Re: [PATCH] futex: cmpxchg_futex_value_locked API change

[Michel Lespinasse <walken@google.com>]

On Thu, Mar 10, 2011 at 07:55:05PM +0100, Thomas Gleixner wrote:
> On Wed, 9 Mar 2011, Michel Lespinasse wrote:
> > On Tue, Mar 08, 2011 at 09:17:11PM +0100, Thomas Gleixner wrote:
> 
> Just looked at it again in detail before picking it up. Can we please
> separate the s/int/u32/ changes from the real API change ?
> 
> > -	pagefault_disable();	/* implies preempt_disable() */
> > +	/* Note that preemption is disabled by futex_atomic_cmpxchg_inatomic
> > +	 * call sites. */
> 
> That wants to be a separate patch as well.

No problems. That makes it 3 patches, will send as replies to this.

> > -int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
> > +int futex_atomic_op_inuser(int encoded_op, u32 __user *uaddr)
> >  {
> >  	int op = (encoded_op >> 28) & 7;
> >  	int cmp = (encoded_op >> 24) & 15;
> > @@ -197,7 +197,7 @@ int futex_atomic_op_inuser(int encoded_op, int __user *uaddr)
> >  	if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28))
> >  		oparg = 1 << oparg;
> >  
> > -	if (!access_ok(VERIFY_WRITE, uaddr, sizeof(int)))
> > +	if (!access_ok(VERIFY_WRITE, uaddr, sizeof(u32)))
> >  		return -EFAULT;
> >  
> >  	pagefault_disable();
> 
> So following the reasoning above, shouldn't that be the same for
> futex_atomic_op_inuser() ?

futex_atomic_op_inuser() is currently called by core futex code with
page faults enabled. I think that's OK - for futex_atomic_cmpxchg_inatomic
I fixed the arm implementation because it was inconsistent with the
other ones, but for futex_atomic_op_inuser() every arch does it
that way.

-- 
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.