Am 14.04.2009 um 12:25 schrieb ard:
On Tue, Apr 14, 2009 at 11:15:42AM +0200, Dr. H. Nikolaus Schaller
wrote:
This brings up one question: what makes us sure that it is indeed a
Kernel bug?
Well, research of Nils did point to something.
Yes, it is always the same handler for SIGILL. I.e. some CPU generated
interrupt.
What looks odd to me is that the FP register had a bad value, i.e. the
subroutine return list on the stack doesn't work any more. Therefore,
it must fail with a SIGILL.
Nils mentioned that almost all KDE tools fail while e.g. the X-Server
never fails. And common to them is that they are prepared for
threading. I don't know if the X-Server uses threading at all (maybe
for animated cursors?) or is only handling events multiplexed through
a single select() call.
But no, as far as I know, there is no clear case if it's user or
kernel space.
It's at least clear that a stable etch distribution cannot run on
the 2.6. kernel, while it can run at the 2.4 kernel, and
therefore we suspect a kernel problem :-).
That is the obvious conclusion. But NPTL vs. Linuxthreads may also be
different and both are also kernel-dependent (different syscalls).
And which one to use may be decided dynamically in the glibc.
Ah, maybe it is simply the missing glibc-jz patch which a stable etch
distro does not necessarily provide.
This patch is quite odd since it should not be necessary in user-
space. Or it is a workaround for a real kernel problem. So we would
have two options:
a) patch glibc for etch
b) patch the kernel so that the glibc patch is not required
Or is simply the NPTL in the glibc broken? In that case the kernel
would
correctly report a stack corruption when trying to return from some
function call.
Therefore we should run the Posix Test Suite!
http://sourceforge.net/project/showfiles.php?group_id=64592&release_id=332108
Great!
The more of these test, the easier it will be to point out the
problem.
After running the tests, we know better if it is the threading library
and really a kernel issue.
Could you try an apt-get gcc on a etch/2.6 system from the command
line and then run the posix test suite? It appears not to have many
dependencies or prerequisites (except make, gcc).
Nikolaus
_______________________________________________
Mipsbook-devel mailing list
Mipsbook-devel@linuxtogo.org
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/mipsbook-devel
