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

Groups > linux.kernel > #1422011 > unrolled thread

[PATCH 0/3] sched: Enhance/fix sched_avg_update()

Back to article view | Back to linux.kernel

Contents

  [PATCH 0/3] sched: Enhance/fix sched_avg_update() Frederic Weisbecker <fweisbec@gmail.com> - 2016-06-14 17:30 +0200
    [NOT-FOR-MERGE-PATCH 3/3] Debug: Alternate old/new version of sched_avg_update calls for profiling Frederic Weisbecker <fweisbec@gmail.com> - 2016-06-14 17:30 +0200
    [PATCH 1/3] sched: Introduce sched_time_avg_ns minimal value Frederic Weisbecker <fweisbec@gmail.com> - 2016-06-14 17:30 +0200

#1422011 — [PATCH 0/3] sched: Enhance/fix sched_avg_update()

Hi,

Here is a fix on sched_time_avg_ms sysctl value and an enhancement on
nohz scenarii.

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
	sched/core

HEAD: 8e6e088b73c2172c45cc3aadab3d4971ae47215b

Thanks,
	Frederic
---

Frederic Weisbecker (3):
      sched: Introduce sched_time_avg_ns minimal value
      sched: Unloop sched avg decaying
      Debug: Alternate old/new version of sched_avg_update calls for profiling


 kernel/sched/core.c  | 18 +++++++++++++++++-
 kernel/sched/fair.c  |  7 ++++++-
 kernel/sched/sched.h |  8 +++++++-
 kernel/sysctl.c      |  3 ++-
 4 files changed, 32 insertions(+), 4 deletions(-)

[toc] | [next] | [standalone]

#1422016 — [NOT-FOR-MERGE-PATCH 3/3] Debug: Alternate old/new version of sched_avg_update calls for profiling

This way we can measure with the function profiler.

	trace_stat/function0:  sched_avg_update_old                   764    223.936 us      0.293 us        0.017 us
	trace_stat/function0:  sched_avg_update                       764    199.818 us      0.261 us        0.006 us
	trace_stat/function1:  sched_avg_update_old                   252    99.225 us       0.393 us        0.106 us
	trace_stat/function1:  sched_avg_update                       252    59.080 us       0.234 us        0.008 us
	trace_stat/function2:  sched_avg_update_old                   606    202.465 us      0.334 us        0.007 us
	trace_stat/function2:  sched_avg_update                       605    184.083 us      0.304 us        0.002 us
	trace_stat/function3:  sched_avg_update_old                   902    261.790 us      0.290 us        0.010 us
	trace_stat/function3:  sched_avg_update                       901    238.253 us      0.264 us        0.004 us
	trace_stat/function4:  sched_avg_update_old                   120    50.391 us       0.419 us        0.095 us
	trace_stat/function4:  sched_avg_update                       120    35.947 us       0.299 us        0.004 us
	trace_stat/function5:  sched_avg_update_old                   581    195.280 us      0.336 us        0.011 us
	trace_stat/function5:  sched_avg_update                       582    173.594 us      0.298 us        0.003 us
	trace_stat/function6:  sched_avg_update_old                   202    45.201 us       0.223 us        0.049 us
	trace_stat/function6:  sched_avg_update                       201    37.892 us       0.188 us        0.013 us
	trace_stat/function7:  sched_avg_update_old                   200    80.731 us       0.403 us        0.477 us
	trace_stat/function7:  sched_avg_update                       200    41.759 us       0.208 us        0.012 us

Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Not-Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 kernel/sched/core.c  | 16 ++++++++++++++++
 kernel/sched/fair.c  |  7 ++++++-
 kernel/sched/sched.h |  8 +++++++-
 3 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 0c0578a..5410e1bb 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -663,6 +663,22 @@ bool sched_can_stop_tick(struct rq *rq)
 }
 #endif /* CONFIG_NO_HZ_FULL */
 
+void sched_avg_update_old(struct rq *rq)
+{
+	s64 period = sched_avg_period();
+
+	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
+		/*
+		 * Inline assembly required to prevent the compiler
+		 * optimising this loop into a divmod call.
+		 * See __iter_div_u64_rem() for another example of this.
+		 */
+		asm("" : "+rm" (rq->age_stamp));
+		rq->age_stamp += period;
+		rq->rt_avg /= 2;
+	}
+}
+
 void sched_avg_update(struct rq *rq)
 {
 	s64 period = sched_avg_period();
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index c6dd8ba..1b487bb 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4607,6 +4607,7 @@ decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
  * see decay_load_misses(). For NOHZ_FULL we get to subtract and add the extra
  * term.
  */
+DEFINE_PER_CPU(int, cpu_lodd);
 static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
 			    unsigned long pending_updates)
 {
@@ -4647,7 +4648,11 @@ static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
 		this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
 	}
 
-	sched_avg_update(this_rq);
+	if (__this_cpu_read(cpu_lodd) % 2)
+		sched_avg_update(this_rq);
+	else
+		sched_avg_update_old(this_rq);
+	__this_cpu_add(cpu_lodd, 1);
 }
 
 /* Used instead of source_load when we know the type == 0 */
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 72f1f30..b7e197b 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1427,6 +1427,7 @@ static inline int hrtick_enabled(struct rq *rq)
 #endif /* CONFIG_SCHED_HRTICK */
 
 #ifdef CONFIG_SMP
+extern void sched_avg_update_old(struct rq *rq);
 extern void sched_avg_update(struct rq *rq);
 
 #ifndef arch_scale_freq_capacity
@@ -1448,10 +1449,15 @@ unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu)
 }
 #endif
 
+DECLARE_PER_CPU(int, cpu_lodd);
 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta)
 {
 	rq->rt_avg += rt_delta * arch_scale_freq_capacity(NULL, cpu_of(rq));
-	sched_avg_update(rq);
+	if (__this_cpu_read(cpu_lodd) % 2)
+		sched_avg_update(rq);
+	else
+		sched_avg_update_old(rq);
+	__this_cpu_add(cpu_lodd, 1);
 }
 #else
 static inline void sched_rt_avg_update(struct rq *rq, u64 rt_delta) { }
-- 
2.7.0

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

#1422017 — [PATCH 1/3] sched: Introduce sched_time_avg_ns minimal value

Writing 0 to sysctl value "kernel.sched_time_avg_ms" triggers a lockup:

	NMI watchdog: Watchdog detected hard LOCKUP on cpu 4
	Modules linked in:
	irq event stamp: 81894
	hardirqs last  enabled at (81893): [<ffffffff8111af88>] rcu_idle_exit+0x68/0xa0
	hardirqs last disabled at (81894): [<ffffffff8113536e>] tick_nohz_idle_exit+0x2e/0x110
	softirqs last  enabled at (81870): [<ffffffff810ab8a1>] _local_bh_enable+0x21/0x50
	softirqs last disabled at (81869): [<ffffffff810ace2b>] irq_enter+0x4b/0x70
	CPU: 4 PID: 0 Comm: swapper/4 Not tainted 4.6.0-rc4+ #269
	Hardware name: MSI MS-7850/Z87-G41 PC Mate(MS-7850), BIOS V1.3 08/18/2013
	0000000000000000 ffff88021fb05b80 ffffffff8141b023 0000000000000000
	0000000000000004 ffff88021fb05ba0 ffffffff81163fcf 0000000000000000
	ffff88021fb05c40 ffff88021fb05be0 ffffffff8119ae7c 00000001810077ce
	Call Trace:
	<NMI>  [<ffffffff8141b023>] dump_stack+0x85/0xc2
	[<ffffffff81163fcf>] watchdog_overflow_callback+0x13f/0x160
	[<ffffffff8119ae7c>] __perf_event_overflow+0x9c/0x1f0
	[<ffffffff8119baa4>] perf_event_overflow+0x14/0x20
	[<ffffffff8100d235>] intel_pmu_handle_irq+0x1d5/0x4a0
	[<ffffffff810062ad>] perf_event_nmi_handler+0x2d/0x50
	[<ffffffff81174397>] ? tracing_generic_entry_update+0x97/0xb0
	[<ffffffff81039cef>] nmi_handle+0xbf/0x2f0
	[<ffffffff81039c35>] ? nmi_handle+0x5/0x2f0
	[<ffffffff81174397>] ? tracing_generic_entry_update+0x97/0xb0
	[<ffffffff8103a171>] default_do_nmi+0x71/0x1b0
	[<ffffffff8103a3c5>] do_nmi+0x115/0x170
	[<ffffffff8194b511>] end_repeat_nmi+0x1a/0x1e
	[<ffffffff81174397>] ? tracing_generic_entry_update+0x97/0xb0
	[<ffffffff81174397>] ? tracing_generic_entry_update+0x97/0xb0
	[<ffffffff81174397>] ? tracing_generic_entry_update+0x97/0xb0
	<<EOE>>  [<ffffffff81177d3f>] ? trace_buffer_lock_reserve+0x3f/0x60
	[<ffffffff810d3d51>] ? sched_avg_update+0x51/0xc0
	[<ffffffff811781e5>] __trace_bputs+0x75/0x100
	[<ffffffff810d3d79>] sched_avg_update+0x79/0xc0
	[<ffffffff810dedc3>] cpu_load_update+0xa3/0xd0
	[<ffffffff810e694a>] cpu_load_update_nohz_stop+0x7a/0x80
	[<ffffffff811353b7>] tick_nohz_idle_exit+0x77/0x110
	[<ffffffff810f0866>] cpu_startup_entry+0x176/0x420
	[<ffffffff810589f4>] start_secondary+0x104/0x110

This is due to the loop in sched_avg_update() that fails to make any
progress when the update delay is 0.

Since this is non-sense to set that value to 0 anyway, force a minimum
of at least 1 millisecond.

Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 kernel/sysctl.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 87b2fc3..b730dd6 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -345,7 +345,8 @@ static struct ctl_table kern_table[] = {
 		.data		= &sysctl_sched_time_avg,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= proc_dointvec,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &one,
 	},
 	{
 		.procname	= "sched_shares_window_ns",
-- 
2.7.0

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web