2A01A0
        dc.l $400036
        dc.l $3601A0
        dc.l $1A00036
        dc.l $1A00036
        dc.l $33FC0003
        dc.l HEAP_3B897+$41A7   ; CTRLRead3
; ---------------------------------------------------------------------------
        bra.w   finish_29ABA
; ---------------------------------------------------------------------------
        tst.w   (CTRL_36494).l
        bne.w   return_29AF6
        moveq   #1,d0
        lea ((HEAP_3B897+$2665)).l,a0
        adda.w  (CTRLRead1).l,a0
        move.b  d7,(a0)
        moveq   #0,d1
        move.b  d7,d1
        move.w  d1,(CTRLRead2).l
        move.w  (CTRLCmds).l,d1
        addq.w  #1,d1
        move.w  d1,(CTRLCmds).l ; Increment CTRL Cmds by 1
        move.w  d0,(CTRLRead3).l
        bra.w   finish_29ABA

Shererd

Is IDA the only disassembler for 68k? – Endotoxin

There are plenty, but most of the ones I've found are for the Amiga platform itself. I'm sure there are several others. – Shererd

Keep in mind that the bulk of code for the Amiga was written to be entirely "relocatable". Using jump tables was pretty much a standard (even for system libraries). The program loader would "relocate" all fixed addresses (a safe operation because there is no virtual memory management), and then all jumps/references were done as offsets. – Holm

Answer 1

    blt.w   finish_29ABA    ; \
    cmpi.w  #$16,d0         ; > These insns check that the index is in range
    bge.w   finish_29ABA    ; /

    add.w   d0,d0           ; since the jump table contains words,
                            ; multiply the index by 2, to get a word index

    move.w  dword_29918(pc,d0.w),d0 ; get a word from the jump table, indexed by d0

    jmp dword_29918+2(pc,d0.w)      ; perform an indirect jump to (PC,d0)

    dword_29918:            ; the following are offsets encoded as words
      ; 0040 0036 0036 0036...

This is usually the result of a C switch statement.

Howlyn

Thank you very much, that was extremely helpful. Everything also makes much more sense when the data is interpreted as words. – Shererd

@Mikaveli, yes, niche, but a pretty large niche. We are many sleeper Amigans out there. :-) – Braid

The offsets being written as dc.l was throwing me at first :) In my own code I have used direct address jump tables. – Regiment