>>> On 16.08.17 at 07:14, <chao....@intel.com> wrote:
> +static inline bool pi_over_limit(unsigned int cpu)
> +{
> + /* Compare w/ constant first to save a division and an add */
> + if ( likely(read_atomic(&per_cpu(vmx_pi_blocking, cpu).counter) <=
> + PI_LIST_FIXED_LIMIT) )
> + return 0;
false
> + else
Pointless else.
> +static unsigned int pi_get_blocking_cpu(unsigned int cpu, unsigned long
> *flags)
> +{
> + spinlock_t *pi_blocking_list_lock;
> +
> + for ( ; ; )
> + {
> + while ( unlikely(pi_over_limit(cpu)) )
> + cpu = cpumask_cycle(cpu, &cpu_online_map);
> +
> + pi_blocking_list_lock = &per_cpu(vmx_pi_blocking, cpu).lock;
> + if ( flags )
> + spin_lock_irqsave(pi_blocking_list_lock, *flags);
> + else
> + spin_lock(pi_blocking_list_lock);
This is ugly, but I think I see why you want it this way. Let's see
what the maintainers think.
Jan
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel
