Re: CPU affinity & IPI latency

From: Davide Libenzi (
Date: Mon Jul 16 2001 - 16:25:57 EST

On 16-Jul-2001 Mike Kravetz wrote:
> On Fri, Jul 13, 2001 at 11:25:21PM -0400, Hubertus Franke wrote:
>> Mike, could we utilize the existing mechanism such as has_cpu.
> I like it. Especially the way you eliminated the situation where
> we would have multiple tasks waiting for schedule. Hope this is
> not a frequent situation!!! The only thing I don't like is the
> use of has_cpu to prevent the task from being scheduled. Right
> now, I can't think of any problems with it. However, in the past
> I have been bit by using fields for purposes other than what they
> were designed for.

How about this ( draft ) :

struct task_struct {
        struct task_struct * wlist_next;

static union {
        struct schedule_data {
                struct task_struct * curr;
                struct task_struct * wlist;
                cycles_t last_schedule;
        } schedule_data;
        char __pad [SMP_CACHE_BYTES];
} aligned_data [NR_CPUS] __cacheline_aligned = { {{&init_task,0}}};

static inline struct task_struct * wpick(aligned_data * ad) {
        struct task_struct * tsk = ad->wlist;
        if (tsk) {
                ad->wlist = tsk->wlist_next;
        return tsk;

static inline void wpush(aligned_data * ad, struct task_struct * tsk) {
        if (task_on_runqueue(tsk))
        tsk->wlist_next = ad->wlist;
        ad->wlist = tsk;

asmlinkage void schedule(void)
        if ((next = wpick(sched_data)))
                goto ...;

In reschedule_idle() when before sending the IPI we do a wpush().
We modify aligned_data->wlist and tsk->wlist_next under runqueue_lock so we
don't need another one.
A slight change is needed to reschedule_idle() to handle the new field.
Pros to this solution are 1) we are not going to give other fields a different
meaning 2) when the idle will call schedule it'll pick the task w/o rescan.

- Davide

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Mon Jul 23 2001 - 21:00:08 EST