Re: [PATCH v4 12/13] fair.c: Use generic rbtree impl in fairscheduler

[Peter Zijlstra]

On Fri, 2012-06-22 at 23:00 -0500, Daniel Santos wrote:
> +static inline long compare_vruntime(u64 *a, u64 *b)
> +{
> +#if __BITS_PER_LONG >= 64
> +       return (long)((s64)*a - (s64)*b);
> +#else
> +/* This is hacky, but is done to reduce instructions -- we wont use this for
> + * rbtree lookups, only inserts, and since our relationship is defined as
> + * non-unique, we only need to return positive if a > b and any other value
> + * means less than.
> + */
> +       return (long)(*a > *b);
> +#endif
> +} 

That's wrong.. suppose: a = 10, b = ULLONG_MAX - 10

In that case (s64)(a - b) = 20, however a > b is false.

And yes, vruntime wrap does happen.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/