Ghidra for amiga disassembling

[copse]

I've tried my first export of my Federation of Free Traders project to assembler, and it's painful to say the least. Ghidra clips mnemonic src/dst columns in the export and the asm appears to be corrupted..

The exported asm (sans clipping corruption) doesn't look too far off Amiga syntaxes. I suspect if I spend an hour or two looking into it, and maybe write my own export plugin I can get something out of it.. but still.

[copse]

Here's an example:

ram:00000930    0ab900008...    eori.l      #0x8000,(currentDoubleBufferOffsetMay...

Actually, there are column widths settable so I guess you just have to bump it up to avoid this clipping. In this case, that's 40 from the "#" to the "...".

[copse]

Ghidra 11.0.1.
How to make a function from a pointer assigned to a table.

In Pool of Radiance an affects table is populated with function handlers by affect type as index (I exported my executable with symbols already defined from Resource hence the label names being Resource-created):

        002308e6 43 fa f7 14     lea        (-0x8ec,PC)=>lbC010FFC,A1
        002308ea 23 c9 00        move.l     A1,(DAT_00287b26).l
                 28 7b 26

Unfortunately these are not recognised as functions by Ghidra. Viewing the decompilation shows something like:

/* WARNING: Removing unreachable block (ram,0x0023003e) */

void UndefinedFunction_0022fffc(void)
...

Going back to the first snippet I double-click on lbC010FFC and view the label.

                             lbC010FFC                                       XREF[1]:     SetupAffectTables:002308e6(*)  
        0022fffc 4e 55 ff fc     link.w     A5,-0x4
        00230000 bf f9 00        cmpa.l     (hSPLower).l,SP
                 28 9d 28

The cursor is on the lbC010FFC label. I type F to create a function, then F again to edit the function signature. Then due to double-binding I get presented with a dialog with "Edit Function" as an option along with something else. I hit enter as "Edit Function" is the default selection.

The way I have found to define the standard affect handler function signature is to paste in:

void Affect00E46E (character_s * char_p, effect_s * effect_p, int param_3)

Then I edit the function name manually to include the 10FFC from the label and hit enter to commit the name and parameters in the dialog, then enter again to commit it as the function signature.

Hope this helps someone!

[copse]

Ghidra 11.0.1.
Here's another minor quirk that doesn't happen too often related to decompilation, local variable types and reuse of stack-based local variables.

byte BuildNearTargets(character_s *char_p,byte max_range)

{
...
  character_s *local_14;
...
  local_14 = (character_s *)PlayerMapPosX_Thunk4(char_p);
  local_10 = PlayerMapPosY_Thunk4(char_p);
...
  if (DAT_0028783a != 0) {
    for (i = 1; bVar2 = result_count, i <= bVar1; i = i + 1) {
      local_14 = (&g_character_array)[*(byte *)((int)&g_map_to_background_tile + (uint)i * 3 + 2)];

In this case map pos x is a long stored at the same stack offset as the character_s pointer assigned below. Ghidra cannot share the variable type and forces it to the later usage.

Related issue: https://github.com/NationalSecurityA...ra/issues/2573

Not a big deal in the scheme of things, but useful to understand. Hope this helps someone!

[alexh]

For Ambermoon I believe they wrote an exporter plugin?

But I think you're right exporting wasn't fun and Pyrdacor said numerous times he wanted to leave Ghidra environment following a successfull disassembly and re-assembly and continue working only on the exported files as soon as possible.

[copse]

Ghidra 11.0.1

There's another likely m68k bug in ghidra decompilation, where the first variable of a struct is an array or pointer it will decompile references to later variables in the struct (even outside the fixed array bounds) as relative to that first variable.

For example:

struct character_s {
    char name[14];
    ...
    ushort x;
    ...
}

The decompilation will look something like:

void f(character_s *char_p) {
    ...
    ... (ushort)*(char_p->name + 30) ...
    ...
}

Perhaps not the exact syntax but you get hte idea.