Wrong syntax in mpn/vax

Paul_Koning at Dell.com Paul_Koning at Dell.com
Tue Jul 17 17:49:35 CEST 2012

On Jul 17, 2012, at 6:30 AM, Torbjorn Granlund wrote:

> <Paul_Koning at Dell.com> writes:
>  I ran into two more issues when I tried "make check".  There's a bug
>  in the NetBSD/vax library loader (ld.elf_so) that affects two of the
>  assembly language files.
> How?

The "jmp Loop(%r5) assembles into the actual address of Loop indexed by R5.  So in a shared library, that address operand is relocated.  The linker warns about a relocation in a .text section, and the loader dies with a segfault.

The fix is to use "jmp Loop[%r5]" -- that makes it a base address of PC-relative additionally indexed by R5.  Then there is no relocation involved (the fixed code is already PIC).

I'm about to report this to NetBSD.
>  Also, there were too many underscores in the function names in those
>  files.  The attached diff is the consolidated change for all this.
> I suppose Ultrix is long dead, so we can ignore it?  At any rate, we
> should use the .asm style and let GMP's configure system tack on any
> needed underscores.  (We could also let an vax/elf.m4 prepend the %
> signs to register names.)

Yup, #include <machine/asm.h> then _C_LABEL(__gmpn_add_n) etc.

>  With those changes, I can run "make check".  It fails one test, the
>  other 32 work.  The failing test is for mpn_get_d.  The reason for the
>  failure is that it goes through a code path that assumes IEEE float,
>  and VAX isn't using IEEE.  There seems to be code in get_d.c for
>  non-IEEE, but that is in the #else path of a #if 1 statement...
> The non-IEEE code is probably suffering from bitrot, since it hasn't
> been tested for a very long time.  I'd like it to be correct, but I
> won't work on it any time soon (due to lack of time and lack of working
> system).
> Perhaps you could work with the NetBSD team to make sure the upcoming
> 6.0 relesase works with simh?  At least, it would be nice if you could
> make sure the bugs that you have found gets fixed.

Yes, will do.  Some of that has already been done.


More information about the gmp-bugs mailing list