Re: [PATCH] atomic_long_t & include/asm-generic/atomic.h V2

[Andrew Morton <akpm@osdl.org>]

Christoph Lameter <clameter@engr.sgi.com> wrote:
>
> V1->V2:
>  - Allow an arch to provide its on atomic_long_t, so that arches such
>    as sh64 can provide their own atomic_long_t.
>  - Make sure that atomic_long_read() always returns (long) even if the
>    arch uses 32 bit for atomic_long_t so that
>     printk("%ld", atomic_long_read(v))
>    always works (not sure how this could work right if atomic_long is
>    long long like for sh4 but I think that is up to the sh64 developers to
>    figure out).
> 
>  Several counters already have the need to use 64 atomic variables on 64
>  bit platforms (see mm_counter_t in sched.h). We have to do ugly ifdefs to
>  fall back to 32 bit atomic on 32 bit platforms.
> 
>  The VM statistics patch that I am working on will also need to make more 
>  extensive use of atomic64.
> 
>  This patch introduces a new type atomic_long_t by providing definitions in
>  asm-generic/atomic.h that works similar to the c "long" type. Its 32 bits on
>  32 bit platforms and 64 bits on 64 bit platforms.

I dunno, this still looks kludgy.  It looks like we couldn't be assed
implementing atomic_long_t in each architecture ;)

How about requiring that all 64-bit archs implement atomic64_t and do:

#if BIT_PER_LONG == 64

typedef atomic64_t atomic_long_t

static inline long atomic_long_read(atomic_long_t *vl)
{
	/* typecast the return value in case arch uses long long */
	return (long)atomic64_read((atomic64_t *)vl);
}

...

static inline void atomic_long_set(atomic_long_t *vl, long i)
{
	/*
	 * Do the cast separately to avoid possible cast-as-lval errors
	 */
	atomic64_t *v = (atomic64_t *)vl;
	atomic64_set(v, i);
}

#else

typedef atomic_t atomic_long_t

static inline long atomic_long_read(atomic_long_t *vl)
{
	return atomic_read((atomic_t *)vl);
}

...

static inline void atomic_long_set(atomic_long_t *vl, long i)
{
	atomic_t *v = (atomic_t *)vl;
	atomic_set(v, i);
}

#endif

?