[Oberon] A2 start. Was "clock_gettime."

[Michael Schierl]

Hello Peter,


Am 25.07.2026 um 23:53 schrieb Michael Schierl:

> It does not matter whether running it with multiarch or inside a 32-bit 
> chroot (Debian sid), it always immediately receives SIGTRAP like in your 
> example. Loading the core file into gdb suggests it got stranded on an 
> INT3 instruction.
> 
> It also does not matter whether it is started with setarch flags (I 
> think I tried all possible ones, but probably I missed the correct 
> combination).

I probably should have tried a 32-bit Debian oldstable chroot in my 
64-bit kernel as well on Monday... As it works fine there. But not on 
Debian stable or unstable.

Anyway, spent some time in edb today tracking down that INT3 instruction.

It maps to this line in Linux.Glue.Mod:

ASSERT(dlsym # NIL);

dlsym is populated by ld-linux.so.2 based on the relocations table in 
the ELF header. So something is wrong in the ELF header for 
ld-linux.so.2 of recent glibc versions. (The ELF header is built by some 
crude "assembly" code in Linux.Glue.Mod PROCEDURE EntryPoint).

It may reassure you that building ELF headers from Assembly is also 
unfamiliar territory for me, it would required me hours of reading the 
spec and comparing it line-by-line with the assembly, so I asked ChatGPT 
(you can follow along at 
<https://chatgpt.com/share/6a6d2167-ab74-83ed-8fa6-63c4041ea7fa> if you 
want to). It led me through some readelf and LD_DEBUG commands until 
claiming that the reason is that it uses a relocation type in the 
relocation table that officially is only supported on 64-bit systems. 
And it seems that more recent glibc is stricter when checking, and 
skipping it, which results in dlsym remaining NIL.

It also provided a patch, which would require more #IFs to make it not 
break 64-bit, but with that patch, I can build a 32-bit command shell 
that at least loads in my 64-bit WSL. Did not test full A2 since setting 
up X11 via wslg is a bit of a hassle right now for me.

Here is the patch:

--- a/source/Linux.Glue.Mod
+++ b/source/Linux.Glue.Mod
@@ -98,9 +98,9 @@ VAR
                 dynamicsection:
                 DD 05H, base + stringtable
                 DD 06H, symboltablebegin + base
-               DD 07H, dlsymrelocation + base
-               DD 08H, dlsymrelocation_end-dlsymrelocation ; size 
(relocationtable)
-               DD 09H, 0CH
+               DD 11H, dlsymrelocation + base   ; DT_REL
+               DD 12H, dlsymrelocation_end-dlsymrelocation ; DT_RELSZ
+               DD 13H, 08H                     ; DT_RELENT
                 DD 0AH, stringtable_end - stringtable; size (stringtable)
                 DD 0BH, 10H

@@ -117,7 +117,7 @@ VAR
                 DB 01H ; relocation type
                 #END
                 DB 01H, 00H, 00H; index of the symbol
-               DD 0H
+               ;; DD 0H
                 dlsymrelocation_end:

                 stringtable:


Feel free to try it and tell me how far it works for you.


Regards,


Michael