Re: [PATCH, v6 2/3] Implement timer slack notifier chain

On Mon, Feb 14, 2011 at 02:32:23PM +0100, Thomas Gleixner wrote:

On Mon, 14 Feb 2011, Kirill A. Shutsemov wrote:

quoted

From: Kirill A. Shutemov <redacted>

Process can change its timer slack using prctl(). Timer slack notifier
call chain allows to react on such change or forbid it.

So we add a notifier call chain and more exports to allow what ?

To allow the cgroup contoller validate the value.

quoted

--- a/kernel/sys.c
+++ b/kernel/sys.c

@@ -1691,15 +1691,10 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
 			error = perf_event_task_enable();
 			break;
 		case PR_GET_TIMERSLACK:
-			error = current->timer_slack_ns;
+			error = prctl_get_timer_slack();

  What's the point of replacing current->timer_slack_ns with a
  function which does exactly the same ?

To keep it consistent. BTW, prctl_get_seccomp() does the same.

quoted

+long prctl_set_timer_slack(long timer_slack_ns)
+{
+	int err;
+
+	/* Reset timer slack to default value */
+	if (timer_slack_ns <= 0) {
+		current->timer_slack_ns = current->default_timer_slack_ns;
+		return 0;

  That does not make any sense at all. Why is setting
  default_timer_slack_ns not subject to validation ?

Hm.. In case of cgroup_timer_slack it's always valid.
But, yes, in general, we should validate it.

  Why is it treaded seperately ?

What do you mean?

quoted

+	}
+
+	err = blocking_notifier_call_chain(&timer_slack_notify_list,
+			timer_slack_ns, NULL);
+	if (err == NOTIFY_DONE)
+		current->timer_slack_ns = timer_slack_ns;
+
+	return notifier_to_errno(err);

Thanks,

	tglx

-- 
 Kirill A. Shutemov