Re: [PATCH v2] introduce atomic_pointer to fix a race condition in cancelable mcs spinlocks

From: Peter Zijlstra
Date: Tue Jun 03 2014 - 11:09:23 EST


On Tue, Jun 03, 2014 at 07:07:27AM -0700, Paul E. McKenney wrote:
> On Tue, Jun 03, 2014 at 09:36:13AM +0200, Peter Zijlstra wrote:
> >
> > #ifdef __CHECKER__
> > #define __atomic __attribute__((address_space(5)))
> > #else
> > #define __atomic
> > #endif
> >
> > #define store(p, v) (*(p) = (typeof(*(p)) __force __atomic)(v))
> > #define load(p) ((typeof(*p) __force)ACCESS_ONCE(*(p)))
> >
> > Along with changes to xchg() and cmpxchg() that require them to take
> > pointers to __atomic.
> >
> > That way we keep the flexibility of xchg() and cmpxchg() for being
> > (mostly) type and size invariant, and get sparse to find wrong usage.
> >
> > Then parisc, sparc32, tile32, metag-lock1 and arc-!llsc can go implement
> > store() however they like.
>
> Should be fun interacting with atomic operations on __rcu variables
> (address space 4). Of course, that is already fun...
>

Hmm, good point, I suppose sparse doesn't like two different
address_space annotations on the same variable ?

/me adds Christpoher Li to the CC list.

ISTR Mikulas actually listing one such, me digs in recent email..

> $ grep -w "fdt->fd" */*.c
> fs/file.c: free_fdmem(fdt->fd);
> fs/file.c: fdt->fd = data;
> fs/file.c: free_fdmem(fdt->fd);
> fs/file.c: struct file * file = xchg(&fdt->fd[i], NULL);

So yes, that's going to be fun, mostly because rcu_assign_pointer()
doesn't actually do the right magic for this to be safe on their
platform(s).


Attachment: pgpsY7BgWfpy3.pgp
Description: PGP signature