Re: [PATCH 09/10] Define PERF_PMU_TXN_READ interface

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Aug 13, 2015 at 01:04:28PM -0700, Sukadev Bhattiprolu wrote:

> | > | +static int perf_read_group(struct perf_event *event,
> | > | +				   u64 read_format, char __user *buf)
> | > | +{
> | > | +	struct perf_event *leader = event->group_leader, *child;
> | > | +	struct perf_event_context *ctx = leader->ctx;
> | > | +	int ret = leader->read_size;

> One other question, We return leader->read_size but allocate/copy_to_user
> the sibling's event->read_size. We consistently use read_format from the
> 'event' being read, rather than its 'group_leader', so we are ok in terms
> of what we copy into values[] for each event in the group.
> 
> But, can the leader's read_format (and hence its read_size) differ from
> its sibling's read_size? If so, in the current code, we return the event's
> read_size but in the new code, we return the leader's read_size.

Hmm, good spotting that. I'm fairly sure I didn't do that on purpose.

I think we should use event->read_size there too and have the lot
consistent. I don't think we require read_format to be uniform across
siblings.