[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250723150235.GG1827@brightrain.aerifal.cx>
Date: Wed, 23 Jul 2025 11:02:36 -0400
From: Rich Felker <dalias@...c.org>
To: J. Neuschäfer <j.neuschaefer@....net>
Cc: musl@...ts.openwall.com, Ariadne Conill <ariadne@...eferenced.org>
Subject: Re: musl/arch/sh: Inconsistency between greg_t (int) and
mcontext_t.gregs (unsigned long[]); libucontext fails to build
On Wed, Jul 23, 2025 at 04:48:45PM +0200, J. Neuschäfer wrote:
> Hello,
>
> I recently tried to build libucontext on top of musl-libc for sh4, and
> ran into the following problem:
>
> arch/sh/makecontext.c: In function ‘libucontext_makecontext’:
> arch/sh/makecontext.c:45:14: error: assignment to ‘libucontext_greg_t *’ {aka ‘int *’} from incompatible pointer type ‘long unsigned int *’ [-Wincompatible-pointer-types]
> 45 | regp = &ucp->uc_mcontext.gregs[4];
> | ^
> make: *** [Makefile:155: arch/sh/makecontext.o] Error 1
>
>
> The corresponding code in libucontext is:
>
> void
> libucontext_makecontext(libucontext_ucontext_t *ucp, void (*func)(void), int argc, ...)
> {
> libucontext_greg_t *sp, *regp;
> [...]
> regp = &ucp->uc_mcontext.gregs[4];
> [...]
> }
>
> The issue is that libucontext expects mcontext_t.gregs to use the same type as
> greg_t (which seems reasonable to expect), but musl-libc doesn't do that:
>
> typedef int greg_t, gregset_t[16];
> [...]
> unsigned long gregs[16];
>
>
> Should this be fixed in musl or in libucontext?
I'm not sure. It probably depends on whether there's a good historical
reason these differ. For the mcontext members it looks like glibc has
them as int rather than long, but still unsigned, so they also have a
signedness mismatch which should not implicitly convert, but at least
it would be well-defined accessing thru the mismatched type if the
pointer conversion were explicit (e.g. adding a (void *) cast).
We probably should do something to improve the situation here, but I
think the easiest immediate fix would just be getting rid of the
useless intermediate var regp in libucontext and doing something like:
ucp->uc_mcontext.gregs[4+i] = va_arg(va, libucontext_greg_t);
instead of:
*regp++ = va_arg(va, libucontext_greg_t);
In general, it's probably a bad idea to use pointers to members of
arch-specific structures where there's no clear authoritative answer
for what the exact types of the members are supposed to be.
Rich
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.
