On Mon, Mar 18, 2019 at 02:21:52PM +0000, Patrick Bellasi wrote:

On 18-Mar 14:10, Peter Zijlstra wrote:

quoted

On Mon, Mar 18, 2019 at 12:18:04PM +0000, Patrick Bellasi wrote:

quoted

On 13-Mar 21:18, Peter Zijlstra wrote:

quoted

On Fri, Feb 08, 2019 at 10:05:42AM +0000, Patrick Bellasi wrote:

quoted

+static void uclamp_fork(struct task_struct *p)
+{
+	unsigned int clamp_id;
+
+	if (unlikely(!p->sched_class->uclamp_enabled))
+		return;
+
+	for (clamp_id = 0; clamp_id < UCLAMP_CNT; ++clamp_id)
+		p->uclamp[clamp_id].active = false;
+}

Because in that case .active == false, and copy_process() will have done
thr right thing?

Don't really get what you mean here? :/

Why don't we have to set .active=false when
!sched_class->uclamp_enabled?

Ok, got it.

In principle because:
- FAIR and RT will have uclamp_enabled
- DL cannot fork

... thus, yes, it seems that the check above is not necessary anyway.

Moreover, as per one of your comments in another message, we still need
to cover the "reset on fork" case for FAIR and RT. Thus, I'm going to
completely remove the support check in uclamp_fork and we always reset
active for all classes.

Right, thanks!