From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753552Ab1GRNe4 (ORCPT ); Mon, 18 Jul 2011 09:34:56 -0400 Received: from casper.infradead.org ([85.118.1.10]:35828 "EHLO casper.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751029Ab1GRNez convert rfc822-to-8bit (ORCPT ); Mon, 18 Jul 2011 09:34:55 -0400 Subject: Re: [PATCH v2 1/6] perf: Add interface to add general events to sysfs From: Peter Zijlstra To: Lin Ming Cc: Ingo Molnar , Andi Kleen , Stephane Eranian , Arnaldo Carvalho de Melo , linux-kernel In-Reply-To: <1310740503-15608-2-git-send-email-ming.m.lin@intel.com> References: <1310740503-15608-1-git-send-email-ming.m.lin@intel.com> <1310740503-15608-2-git-send-email-ming.m.lin@intel.com> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8BIT Date: Mon, 18 Jul 2011 15:34:28 +0200 Message-ID: <1310996068.13765.67.camel@twins> Mime-Version: 1.0 X-Mailer: Evolution 2.30.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 2011-07-15 at 14:34 +0000, Lin Ming wrote: > PMU can implement ::add_events() to add its general events to sysfs. > > For example, > > /sys/bus/event_source/devices/uncore/events > └── cycle > > Signed-off-by: Lin Ming > --- > include/linux/perf_event.h | 9 +++++++ > kernel/events/core.c | 52 ++++++++++++++++++++++++++++++++++++++++++++ > 2 files changed, 61 insertions(+), 0 deletions(-) > > diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h > index 3f2711c..14337a3 100644 > --- a/include/linux/perf_event.h > +++ b/include/linux/perf_event.h > @@ -676,6 +676,14 @@ struct pmu { > * for each successful ->add() during the transaction. > */ > void (*cancel_txn) (struct pmu *pmu); /* optional */ > + > + void (*add_events) (void); > +}; I'd rather not have a function pointer here, and all the kobj thingies can be hooking into the existing struct device, right? > +struct pmu_event { > + char name[64]; > + u64 config; > + struct kobj_attribute attr; > }; Can't this live in kerne/event/internal.h? > /** > @@ -941,6 +949,7 @@ struct perf_output_handle { > > extern int perf_pmu_register(struct pmu *pmu, char *name, int type); > extern void perf_pmu_unregister(struct pmu *pmu); > +extern void perf_pmu_add_events(struct pmu *pmu, struct pmu_event events[], int count); > > extern int perf_num_counters(void); > extern const char *perf_pmu_name(void); > diff --git a/kernel/events/core.c b/kernel/events/core.c > index 0567e32..9f4f463 100644 > --- a/kernel/events/core.c > +++ b/kernel/events/core.c > @@ -5571,6 +5571,8 @@ static int pmu_dev_alloc(struct pmu *pmu) > if (ret) > goto free_dev; > > + if (pmu->add_events) > + pmu->add_events(); > out: > return ret; > > @@ -7035,4 +7037,54 @@ struct cgroup_subsys perf_subsys = { > .exit = perf_cgroup_exit, > .attach_task = perf_cgroup_attach_task, > }; > + > +static ssize_t pmu_event_show(struct kobject *kobj, > + struct kobj_attribute *attr, char *buf) > +{ > + struct pmu_event *event = container_of(attr, struct pmu_event, attr); > + int size; > + > + size = sprintf(buf, "%lld\n", event->config); > + > + return size; > +} > + > +void perf_pmu_add_events(struct pmu *pmu, struct pmu_event events[], int count) > +{ > + struct attribute_group *attr_group = NULL; > + struct attribute **all_attrs = NULL; > + struct kobj_attribute *event_attr; > + struct pmu_event *event; > + int i; > + > + attr_group = kzalloc(sizeof(*attr_group), GFP_KERNEL); > + if (!attr_group) > + return; > + attr_group->name = "events"; > + > + all_attrs = kzalloc(sizeof(*all_attrs) * (count + 1), GFP_KERNEL); > + if (!all_attrs) > + goto fail; > + > + for (i = 0; i < count; i++) { > + event = &events[i]; > + event_attr = &event->attr; > + > + sysfs_attr_init(&event_attr->attr); > + event_attr->attr.name = event->name; > + event_attr->attr.mode = 0444; > + event_attr->show = pmu_event_show; > + > + all_attrs[i] = &event_attr->attr; > + } > + > + attr_group->attrs = all_attrs; > + if (!sysfs_create_group(&pmu->dev->kobj, attr_group)) > + return; > + > +fail: > + kfree(attr_group); > + kfree(all_attrs); > +} > + > #endif /* CONFIG_CGROUP_PERF */ Uhm, why does this live under CONFIG_CGROUP_PERF? Also, I would prefer an interface like: int perf_pmu_add_event(struct pmu *pmu, const char *name, u64 config) { } which would create the events directory if not already present and allocate and add a pmu_sysfs_event thingy.