GDB Stub with ARM BE8 target.

Luis Machado luis.machado@arm.com
Wed Jun 1 08:55:26 GMT 2022


Hi Jon,

On 5/31/22 00:03, Jon Wilson via Gdb wrote:
> I am trying to integrate a GDB stub with a ARM BE8 target. However, it would appear that using "set endian big" defaults to BE32 architecture. I cannot seem to find a way to have the debugger interpret memory according to the BE8 architecture. The only support for BE8 mode I can see seems to be for when using BE8 core dumps or processes (https://github.com/bminor/binutils-gdb/blob/9606310bbbf8a2d529b5c2aa71cac48a46f65869/gdb/arm-tdep.c#L9781). I can't seem to find any documentation on the topic (which is pretty forgivable given how niche it is).

The "set endian" command might not do what you'd expect, as it will set the endianness for data accesses only. The "code access" endianness
will be set by the binary's ELF flag EF_ARM_BE8. So once you load a BE8 binary, GDB will know how to deal with it.

> 
> I looked at handling the endianness in the stub itself (e.g. inverting the endianness of memory based on permissions), but given the code stream can contain both instructions (which should be little endian in BE8) as well as data (which should be interpretted as big endian), this is not possible.
> 
> Have I missed an option somewhere? If this support is missing, then how should it be added? Should it be handled in a similar way to how GDB handles ARM versus Thumb mode? (https://github.com/bminor/binutils-gdb/blob/9606310bbbf8a2d529b5c2aa71cac48a46f65869/gdb/arm-tdep.c#L10589)?

I don't think you've missed an option. Technically GDB should support BE8 mode, but in practice, since it is less exercised, it may have regressed and may need to be fixed.

Basically GDB needs to be aware of data/code endianess when dealing with memory. There might be places that are not dealing with this correctly. Detection of BE8 should happen
through the ELF flag mentioned above, so you wouldn't need to fiddle with GDB option or thumb-like heuristics.

> 
> I'd be very grateful for any advice.
> 
> Kind Regards.
> 
> 
> Jon
> 



More information about the Gdb mailing list