Re: 2.6.13-rc5-mm1: oops when starting nscd on AMD64

[Alexander Nyberg <alexn@telia.com>]

> > > I don't think it was supposed to do that.
> >  > 
> >  > Quite possibly it's something to do with the new debugging code - could you
> >  > please take a copy of the offending config, send it over and then try
> >  > removing debug options, see if the crash goes away?  CONFIG_DEBUG_PREEMPT
> >  > would be the first to try..
> > 
> >  The (offending) .config is attached and here's what happens without CONFIG_DEBUG_PREEMPT
> >  (the other debug options being unchanged):
> 
> Yes, my emt64 machine keels over with your .config too.  Maybe it's due to
> CONFIG_SMP=n, not sure.
> 
> Bisection searching shows that the bug was introduced by
> slab-leak-detector-give-longer-traces.patch.
> 

I was afraid it was when I first saw it but I couldn't reproduce (and
still can't).

> Call Trace:<ffffffff801a17bb>{sys_epoll_create+568} <ffffffff8018b1f7>{vfs_readdir+167}
>        <ffffffff80231000>{add_preempt_count+93} <ffffffff8010e8fa>{system_call+126}    
>                                                                                    

For some reason your compilers inline heavier than mine do, which makes
this:

kmem_cache_alloc
sys_epoll_create	(__builtin_return_address(0))
system_call		(__builtin_return_address(1))
			(__builtin_return_address(2))

and off the stack we go...

I guess it was naive to even try to use this for more than the first
caller, sorry. Please throw that thing away and I'll do some backtracing
similar to CONFIG_PAGE_OWNER