From: Peter Zijlstra <a.p.zijlstra@chello.nl>
To: Arun Sharma <asharma@fb.com>
Cc: Stephane Eranian <eranian@google.com>,
mingo@elte.hu, William Cohen <wcohen@redhat.com>,
Vince Weaver <vince@deater.net>,
linux-kernel@vger.kernel.org
Subject: Re: [RFC][PATCH 0/6] perf: x86 RDPMC and RDTSC support
Date: Tue, 06 Dec 2011 10:42:29 +0100 [thread overview]
Message-ID: <1323164549.32012.50.camel@twins> (raw)
In-Reply-To: <4EDD50F5.30300@fb.com>
On Mon, 2011-12-05 at 15:17 -0800, Arun Sharma wrote:
>
> I suspect that some optimizations are possible in perf_rotate_context
> that don't involve enabling/disabling the PMU via an IPI for simple
> cases that involve one or two hardware events (eg: fixed counters).
I've got a patch queued that does exactly that.
---
Subject: perf: Avoid a useless pmu_disable() in the perf-tick
From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Wed Nov 16 14:38:16 CET 2011
Gleb writes:
> Currently pmu is disabled and re-enabled on each timer interrupt even
> when no rotation or frequency adjustment is needed. On Intel CPU this
> results in two writes into PERF_GLOBAL_CTRL MSR per tick. On bare metal
> it does not cause significant slowdown, but when running perf in a virtual
> machine it leads to 20% slowdown on my machine.
Cure this by keeping a perf_event_context::nr_freq counter that counts the
number of active events that require frequency adjustments and use this in a
similar fashion to the already existing nr_events != nr_active test in
perf_rotate_context().
By being able to exclude both rotation and frequency adjustments a-priory for
the common case we can avoid the otherwise superfluous PMU disable.
Suggested-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
include/linux/perf_event.h | 1
kernel/events/core.c | 50 +++++++++++++++++++++++++++++----------------
2 files changed, 34 insertions(+), 17 deletions(-)
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -889,6 +889,7 @@ struct perf_event_context {
int nr_active;
int is_active;
int nr_stat;
+ int nr_freq;
int rotate_disable;
atomic_t refcount;
struct task_struct *task;
Index: linux-2.6/kernel/events/core.c
===================================================================
--- linux-2.6.orig/kernel/events/core.c
+++ linux-2.6/kernel/events/core.c
@@ -1127,6 +1127,8 @@ event_sched_out(struct perf_event *event
if (!is_software_event(event))
cpuctx->active_oncpu--;
ctx->nr_active--;
+ if (event->attr.freq && event->attr.sample_freq)
+ ctx->nr_freq--;
if (event->attr.exclusive || !cpuctx->active_oncpu)
cpuctx->exclusive = 0;
}
@@ -1404,6 +1406,8 @@ event_sched_in(struct perf_event *event,
if (!is_software_event(event))
cpuctx->active_oncpu++;
ctx->nr_active++;
+ if (event->attr.freq && event->attr.sample_freq)
+ ctx->nr_freq++;
if (event->attr.exclusive)
cpuctx->exclusive = 1;
@@ -2326,6 +2330,9 @@ static void perf_ctx_adjust_freq(struct
u64 interrupts, now;
s64 delta;
+ if (!ctx->nr_freq)
+ return;
+
list_for_each_entry_rcu(event, &ctx->event_list, event_entry) {
if (event->state != PERF_EVENT_STATE_ACTIVE)
continue;
@@ -2381,12 +2388,14 @@ static void perf_rotate_context(struct p
{
u64 interval = (u64)cpuctx->jiffies_interval * TICK_NSEC;
struct perf_event_context *ctx = NULL;
- int rotate = 0, remove = 1;
+ int rotate = 0, remove = 1, freq = 0;
if (cpuctx->ctx.nr_events) {
remove = 0;
if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
rotate = 1;
+ if (cpuctx->ctx.nr_freq)
+ freq = 1;
}
ctx = cpuctx->task_ctx;
@@ -2394,33 +2403,40 @@ static void perf_rotate_context(struct p
remove = 0;
if (ctx->nr_events != ctx->nr_active)
rotate = 1;
+ if (ctx->nr_freq)
+ freq = 1;
}
+ if (!rotate && !freq)
+ goto done;
+
perf_ctx_lock(cpuctx, cpuctx->task_ctx);
perf_pmu_disable(cpuctx->ctx.pmu);
- perf_ctx_adjust_freq(&cpuctx->ctx, interval);
- if (ctx)
- perf_ctx_adjust_freq(ctx, interval);
-
- if (!rotate)
- goto done;
- cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
- if (ctx)
- ctx_sched_out(ctx, cpuctx, EVENT_FLEXIBLE);
+ if (freq) {
+ perf_ctx_adjust_freq(&cpuctx->ctx, interval);
+ if (ctx)
+ perf_ctx_adjust_freq(ctx, interval);
+ }
+
+ if (rotate) {
+ cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
+ if (ctx)
+ ctx_sched_out(ctx, cpuctx, EVENT_FLEXIBLE);
+
+ rotate_ctx(&cpuctx->ctx);
+ if (ctx)
+ rotate_ctx(ctx);
- rotate_ctx(&cpuctx->ctx);
- if (ctx)
- rotate_ctx(ctx);
+ perf_event_sched_in(cpuctx, ctx, current);
+ }
- perf_event_sched_in(cpuctx, ctx, current);
+ perf_pmu_enable(cpuctx->ctx.pmu);
+ perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
done:
if (remove)
list_del_init(&cpuctx->rotation_list);
-
- perf_pmu_enable(cpuctx->ctx.pmu);
- perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
}
void perf_event_task_tick(void)
next prev parent reply other threads:[~2011-12-06 9:43 UTC|newest]
Thread overview: 48+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-11-21 14:51 [RFC][PATCH 0/6] perf: x86 RDPMC and RDTSC support Peter Zijlstra
2011-11-21 14:51 ` [RFC][PATCH 1/6] perf: Update the mmap control page on mmap() Peter Zijlstra
2011-11-21 14:51 ` [RFC][PATCH 2/6] perf, arch: Rework perf_event_index() Peter Zijlstra
2011-11-21 16:22 ` Eric B Munson
2011-11-21 17:23 ` Will Deacon
2011-11-21 19:18 ` Peter Zijlstra
2011-11-21 20:31 ` Will Deacon
2011-11-21 20:35 ` Peter Zijlstra
2011-11-21 22:43 ` Will Deacon
2011-11-22 11:26 ` Peter Zijlstra
2011-11-22 11:47 ` Will Deacon
2011-11-22 11:49 ` Oleg Strikov
2011-11-22 11:52 ` Will Deacon
2011-11-22 11:56 ` Oleg Strikov
2011-11-22 12:00 ` Oleg Strikov
2011-11-22 12:14 ` Will Deacon
2011-11-22 12:25 ` Oleg Strikov
2011-11-22 11:51 ` Peter Zijlstra
2011-11-22 11:54 ` Will Deacon
2011-11-22 11:48 ` Oleg Strikov
2011-11-21 14:51 ` [RFC][PATCH 3/6] perf, x86: Implement userspace RDPMC Peter Zijlstra
2011-11-21 14:51 ` [RFC][PATCH 4/6] perf, x86: Provide means of disabling " Peter Zijlstra
2011-11-21 14:51 ` [RFC][PATCH 5/6] perf: Extend the mmap control page with time (TSC) fields Peter Zijlstra
2011-12-28 17:55 ` Stephane Eranian
2011-11-21 14:51 ` [RFC][PATCH 6/6] perf, tools: X86 RDPMC, RDTSC test Peter Zijlstra
2011-11-21 15:29 ` Stephane Eranian
2011-11-21 15:37 ` Peter Zijlstra
2011-11-21 16:59 ` Peter Zijlstra
2011-11-21 17:42 ` Stephane Eranian
2011-11-21 15:02 ` [RFC][PATCH 0/6] perf: x86 RDPMC and RDTSC support Vince Weaver
2011-11-21 16:05 ` William Cohen
2011-11-21 16:08 ` William Cohen
2011-12-02 19:26 ` Arun Sharma
2011-12-02 22:22 ` Stephane Eranian
2011-12-05 20:16 ` Arun Sharma
2011-12-05 23:17 ` Arun Sharma
2011-12-06 1:38 ` Stephane Eranian
2011-12-06 9:42 ` Peter Zijlstra [this message]
2011-12-06 21:53 ` Arun Sharma
2011-12-16 22:36 ` Vince Weaver
2011-12-21 12:58 ` Peter Zijlstra
2011-12-21 13:15 ` Ingo Molnar
2011-12-23 20:12 ` Vince Weaver
2011-12-21 15:04 ` Vince Weaver
2011-12-21 21:32 ` Vince Weaver
2011-12-21 21:41 ` Peter Zijlstra
2011-12-21 22:19 ` Vince Weaver
2011-12-21 22:32 ` Peter Zijlstra
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=1323164549.32012.50.camel@twins \
--to=a.p.zijlstra@chello.nl \
--cc=asharma@fb.com \
--cc=eranian@google.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=vince@deater.net \
--cc=wcohen@redhat.com \
/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