Re: [PATCH 2/2] Introduce __cond_lock_err

[Josh Triplett]

On Thu, Dec 21, 2017 at 05:10:00PM -0800, Matthew Wilcox wrote:
> On Thu, Dec 21, 2017 at 02:48:10PM -0700, Ross Zwisler wrote:
> > > +++ b/include/linux/compiler_types.h
> > > @@ -16,6 +16,7 @@
> > >  # define __acquire(x)	__context__(x,1)
> > >  # define __release(x)	__context__(x,-1)
> > >  # define __cond_lock(x,c)	((c) ? ({ __acquire(x); 1; }) : 0)
> > > +# define __cond_lock_err(x,c)	((c) ? 1 : ({ __acquire(x); 0; }))
> > 					       ^
> > I think we actually want this to return c here ^
> > 
> > The old code saved off the actual return value from __follow_pte_pmd() (say,
> > -EINVAL) in 'res', and that was what was returned on error from both
> > follow_pte_pmd() and follow_pte().  The value of 1 returned by __cond_lock()
> > was just discarded (after we cast it to void for some reason).
> > 
> > With this new code we actually return the value from __cond_lock_err(), which
> > means that instead of returning -EINVAL, we'll return 1 on error.
> 
> Yes, but this define is only #if __CHECKER__, so it doesn't matter what we
> return as this code will never run.

It does matter slightly, as Sparse does some (very limited) value-based
analyses. Let's future-proof it.

> That said, if sparse supports the GNU syntax of ?: then I have no
> objection to doing that.

Sparse does support that syntax.

- Josh Triplett