From: Thomas Gleixner <tglx@linutronix.de>
To: Ian Munsie <imunsie@au1.ibm.com>
Cc: linux-kernel <linux-kernel@vger.kernel.org>,
Arnaldo Carvalho de Melo <acme@ghostprotocols.net>,
Ingo Molnar <mingo@elte.hu>,
Frederic Weisbecker <fweisbec@gmail.com>,
Mike Galbraith <efault@gmx.de>, Paul Mackerras <paulus@samba.org>,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Stephane Eranian <eranian@google.com>
Subject: Re: [PATCH 3/3] perf record/report: Process events in order
Date: Mon, 6 Dec 2010 15:41:22 +0100 (CET) [thread overview]
Message-ID: <alpine.LFD.2.00.1012061447080.2653@localhost6.localdomain6> (raw)
In-Reply-To: <1291640985-sup-4443@au1.ibm.com>
On Tue, 7 Dec 2010, Ian Munsie wrote:
> Excerpts from Thomas Gleixner's message of Tue Dec 07 00:04:20 +1100 2010:
> > > I just moved them into this routine to keep all the dispatching in one
> > > place, whether delayed or not. These particular events will still be
> > > processed immediately when encountered in the file. Only >=
> > > PERF_RECORD_MMAP && <= PERF_RECORD_SAMPLE will be delayed via the
> > > perf_session__process_timed function.
> >
> > Gah. This is nasty. I really prefer the explicit split of instant
> > processed and possibly delayed events. That makes the code clear and
> > easy to extend. I just want to add a new event type at the right place
> > and not worry about magic comparisions in some other place.
>
> Fair enough, I'll split them back out.
>
> > > For instance, suppose we ran this on an old kernel without support for
> > > timestamps on every event (so timestamps are only on sample events):
> > >
> > > perf record -T
> > > perf report
> > >
> > > If perf report tried to process the events in order, all the events
> > > without timestamps would be processed first -- including the
> > > PERF_RECORD_EXIT event, which would cause every sample not to be
> > > attributed. Falling back means we should get no worse than the old
> > > behaviour, while an upgraded kernel will provide the timestamps and
> > > should not fall back.
> >
> > Ok, but you'll break existing code which does only care about sample
> > ordering if you do that at the session level unconditionally.
>
> Good point. I'll give this some thought overnight.
Does the patch below on top of my previous one work for you ?
Thanks,
tglx
---
tools/perf/util/session.c | 70 ++++++++++++++++++++++++++++------------------
tools/perf/util/session.h | 1
2 files changed, 44 insertions(+), 27 deletions(-)
Index: linux-2.6-tip/tools/perf/util/session.c
===================================================================
--- linux-2.6-tip.orig/tools/perf/util/session.c
+++ linux-2.6-tip/tools/perf/util/session.c
@@ -709,6 +709,29 @@ static int perf_session_deliver_event(st
}
}
+static int preprocess_sample_record(struct perf_session *session,
+ event_t *event, struct sample_data *sample)
+{
+ if (event->header.type != PERF_RECORD_SAMPLE)
+ return 0;
+
+ dump_printf("(IP, %d): %d/%d: %#Lx period: %Ld\n", event->header.misc,
+ sample->pid, sample->tid, sample->ip, sample->period);
+
+ if (!(session->sample_type & PERF_SAMPLE_CALLCHAIN))
+ return 0;
+
+ if (!ip_callchain__valid(sample->callchain, event)) {
+ pr_debug("call-chain problem with event, skipping it.\n");
+ ++session->hists.stats.nr_invalid_chains;
+ session->hists.stats.total_invalid_chains += sample->period;
+ return -EINVAL;
+ }
+
+ callchain__dump(sample);
+ return 0;
+}
+
static int perf_session__process_event(struct perf_session *session,
event_t *event,
struct perf_event_ops *ops,
@@ -722,13 +745,6 @@ static int perf_session__process_event(s
if (session->header.needs_swap && event__swap_ops[event->header.type])
event__swap_ops[event->header.type](event);
- if (event->header.type >= PERF_RECORD_MMAP &&
- event->header.type <= PERF_RECORD_SAMPLE) {
- event__parse_sample(event, session, &sample);
- if (dump_trace)
- perf_session__print_tstamp(session, event, &sample);
- }
-
if (event->header.type < PERF_RECORD_HEADER_MAX) {
dump_printf("%#Lx [%#x]: PERF_RECORD_%s",
file_offset, event->header.size,
@@ -738,27 +754,16 @@ static int perf_session__process_event(s
/* These events are processed right away */
switch (event->header.type) {
- case PERF_RECORD_SAMPLE:
- dump_printf("(IP, %d): %d/%d: %#Lx period: %Ld\n",
- event->header.misc,
- sample.pid, sample.tid, sample.ip, sample.period);
-
- if (session->sample_type & PERF_SAMPLE_CALLCHAIN) {
- if (!ip_callchain__valid(sample.callchain, event)) {
- pr_debug("call-chain problem with event, "
- "skipping it.\n");
- ++session->hists.stats.nr_invalid_chains;
- session->hists.stats.total_invalid_chains +=
- sample.period;
- return 0;
- }
-
- callchain__dump(&sample);
- }
- break;
-
case PERF_RECORD_HEADER_ATTR:
- return ops->attr(event, session);
+ /* This updates session->sample_id_all */
+ ret = ops->attr(event, session);
+ if (ops->ordering_requires_timestamps &&
+ ops->ordered_samples && !session->sample_id_all) {
+ session->ordered_samples.next_flush = ULLONG_MAX;
+ flush_sample_queue(session, ops);
+ ops->ordered_samples = false;
+ }
+ return ret;
case PERF_RECORD_HEADER_EVENT_TYPE:
return ops->event_type(event, session);
case PERF_RECORD_HEADER_TRACING_DATA:
@@ -773,6 +778,17 @@ static int perf_session__process_event(s
break;
}
+ /*
+ * For all non synthetized events we get the sample data
+ */
+ event__parse_sample(event, session, &sample);
+ if (dump_trace)
+ perf_session__print_tstamp(session, event, &sample);
+
+ /* Preprocess sample records - precheck callchains */
+ if (preprocess_sample_record(session, event, &sample))
+ return 0;
+
if (ops->ordered_samples) {
ret = perf_session_queue_event(session, event, &sample);
if (ret != -ETIME)
Index: linux-2.6-tip/tools/perf/util/session.h
===================================================================
--- linux-2.6-tip.orig/tools/perf/util/session.h
+++ linux-2.6-tip/tools/perf/util/session.h
@@ -82,6 +82,7 @@ struct perf_event_ops {
build_id;
event_op2 finished_round;
bool ordered_samples;
+ bool ordering_requires_timestamps;
};
struct perf_session *perf_session__new(const char *filename, int mode, bool force, bool repipe);
next prev parent reply other threads:[~2010-12-06 14:41 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-12-04 22:26 [GIT PULL 0/5] perf/core: Support for PERF_SAMPLE_ in all events Arnaldo Carvalho de Melo
2010-12-04 22:26 ` [PATCH 1/5] perf events: Fix event inherit fallout of precalculated headers Arnaldo Carvalho de Melo
2010-12-04 22:26 ` [PATCH 2/5] perf events: Separate the routines handling the PERF_SAMPLE_ identity fields Arnaldo Carvalho de Melo
2010-12-04 22:26 ` [PATCH 3/5] perf events: Make sample_type identity fields available in all PERF_RECORD_ events Arnaldo Carvalho de Melo
2010-12-04 22:26 ` [PATCH 4/5] perf session: Parse sample earlier Arnaldo Carvalho de Melo
2010-12-04 22:26 ` [PATCH 5/5] perf tools: Ask for ID PERF_SAMPLE_ info on all PERF_RECORD_ events Arnaldo Carvalho de Melo
2010-12-05 13:32 ` [Patch] perf: session: Sort all events if ordered_samples=true Thomas Gleixner
2010-12-06 2:37 ` Process events in order if recording from multiple CPUs Ian Munsie
2010-12-06 2:37 ` [PATCH 1/3] perf tool: Better displaying of unresolved DSOs and symbols Ian Munsie
2010-12-07 6:56 ` [tip:perf/core] perf hist: " tip-bot for Ian Munsie
2010-12-06 2:37 ` [PATCH 2/3] perf: Move all output for perf report -D into trace_event Ian Munsie
2010-12-06 2:37 ` [PATCH 3/3] perf record/report: Process events in order Ian Munsie
2010-12-06 9:20 ` Thomas Gleixner
2010-12-06 12:42 ` Ian Munsie
2010-12-06 13:04 ` Thomas Gleixner
2010-12-06 13:11 ` Ian Munsie
2010-12-06 14:41 ` Thomas Gleixner [this message]
2010-12-07 1:15 ` Ian Munsie
2010-12-07 10:47 ` Thomas Gleixner
2010-12-07 10:54 ` Arnaldo Carvalho de Melo
2010-12-07 12:24 ` Thomas Gleixner
2010-12-07 6:57 ` [tip:perf/core] perf session: Sort all events if ordered_samples=true tip-bot for Thomas Gleixner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.LFD.2.00.1012061447080.2653@localhost6.localdomain6 \
--to=tglx@linutronix.de \
--cc=a.p.zijlstra@chello.nl \
--cc=acme@ghostprotocols.net \
--cc=efault@gmx.de \
--cc=eranian@google.com \
--cc=fweisbec@gmail.com \
--cc=imunsie@au1.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=paulus@samba.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®