From: Peter Zijlstra <peterz@infradead.org>
To: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: mingo@redhat.com, vincent.guittot@linaro.org,
daniel.lezcano@linaro.org,
Dietmar Eggemann <Dietmar.Eggemann@arm.com>,
yuyang.du@intel.com, mturquette@baylibre.com, rjw@rjwysocki.net,
Juri Lelli <Juri.Lelli@arm.com>,
sgurrappadi@nvidia.com, pang.xunlei@zte.com.cn,
linux-kernel@vger.kernel.org, linux-pm@vger.kernel.org
Subject: Re: [RFCv5 PATCH 38/46] sched: scheduler-driven cpu frequency selection
Date: Sat, 15 Aug 2015 14:35:31 +0200 [thread overview]
Message-ID: <20150815123531.GE10304@worktop.programming.kicks-ass.net> (raw)
In-Reply-To: <1436293469-25707-39-git-send-email-morten.rasmussen@arm.com>
On Tue, Jul 07, 2015 at 07:24:21PM +0100, Morten Rasmussen wrote:
> diff --git a/kernel/sched/cpufreq_sched.c b/kernel/sched/cpufreq_sched.c
> new file mode 100644
> index 0000000..5020f24
> --- /dev/null
> +++ b/kernel/sched/cpufreq_sched.c
> @@ -0,0 +1,308 @@
> +/*
> + * Copyright (C) 2015 Michael Turquette <mturquette@linaro.org>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + */
> +
> +#include <linux/cpufreq.h>
> +#include <linux/module.h>
> +#include <linux/kthread.h>
> +#include <linux/percpu.h>
> +#include <linux/irq_work.h>
> +
> +#include "sched.h"
> +
> +#define THROTTLE_NSEC 50000000 /* 50ms default */
> +
> +static DEFINE_PER_CPU(unsigned long, pcpu_capacity);
> +static DEFINE_PER_CPU(struct cpufreq_policy *, pcpu_policy);
> +
> +/**
> + * gov_data - per-policy data internal to the governor
> + * @throttle: next throttling period expiry. Derived from throttle_nsec
> + * @throttle_nsec: throttle period length in nanoseconds
> + * @task: worker thread for dvfs transition that may block/sleep
> + * @irq_work: callback used to wake up worker thread
> + * @freq: new frequency stored in *_sched_update_cpu and used in *_sched_thread
> + *
> + * struct gov_data is the per-policy cpufreq_sched-specific data structure. A
> + * per-policy instance of it is created when the cpufreq_sched governor receives
> + * the CPUFREQ_GOV_START condition and a pointer to it exists in the gov_data
> + * member of struct cpufreq_policy.
> + *
> + * Readers of this data must call down_read(policy->rwsem). Writers must
> + * call down_write(policy->rwsem).
> + */
> +struct gov_data {
> + ktime_t throttle;
> + unsigned int throttle_nsec;
> + struct task_struct *task;
> + struct irq_work irq_work;
> + struct cpufreq_policy *policy;
> + unsigned int freq;
> +};
> +
> +static void cpufreq_sched_try_driver_target(struct cpufreq_policy *policy, unsigned int freq)
> +{
> + struct gov_data *gd = policy->governor_data;
> +
> + /* avoid race with cpufreq_sched_stop */
> + if (!down_write_trylock(&policy->rwsem))
> + return;
> +
> + __cpufreq_driver_target(policy, freq, CPUFREQ_RELATION_L);
> +
> + gd->throttle = ktime_add_ns(ktime_get(), gd->throttle_nsec);
> + up_write(&policy->rwsem);
> +}
That locking truly is disgusting.. why can't we change that?
> +static int cpufreq_sched_thread(void *data)
> +{
> +
> + ret = set_cpus_allowed_ptr(gd->task, policy->related_cpus);
That's not sufficient, you really want to have called kthread_bind() on
these threads, otherwise userspace can change affinity on you.
> +
> + do_exit(0);
I thought kthreads only needed to return...
> +}
> +void cpufreq_sched_set_cap(int cpu, unsigned long capacity)
> +{
> + unsigned int freq_new, cpu_tmp;
> + struct cpufreq_policy *policy;
> + struct gov_data *gd;
> + unsigned long capacity_max = 0;
> +
> + /* update per-cpu capacity request */
> + __this_cpu_write(pcpu_capacity, capacity);
> +
> + policy = cpufreq_cpu_get(cpu);
So this does a down_read_trylock(&cpufreq_rwsem) and a
read_lock_irqsave(&cpufreq_driver_lock), all while holding scheduler
locks.
> + if (cpufreq_driver_might_sleep())
> + irq_work_queue_on(&gd->irq_work, cpu);
> + else
> + cpufreq_sched_try_driver_target(policy, freq_new);
This will then do a down_write_trylock(&policy->rwsem)
> +
> +out:
> + cpufreq_cpu_put(policy);
> + return;
> +}
That is just insane... surely we can replace all that with a wee bit of
RCU logic.
So something like:
DEFINE_MUTEX(cpufreq_mutex);
struct cpufreq_driver *cpufreq_driver;
struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
{
struct cpufreq_driver *driver;
struct cpufreq_policy *policy;
rcu_read_lock();
driver = rcu_dereference(cpufreq_driver);
if (!driver)
goto err;
policy = per_cpu_ptr(driver->policy, cpu);
if (!policy)
goto err;
return policy;
err:
rcu_read_unlock();
return NULL;
}
void cpufreq_cpu_put(struct cpufreq_policy *policy)
{
rcu_read_unlock();
}
void cpufreq_set_driver(struct cpufreq_driver *driver)
{
mutex_lock(&cpufreq_mutex);
rcu_assign_pointer(cpufreq_driver, NULL);
/*
* Wait for everyone to observe the lack of driver; iow. until
* its unused.
*/
synchronize_rcu();
/*
* Now that ye olde driver be gone, install a new one.
*/
if (driver)
rcu_assign_pointer(cpufreq_driver, driver);
mutex_unlock(&cpufreq_mutex);
}
No need for cpufreq_rwsem or cpufreq_driver_lock..
Hmm?
next prev parent reply other threads:[~2015-08-15 19:54 UTC|newest]
Thread overview: 155+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-07 18:23 [RFCv5 PATCH 00/46] sched: Energy cost model for energy-aware scheduling Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 01/46] arm: Frequency invariant scheduler load-tracking support Morten Rasmussen
2015-07-21 15:41 ` [RFCv5, " Leo Yan
2015-07-22 13:31 ` Morten Rasmussen
2015-07-22 14:59 ` Leo Yan
2015-07-23 11:06 ` Morten Rasmussen
2015-07-23 14:22 ` Leo Yan
2015-07-24 9:43 ` Morten Rasmussen
2015-08-03 9:22 ` [RFCv5 PATCH " Vincent Guittot
2015-08-17 15:59 ` Dietmar Eggemann
2015-08-11 9:27 ` Peter Zijlstra
2015-08-14 16:08 ` Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 02/46] sched: Make load tracking frequency scale-invariant Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 03/46] arm: vexpress: Add CPU clock-frequencies to TC2 device-tree Morten Rasmussen
2015-07-08 12:36 ` Jon Medhurst (Tixy)
2015-07-10 13:35 ` Dietmar Eggemann
2015-07-07 18:23 ` [RFCv5 PATCH 04/46] sched: Convert arch_scale_cpu_capacity() from weak function to #define Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 05/46] arm: Update arch_scale_cpu_capacity() to reflect change to define Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 06/46] sched: Make usage tracking cpu scale-invariant Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 07/46] arm: Cpu invariant scheduler load-tracking support Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 08/46] sched: Get rid of scaling usage by cpu_capacity_orig Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 09/46] sched: Track blocked utilization contributions Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 10/46] sched: Include blocked utilization in usage tracking Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 11/46] sched: Remove blocked load and utilization contributions of dying tasks Morten Rasmussen
2015-07-22 6:51 ` Leo Yan
2015-07-22 13:45 ` Morten Rasmussen
2015-08-11 11:39 ` Peter Zijlstra
2015-08-11 14:58 ` Morten Rasmussen
2015-08-11 17:23 ` Peter Zijlstra
2015-08-12 9:08 ` Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 12/46] sched: Initialize CFS task load and usage before placing task on rq Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 13/46] sched: Documentation for scheduler energy cost model Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 14/46] sched: Make energy awareness a sched feature Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 15/46] sched: Introduce energy data structures Morten Rasmussen
2015-07-07 18:23 ` [RFCv5 PATCH 16/46] sched: Allocate and initialize " Morten Rasmussen
2015-08-12 10:04 ` Peter Zijlstra
2015-08-12 17:08 ` Dietmar Eggemann
2015-08-12 10:17 ` Peter Zijlstra
2015-08-12 17:09 ` Dietmar Eggemann
2015-08-12 17:23 ` Peter Zijlstra
2015-07-07 18:24 ` [RFCv5 PATCH 17/46] sched: Introduce SD_SHARE_CAP_STATES sched_domain flag Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 18/46] arm: topology: Define TC2 energy and provide it to the scheduler Morten Rasmussen
2015-08-12 10:33 ` Peter Zijlstra
2015-08-12 18:47 ` Dietmar Eggemann
2015-08-17 9:19 ` [RFCv5, " Leo Yan
2015-08-20 19:19 ` Dietmar Eggemann
2015-07-07 18:24 ` [RFCv5 PATCH 19/46] sched: Compute cpu capacity available at current frequency Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 20/46] sched: Relocated get_cpu_usage() and change return type Morten Rasmussen
2015-08-12 10:59 ` Peter Zijlstra
2015-08-12 14:40 ` Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 21/46] sched: Highest energy aware balancing sched_domain level pointer Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 22/46] sched: Calculate energy consumption of sched_group Morten Rasmussen
2015-08-13 15:34 ` Peter Zijlstra
2015-08-14 10:28 ` Morten Rasmussen
2015-09-02 17:19 ` Leo Yan
2015-09-17 16:41 ` Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 23/46] sched: Extend sched_group_energy to test load-balancing decisions Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 24/46] sched: Estimate energy impact of scheduling decisions Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 25/46] sched: Add over-utilization/tipping point indicator Morten Rasmussen
2015-08-13 17:35 ` Peter Zijlstra
2015-08-14 13:02 ` Morten Rasmussen
2015-09-29 20:08 ` Steve Muckle
2015-10-09 12:49 ` Morten Rasmussen
2015-08-17 13:10 ` Leo Yan
2015-07-07 18:24 ` [RFCv5 PATCH 26/46] sched: Store system-wide maximum cpu capacity in root domain Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 27/46] sched, cpuidle: Track cpuidle state index in the scheduler Morten Rasmussen
2015-07-21 6:41 ` Leo Yan
2015-07-21 15:16 ` Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 28/46] sched: Count number of shallower idle-states in struct sched_group_energy Morten Rasmussen
2015-08-13 18:10 ` Peter Zijlstra
2015-08-14 19:08 ` Sai Gurrappadi
2015-07-07 18:24 ` [RFCv5 PATCH 29/46] sched: Determine the current sched_group idle-state Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 30/46] sched: Add cpu capacity awareness to wakeup balancing Morten Rasmussen
2015-08-13 18:24 ` Peter Zijlstra
2015-08-14 16:20 ` Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 31/46] sched: Consider spare cpu capacity at task wake-up Morten Rasmussen
2015-07-21 0:37 ` Sai Gurrappadi
2015-07-21 15:12 ` Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 32/46] sched: Energy-aware wake-up task placement Morten Rasmussen
2015-07-17 0:10 ` Sai Gurrappadi
2015-07-20 15:38 ` Morten Rasmussen
2015-08-17 16:23 ` Leo Yan
2015-09-02 17:11 ` Leo Yan
2015-09-18 10:34 ` Dietmar Eggemann
2015-09-20 18:39 ` Steve Muckle
2015-09-20 22:03 ` Leo Yan
2015-09-29 0:15 ` Steve Muckle
2015-07-07 18:24 ` [RFCv5 PATCH 33/46] sched: Consider a not over-utilized energy-aware system as balanced Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 34/46] sched: Enable idle balance to pull single task towards cpu with higher capacity Morten Rasmussen
2015-08-15 9:15 ` Peter Zijlstra
2015-07-07 18:24 ` [RFCv5 PATCH 35/46] sched: Disable energy-unfriendly nohz kicks Morten Rasmussen
2015-08-15 9:33 ` Peter Zijlstra
2015-07-07 18:24 ` [RFCv5 PATCH 36/46] sched: Prevent unnecessary active balance of single task in sched group Morten Rasmussen
2015-08-15 9:46 ` Peter Zijlstra
2015-07-07 18:24 ` [RFCv5 PATCH 37/46] cpufreq: introduce cpufreq_driver_might_sleep Morten Rasmussen
2015-07-08 15:09 ` Michael Turquette
2015-07-07 18:24 ` [RFCv5 PATCH 38/46] sched: scheduler-driven cpu frequency selection Morten Rasmussen
2015-07-08 15:09 ` Michael Turquette
2015-08-11 2:14 ` Leo Yan
2015-08-11 8:59 ` Juri Lelli
2015-08-15 12:35 ` Peter Zijlstra [this message]
2015-09-04 13:27 ` Juri Lelli
2015-09-14 15:57 ` Juri Lelli
2015-09-15 13:45 ` Peter Zijlstra
2015-09-15 16:22 ` Juri Lelli
2015-08-15 13:05 ` Peter Zijlstra
2015-08-25 10:45 ` Juri Lelli
2015-10-08 0:14 ` Steve Muckle
2015-10-08 9:41 ` Juri Lelli
2015-09-28 16:48 ` Punit Agrawal
2015-09-29 0:26 ` Steve Muckle
2015-07-07 18:24 ` [RFCv5 PATCH 39/46] sched/cpufreq_sched: use static key for " Morten Rasmussen
2015-07-08 15:19 ` Michael Turquette
2015-07-10 9:50 ` Juri Lelli
2015-08-15 12:40 ` Peter Zijlstra
2015-07-07 18:24 ` [RFCv5 PATCH 40/46] sched/cpufreq_sched: compute freq_new based on capacity_orig_of() Morten Rasmussen
2015-07-08 15:22 ` Michael Turquette
2015-07-09 16:21 ` Juri Lelli
2015-08-15 12:46 ` Peter Zijlstra
2015-08-16 4:03 ` Michael Turquette
2015-08-16 20:24 ` Peter Zijlstra
2015-08-17 12:19 ` Juri Lelli
2015-10-13 19:47 ` Steve Muckle
2015-07-07 18:24 ` [RFCv5 PATCH 41/46] sched/fair: add triggers for OPP change requests Morten Rasmussen
2015-07-08 15:42 ` Michael Turquette
2015-07-09 16:52 ` Juri Lelli
2015-08-04 13:41 ` Vincent Guittot
2015-08-10 13:43 ` Juri Lelli
2015-08-10 15:07 ` Vincent Guittot
2015-08-11 9:08 ` Juri Lelli
2015-08-11 11:41 ` Vincent Guittot
2015-08-11 15:07 ` Juri Lelli
2015-08-11 16:37 ` Vincent Guittot
2015-08-12 15:15 ` Juri Lelli
2015-08-13 12:08 ` Vincent Guittot
2015-08-14 11:39 ` Juri Lelli
2015-08-17 9:43 ` Vincent Guittot
2015-08-15 12:48 ` Peter Zijlstra
2015-08-16 3:50 ` Michael Turquette
2015-08-17 18:22 ` Rafael J. Wysocki
2015-07-07 18:24 ` [RFCv5 PATCH 42/46] sched/{core,fair}: trigger OPP change request on fork() Morten Rasmussen
2015-07-07 18:24 ` [RFCv5 PATCH 43/46] sched/{fair,cpufreq_sched}: add reset_capacity interface Morten Rasmussen
2015-10-08 20:40 ` Steve Muckle
2015-10-09 9:14 ` Juri Lelli
2015-10-12 19:02 ` Steve Muckle
2015-07-07 18:24 ` [RFCv5 PATCH 44/46] sched/fair: jump to max OPP when crossing UP threshold Morten Rasmussen
2015-07-08 16:40 ` Michael Turquette
2015-07-08 16:47 ` Michael Turquette
2015-07-10 10:17 ` Juri Lelli
2015-07-07 18:24 ` [RFCv5 PATCH 45/46] sched/cpufreq_sched: modify pcpu_capacity handling Morten Rasmussen
2015-07-08 16:42 ` Michael Turquette
2015-07-09 16:55 ` Juri Lelli
2015-08-16 20:35 ` Peter Zijlstra
2015-08-17 11:16 ` Juri Lelli
2015-07-07 18:24 ` [RFCv5 PATCH 46/46] sched/fair: cpufreq_sched triggers for load balancing Morten Rasmussen
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=20150815123531.GE10304@worktop.programming.kicks-ass.net \
--to=peterz@infradead.org \
--cc=Dietmar.Eggemann@arm.com \
--cc=Juri.Lelli@arm.com \
--cc=daniel.lezcano@linaro.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pm@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=morten.rasmussen@arm.com \
--cc=mturquette@baylibre.com \
--cc=pang.xunlei@zte.com.cn \
--cc=rjw@rjwysocki.net \
--cc=sgurrappadi@nvidia.com \
--cc=vincent.guittot@linaro.org \
--cc=yuyang.du@intel.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;
as well as URLs for NNTP newsgroup(s).