Messages in this thread Patch in this message | | | Date | Mon, 30 Nov 2015 10:38:22 +0100 | From | Peter Zijlstra <> | Subject | Re: [PATCH] sched: change nr_uninterruptible to be signed |
| |
On Fri, Nov 27, 2015 at 12:09:44PM +0800, yalin wang wrote: > nr_uninterruptible will be negative during running, > this happened when dequeue a TASK_UNINTERRUPTIBLE task > from rq1 and then wake up the task and queue it to rq2, > then rq2->nr_uninterruptible-- will reuslt in negative value > sometimes.
Why!? Aren't isn't signed stuff more likely affected by over/underflow muck? I'd be more inclined to remove the signed muck altogether like:
That relies on C having defined overflow semantics for unsigned types and the (hard) assumption that the hardware uses 2s-complement (which the kernel assumes in many many places already).
diff --git a/kernel/sched/loadavg.c b/kernel/sched/loadavg.c index ef7159012cf3..62cff13ed120 100644 --- a/kernel/sched/loadavg.c +++ b/kernel/sched/loadavg.c @@ -80,13 +80,14 @@ void get_avenrun(unsigned long *loads, unsigned long offset, int shift) long calc_load_fold_active(struct rq *this_rq) { - long nr_active, delta = 0; + unsigned long nr_active; + long delta = 0; nr_active = this_rq->nr_running; - nr_active += (long)this_rq->nr_uninterruptible; + nr_active += this_rq->nr_uninterruptible; if (nr_active != this_rq->calc_load_active) { - delta = nr_active - this_rq->calc_load_active; + delta = (long)(nr_active - this_rq->calc_load_active); this_rq->calc_load_active = nr_active; } diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 2eb2002aa336..2f9b4ce759dc 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -652,7 +652,7 @@ struct rq { /* calc_load related fields */ unsigned long calc_load_update; - long calc_load_active; + unsigned long calc_load_active; #ifdef CONFIG_SCHED_HRTICK #ifdef CONFIG_SMP
| |