On Thu, 25 Jan 2007, Mikael Pettersson wrote:
> Sergei Golovan writes:
> > As one can see, erlang Debian package (R11B-2 relese, 11.b.2 in Debian
> > versioning) fails to build on m68k now
> >
> (http://buildd.debian.org/fetch.cgi?&pkg=erlang&ver=1%3A11.b.2-4&arch=m68k&stamp=1167824208&file=log).
> >
> > After some research, we've found that erlang relies on the alignment
> > of some global variable. The following patch fixes it:
> > ------------------------------------------------------------------
> > --- erlang-11.b.2.orig/erts/emulator/beam/binary.c
> > +++ erlang-11.b.2/erts/emulator/beam/binary.c
> > @@ -31,7 +31,7 @@
> > #include "erl_bits.h"
> >
> > Uint erts_allocated_binaries;
> > -erts_mtx_t erts_bin_alloc_mtx;
> > +erts_mtx_t erts_bin_alloc_mtx __attribute__ ((aligned (4)));
> >
> > void
> > erts_init_binary(void)
> > ------------------------------------------------------------------
> >
> > So, the question is the following: Are global variables supposed to be
> > always aligned? Is this a bug in GCC that it doesn't align the
> > variable?
>
> Definitely smells like a toolchain bug to me. In this case, since you've
> enabled threads, erts_mtx_t is a properly declared struct type containing
> among other things a pthread_mutex_t, an int, and a couple of pointers.
> I don't remember how strict M68K is about alignment, but presumably it wants
> non-arbitrary alignment for word-sized fields, especially perhaps if they're
> used for atomic accesses.
>
> The error could be in gcc, gas, ld, or even libc.
The m68k ABI specifies that 32-bit objects must be aligned to a 16-bit address.
I.e. ints and pointers are not guaranteed to be 32-bit aligned.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [EMAIL PROTECTED]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
