Re: [PATCH v3 1/n] perf/core: addressing 4x slowdown during per-process profiling of STREAM benchmark on Intel Xeon Phi

[Alexey Budankov]

On 20.06.2017 19:37, Mark Rutland wrote:
> On Tue, Jun 20, 2017 at 06:22:56PM +0300, Alexey Budankov wrote:
> > On 20.06.2017 16:36, Mark Rutland wrote:
> > > On Mon, Jun 19, 2017 at 11:31:59PM +0300, Alexey Budankov wrote:
> > > > On 15.06.2017 22:56, Mark Rutland wrote:
> > > > > On Thu, Jun 15, 2017 at 08:41:42PM +0300, Alexey Budankov wrote:
> > > > > > +static int
> > > > > > +perf_cpu_tree_iterate(struct rb_root *tree,
> > > > > > +		perf_cpu_tree_callback_t callback, void *data)
> > > > > > +{
> > > > > > +	int ret = 0;
> > > > > > +	struct rb_node *node;
> > > > > > +	struct perf_event *event;
> > > > > > +
> > > > > > +	WARN_ON_ONCE(!tree);
> > > > > > +
> > > > > > +	for (node = rb_first(tree); node; node = rb_next(node)) {
> > > > > > +		struct perf_event *node_event = container_of(node,
> > > > > > +				struct perf_event, group_node);
> > > > > > +
> > > > > > +		list_for_each_entry(event, &node_event->group_list,
> > > > > > +				group_list_entry) {
> > > > > > +			ret = callback(event, data);
> > > > > > +			if (ret)
> > > > > > +				return ret;
> > > > > > +		}
> > > > > > +	}
> > > > > > +
> > > > > > +	return 0;
> > > > > >   }
> > > > >
> > > > > If you need to iterate over every event, you can use the list that
> > > > > threads the whole tree.
> > > >
> > > > Could you please explain more on that?
> > >
> > > In Peter's original suggestion, we'd use a threaded tree rather than a
> > > tree of lists.
> > >
> > > i.e. you'd have something like:
> > >
> > > struct threaded_rb_node {
> > > 	struct rb_node   node;
> > > 	struct list_head head;
> > > };
> >
> > Is this for every group leader?
>
> Yes; *every* group leader would be directly in the threaded rb tree.

In this case the tree's key heeds to be something trickier than just 
event->cpu. To avoid that complication group_list is introduced. BTW, 
addressing perf_event_tree_delete issue doesn't look like a big change now:

static void
perf_cpu_tree_delete(struct rb_root *tree, struct perf_event *event)
{
	struct perf_event *next;

	WARN_ON_ONCE(!tree || !event);

	list_del_init(&event->group_entry);

	if (!RB_EMPTY_NODE(&event->group_node)) {
		if (!list_empty(&event->group_list)) {
			next = list_first_entry(&event->group_list,
					struct perf_event, group_entry);
			list_replace_init(&event->group_list,
					&next->group_list);
			rb_replace_node(&event->group_node,
					&next->group_node, tree);
		} else {
			rb_erase(&event->group_node, tree);
		}
		RB_CLEAR_NODE(&event->group_node);
	}
}

> > Which objects does the head keep?
>
> Sorry, I'm not sure how to answer that. Did the above clarify?
>
> If not, could you rephrase the question?
>
> Thanks,
> Mark.