Re: [RFC PATCH 2/5] thermal: cpu_cooling: Add notifications support for the clients

From: Eduardo Valentin
Date: Thu May 15 2014 - 14:40:52 EST


Hello Amit,


On Thu, May 08, 2014 at 08:07:57PM +0530, Amit Daniel Kachhap wrote:
> This patch adds notification support for those clients of cpu_cooling
> APIs which may want to do something interesting after receiving these
> cpu_cooling events. The notifier structure passed is of both Set/Get type.
> The notfications events can be of type,
> 1. CPU_COOLING_SET_STATE_PRE
> 2. CPU_COOLING_SET_STATE_POST
> 3. CPU_COOLING_GET_CUR_STATE
> 4. CPU_COOLING_GET_MAX_STATE

What do you think about expanding this feature to all cooling devices?

I mean, why should we restrict to cpu(freq) cooling?

The clock cooling device I posted (and that I plan to merge soon) should
also benefit of this infrastructure. The clock cooling is for the
context of other processing units, that do not benefit of cpufreq.


>
> The advantages of these notfications is to differentiate between different
> P states in the cpufreq table and the cooling states. The clients of these
> events may group few P states into 1 cooling states. Also some more cooling
> states can be enabled when the maximum of P state is reached. Post notifications
> can be used for those cases.
>
> Signed-off-by: Amit Daniel Kachhap <amit.daniel@xxxxxxxxxxx>
> ---
> drivers/thermal/cpu_cooling.c | 99 +++++++++++++++++++++++++++++++++++++++-
> include/linux/cpu_cooling.h | 55 +++++++++++++++++++++++
> 2 files changed, 151 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/thermal/cpu_cooling.c b/drivers/thermal/cpu_cooling.c
> index 21f44d4..e2aeb36 100644
> --- a/drivers/thermal/cpu_cooling.c
> +++ b/drivers/thermal/cpu_cooling.c
> @@ -50,6 +50,7 @@ struct cpufreq_cooling_device {
> unsigned int cpufreq_state;
> unsigned int cpufreq_val;
> struct cpumask allowed_cpus;
> + struct cpufreq_cooling_status request_status;
> struct list_head node;
> };
> static DEFINE_IDR(cpufreq_idr);
> @@ -59,6 +60,8 @@ static DEFINE_MUTEX(cooling_cpufreq_lock);
> #define NOTIFY_INVALID NULL
> static struct cpufreq_cooling_device *notify_device;
>
> +/* Notfier list to validates/updates the cpufreq cooling states */
> +static BLOCKING_NOTIFIER_HEAD(cpufreq_cooling_state_notifier_list);
> /* A list to hold all the cpufreq cooling devices registered */
> static LIST_HEAD(cpufreq_cooling_list);
>
> @@ -266,6 +269,21 @@ static unsigned int get_cpu_frequency(unsigned int cpu, unsigned long level)
> return freq;
> }
>
> +static int
> +cpufreq_cooling_notify_states(struct cpufreq_cooling_status *request,
> + enum cpu_cooling_state_ops op)
> +{
> + /* Invoke the notifiers which have registered for this state change */
> + if (op == CPU_COOLING_SET_STATE_PRE ||
> + op == CPU_COOLING_SET_STATE_POST ||
> + op == CPU_COOLING_GET_MAX_STATE ||
> + op == CPU_COOLING_GET_CUR_STATE) {
> + blocking_notifier_call_chain(
> + &cpufreq_cooling_state_notifier_list, op, request);
> + }
> + return 0;
> +}
> +
> /**
> * cpufreq_apply_cooling - function to apply frequency clipping.
> * @cpufreq_device: cpufreq_cooling_device pointer containing frequency
> @@ -285,9 +303,18 @@ static int cpufreq_apply_cooling(struct cpufreq_cooling_device *cpufreq_device,
> struct cpumask *mask = &cpufreq_device->allowed_cpus;
> unsigned int cpu = cpumask_any(mask);
>
> + cpufreq_device->request_status.cur_state =
> + cpufreq_device->cpufreq_state;
> + cpufreq_device->request_status.new_state = cooling_state;
> +
> + cpufreq_cooling_notify_states(&cpufreq_device->request_status,
> + CPU_COOLING_SET_STATE_PRE);
> +

Can a listener block/abort/postpone a state transition?

> + cooling_state = cpufreq_device->request_status.new_state;
>

So you are proposing listeners can have control on target cpufreq state,
right? Or did I get it wrong?

I think we should have this better documented. Besides, if that is the
case, what happens if we have several listeners?

> /* Check if the old cooling action is same as new cooling action */
> - if (cpufreq_device->cpufreq_state == cooling_state)
> + if (cpufreq_device->cpufreq_state ==
> + cpufreq_device->request_status.new_state)
> return 0;
>
> clip_freq = get_cpu_frequency(cpu, cooling_state);
> @@ -304,7 +331,8 @@ static int cpufreq_apply_cooling(struct cpufreq_cooling_device *cpufreq_device,
> }
>
> notify_device = NOTIFY_INVALID;
> -
> + cpufreq_cooling_notify_states(&cpufreq_device->request_status,
> + CPU_COOLING_SET_STATE_POST);
> return 0;
> }
>
> @@ -383,6 +411,11 @@ static int cpufreq_get_max_state(struct thermal_cooling_device *cdev,
> if (count > 0)
> *state = count;
>
> + cpufreq_device->request_status.max_state = count;
> + cpufreq_cooling_notify_states(&cpufreq_device->request_status,
> + CPU_COOLING_GET_MAX_STATE);
> + *state = cpufreq_device->request_status.max_state;
> +

same question here, who determines the priorities when several listeners
are present in the system?

> return ret;
> }
>
> @@ -410,8 +443,12 @@ static int cpufreq_get_cur_state(struct thermal_cooling_device *cdev,
> /* Cooling device pointer not found */
> return -EINVAL;
> }
> + cpufreq_device->request_status.cur_state =
> + cpufreq_device->cpufreq_state;
> + cpufreq_cooling_notify_states(&cpufreq_device->request_status,
> + CPU_COOLING_GET_CUR_STATE);
>

ditto.

> - *state = cpufreq_device->cpufreq_state;
> + *state = cpufreq_device->request_status.cur_state;
>
> return 0;
> }
> @@ -520,6 +557,7 @@ __cpufreq_cooling_register(struct device_node *np,
> }
> cpufreq_dev->cool_dev = cool_dev;
> cpufreq_dev->cpufreq_state = 0;
> + cpufreq_dev->request_status.devdata = devdata;
> mutex_lock(&cooling_cpufreq_lock);
>
> /* Register the notifier for first cpufreq cooling device */
> @@ -614,3 +652,58 @@ void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev)
> kfree(cpufreq_dev);
> }
> EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister);
> +
> +/**
> + * cpufreq_cooling_register_notifier - register a driver with cpufreq cooling.
> + * @nb: notifier function to register.
> + * @list: CPU_COOLING_STATE_NOTIFIER type.
> + *
> + * Add a driver to receive cpufreq cooling notifications like current state,
> + * max state and set state. The drivers after reading the events can perform
> + * some mapping between the P states and cooling states like grouping some P
> + * states into 1 cooling state.
> + *
> + * Return: 0 (success)
> + */
> +int cpufreq_cooling_register_notifier(struct notifier_block *nb,
> + unsigned int list)
> +{
> + int ret = 0;
> + switch (list) {
> + case CPU_COOLING_STATE_NOTIFIER:
> + ret = blocking_notifier_chain_register(
> + &cpufreq_cooling_state_notifier_list, nb);
> + break;
> + default:
> + ret = -EINVAL;
> + }
> + return ret;
> +}
> +EXPORT_SYMBOL_GPL(cpufreq_cooling_register_notifier);
> +
> +/**
> + * cpufreq_cooling_unregister_notifier - unregisters a driver with cpufreq
> + * cooling.
> + * @nb: notifier function to unregister.
> + * @list: CPU_COOLING_STATE_NOTIFIER type.
> + *
> + * Removes a driver to receive further cpufreq cooling notifications.
> + *
> + * Return: 0 (success)
> + */
> +int cpufreq_cooling_unregister_notifier(struct notifier_block *nb,
> + unsigned int list)
> +{
> + int ret = 0;
> + switch (list) {
> + case CPU_COOLING_STATE_NOTIFIER:
> + ret = blocking_notifier_chain_unregister(
> + &cpufreq_cooling_state_notifier_list, nb);
> + break;
> + default:
> + ret = -EINVAL;
> + }
> +
> + return ret;
> +}
> +EXPORT_SYMBOL_GPL(cpufreq_cooling_unregister_notifier);

The exported symbols must be documented under Documentation/ too.

> diff --git a/include/linux/cpu_cooling.h b/include/linux/cpu_cooling.h
> index aaef7d8..f786935 100644
> --- a/include/linux/cpu_cooling.h
> +++ b/include/linux/cpu_cooling.h
> @@ -28,6 +28,22 @@
> #include <linux/thermal.h>
> #include <linux/cpumask.h>
>
> +#define CPU_COOLING_STATE_NOTIFIER (0)
> +
> +enum cpu_cooling_state_ops {
> + CPU_COOLING_SET_STATE_PRE,
> + CPU_COOLING_SET_STATE_POST,
> + CPU_COOLING_GET_CUR_STATE,
> + CPU_COOLING_GET_MAX_STATE,
> +};
> +
> +struct cpufreq_cooling_status {
> + unsigned long cur_state;
> + unsigned long new_state;
> + unsigned long max_state;
> + void *devdata;
> +};
> +
> #ifdef CONFIG_CPU_THERMAL
> /**
> * cpufreq_cooling_register - function to create cpufreq cooling device.
> @@ -62,6 +78,34 @@ of_cpufreq_cooling_register(struct device_node *np,
> void cpufreq_cooling_unregister(struct thermal_cooling_device *cdev);
>
> unsigned long cpufreq_cooling_get_level(unsigned int cpu, unsigned int freq);
> +
> +/**
> + * cpufreq_cooling_register_notifier - register a driver with cpufreq cooling.
> + * @nb: notifier function to register.
> + * @list: CPU_COOLING_STATE_NOTIFIER type.
> + *
> + * Add a driver to receive cpufreq cooling notifications like current state,
> + * max state and set state. The drivers after reading the events can perform
> + * some mapping between the P states and cooling states like grouping some P
> + * states into 1 cooling state.
> + *
> + * Return: 0 (success)
> + */
> +int cpufreq_cooling_register_notifier(struct notifier_block *nb,
> + unsigned int list);
> +
> +/**
> + * cpufreq_cooling_unregister_notifier - unregisters a driver with cpufreq
> + * cooling.
> + * @nb: notifier function to unregister.
> + * @list: CPU_COOLING_STATE_NOTIFIER type.
> + *
> + * Removes a driver to receive further cpufreq cooling notifications.
> + *
> + * Return: 0 (success)
> + */
> +int cpufreq_cooling_unregister_notifier(struct notifier_block *nb,
> + unsigned int list);
> #else /* !CONFIG_CPU_THERMAL */
> static inline struct thermal_cooling_device *
> cpufreq_cooling_register(const struct cpumask *clip_cpus, void *devdata)
> @@ -84,6 +128,17 @@ unsigned long cpufreq_cooling_get_level(unsigned int cpu, unsigned int freq)
> {
> return THERMAL_CSTATE_INVALID;
> }
> +static inline int
> +cpufreq_cooling_register_notifier(struct notifier_block *nb, unsigned int list)
> +{
> + return 0;
> +}
> +static inline int
> +cpufreq_cooling_unregister_notifier(struct notifier_block *nb,
> + unsigned int list)
> +{
> + return 0;
> +}
> #endif /* CONFIG_CPU_THERMAL */
>
> #endif /* __CPU_COOLING_H__ */
> --
> 1.7.1
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/