hrtimer: make the nanosleep() syscall use the per process slack
Arjan van de Ven [Mon, 8 Sep 2008 15:58:59 +0000 (08:58 -0700)]
This patch makes the nanosleep() system call use the per process
slack value; with this users are able to externally control existing
applications to reduce the wakeup rate.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>

kernel/hrtimer.c

index a022209..9a4c901 100644 (file)
@@ -1563,9 +1563,14 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
        struct restart_block *restart;
        struct hrtimer_sleeper t;
        int ret = 0;
+       unsigned long slack;
+
+       slack = current->timer_slack_ns;
+       if (rt_task(current))
+               slack = 0;
 
        hrtimer_init_on_stack(&t.timer, clockid, mode);
-       hrtimer_set_expires(&t.timer, timespec_to_ktime(*rqtp));
+       hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack);
        if (do_nanosleep(&t, mode))
                goto out;