From: Daniel Lezcano <daniel.lezcano@linaro.org>
To: Viresh Kumar <viresh.kumar@linaro.org>, rjw@sisk.pl
Cc: linaro-kernel@lists.linaro.org, patches@linaro.org,
linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 19/21] cpuidle: create list of registered drivers
Date: Thu, 26 Sep 2013 00:30:51 +0200 [thread overview]
Message-ID: <5243641B.6050604@linaro.org> (raw)
In-Reply-To: <7e9565e323c1e065eeb2a2a1075967bd21ee6066.1379779777.git.viresh.kumar@linaro.org>
On 09/22/2013 03:21 AM, Viresh Kumar wrote:
> Currently we have multiple definitions of few routines based on following config
> option: CONFIG_CPU_IDLE_MULTIPLE_DRIVERS.
>
> These are present to save space by not creating per-cpu variable for platforms
> which need only one cpuidle driver to be registered for all CPUs.
>
> But this setup has a problem. For ARM multi-platform kernel use case this option
> will get enabled and so we will have per-cpu variables even for platforms that
> don't need it.
>
> The bigger problem is two separate code paths for such platforms for single &
> multi platform kernels. Which doesn't sound good.
>
> A better way of solving this problem would be to create cpuidle driver's list
> that can be used to manage all information we need. Then we don't really have to
> write any special code for handling platforms with
> CONFIG_CPU_IDLE_MULTIPLE_DRIVERS option set.
>
> This patch does it.
If you introduce a list, you will have to introduce a lock to protect
it. This lock will be in the fast path cpuidle_idle_call with the
get_driver function and conforming to the comment: "NOTE: no locks or
semaphores should be used here".
A lock has been introduced in this function already and the system hangs
with 1024 cpus.
> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
> ---
> drivers/cpuidle/driver.c | 106 ++++++++++++-----------------------------------
> include/linux/cpuidle.h | 1 +
> 2 files changed, 27 insertions(+), 80 deletions(-)
>
> diff --git a/drivers/cpuidle/driver.c b/drivers/cpuidle/driver.c
> index a4a93b4..320b4ec 100644
> --- a/drivers/cpuidle/driver.c
> +++ b/drivers/cpuidle/driver.c
> @@ -18,10 +18,19 @@
> #include "cpuidle.h"
>
> DEFINE_SPINLOCK(cpuidle_driver_lock);
> +static LIST_HEAD(cpuidle_detected_drivers);
>
> -#ifdef CONFIG_CPU_IDLE_MULTIPLE_DRIVERS
> +static inline struct cpuidle_driver *
> +__cpuidle_get_driver(const struct cpumask *cpumask)
> +{
> + struct cpuidle_driver *drv;
> +
> + list_for_each_entry(drv, &cpuidle_detected_drivers, driver_list)
> + if (cpumask_intersects(drv->cpumask, cpumask))
> + return drv;
>
> -static DEFINE_PER_CPU(struct cpuidle_driver *, cpuidle_drivers);
> + return NULL;
> +}
>
> /**
> * __cpuidle_get_cpu_driver - return the cpuidle driver tied to a CPU.
> @@ -32,103 +41,39 @@ static DEFINE_PER_CPU(struct cpuidle_driver *, cpuidle_drivers);
> */
> static inline struct cpuidle_driver *__cpuidle_get_cpu_driver(int cpu)
> {
> - return per_cpu(cpuidle_drivers, cpu);
> + return __cpuidle_get_driver(cpumask_of(cpu));
> }
>
> /**
> - * __cpuidle_unset_driver - unset per CPU driver variables.
> + * __cpuidle_add_driver - adds a cpuidle driver to list.
> * @drv: a valid pointer to a struct cpuidle_driver
> *
> - * For each CPU in the driver's CPU mask, unset the registered driver per CPU
> - * variable. If @drv is different from the registered driver, the corresponding
> - * variable is not cleared.
> - */
> -static inline void __cpuidle_unset_driver(struct cpuidle_driver *drv)
> -{
> - int cpu;
> -
> - for_each_cpu(cpu, drv->cpumask) {
> -
> - if (drv != __cpuidle_get_cpu_driver(cpu))
> - continue;
> -
> - per_cpu(cpuidle_drivers, cpu) = NULL;
> - }
> -}
> -
> -/**
> - * __cpuidle_set_driver - set per CPU driver variables for the given driver.
> - * @drv: a valid pointer to a struct cpuidle_driver
> - *
> - * For each CPU in the driver's cpumask, unset the registered driver per CPU
> - * to @drv.
> + * Adds cpuidle driver to cpuidle_detected_drivers list if no driver is already
> + * registered for any CPUs present in drv->cpumask.
> *
> * Returns 0 on success, -EBUSY if the CPUs have driver(s) already.
> */
> -static inline int __cpuidle_set_driver(struct cpuidle_driver *drv)
> -{
> - int cpu;
> -
> - for_each_cpu(cpu, drv->cpumask) {
> -
> - if (__cpuidle_get_cpu_driver(cpu)) {
> - __cpuidle_unset_driver(drv);
> - return -EBUSY;
> - }
> -
> - per_cpu(cpuidle_drivers, cpu) = drv;
> - }
> -
> - return 0;
> -}
> -
> -#else
> -
> -static struct cpuidle_driver *cpuidle_curr_driver;
> -
> -/**
> - * __cpuidle_get_cpu_driver - return the global cpuidle driver pointer.
> - * @cpu: ignored without the multiple driver support
> - *
> - * Return a pointer to a struct cpuidle_driver object or NULL if no driver was
> - * previously registered.
> - */
> -static inline struct cpuidle_driver *__cpuidle_get_cpu_driver(int cpu)
> -{
> - return cpuidle_curr_driver;
> -}
> -
> -/**
> - * __cpuidle_set_driver - assign the global cpuidle driver variable.
> - * @drv: pointer to a struct cpuidle_driver object
> - *
> - * Returns 0 on success, -EBUSY if the driver is already registered.
> - */
> -static inline int __cpuidle_set_driver(struct cpuidle_driver *drv)
> +static inline int __cpuidle_add_driver(struct cpuidle_driver *drv)
> {
> - if (cpuidle_curr_driver)
> + if (__cpuidle_get_driver(drv->cpumask))
> return -EBUSY;
>
> - cpuidle_curr_driver = drv;
> + list_add(&drv->driver_list, &cpuidle_detected_drivers);
>
> return 0;
> }
>
> /**
> - * __cpuidle_unset_driver - unset the global cpuidle driver variable.
> - * @drv: a pointer to a struct cpuidle_driver
> + * __cpuidle_remove_driver - remove cpuidle driver from list.
> + * @drv: a valid pointer to a struct cpuidle_driver
> *
> - * Reset the global cpuidle variable to NULL. If @drv does not match the
> - * registered driver, do nothing.
> + * Removes cpuidle driver from cpuidle_detected_drivers list.
> */
> -static inline void __cpuidle_unset_driver(struct cpuidle_driver *drv)
> +static inline void __cpuidle_remove_driver(struct cpuidle_driver *drv)
> {
> - if (drv == cpuidle_curr_driver)
> - cpuidle_curr_driver = NULL;
> + list_del(&drv->driver_list);
> }
>
> -#endif
> -
> /**
> * cpuidle_setup_broadcast_timer - enable/disable the broadcast timer
> * @arg: a void pointer used to match the SMP cross call API
> @@ -158,6 +103,7 @@ static void __cpuidle_driver_init(struct cpuidle_driver *drv)
> int i;
>
> drv->refcnt = 0;
> + INIT_LIST_HEAD(&drv->driver_list);
>
> /*
> * Use all possible CPUs as the default, because if the kernel boots
> @@ -244,7 +190,7 @@ static int __cpuidle_register_driver(struct cpuidle_driver *drv)
>
> __cpuidle_driver_init(drv);
>
> - ret = __cpuidle_set_driver(drv);
> + ret = __cpuidle_add_driver(drv);
> if (ret)
> return ret;
>
> @@ -277,7 +223,7 @@ static void __cpuidle_unregister_driver(struct cpuidle_driver *drv)
> (void *)CLOCK_EVT_NOTIFY_BROADCAST_OFF, 1);
> }
>
> - __cpuidle_unset_driver(drv);
> + __cpuidle_remove_driver(drv);
> }
>
> /**
> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> index 0f0da17..81b74d2 100644
> --- a/include/linux/cpuidle.h
> +++ b/include/linux/cpuidle.h
> @@ -129,6 +129,7 @@ struct cpuidle_driver {
>
> /* the driver handles the cpus in cpumask */
> struct cpumask *cpumask;
> + struct list_head driver_list;
> };
>
> #ifdef CONFIG_CPU_IDLE
>
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
next prev parent reply other threads:[~2013-09-25 22:30 UTC|newest]
Thread overview: 70+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-09-22 1:20 [PATCH 00/21] CPUIdle: Minor cleanups for 3.13 Viresh Kumar
2013-09-22 1:20 ` [PATCH 01/21] cpuidle: fix indentation of cpumask Viresh Kumar
2013-09-22 1:20 ` [PATCH 02/21] cpuidle: Fix comments in cpuidle core Viresh Kumar
2013-09-22 1:20 ` [PATCH 03/21] cpuidle: make __cpuidle_get_cpu_driver() inline Viresh Kumar
2013-09-25 21:27 ` Daniel Lezcano
2013-09-22 1:20 ` [PATCH 04/21] cpuidle: make __cpuidle_device_init() return void Viresh Kumar
2013-09-22 1:20 ` [PATCH 05/21] cpuidle: make __cpuidle_driver_init() " Viresh Kumar
2013-09-23 9:58 ` Hongbo Zhang
2013-09-23 10:02 ` Viresh Kumar
2013-09-22 1:20 ` [PATCH 06/21] cpuidle: rearrange code in __cpuidle_driver_init() Viresh Kumar
2013-09-25 21:40 ` Daniel Lezcano
2013-09-26 5:01 ` Viresh Kumar
2013-09-22 1:20 ` [PATCH 07/21] cpuidle: rearrange __cpuidle_register_device() to keep minimal exit points Viresh Kumar
2013-09-25 21:49 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 08/21] cpuidle: use cpuidle_disabled() instead of "off" Viresh Kumar
2013-09-25 21:52 ` Daniel Lezcano
2013-09-26 5:06 ` Viresh Kumar
2013-09-26 8:25 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 09/21] cpuidle: merge two if() statements for checking error cases Viresh Kumar
2013-09-25 21:52 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 10/21] cpuidle: reduce code duplication inside cpuidle_idle_call() Viresh Kumar
2013-09-25 22:01 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 11/21] cpuidle: replace multiline statements with single line in cpuidle_idle_call() Viresh Kumar
2013-09-25 22:03 ` Daniel Lezcano
2013-09-26 5:51 ` Viresh Kumar
2013-09-26 7:55 ` Daniel Lezcano
2013-09-26 8:11 ` Viresh Kumar
2013-09-22 1:21 ` [PATCH 12/21] cpuidle: call cpuidle_get_driver() from after taking cpuidle_driver_lock Viresh Kumar
2013-09-25 22:04 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 13/21] cpuidle: use drv instead of cpuidle_driver in show_current_driver() Viresh Kumar
2013-09-25 22:05 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 14/21] cpuidle: coupled: don't compare cpu masks unnecessarily Viresh Kumar
2013-09-25 22:06 ` Daniel Lezcano
2013-09-26 0:25 ` Colin Cross
2013-09-26 6:36 ` Viresh Kumar
2013-09-26 6:50 ` Colin Cross
2013-09-22 1:21 ` [PATCH 15/21] cpuidle: free all state kobjects from cpuidle_free_state_kobj() Viresh Kumar
2013-09-25 22:09 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 16/21] cpuidle: avoid unnecessary kzalloc/free of struct cpuidle_device_kobj Viresh Kumar
2013-09-25 22:12 ` Daniel Lezcano
2013-09-26 6:05 ` Viresh Kumar
2013-09-26 8:30 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 17/21] cpuidle: avoid unnecessary kzalloc/free of struct cpuidle_driver_kobj Viresh Kumar
2013-09-25 22:16 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 18/21] cpuidle: don't call poll_idle_init() for every cpu Viresh Kumar
2013-09-25 22:22 ` Daniel Lezcano
2013-09-26 6:09 ` Viresh Kumar
2013-09-26 8:28 ` Daniel Lezcano
2013-10-03 10:33 ` Viresh Kumar
2013-10-03 11:46 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 19/21] cpuidle: create list of registered drivers Viresh Kumar
2013-09-25 22:30 ` Daniel Lezcano [this message]
2013-09-26 6:17 ` Viresh Kumar
2013-09-26 8:19 ` Daniel Lezcano
2013-09-28 21:33 ` Paul E. McKenney
2013-09-30 18:37 ` Daniel Lezcano
2013-10-03 4:38 ` Viresh Kumar
2013-10-03 10:47 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 20/21] cpuidle: don't calculate time-diff if entered_state == 0 Viresh Kumar
2013-09-25 22:38 ` Daniel Lezcano
2013-09-26 6:24 ` Viresh Kumar
2013-09-26 8:25 ` Daniel Lezcano
2013-09-26 8:28 ` Viresh Kumar
2013-09-26 8:33 ` Daniel Lezcano
2013-09-22 1:21 ` [PATCH 21/21] cpuidle: change governor from within cpuidle_replace_governor() Viresh Kumar
2013-09-25 22:50 ` Daniel Lezcano
2013-09-26 6:37 ` Viresh Kumar
2013-09-26 8:20 ` Daniel Lezcano
2013-10-03 10:36 ` Viresh Kumar
2013-10-03 11:58 ` Daniel Lezcano
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=5243641B.6050604@linaro.org \
--to=daniel.lezcano@linaro.org \
--cc=linaro-kernel@lists.linaro.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pm@vger.kernel.org \
--cc=patches@linaro.org \
--cc=rjw@sisk.pl \
--cc=viresh.kumar@linaro.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;
as well as URLs for NNTP newsgroup(s).