Re: [PATCH 1/9] perf: Remove redundant parent context check from context_equiv

From: Peter Zijlstra <peterz@infradead.org>
To: Jiri Olsa <jolsa@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>,
	linux-kernel@vger.kernel.org, Andi Kleen <andi@firstfloor.org>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Corey Ashford <cjashfor@linux.vnet.ibm.com>,
	David Ahern <dsahern@gmail.com>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	Ingo Molnar <mingo@kernel.org>,
	"Jen-Cheng(Tommy) Huang" <tommy24@gatech.edu>,
	Namhyung Kim <namhyung@kernel.org>,
	Paul Mackerras <paulus@samba.org>,
	Stephane Eranian <eranian@google.com>
Subject: Re: [PATCH 1/9] perf: Remove redundant parent context check from context_equiv
Date: Mon, 8 Sep 2014 17:13:05 +0200	[thread overview]
Message-ID: <20140908151305.GU3588@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <20140908133428.GG6758@twins.programming.kicks-ass.net>

[-- Attachment #1: Type: text/plain, Size: 1824 bytes --]

On Mon, Sep 08, 2014 at 03:34:28PM +0200, Peter Zijlstra wrote:
> On Mon, Sep 08, 2014 at 02:01:41PM +0200, Jiri Olsa wrote:
> > On Mon, Sep 08, 2014 at 12:01:22PM +0200, Peter Zijlstra wrote:
> > > On Mon, Sep 08, 2014 at 11:48:55AM +0200, Peter Zijlstra wrote:
> > > 
> > > > > The thing is; I don't understand those reasons. That commit log doesn't
> > > > > explain.
> > > > 
> > > > Ah wait, I finally see. I think we want to fix that exit path, not
> > > > disallow the cloning.
> > > > 
> > > > The thing is, by not allowing this optimization simple things like eg.
> > > > pipe-test say very expensive.
> > > 
> > > So its 179033b3e064 ("perf: Add PERF_EVENT_STATE_EXIT state for events
> > > with exited task") that introduces the problem. Before that things would
> > > work correctly afaict.
> > 
> > hum, I dont think so.. because the perf_remove_from_context set event
> > to PERF_EVENT_STATE_OFF state anyway.. thus making any new cloned events
> > disabled
> 
> Urgh, see I knew I was missing something.
> 
> Can't we fix that? Lemme check to see what relies on this.

2e2af50b1fab ("perf_events: Disable events when we detach them")

Seems to be about it. And I think we should solve that differently, but
the best I can come up with ties into the event->ctx mess we have in
that other thread.

The thing is, IOC_ENABLE/DISABLE and read() and such should act
(sanely and) independent from the attached state.

Its just that the whole event->ctx migration mess is making this
somewhat hard atm.

So things like perf_event_read() should not only check ctx->is_active
but also worry about event->attach_state & PERF_ATTACH_CONTEXT.

Now the biggest problem is that we cannot tell if its a temporary state
(move_group / migrate_context) or permanent (exit)... 

Urgh

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]