Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aTPXxEn3TIqdIOGa@intrepid>
Date: Sat, 6 Dec 2025 08:14:12 +0100
From: Markus Wichmann <nullplan@....net>
To: musl@...ts.openwall.com
Subject: Re: How supported is Thumb 1?

Am Fri, Dec 05, 2025 at 07:40:18PM -0500 schrieb Rich Felker:
> Can you test with -march set to whatever the cortex-m model that's
> based on armv6 is? I believe there's one that's only "thumb1" (no
> 32-bit arm isa) but that actually supports everything we use, and this
> is what the "thumb1" support was added for. At present there's
> probably very little point in supporting it since we don't have fdpic
> support on arm, but the intent was to support that. And technically it
> does work now as long as you use pie; just it won't be shareable text.
> 
> Rich

Only Thumb1? That is not supportible, at least not according to the
reference manual. Only Thumb2 would be, though. And indeed, with some
googling I found this nice source here:
https://upload.wikimedia.org/wikiversity/en/4/40/ARM.2ASM.Thumb.20241123.pdf

Which claims that the Cortex-M0 chips are based on ARMv6T2 (and they
only get newer from there).

But in any case, the issues I've identified here are specific to ARMv4.
v5 and above have BLX in both modes, circumventing the issue entirely.
My point was more that if ARMv4 Thumb (which is the most restrictive
instruction set) is not supposed to be supported, then maybe it is
better to break it in the configure script, rather than produce broken
code.

Ciao,
Markus

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.