On Fri, Mar 31, 2017 at 07:43:20PM +0530, Anurup M wrote:

On Thursday 30 March 2017 04:16 PM, Mark Rutland wrote:

quoted

quoted

quoted

quoted

quoted

+	/*

quoted

quoted

quoted

quoted

+	 * We must NOT create groups containing mixed PMUs, although
+	 * software events are acceptable
+	 */
+	if (event->group_leader->pmu != event->pmu &&
+	    !is_software_event(event->group_leader))
+		return -EINVAL;
+
+	list_for_each_entry(sibling, &event->group_leader->sibling_list,
+			    group_entry)
+		if (sibling->pmu != event->pmu && !is_software_event(sibling))
+			return -EINVAL;

Please also check the number of counters.

Sorry, I could not follow this comment correctly. Could you please explain ?
I check the available counters and update used mask in pmu_add -->
get_event_index

What I meant was that here we should ensure that a group does not
contain more events than can fit into counters.

For example, if the HW had two counters, we should reject any group with
more than two events. Such groups can never be scheduled, and make no
sense.

I have referred drivers/bus/arm-cci.c and could find validate_group
and validate_event functions,
which create a fake_pmu to check the available counters for the
events in the group.
Is that the same way which is expected here? Please comment.

Something like that.

I think it's simplest to have a validate_group() function, which counts
the number of counters used. See my suggestion in [1].

Thanks,
Mark.

[1] https://lkml.kernel.org/r/20170331135955.GB6488 at leverpostej