new syscall stub support for ia64 libc

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

David Mosberger wrote:

> I hacked a 2.6 kernel to not pass the sysinfo aux vector entries.  The
> build worked fine for me.  There is nothing unusual on your end, I
> assume?

The kernel work fine with NPTL and the current libc before the patch.
That should qualify as nothing unusual.  There certainly is not code
which sets AT_SYSINFO etc.

What I found so far is that it i fails inside libc, not ld.so.  I.e.,
the startup already left ld.so when it bombs.  In which, the test
program I have (io/pwd, one of the simplest programs there is) fails in
the printf code.  The fstat(1,...) is executed, then it fails.  In a
correct program the next step would be an mmap() syscall.  mmap itself
isn't broken since ld.so already performed a ton of those calls (though
not exactly the same code, it uses the copy in ld.so itself).

- -- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (GNU/Linux)

iD8DBQE/tEyj2ijCOnn/RHQRAhW6AKDF6gLufh9j1aDildXCxHRcqKzDBgCfVnRW
zJIH1hY/S0/0ryyasVegiVk=
=aKCO
-----END PGP SIGNATURE-----