Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1538591 > unrolled thread

[PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2)

Started byNamhyung Kim <namhyung@kernel.org>
First post2016-12-08 15:50 +0100
Last post2016-12-10 17:40 +0100
Articles 4 — 2 participants

Back to article view | Back to linux.kernel

Contents

  [PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2) Namhyung Kim <namhyung@kernel.org> - 2016-12-08 15:50 +0100
    [PATCH v2 2/6] perf sched timehist: Introduce struct idle_time_data Namhyung Kim <namhyung@kernel.org> - 2016-12-08 16:00 +0100
    [PATCH v2 3/6] perf sched timehist: Save callchain when entering idle Namhyung Kim <namhyung@kernel.org> - 2016-12-08 16:00 +0100
    Re: [PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option  (v2) David Ahern <dsahern@gmail.com> - 2016-12-10 17:40 +0100

#1538591 — [PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2)

FromNamhyung Kim <namhyung@kernel.org>
Date2016-12-08 15:50 +0100
Subject[PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2)
Message-ID<sM8bf-7j1-3@gated-at.bofh.it>
Hi,

This patchset implements the idle hist feature which analyze reason of system
idle.  Sometimes I need to investigate what makes CPUs to go idle even though
I have jobs to do.  It may be due to I/O, waiting on lock or whatever.

To identify the reasons it only accounts events related to idle task.  Also it
shows callchains when entering to idle and time how long it's in the idle.
Although it's not perfect, it works well to help finding the reasons.

With --idle-hist option, only idle task's runtime is accounted to previous
task (which makes the cpu go to idle).  With --summary(-only) option, you can
see the total idle stat by task or callchains.

 * v2 changes
  - check prev_pid instead of sample->pid for idle task (due to race in exit path)
  - factor idle task initialization  (David)

The example output looks like below:

  Idle-time summary
                            comm  parent  sched-out    idle-time   min-idle    avg-idle    max-idle  stddev  migrations
                                            (count)       (msec)     (msec)      (msec)      (msec)       %
  ---------------------------------------------------------------------------------------------------------------------
                  rcu_preempt[7]       2         95      550.872      0.011       5.798      23.146    7.63       0
                 migration/1[16]       2          1       15.558     15.558      15.558      15.558    0.00       0
                  khugepaged[39]       2          1        3.062      3.062       3.062       3.062    0.00       0
               kworker/0:1H[124]       2          2        4.728      0.611       2.364       4.116   74.12       0
            systemd-journal[167]       1          1        4.510      4.510       4.510       4.510    0.00       0
              kworker/u16:3[558]       2         13       74.737      0.080       5.749      12.960   21.96       0
             irq/34-iwlwifi[628]       2         21      118.403      0.032       5.638      23.990   24.00       0
              kworker/u17:0[673]       2          1        3.523      3.523       3.523       3.523    0.00       0
                dbus-daemon[722]       1          1        6.743      6.743       6.743       6.743    0.00       0
                    ifplugd[741]       1          1       58.826     58.826      58.826      58.826    0.00       0
            wpa_supplicant[1490]       1          1       13.302     13.302      13.302      13.302    0.00       0
               wpa_actiond[1492]       1          2        4.064      0.168       2.032       3.896   91.72       0
                   dockerd[1500]       1          1        0.055      0.055       0.055       0.055    0.00       0
              dockerd[1516/1500]    1500          1        5.732      5.732       5.732       5.732    0.00       0
              dockerd[1571/1500]    1500          1        0.020      0.020       0.020       0.020    0.00       0
  ...
  
  Idle stats:
      CPU  0 idle for    902.195  msec
      CPU  1 idle for    899.932  msec
      CPU  2 idle for    941.218  msec
      CPU  3 idle for    822.453  msec
  
  Idle stats by callchain:
    CPU  0:   902.195  msec
    Idle time (msec)     Count  Callchains
    ----------------  --------  --------------------------------------------------
             370.589        69  futex_wait_queue_me <- futex_wait <- do_futex <- sys_futex <- entry_SYSCALL_64_fastpath
             178.799        17  worker_thread <- kthread <- ret_from_fork
             128.352        17  schedule_timeout <- rcu_gp_kthread <- kthread <- ret_from_fork
             125.111        19  schedule_hrtimeout_range_clock <- schedule_hrtimeout_range <- poll_schedule_timeout <- do_select <- core_sys_select
              71.599        50  schedule_hrtimeout_range_clock <- schedule_hrtimeout_range <- poll_schedule_timeout <- do_sys_poll <- sys_poll
              23.146         1  rcu_gp_kthread <- kthread <- ret_from_fork
               4.510         1  schedule_hrtimeout_range_clock <- schedule_hrtimeout_range <- ep_poll <- sys_epoll_wait <- do_syscall_64
               0.085         1  schedule_hrtimeout_range_clock <- schedule_hrtimeout_range <- poll_schedule_timeout <- do_sys_poll <- do_restart_poll
  ...


The code is available at 'perf/timehist-idle-v2' branch in my tree:

  git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git 

Any comments are welcome!

Thanks,
Namhyung


Namhyung Kim (6):
  perf sched timehist: Split is_idle_sample()
  perf sched timehist: Introduce struct idle_time_data
  perf sched timehist: Save callchain when entering idle
  perf sched timehist: Skip non-idle events when necessary
  perf sched timehist: Add -I/--idle-hist option
  perf sched timehist: Show callchains for idle stat

 tools/perf/Documentation/perf-sched.txt |   4 +
 tools/perf/builtin-sched.c              | 252 +++++++++++++++++++++++++++-----
 2 files changed, 222 insertions(+), 34 deletions(-)

-- 
2.10.1

[toc] | [next] | [standalone]

#1538594 — [PATCH v2 2/6] perf sched timehist: Introduce struct idle_time_data

FromNamhyung Kim <namhyung@kernel.org>
Date2016-12-08 16:00 +0100
Subject[PATCH v2 2/6] perf sched timehist: Introduce struct idle_time_data
Message-ID<sM8kW-7mt-7@gated-at.bofh.it>
In reply to#1538591
The struct idle_time_data is to keep idle stats with callchains entering
to the idle task.  The normal thread_runtime calculation is done
transparently since it extends the struct thread_runtime.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/builtin-sched.c | 37 +++++++++++++++++++++++++++++++++----
 1 file changed, 33 insertions(+), 4 deletions(-)

diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
index 46aa4705f321..6f0687dd392d 100644
--- a/tools/perf/builtin-sched.c
+++ b/tools/perf/builtin-sched.c
@@ -230,6 +230,15 @@ struct evsel_runtime {
 	u32 ncpu;       /* highest cpu slot allocated */
 };
 
+/* per cpu idle time data */
+struct idle_thread_runtime {
+	struct thread_runtime tr;
+	struct thread *last_thread;
+	struct rb_root sorted_root;
+	struct callchain_root callchain;
+	struct callchain_cursor cursor;
+};
+
 /* track idle times per cpu */
 static struct thread **idle_threads;
 static int idle_max_cpu;
@@ -1997,13 +2006,31 @@ static void save_task_callchain(struct perf_sched *sched,
 	}
 }
 
+static int init_idle_thread(struct thread *thread)
+{
+	struct idle_thread_runtime *itr;
+
+	thread__set_comm(thread, idle_comm, 0);
+
+	itr = zalloc(sizeof(*itr));
+	if (itr == NULL)
+		return -ENOMEM;
+
+	init_stats(&itr->tr.run_stats);
+	callchain_init(&itr->callchain);
+	callchain_cursor_reset(&itr->cursor);
+	thread__set_priv(thread, itr);
+
+	return 0;
+}
+
 /*
  * Track idle stats per cpu by maintaining a local thread
  * struct for the idle task on each cpu.
  */
 static int init_idle_threads(int ncpu)
 {
-	int i;
+	int i, ret;
 
 	idle_threads = zalloc(ncpu * sizeof(struct thread *));
 	if (!idle_threads)
@@ -2017,7 +2044,9 @@ static int init_idle_threads(int ncpu)
 		if (idle_threads[i] == NULL)
 			return -ENOMEM;
 
-		thread__set_comm(idle_threads[i], idle_comm, 0);
+		ret = init_idle_thread(idle_threads[i]);
+		if (ret < 0)
+			return ret;
 	}
 
 	return 0;
@@ -2064,8 +2093,8 @@ static struct thread *get_idle_thread(int cpu)
 		idle_threads[cpu] = thread__new(0, 0);
 
 		if (idle_threads[cpu]) {
-			idle_threads[cpu]->tid = 0;
-			thread__set_comm(idle_threads[cpu], idle_comm, 0);
+			if (init_idle_thread(idle_threads[cpu]) < 0)
+				return NULL;
 		}
 	}
 
-- 
2.10.1

[toc] | [prev] | [next] | [standalone]

#1538598 — [PATCH v2 3/6] perf sched timehist: Save callchain when entering idle

FromNamhyung Kim <namhyung@kernel.org>
Date2016-12-08 16:00 +0100
Subject[PATCH v2 3/6] perf sched timehist: Save callchain when entering idle
Message-ID<sM8kW-7mt-31@gated-at.bofh.it>
In reply to#1538591
In order to investigate reason of idle, it needs to keep the callchains
when entering to idle.  This can be identified sched_switch event having
next_pid as 0.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 tools/perf/builtin-sched.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
index 6f0687dd392d..4c5c58ce8958 100644
--- a/tools/perf/builtin-sched.c
+++ b/tools/perf/builtin-sched.c
@@ -200,6 +200,7 @@ struct perf_sched {
 	/* options for timehist command */
 	bool		summary;
 	bool		summary_only;
+	bool		idle_hist;
 	bool		show_callchain;
 	unsigned int	max_stack;
 	bool		show_cpu_visual;
@@ -2101,6 +2102,29 @@ static struct thread *get_idle_thread(int cpu)
 	return idle_threads[cpu];
 }
 
+static void save_idle_callchain(struct thread *thread,
+				struct perf_sample *sample)
+{
+	struct thread *idle;
+	struct idle_thread_runtime *itr;
+
+	if (!symbol_conf.use_callchain || sample->callchain == NULL)
+		return;
+
+	idle = get_idle_thread(sample->cpu);
+	if (idle == NULL) {
+		pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
+		return;
+	}
+
+	itr = thread__priv(idle);
+	if (itr == NULL)
+		return;
+
+	itr->last_thread = thread;
+	callchain_cursor__copy(&itr->cursor, &callchain_cursor);
+}
+
 /*
  * handle runtime stats saved per thread
  */
@@ -2154,6 +2178,11 @@ static struct thread *timehist_get_thread(struct perf_sched *sched,
 		}
 
 		save_task_callchain(sched, sample, evsel, machine);
+		if (sched->idle_hist) {
+			/* copy task callchain when entering to idle */
+			if (perf_evsel__intval(evsel, sample, "next_pid") == 0)
+				save_idle_callchain(thread, sample);
+		}
 	}
 
 	return thread;
-- 
2.10.1

[toc] | [prev] | [next] | [standalone]

#1539858 — Re: [PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2)

FromDavid Ahern <dsahern@gmail.com>
Date2016-12-10 17:40 +0100
SubjectRe: [PATCHSET 0/6] perf sched timehist: Introduce --idle-hist option (v2)
Message-ID<sMSQO-4zr-11@gated-at.bofh.it>
In reply to#1538591
On 12/8/16 7:47 AM, Namhyung Kim wrote:
> Hi,
> 
> This patchset implements the idle hist feature which analyze reason of system
> idle.  Sometimes I need to investigate what makes CPUs to go idle even though
> I have jobs to do.  It may be due to I/O, waiting on lock or whatever.
> 

...

> Namhyung Kim (6):
>   perf sched timehist: Split is_idle_sample()
>   perf sched timehist: Introduce struct idle_time_data
>   perf sched timehist: Save callchain when entering idle
>   perf sched timehist: Skip non-idle events when necessary
>   perf sched timehist: Add -I/--idle-hist option
>   perf sched timehist: Show callchains for idle stat
> 
>  tools/perf/Documentation/perf-sched.txt |   4 +
>  tools/perf/builtin-sched.c              | 252 +++++++++++++++++++++++++++-----
>  2 files changed, 222 insertions(+), 34 deletions(-)
> 

LGTM

Acked-by: David Ahern <dsahern@gmail.com>

Suggested improvement: Add the length of the time window. ie., data collected (or analyzed over if --time is used) for N.M seconds. Puts the amount of idle time into perspective.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web