Re: [PATCH v0 3/5] perf: Introduce instruction trace filtering

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Dec 11, 2015 at 04:28:54PM +0100, Peter Zijlstra wrote:
> On Fri, Dec 11, 2015 at 03:36:36PM +0200, Alexander Shishkin wrote:
> 
> > @@ -9063,6 +9621,18 @@ inherit_event(struct perf_event *parent_event,
> >  	get_ctx(child_ctx);
> >  
> >  	/*
> > +	 * Clone itrace filters from the parent, if any
> > +	 */
> > +	if (has_itrace_filter(child_event)) {
> > +		if (perf_itrace_filters_clone(child_event, parent_event,
> > +					      child)) {
> > +			put_ctx(child_ctx);
> > +			free_event(child_event);
> > +			return NULL;
> 
> So inherit_event()'s return policy is somewhat opaque, there's 3
> possible returns:
> 
>  1) a valid struct perf_event pointer; the clone was successful
>  2) ERR_PTR(err), the clone failed, abort inherit_group, fail fork()
>  3) NULL, the clone failed, ignore, continue
> 
> We return NULL under two special cases:
> 
>  - the original event doesn't exist anymore, we're an orphan, do not make
>    more orphans.
> 
>  - the parent event is dying
> 
> 
> I'm fairly sure this return should be in the 2) category. If we cannot
> fully clone the event something bad happened, we should not ignore it.

On second thought; we should not inherit the filters at all.

We should always use event->parent (if exists) for filters. Otherwise
inherited events will get different filters if you change the filter
after clone.