[Bug binutils/22859] New: gas seems to misassemble cmp r,$label+offset on PDP-11 with reproducer

[etchedpixels at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=22859

            Bug ID: 22859
           Summary: gas seems to misassemble cmp r,$label+offset on PDP-11
                    with reproducer
           Product: binutils
           Version: 2.30
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: binutils
          Assignee: unassigned at sourceware dot org
          Reporter: etchedpixels at gmail dot com
  Target Milestone: ---

(Short reproducer below explanation)

I hit this trying to build my OS with GNU binutils 2.30 (it seems to work with
ancient binutils)
Linux 64bit cross build --target=pdp11-aout

ld errors with

sed.o:sed.o:(.text+0x7a4): relocation truncated to fit: 16 against `.data'
sed.o:sed.o:(.text+0x7be): relocation truncated to fit: 16 against `.data'
sed.o:sed.o:(.text+0x1b16): relocation truncated to fit: 16 against `.data'

Doing an objdump on the module shows

    79c:        15f7 0001 1e14  mov     $1, $25b6 <_lastline>
     7a2:       2097 6c5c       cmp     r2, $66134
     7a6:       8306            blos    7b4 <_getline.part.1+0x40>
     7a8:       0a26            clr     -(sp)
     7aa:       09f7 f8f6       jsr     pc, a4 <_truncated>
     7ae:       0ac2            dec     r2
     7b0:       65c6 0002       add     $2, sp
     7b4:       8a0a            clrb    (r2)
     7b6:       1080            mov     r2, r0
     7b8:       1582            mov     (sp)+, r2
     7ba:       0087            rts     pc
     7bc:       2097 6c5c       cmp     r2, $66134
     7c0:       8202            bhi     7c6 <_getline.part.1+0x52>
     7c2:       904a            movb    r1, (r2)
     7c4:       0a82            inc     r2

So the linker error appears reasonable. However the matching assembler input
for gas is

        mov $01, _lastline
L_129:
        cmp r2,$_linebuf+01750
        blos L_130
        clr -(sp)
        jsr pc, _truncated
        dec r2
        add $02, sp
L_130:
        clrb (r2)
        mov r2, r0
        mov (sp)+, r2
        rts pc
L_125:
        cmp r2,$_linebuf+01750
        bhi L_127
        movb r1, (r2)
        inc r2

Which leads me assume that cmp register, $label+offset is being mis-assembled.


If I try and assemble the following
                .text
                .even
whoops:         cmp r2, $boing+1

                .data
boing:          .word 0


then sure enough if I objdump it

00000000 <whoops>:
   0:   2097 0005       cmp     r2, $5

-- 
You are receiving this mail because:
You are on the CC list for the bug.