Re: [PATCH 3/4] IPMI: convert locked counters to atomics

[Andrew Morton <akpm@linux-foundation.org>]

On Thu, 14 Feb 2008 13:33:10 -0600
Corey Minyard <minyard@acm.org> wrote:

> Andrew Morton wrote:
> >> +	for (i = 0; i < IPMI_NUM_STATS; i++)
> >> +		atomic_set(&intf->stats[i], 0);
> >>     
> >
> > And this is why it would be very hard for any architecture to ever
> > implement atomic_t as
> >
> > struct atomic_t {
> > 	int counter;
> > 	spinlock_t lock;
> > };
> >
> > The interface assumes that atomic_set() fully initialises the atomic_t, and
> > that atomic_set() can be used agaisnt both an uninitialised atomic_t and
> > against an already-initialised atomic_t.  IOW, we don't have atomic_init().
> >
> > So would our hypothetical future architcture's atomic_set() do spin_lock(),
> > or would it do spin_lock_init()?  Either one is wrong in many atomic_set
> > callsites.
> >
> > Oh well.
> >   
> Yeah, I thought the same thing when I did this.  Do we start working
> on an atomic_init()?  It would be easy enough to set it to atomic_set()
> for current architectures.

I suppose we should, but I can't say I'm terribly excited by the prospect ;)