Re: + futex-runtime-enable-pi-and-robust-functionality.patch added to -mm tree

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Sat, Feb 16, 2008 at 02:05:13PM +0100, Thomas Gleixner wrote:
> On Sat, 16 Feb 2008, Heiko Carstens wrote:
> > > +	/*
> > > +	 * This will fail and we want it. Some arch implementations do
> > > +	 * runtime detection of the futex_atomic_cmpxchg_inatomic()
> > > +	 * functionality. We want to know that before we call in any
> > > +	 * of the complex code paths. Also we want to prevent
> > > +	 * registration of robust lists in that case. NULL is
> > > +	 * guaranteed to fault and we get -EFAULT on functional
> > > +	 * implementation, the non functional ones will return
> > > +	 * -ENOSYS.
> > > +	 */
> > > +	curval = cmpxchg_futex_value_locked(NULL, 0, 0);
> > > +	if (curval == -EFAULT)
> > > +		futex_cmpxchg_enabled = 1;
> > > +
> > 
> > Why should that fail? You're accessing a kernel space address here and no
> > user space address.
> 
> Well, NULL pointer dereferencing is supposed to fail, isn't it ?

I wasn't sure that this is true for all architectures, but...

> > Indeed it does fail with an Oops on s390 since we enable low address
> > protection in the kernel so we get an exception if something within the
> > kernel writes to the first 512 bytes of the kernel address space.
> > Otherwise it would have silently passed the test...
> 
> NULL pointer dereferencing faults on all architectures, at least it
> should, but we explicitely disable pagefaults and recover via the
> extable fixup, which is in S390 as well. That returns -EFAULT and
> signals that there is a working implementation, while those which have
> no support return -ENOSYS, which keeps the robust/pi stuff disabled.

...one of our exception table entries has an off-by-one bug.
Never mind, I'll go and fix our own stuff instead ;)