klibc - May 2006 - vfork support: need help on arm, parisc, s390

Hello all,

I really want to support vfork() in klibc, mostly because uClinux *has* 
to use it.  Unfortunately vfork() isn't allowed to use the stack *at 
all* across the system call -- including return address -- which means 
it needs an assembly wrapper on most architectures.

I have tried implementing wrappers for most architectures, but I don't 
know parisc or s390/s390x well enough, and both architectures do seem to 
need it.

So if you have either, I would appreciate it if you could pull the git 
version of the standalone klibc tree, and implement the appropriate vfork.S.

I also need help *testing* the ARM wrapper I do have, since I currently 
have no access to ARM.

As before, I have only compile-tested m68k.  I don't have an m68k box or 
emulator.

	-hpa

* H. Peter Anvin <hpa@zytor.com> [2006-05-30
23:00]:
> I also need help *testing* the ARM wrapper I do have, since I
> currently have no access to ARM.
  KLIBCAS usr/klibc/arch/arm/vfork.o
usr/klibc/arch/arm/vfork.S: Assembler messages:
usr/klibc/arch/arm/vfork.S:14: Error: cannot represent SWI relocation
in this object file format
make[2]: *** [usr/klibc/arch/arm/vfork.o] Error 1

This is on a Debian ARM box, little-endian with standard ABI (i.e. not
EABI)
-- 
Martin Michlmayr
http://www.cyrius.com/

Martin Michlmayr wrote:
> * H. Peter Anvin <hpa@zytor.com> [2006-05-30 23:00]:
>> I also need help *testing* the ARM wrapper I do have, since I
>> currently have no access to ARM.
> 
>   KLIBCAS usr/klibc/arch/arm/vfork.o
> usr/klibc/arch/arm/vfork.S: Assembler messages:
> usr/klibc/arch/arm/vfork.S:14: Error: cannot represent SWI relocation
> in this object file format
> make[2]: *** [usr/klibc/arch/arm/vfork.o] Error 1
> 
> This is on a Debian ARM box, little-endian with standard ABI (i.e. not
> EABI)
Where can I find documentation on these ABIs?

	-hpa

On Wed, May 31, 2006 at 12:00:02AM -0700, H. Peter Anvin
wrote:
> Hello all,
> 
> I really want to support vfork() in klibc, mostly because uClinux *has*
> to use it.  Unfortunately vfork() isn't allowed to use the stack *at
all*
> across the system call -- including return address -- which 
> means it needs an assembly wrapper on most architectures.
> 
> I have tried implementing wrappers for most architectures, but I don't
> know parisc or s390/s390x well enough, and both architectures do seem to
> need it.
Will have a look at s390/s390x but might take a few days since I'm busy
with other stuff right now...