From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932394Ab0EKJwW (ORCPT ); Tue, 11 May 2010 05:52:22 -0400 Received: from casper.infradead.org ([85.118.1.10]:39697 "EHLO casper.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756965Ab0EKJwU convert rfc822-to-8bit (ORCPT ); Tue, 11 May 2010 05:52:20 -0400 Subject: [PATCH] perf: Fix exit() vs PERF_FORMAT_GROUP -v2 From: Peter Zijlstra To: Ingo Molnar Cc: Paul Mackerras , Corey J Ashford , stephane eranian , LKML Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8BIT Date: Tue, 11 May 2010 11:51:53 +0200 Message-ID: <1273571513.5605.3527.camel@twins> Mime-Version: 1.0 X-Mailer: Evolution 2.28.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Both Stephane and Corey reported that PERF_FORMAT_GROUP didn't work as expected if the task the counters were attached to quit before the read() call. The cause is that we unconditionally destroy the grouping when we remove counters from their context. Fix this by splitting off the group destroy from the list removal such that perf_event_remove_from_context() does not do this and change perf_event_release() to do so. Reported-by: Corey Ashford Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra --- kernel/perf_event.c | 19 ++++++++++++++++--- 1 files changed, 16 insertions(+), 3 deletions(-) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index a904746..c97e825 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -308,8 +308,6 @@ list_add_event(struct perf_event *event, struct perf_event_context *ctx) static void list_del_event(struct perf_event *event, struct perf_event_context *ctx) { - struct perf_event *sibling, *tmp; - if (list_empty(&event->group_entry)) return; ctx->nr_events--; @@ -333,6 +331,12 @@ list_del_event(struct perf_event *event, struct perf_event_context *ctx) */ if (event->state > PERF_EVENT_STATE_OFF) event->state = PERF_EVENT_STATE_OFF; +} + +static void +perf_destroy_group(struct perf_event *event, struct perf_event_context *ctx) +{ + struct perf_event *sibling, *tmp; /* * If this was a group event with sibling events then @@ -1868,6 +1872,12 @@ int perf_event_release_kernel(struct perf_event *event) { struct perf_event_context *ctx = event->ctx; + /* + * Remove from the PMU, can't get re-enabled since we got + * here because the last ref went. + */ + perf_event_disable(event); + WARN_ON_ONCE(ctx->parent_ctx); /* * There are two ways this annotation is useful: @@ -1882,7 +1892,10 @@ int perf_event_release_kernel(struct perf_event *event) * to trigger the AB-BA case. */ mutex_lock_nested(&ctx->mutex, SINGLE_DEPTH_NESTING); - perf_event_remove_from_context(event); + raw_spin_lock_irq(&ctx->lock); + list_del_event(event, ctx); + perf_destroy_group(event, ctx); + raw_spin_unlock_irq(&ctx->lock); mutex_unlock(&ctx->mutex); mutex_lock(&event->owner->perf_event_mutex);