Re: [Xen-devel] [PATCH v2 2/3] xen/atomic: Implement atomic_{inc, dec}_bounded()

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 02.07.14 at 15:47, <andrew.cooper3@xxxxxxxxxx> wrote:
> These will increment/decremented an atomic_t, while ensuring that the 
> atomic_t
> doesn't hit a specific bound.
> 
> This involves introducing atomic_cmpxchg() for x86, which previously only
> existed on ARM.
> 
> Suggested-by: Don Slutz <dslutz@xxxxxxxxxxx>
> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> CC: Keir Fraser <keir@xxxxxxx>
> CC: Jan Beulich <JBeulich@xxxxxxxx>

You should be getting used to Cc all "REST" maintainers on respective
changes.

> --- /dev/null
> +++ b/xen/common/atomic.c
> @@ -0,0 +1,35 @@
> +#include <xen/atomic.h>
> +
> +bool_t atomic_inc_bounded(atomic_t *v, int bound)
> +{
> +    int old, new, prev = atomic_read(v);
> +
> +    do
> +    {
> +        old = prev;
> +        new = old + 1;
> +        if ( new >= bound )
> +            return 0;
> +
> +        prev = atomic_cmpxchg(v, old, new);
> +    } while ( prev != old );
> +
> +    return 1;
> +}
> +
> +bool_t atomic_dec_bounded(atomic_t *v, int bound)
> +{
> +    int old, new, prev = atomic_read(v);
> +
> +    do
> +    {
> +        old = prev;
> +        new = old - 1;
> +        if ( new <= bound )
> +            return 0;
> +
> +        prev = atomic_cmpxchg(v, old, new);
> +    } while ( prev != old );
> +
> +    return 1;
> +}

Same question here: Do we really need these? There are various uses
of cmpxchg() in common code already, and in patch 3 you don't really
need the cmpxchg to happen on an atomic_t, so plain cmpxchg() with
the little bit of extra logic open coded would seem fine to me.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel