Re: [PATCH v6 1/3] perf/core: use rb trees for pinned/flexible groups

[Alexey Budankov]

On 29.08.2017 16:51, Alexander Shishkin wrote:
> Alexey Budankov <alexey.budan...@linux.intel.com> writes:
> 
>> Now I figured that not all indexed events are always located under 
>> the root with the same cpu, and it depends on the order of insertion
>> e.g. with insertion order 01,02,03,14,15,16 we get this:
>>
>>      02
>>     /  \
>>    01  14
>>       /  \
>>      03  15
>>            \
>>            16
>>
>> and it is unclear how to iterate cpu==0 part of tree in this case.
> 
> Using this example, rb_next() should take you through the nodes in this
> order (assuming you start with 01): 01, 02, 03, 14, etc. So you iterate
> while event->cpu==cpu using rb_next() and you should be fine.

Well, indeed we get the most left leaf (03) in rb_next() for the case above.

> 
>> Iterating cpu specific subtree like this:
>>
>> #define for_each_group_event(event, group, cpu, pmu, field)   \
>>      for (event = rb_entry_safe(group_first(group, cpu, pmu), \
>>                                 typeof(*event), field);       \
>>           event && event->cpu == cpu && event->pmu == pmu;    \
>>           event = rb_entry_safe(rb_next(&event->field),       \
>>                                 typeof(*event), field))
> 
> Afaict, this assumes that you are also ordering on event->pmu, which
> should be reflected in your _less function. And also assuming that
> group_first() is doing the right thing. Can we see the code?

I didn't do ordering by PMU for this patch set. Yet more I implemented 
groups_first() like this:

static struct perf_event *
perf_event_groups_first(struct perf_event_groups *groups, int cpu)
{
        struct perf_event *node_event = NULL;
        struct rb_node *node = NULL;

        node = groups->tree.rb_node;

        while (node) {
                node_event = container_of(node,
                                struct perf_event, group_node);

                if (cpu < node_event->cpu) {
                        node = node->rb_left;
                } else if (cpu > node_event->cpu) {
                        node = node->rb_right;
                } else {
                        node = node->rb_left;
                }
        }

        return node_event;
}

and it doesn't work as expected for case above with cpu == 1.

I corrected the code above to this:

static struct perf_event *
perf_event_groups_first(struct perf_event_groups *groups, int cpu)
{
        struct perf_event *node_event = NULL, *match = NULL;
        struct rb_node *node = NULL;

        node = groups->tree.rb_node;

        while (node) {
                node_event = container_of(node,
                                struct perf_event, group_node);

                if (cpu < node_event->cpu) {
                        node = node->rb_left;
                } else if (cpu > node_event->cpu) {
                        node = node->rb_right;
                } else {
                        match = node_event;
                        node = node->rb_left;
                }
        }

        return match;
}

but now struggling with silent oopses which I guess are not 
related to multiplexing at all.

Please look at v8 for a while. It addresses your comments for v7.

> 
> Regards,
> --
> Alex
> 

Thanks,
Alexey