Who maintains or makes fixes for ld for AIX?

Who maintains or makes fixes for ld for AIX?

[Kevin Heatwole]

I'm fairly new to the GCC scene, but I have a task to make a 
cross-compiler hosted on LinuxPPC and targeting AIX.

I'm running into some bugs that I have yet to track down, but appear to 
be in either GAS or ld as it deals with entries in the TOC.  For some 
reason, there are some entries that should just be 32-bit addresses in 
the TOC and the image ends up with extra an 32-bit zero in between some 
entries (almost as if the code thinks an address is 64-bit in some 
instances, but I am dealing with all 32-bit addresses).  The resulting 
a.out doesn't run and the TOC offsets are really garbage (almost as if 
the linker thought it had located all 32-bit addresses, but had left 
unintended gaps in some of them and didn't figure in the gaps in its 
offset calculations).

Anyway, I'd like to exchange a little email with whomever is maintaining 
ld and as as a cross to AIX.  I'm willing to fix the problems but I 
don't want to reinvent the wheel.

One possibility is that I'm using a snapshot build of binutils (the 
released builds just didn't work hardly at all), so maybe I need to step 
up to a new snapshot.

Anyway, if there is anyone out there that knows this stuff, I'd 
appreciate a contact.

Thanks,

Kevin

Re: Who maintains or makes fixes for ld for AIX?

[Geoff Keating]

> Date: Mon, 29 Oct 2001 12:46:55 -0500
> From: Kevin Heatwole <kdh@ocsystems.com>
> 
> I'm fairly new to the GCC scene, but I have a task to make a 
> cross-compiler hosted on LinuxPPC and targeting AIX.
> 
> I'm running into some bugs that I have yet to track down, but appear to 
> be in either GAS or ld as it deals with entries in the TOC.  For some 
> reason, there are some entries that should just be 32-bit addresses in 
> the TOC and the image ends up with extra an 32-bit zero in between some 
> entries (almost as if the code thinks an address is 64-bit in some 
> instances, but I am dealing with all 32-bit addresses).  The resulting 
> a.out doesn't run and the TOC offsets are really garbage (almost as if 
> the linker thought it had located all 32-bit addresses, but had left 
> unintended gaps in some of them and didn't figure in the gaps in its 
> offset calculations).
> 
> Anyway, I'd like to exchange a little email with whomever is maintaining 
> ld and as as a cross to AIX.  I'm willing to fix the problems but I 
> don't want to reinvent the wheel.

I don't believe anyone has yet tried using these tools in a _cross_ to
AIX.  They were developed for use natively on the machine.  However,
I can't think of any fundamental reason why they shouldn't work, so
it's probably just bugs, not completely missing features.

A good person to ask is Tom Rix, <trix@redhat.com>.

-- 
- Geoffrey Keating <geoffk@geoffk.org>

Re: Who maintains or makes fixes for ld for AIX?

[Tom Rix]

Geoff Keating wrote:

> > Date: Mon, 29 Oct 2001 12:46:55 -0500
> > From: Kevin Heatwole <kdh@ocsystems.com>
> >
> > I'm fairly new to the GCC scene, but I have a task to make a
> > cross-compiler hosted on LinuxPPC and targeting AIX.
> >
> > I'm running into some bugs that I have yet to track down, but appear to
> > be in either GAS or ld as it deals with entries in the TOC.  For some
> > reason, there are some entries that should just be 32-bit addresses in
> > the TOC and the image ends up with extra an 32-bit zero in between some
> > entries (almost as if the code thinks an address is 64-bit in some
> > instances, but I am dealing with all 32-bit addresses).  The resulting
> > a.out doesn't run and the TOC offsets are really garbage (almost as if
> > the linker thought it had located all 32-bit addresses, but had left
> > unintended gaps in some of them and didn't figure in the gaps in its
> > offset calculations).
> >
> > Anyway, I'd like to exchange a little email with whomever is maintaining
> > ld and as as a cross to AIX.  I'm willing to fix the problems but I
> > don't want to reinvent the wheel.
>
> I don't believe anyone has yet tried using these tools in a _cross_ to
> AIX.  They were developed for use natively on the machine.  However,
> I can't think of any fundamental reason why they shouldn't work, so
> it's probably just bugs, not completely missing features.
>
> A good person to ask is Tom Rix, <trix@redhat.com>.

I have done some hacking to do a x86 linux-x- ppc aix binutils.   The only
big problem you will have is with bfd & core files.   The core routines are
very dependent on the aix system headers.   I spent a little time trying to
unravel the structures but it was a real pain and I moved on.

x86 linux x ppc aix gcc
No big problems.
problem doing -static using /lib/syscalls.exp and friends
need crt0.o

I did this just to eyeball the binaries.  I don't think it ever ran anything
because I have a real aix for that.

Agreeing with Geoff that it is likely bugs.

Tom


--
Tom Rix
GCC Engineer
trix@redhat.com
256.704.9201

Re: Who maintains or makes fixes for ld for AIX?

[Daniel Jacobowitz]

On Wed, Oct 31, 2001 at 03:39:02PM -0600, Tom Rix wrote:
> I have done some hacking to do a x86 linux-x- ppc aix binutils.   The only
> big problem you will have is with bfd & core files.   The core routines are
> very dependent on the aix system headers.   I spent a little time trying to
> unravel the structures but it was a real pain and I moved on.

Have you tried recently?  Might want to look again; all you should need
to do is extract two or three constants (er... I think.  Does AIX use
ELF core dumps?)

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: Who maintains or makes fixes for ld for AIX?

[Tom Rix]

Daniel Jacobowitz wrote:

> On Wed, Oct 31, 2001 at 03:39:02PM -0600, Tom Rix wrote:
> > I have done some hacking to do a x86 linux-x- ppc aix binutils.   The only
> > big problem you will have is with bfd & core files.   The core routines are
> > very dependent on the aix system headers.   I spent a little time trying to
> > unravel the structures but it was a real pain and I moved on.
>
> Have you tried recently?  Might want to look again; all you should need
> to do is extract two or three constants (er... I think.  Does AIX use
> ELF core dumps?)

I'll check it out tomarrow.  From what I remember it was a couple of "simple"
macros but they depended on the process structure and that got hairy real
quick.   I think some work was done to set the core functions pointers to stubs
in cross configs  but have not varified.

No AIX is xcoff based.

An elf loader of AIX would be a cool hack.


>
>
> --
> Daniel Jacobowitz                           Carnegie Mellon University
> MontaVista Software                         Debian GNU/Linux Developer

--
Tom Rix
GCC Engineer
trix@redhat.com
256.704.9201