diff mbox series

[3/5] cpufreq: qcom-cpufreq-hw: Add dcvs interrupt support

Message ID 20210608222926.2707768-4-thara.gopinath@linaro.org (mailing list archive)
State Not Applicable, archived
Headers show
Series Introduce LMh driver for Qualcomm SoCs | expand

Commit Message

Thara Gopinath June 8, 2021, 10:29 p.m. UTC
Add interrupt support to notify the kernel of h/w initiated frequency
throttling by LMh. Convey this to scheduler via thermal presssure
interface.

Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
---
 drivers/cpufreq/qcom-cpufreq-hw.c | 100 ++++++++++++++++++++++++++++++
 1 file changed, 100 insertions(+)

Comments

Viresh Kumar June 14, 2021, 10:31 a.m. UTC | #1
On 08-06-21, 18:29, Thara Gopinath wrote:
> Add interrupt support to notify the kernel of h/w initiated frequency
> throttling by LMh. Convey this to scheduler via thermal presssure
> interface.
> 
> Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
> ---
>  drivers/cpufreq/qcom-cpufreq-hw.c | 100 ++++++++++++++++++++++++++++++
>  1 file changed, 100 insertions(+)
> 
> diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
> index f86859bf76f1..95e17330aa9d 100644
> --- a/drivers/cpufreq/qcom-cpufreq-hw.c
> +++ b/drivers/cpufreq/qcom-cpufreq-hw.c
> @@ -13,6 +13,7 @@
>  #include <linux/of_platform.h>
>  #include <linux/pm_opp.h>
>  #include <linux/slab.h>
> +#include <linux/interrupt.h>
>  
>  #define LUT_MAX_ENTRIES			40U
>  #define LUT_SRC				GENMASK(31, 30)
> @@ -22,10 +23,13 @@
>  #define CLK_HW_DIV			2
>  #define LUT_TURBO_IND			1
>  
> +#define HZ_PER_KHZ			1000
> +
>  struct qcom_cpufreq_soc_data {
>  	u32 reg_enable;
>  	u32 reg_freq_lut;
>  	u32 reg_volt_lut;
> +	u32 reg_current_vote;
>  	u32 reg_perf_state;
>  	u8 lut_row_size;
>  };
> @@ -33,7 +37,11 @@ struct qcom_cpufreq_soc_data {
>  struct qcom_cpufreq_data {
>  	void __iomem *base;
>  	struct resource *res;
> +	struct delayed_work lmh_dcvs_poll_work;
>  	const struct qcom_cpufreq_soc_data *soc_data;
> +	cpumask_var_t cpus;
> +	unsigned long throttled_freq;
> +	int lmh_dcvs_irq;
>  };
>  
>  static unsigned long cpu_hw_rate, xo_rate;
> @@ -251,10 +259,79 @@ static void qcom_get_related_cpus(int index, struct cpumask *m)
>  	}
>  }
>  
> +static inline unsigned long qcom_lmh_vote_to_freq(u32 val)
> +{
> +	return (val & 0x3FF) * 19200;
> +}
> +
> +static void qcom_lmh_dcvs_notify(struct qcom_cpufreq_data *data)
> +{
> +	struct cpufreq_policy policy;
> +	struct dev_pm_opp *opp;
> +	struct device *dev;
> +	unsigned long max_capacity, capacity, freq_hz;
> +	unsigned int val, freq;
> +
> +	val = readl_relaxed(data->base + data->soc_data->reg_current_vote);
> +	freq = qcom_lmh_vote_to_freq(val);
> +	freq_hz = freq * HZ_PER_KHZ;
> +
> +	/* Do I need to calculate ceil and floor ? */

You don't know ?

> +	dev = get_cpu_device(cpumask_first(data->cpus));
> +	opp = dev_pm_opp_find_freq_floor(dev, &freq_hz);
> +	if (IS_ERR(opp) && PTR_ERR(opp) == -ERANGE)
> +		opp = dev_pm_opp_find_freq_ceil(dev, &freq_hz);
> +
> +	data->throttled_freq = freq_hz / HZ_PER_KHZ;
> +

What exactly are we trying to do here ? A comment would be good as
well.

> +	cpufreq_get_policy(&policy, cpumask_first(data->cpus));
> +
> +	/* Update thermal pressure */
> +	max_capacity = arch_scale_cpu_capacity(cpumask_first(data->cpus));

Set capacity of a single CPU from a policy ?

> +	capacity = data->throttled_freq * max_capacity;
> +	capacity /= policy.cpuinfo.max_freq;
> +	/* Don't pass boost capacity to scheduler */
> +	if (capacity > max_capacity)
> +		capacity = max_capacity;
> +	arch_set_thermal_pressure(data->cpus, max_capacity - capacity);

You should really be using policy->cpus instead of allocating
data->cpus..

> +}
> +
> +static void qcom_lmh_dcvs_poll(struct work_struct *work)
> +{
> +	struct qcom_cpufreq_data *data;
> +
> +	data = container_of(work, struct qcom_cpufreq_data, lmh_dcvs_poll_work.work);
> +
> +	qcom_lmh_dcvs_notify(data);

You should really move the below stuff the disable_irq_nosync(), it
will make your life easier.

> +	/**
> +	 * If h/w throttled frequency is higher than what cpufreq has requested for, stop
> +	 * polling and switch back to interrupt mechanism
> +	 */
> +	if (data->throttled_freq >= qcom_cpufreq_hw_get(cpumask_first(data->cpus)))
> +		/* Clear the existing interrupts and enable it back */
> +		enable_irq(data->lmh_dcvs_irq);
> +	else
> +		mod_delayed_work(system_highpri_wq, &data->lmh_dcvs_poll_work,
> +				 msecs_to_jiffies(10));
> +}
> +
> +static irqreturn_t qcom_lmh_dcvs_handle_irq(int irq, void *data)
> +{
> +	struct qcom_cpufreq_data *c_data = data;
> +
> +	/* Disable interrupt and enable polling */
> +	disable_irq_nosync(c_data->lmh_dcvs_irq);
> +	qcom_lmh_dcvs_notify(c_data);
> +	mod_delayed_work(system_highpri_wq, &c_data->lmh_dcvs_poll_work, msecs_to_jiffies(10));
> +
> +	return 0;
> +}
> +
>  static const struct qcom_cpufreq_soc_data qcom_soc_data = {
>  	.reg_enable = 0x0,
>  	.reg_freq_lut = 0x110,
>  	.reg_volt_lut = 0x114,
> +	.reg_current_vote = 0x704,

Should this be a different patch ?

>  	.reg_perf_state = 0x920,
>  	.lut_row_size = 32,
>  };
> @@ -285,6 +362,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  	void __iomem *base;
>  	struct qcom_cpufreq_data *data;
>  	int ret, index;
> +	bool lmh_mitigation_enabled = false;

You just overwrite it below, no need to initialize it.

>  
>  	cpu_dev = get_cpu_device(policy->cpu);
>  	if (!cpu_dev) {
> @@ -305,6 +383,8 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  
>  	index = args.args[0];
>  
> +	lmh_mitigation_enabled = of_property_read_bool(pdev->dev.of_node, "qcom,support-lmh");
> +
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, index);
>  	if (!res) {
>  		dev_err(dev, "failed to get mem resource %d\n", index);
> @@ -329,6 +409,11 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  		goto unmap_base;
>  	}
>  
> +	if (!alloc_cpumask_var(&data->cpus, GFP_KERNEL)) {
> +		ret = -ENOMEM;
> +		goto unmap_base;
> +	}
> +
>  	data->soc_data = of_device_get_match_data(&pdev->dev);
>  	data->base = base;
>  	data->res = res;
> @@ -347,6 +432,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  		goto error;
>  	}
>  
> +	cpumask_copy(data->cpus, policy->cpus);
>  	policy->driver_data = data;
>  
>  	ret = qcom_cpufreq_hw_read_lut(cpu_dev, policy);
> @@ -370,6 +456,20 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
>  	}
>  
> +	if (lmh_mitigation_enabled) {

Shouldn't you move the allocation and setting of data->cpus here ? I
suggest creating a separate routine for all initialization around this
stuff.

> +		data->lmh_dcvs_irq = platform_get_irq(pdev, index);
> +		if (data->lmh_dcvs_irq < 0) {
> +			ret = data->lmh_dcvs_irq;
> +			goto error;
> +		}
> +		ret = devm_request_irq(dev, data->lmh_dcvs_irq, qcom_lmh_dcvs_handle_irq,
> +				       0, "dcvsh-irq", data);

I would rather pass policy as data here.

> +		if (ret) {
> +			dev_err(dev, "Error %d registering irq %x\n", ret, data->lmh_dcvs_irq);
> +			goto error;
> +		}
> +		INIT_DEFERRABLE_WORK(&data->lmh_dcvs_poll_work, qcom_lmh_dcvs_poll);
> +	}
>  	return 0;
>  error:
>  	kfree(data);
Thara Gopinath June 15, 2021, 1:58 a.m. UTC | #2
On 6/14/21 6:31 AM, Viresh Kumar wrote:
> On 08-06-21, 18:29, Thara Gopinath wrote:
>> Add interrupt support to notify the kernel of h/w initiated frequency
>> throttling by LMh. Convey this to scheduler via thermal presssure
>> interface.
>>
>> Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
>> ---
>>   drivers/cpufreq/qcom-cpufreq-hw.c | 100 ++++++++++++++++++++++++++++++
>>   1 file changed, 100 insertions(+)
>>
>> diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
>> index f86859bf76f1..95e17330aa9d 100644
>> --- a/drivers/cpufreq/qcom-cpufreq-hw.c
>> +++ b/drivers/cpufreq/qcom-cpufreq-hw.c
>> @@ -13,6 +13,7 @@
>>   #include <linux/of_platform.h>
>>   #include <linux/pm_opp.h>
>>   #include <linux/slab.h>
>> +#include <linux/interrupt.h>
>>   
>>   #define LUT_MAX_ENTRIES			40U
>>   #define LUT_SRC				GENMASK(31, 30)
>> @@ -22,10 +23,13 @@
>>   #define CLK_HW_DIV			2
>>   #define LUT_TURBO_IND			1
>>   
>> +#define HZ_PER_KHZ			1000
>> +
>>   struct qcom_cpufreq_soc_data {
>>   	u32 reg_enable;
>>   	u32 reg_freq_lut;
>>   	u32 reg_volt_lut;
>> +	u32 reg_current_vote;
>>   	u32 reg_perf_state;
>>   	u8 lut_row_size;
>>   };
>> @@ -33,7 +37,11 @@ struct qcom_cpufreq_soc_data {
>>   struct qcom_cpufreq_data {
>>   	void __iomem *base;
>>   	struct resource *res;
>> +	struct delayed_work lmh_dcvs_poll_work;
>>   	const struct qcom_cpufreq_soc_data *soc_data;
>> +	cpumask_var_t cpus;
>> +	unsigned long throttled_freq;
>> +	int lmh_dcvs_irq;
>>   };
>>   
>>   static unsigned long cpu_hw_rate, xo_rate;
>> @@ -251,10 +259,79 @@ static void qcom_get_related_cpus(int index, struct cpumask *m)
>>   	}
>>   }
>>   
>> +static inline unsigned long qcom_lmh_vote_to_freq(u32 val)
>> +{
>> +	return (val & 0x3FF) * 19200;
>> +}
>> +
>> +static void qcom_lmh_dcvs_notify(struct qcom_cpufreq_data *data)
>> +{
>> +	struct cpufreq_policy policy;
>> +	struct dev_pm_opp *opp;
>> +	struct device *dev;
>> +	unsigned long max_capacity, capacity, freq_hz;
>> +	unsigned int val, freq;
>> +
>> +	val = readl_relaxed(data->base + data->soc_data->reg_current_vote);
>> +	freq = qcom_lmh_vote_to_freq(val);
>> +	freq_hz = freq * HZ_PER_KHZ;
>> +
>> +	/* Do I need to calculate ceil and floor ? */
> 
> You don't know ?

stray comment! Will remove it.

> 
>> +	dev = get_cpu_device(cpumask_first(data->cpus));
>> +	opp = dev_pm_opp_find_freq_floor(dev, &freq_hz);
>> +	if (IS_ERR(opp) && PTR_ERR(opp) == -ERANGE)
>> +		opp = dev_pm_opp_find_freq_ceil(dev, &freq_hz);
>> +
>> +	data->throttled_freq = freq_hz / HZ_PER_KHZ;
>> +
> 
> What exactly are we trying to do here ? A comment would be good as
> well.

You want me to put a comment saying converting frequency in hz to khz ?

> 
>> +	cpufreq_get_policy(&policy, cpumask_first(data->cpus));
>> +
>> +	/* Update thermal pressure */
>> +	max_capacity = arch_scale_cpu_capacity(cpumask_first(data->cpus));
> 
> Set capacity of a single CPU from a policy ?

Get maximum capacity of a cpu.

> 
>> +	capacity = data->throttled_freq * max_capacity;
>> +	capacity /= policy.cpuinfo.max_freq;
>> +	/* Don't pass boost capacity to scheduler */
>> +	if (capacity > max_capacity)
>> +		capacity = max_capacity;
>> +	arch_set_thermal_pressure(data->cpus, max_capacity - capacity);
> 
> You should really be using policy->cpus instead of allocating
> data->cpus..

Yes I should be. But I still need data->cpus to get the policy.

> 
>> +}
>> +
>> +static void qcom_lmh_dcvs_poll(struct work_struct *work)
>> +{
>> +	struct qcom_cpufreq_data *data;
>> +
>> +	data = container_of(work, struct qcom_cpufreq_data, lmh_dcvs_poll_work.work);
>> +
>> +	qcom_lmh_dcvs_notify(data);
> 
> You should really move the below stuff the disable_irq_nosync(), it
> will make your life easier.

I don't understand your comment here. I want to disable irq. call 
notify. Start polling. And in polling I want to call notify and if the 
thermal event has passed stop polling else continue polling.
> 
>> +	/**
>> +	 * If h/w throttled frequency is higher than what cpufreq has requested for, stop
>> +	 * polling and switch back to interrupt mechanism
>> +	 */
>> +	if (data->throttled_freq >= qcom_cpufreq_hw_get(cpumask_first(data->cpus)))
>> +		/* Clear the existing interrupts and enable it back */
>> +		enable_irq(data->lmh_dcvs_irq);
>> +	else
>> +		mod_delayed_work(system_highpri_wq, &data->lmh_dcvs_poll_work,
>> +				 msecs_to_jiffies(10));
>> +}
>> +
>> +static irqreturn_t qcom_lmh_dcvs_handle_irq(int irq, void *data)
>> +{
>> +	struct qcom_cpufreq_data *c_data = data;
>> +
>> +	/* Disable interrupt and enable polling */
>> +	disable_irq_nosync(c_data->lmh_dcvs_irq);
>> +	qcom_lmh_dcvs_notify(c_data);
>> +	mod_delayed_work(system_highpri_wq, &c_data->lmh_dcvs_poll_work, msecs_to_jiffies(10));
>> +
>> +	return 0;
>> +}
>> +
>>   static const struct qcom_cpufreq_soc_data qcom_soc_data = {
>>   	.reg_enable = 0x0,
>>   	.reg_freq_lut = 0x110,
>>   	.reg_volt_lut = 0x114,
>> +	.reg_current_vote = 0x704,
> 
> Should this be a different patch ?

Why ? This is the register to read the throttled frequency.

> 
>>   	.reg_perf_state = 0x920,
>>   	.lut_row_size = 32,
>>   };
>> @@ -285,6 +362,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   	void __iomem *base;
>>   	struct qcom_cpufreq_data *data;
>>   	int ret, index;
>> +	bool lmh_mitigation_enabled = false;
> 
> You just overwrite it below, no need to initialize it.

Sure.

> 
>>   
>>   	cpu_dev = get_cpu_device(policy->cpu);
>>   	if (!cpu_dev) {
>> @@ -305,6 +383,8 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   
>>   	index = args.args[0];
>>   
>> +	lmh_mitigation_enabled = of_property_read_bool(pdev->dev.of_node, "qcom,support-lmh");
>> +
>>   	res = platform_get_resource(pdev, IORESOURCE_MEM, index);
>>   	if (!res) {
>>   		dev_err(dev, "failed to get mem resource %d\n", index);
>> @@ -329,6 +409,11 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   		goto unmap_base;
>>   	}
>>   
>> +	if (!alloc_cpumask_var(&data->cpus, GFP_KERNEL)) {
>> +		ret = -ENOMEM;
>> +		goto unmap_base;
>> +	}
>> +
>>   	data->soc_data = of_device_get_match_data(&pdev->dev);
>>   	data->base = base;
>>   	data->res = res;
>> @@ -347,6 +432,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   		goto error;
>>   	}
>>   
>> +	cpumask_copy(data->cpus, policy->cpus);
>>   	policy->driver_data = data;
>>   
>>   	ret = qcom_cpufreq_hw_read_lut(cpu_dev, policy);
>> @@ -370,6 +456,20 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
>>   	}
>>   
>> +	if (lmh_mitigation_enabled) {
> 
> Shouldn't you move the allocation and setting of data->cpus here ? I
> suggest creating a separate routine for all initialization around this
> stuff.

I should considering nothing else is using data->cpus. Yes I will create 
a separate init function.

> 
>> +		data->lmh_dcvs_irq = platform_get_irq(pdev, index);
>> +		if (data->lmh_dcvs_irq < 0) {
>> +			ret = data->lmh_dcvs_irq;
>> +			goto error;
>> +		}
>> +		ret = devm_request_irq(dev, data->lmh_dcvs_irq, qcom_lmh_dcvs_handle_irq,
>> +				       0, "dcvsh-irq", data);
> 
> I would rather pass policy as data here.

So policy for a cpu can change runtime, right ?

> 
>> +		if (ret) {
>> +			dev_err(dev, "Error %d registering irq %x\n", ret, data->lmh_dcvs_irq);
>> +			goto error;
>> +		}
>> +		INIT_DEFERRABLE_WORK(&data->lmh_dcvs_poll_work, qcom_lmh_dcvs_poll);
>> +	}
>>   	return 0;
>>   error:
>>   	kfree(data);
>
Viresh Kumar June 15, 2021, 5:16 a.m. UTC | #3
On 14-06-21, 21:58, Thara Gopinath wrote:
> On 6/14/21 6:31 AM, Viresh Kumar wrote:
> > On 08-06-21, 18:29, Thara Gopinath wrote:
> > > +static void qcom_lmh_dcvs_notify(struct qcom_cpufreq_data *data)
> > > +{
> > > +	struct cpufreq_policy policy;
> > > +	struct dev_pm_opp *opp;
> > > +	struct device *dev;
> > > +	unsigned long max_capacity, capacity, freq_hz;
> > > +	unsigned int val, freq;
> > > +
> > > +	val = readl_relaxed(data->base + data->soc_data->reg_current_vote);
> > > +	freq = qcom_lmh_vote_to_freq(val);
> > > +	freq_hz = freq * HZ_PER_KHZ;
> > > +
> > > +	/* Do I need to calculate ceil and floor ? */
> > 
> > You don't know ?
> 
> stray comment! Will remove it.
> 
> > 
> > > +	dev = get_cpu_device(cpumask_first(data->cpus));
> > > +	opp = dev_pm_opp_find_freq_floor(dev, &freq_hz);
> > > +	if (IS_ERR(opp) && PTR_ERR(opp) == -ERANGE)
> > > +		opp = dev_pm_opp_find_freq_ceil(dev, &freq_hz);
> > > +
> > > +	data->throttled_freq = freq_hz / HZ_PER_KHZ;
> > > +
> > 
> > What exactly are we trying to do here ? A comment would be good as
> > well.
> 
> You want me to put a comment saying converting frequency in hz to khz ?

Not that, but for the entire routine. What exactly are we looking to
do here and why?

> > 
> > > +	cpufreq_get_policy(&policy, cpumask_first(data->cpus));
> > > +
> > > +	/* Update thermal pressure */
> > > +	max_capacity = arch_scale_cpu_capacity(cpumask_first(data->cpus));
> > 
> > Set capacity of a single CPU from a policy ?
> 
> Get maximum capacity of a cpu.

Ahh, I thought you are setting it :(

> > 
> > > +	capacity = data->throttled_freq * max_capacity;
> > > +	capacity /= policy.cpuinfo.max_freq;
> > > +	/* Don't pass boost capacity to scheduler */
> > > +	if (capacity > max_capacity)
> > > +		capacity = max_capacity;
> > > +	arch_set_thermal_pressure(data->cpus, max_capacity - capacity);
> > 
> > You should really be using policy->cpus instead of allocating
> > data->cpus..
> 
> Yes I should be. But I still need data->cpus to get the policy.

From the comment which comes later on, you shall get the policy here
anyway.
 
> > > +}
> > > +
> > > +static void qcom_lmh_dcvs_poll(struct work_struct *work)
> > > +{
> > > +	struct qcom_cpufreq_data *data;
> > > +
> > > +	data = container_of(work, struct qcom_cpufreq_data, lmh_dcvs_poll_work.work);
> > > +
> > > +	qcom_lmh_dcvs_notify(data);
> > 
> > You should really move the below stuff the disable_irq_nosync(), it
> > will make your life easier.

Damn, s/disable_irq_nosync/qcom_lmh_dcvs_notify/

> I don't understand your comment here. I want to disable irq. call notify.
> Start polling. And in polling I want to call notify and if the thermal event
> has passed stop polling else continue polling.

Yeah, I messed up in the comment. I was asking to move the enable-irq
and mod_delayed_work to qcom_lmh_dcvs_notify() itself.

> > > +	/**
> > > +	 * If h/w throttled frequency is higher than what cpufreq has requested for, stop
> > > +	 * polling and switch back to interrupt mechanism
> > > +	 */
> > > +	if (data->throttled_freq >= qcom_cpufreq_hw_get(cpumask_first(data->cpus)))
> > > +		/* Clear the existing interrupts and enable it back */
> > > +		enable_irq(data->lmh_dcvs_irq);
> > > +	else
> > > +		mod_delayed_work(system_highpri_wq, &data->lmh_dcvs_poll_work,
> > > +				 msecs_to_jiffies(10));
> > > +}
> > > +
> > > +static irqreturn_t qcom_lmh_dcvs_handle_irq(int irq, void *data)
> > > +{
> > > +	struct qcom_cpufreq_data *c_data = data;
> > > +
> > > +	/* Disable interrupt and enable polling */
> > > +	disable_irq_nosync(c_data->lmh_dcvs_irq);
> > > +	qcom_lmh_dcvs_notify(c_data);
> > > +	mod_delayed_work(system_highpri_wq, &c_data->lmh_dcvs_poll_work, msecs_to_jiffies(10));
> > > +
> > > +	return 0;
> > > +}
> > > +
> > >   static const struct qcom_cpufreq_soc_data qcom_soc_data = {
> > >   	.reg_enable = 0x0,
> > >   	.reg_freq_lut = 0x110,
> > >   	.reg_volt_lut = 0x114,
> > > +	.reg_current_vote = 0x704,
> > 
> > Should this be a different patch ?
> 
> Why ? This is the register to read the throttled frequency.

Okay, it looked this is separate. Leave it.

> > > +		ret = devm_request_irq(dev, data->lmh_dcvs_irq, qcom_lmh_dcvs_handle_irq,
> > > +				       0, "dcvsh-irq", data);
> > 
> > I would rather pass policy as data here.
> 
> So policy for a cpu can change runtime, right ?

No, it is allocated just once.
Bjorn Andersson June 18, 2021, 6:16 p.m. UTC | #4
On Tue 08 Jun 17:29 CDT 2021, Thara Gopinath wrote:
> diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
[..]
> @@ -305,6 +383,8 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  
>  	index = args.args[0];
>  
> +	lmh_mitigation_enabled = of_property_read_bool(pdev->dev.of_node, "qcom,support-lmh");

Rather than adding a new interrupt _and_ a flag to tell the driver that
this new interrupt should be used, wouldn't it be sufficient to just see
if the interrupt is specified?

> +
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, index);
>  	if (!res) {
>  		dev_err(dev, "failed to get mem resource %d\n", index);
> @@ -329,6 +409,11 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  		goto unmap_base;
>  	}
>  
> +	if (!alloc_cpumask_var(&data->cpus, GFP_KERNEL)) {
> +		ret = -ENOMEM;
> +		goto unmap_base;
> +	}
> +
>  	data->soc_data = of_device_get_match_data(&pdev->dev);
>  	data->base = base;
>  	data->res = res;
> @@ -347,6 +432,7 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  		goto error;
>  	}
>  
> +	cpumask_copy(data->cpus, policy->cpus);
>  	policy->driver_data = data;
>  
>  	ret = qcom_cpufreq_hw_read_lut(cpu_dev, policy);
> @@ -370,6 +456,20 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>  			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
>  	}
>  
> +	if (lmh_mitigation_enabled) {
> +		data->lmh_dcvs_irq = platform_get_irq(pdev, index);
> +		if (data->lmh_dcvs_irq < 0) {

This will be -ENXIO if the interrupt isn't specified and <0 for other
errors, so you should be able to distinguish the two failure cases.

Regards,
Bjorn

> +			ret = data->lmh_dcvs_irq;
> +			goto error;
> +		}
> +		ret = devm_request_irq(dev, data->lmh_dcvs_irq, qcom_lmh_dcvs_handle_irq,
> +				       0, "dcvsh-irq", data);
> +		if (ret) {
> +			dev_err(dev, "Error %d registering irq %x\n", ret, data->lmh_dcvs_irq);
> +			goto error;
> +		}
> +		INIT_DEFERRABLE_WORK(&data->lmh_dcvs_poll_work, qcom_lmh_dcvs_poll);
> +	}
>  	return 0;
>  error:
>  	kfree(data);
> -- 
> 2.25.1
>
Thara Gopinath June 18, 2021, 9:55 p.m. UTC | #5
On 6/18/21 2:16 PM, Bjorn Andersson wrote:
> On Tue 08 Jun 17:29 CDT 2021, Thara Gopinath wrote:
>> diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
> [..]
>> @@ -305,6 +383,8 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>>   
>>   	index = args.args[0];
>>   
>> +	lmh_mitigation_enabled = of_property_read_bool(pdev->dev.of_node, "qcom,support-lmh");
> 
> Rather than adding a new interrupt _and_ a flag to tell the driver that
> this new interrupt should be used, wouldn't it be sufficient to just see
> if the interrupt is specified?

Yes. you are right. It should be. Though when I wrote it there was some 
reason which I forget now. I will remove it.
diff mbox series

Patch

diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
index f86859bf76f1..95e17330aa9d 100644
--- a/drivers/cpufreq/qcom-cpufreq-hw.c
+++ b/drivers/cpufreq/qcom-cpufreq-hw.c
@@ -13,6 +13,7 @@ 
 #include <linux/of_platform.h>
 #include <linux/pm_opp.h>
 #include <linux/slab.h>
+#include <linux/interrupt.h>
 
 #define LUT_MAX_ENTRIES			40U
 #define LUT_SRC				GENMASK(31, 30)
@@ -22,10 +23,13 @@ 
 #define CLK_HW_DIV			2
 #define LUT_TURBO_IND			1
 
+#define HZ_PER_KHZ			1000
+
 struct qcom_cpufreq_soc_data {
 	u32 reg_enable;
 	u32 reg_freq_lut;
 	u32 reg_volt_lut;
+	u32 reg_current_vote;
 	u32 reg_perf_state;
 	u8 lut_row_size;
 };
@@ -33,7 +37,11 @@  struct qcom_cpufreq_soc_data {
 struct qcom_cpufreq_data {
 	void __iomem *base;
 	struct resource *res;
+	struct delayed_work lmh_dcvs_poll_work;
 	const struct qcom_cpufreq_soc_data *soc_data;
+	cpumask_var_t cpus;
+	unsigned long throttled_freq;
+	int lmh_dcvs_irq;
 };
 
 static unsigned long cpu_hw_rate, xo_rate;
@@ -251,10 +259,79 @@  static void qcom_get_related_cpus(int index, struct cpumask *m)
 	}
 }
 
+static inline unsigned long qcom_lmh_vote_to_freq(u32 val)
+{
+	return (val & 0x3FF) * 19200;
+}
+
+static void qcom_lmh_dcvs_notify(struct qcom_cpufreq_data *data)
+{
+	struct cpufreq_policy policy;
+	struct dev_pm_opp *opp;
+	struct device *dev;
+	unsigned long max_capacity, capacity, freq_hz;
+	unsigned int val, freq;
+
+	val = readl_relaxed(data->base + data->soc_data->reg_current_vote);
+	freq = qcom_lmh_vote_to_freq(val);
+	freq_hz = freq * HZ_PER_KHZ;
+
+	/* Do I need to calculate ceil and floor ? */
+	dev = get_cpu_device(cpumask_first(data->cpus));
+	opp = dev_pm_opp_find_freq_floor(dev, &freq_hz);
+	if (IS_ERR(opp) && PTR_ERR(opp) == -ERANGE)
+		opp = dev_pm_opp_find_freq_ceil(dev, &freq_hz);
+
+	data->throttled_freq = freq_hz / HZ_PER_KHZ;
+
+	cpufreq_get_policy(&policy, cpumask_first(data->cpus));
+
+	/* Update thermal pressure */
+	max_capacity = arch_scale_cpu_capacity(cpumask_first(data->cpus));
+	capacity = data->throttled_freq * max_capacity;
+	capacity /= policy.cpuinfo.max_freq;
+	/* Don't pass boost capacity to scheduler */
+	if (capacity > max_capacity)
+		capacity = max_capacity;
+	arch_set_thermal_pressure(data->cpus, max_capacity - capacity);
+}
+
+static void qcom_lmh_dcvs_poll(struct work_struct *work)
+{
+	struct qcom_cpufreq_data *data;
+
+	data = container_of(work, struct qcom_cpufreq_data, lmh_dcvs_poll_work.work);
+
+	qcom_lmh_dcvs_notify(data);
+	/**
+	 * If h/w throttled frequency is higher than what cpufreq has requested for, stop
+	 * polling and switch back to interrupt mechanism
+	 */
+	if (data->throttled_freq >= qcom_cpufreq_hw_get(cpumask_first(data->cpus)))
+		/* Clear the existing interrupts and enable it back */
+		enable_irq(data->lmh_dcvs_irq);
+	else
+		mod_delayed_work(system_highpri_wq, &data->lmh_dcvs_poll_work,
+				 msecs_to_jiffies(10));
+}
+
+static irqreturn_t qcom_lmh_dcvs_handle_irq(int irq, void *data)
+{
+	struct qcom_cpufreq_data *c_data = data;
+
+	/* Disable interrupt and enable polling */
+	disable_irq_nosync(c_data->lmh_dcvs_irq);
+	qcom_lmh_dcvs_notify(c_data);
+	mod_delayed_work(system_highpri_wq, &c_data->lmh_dcvs_poll_work, msecs_to_jiffies(10));
+
+	return 0;
+}
+
 static const struct qcom_cpufreq_soc_data qcom_soc_data = {
 	.reg_enable = 0x0,
 	.reg_freq_lut = 0x110,
 	.reg_volt_lut = 0x114,
+	.reg_current_vote = 0x704,
 	.reg_perf_state = 0x920,
 	.lut_row_size = 32,
 };
@@ -285,6 +362,7 @@  static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
 	void __iomem *base;
 	struct qcom_cpufreq_data *data;
 	int ret, index;
+	bool lmh_mitigation_enabled = false;
 
 	cpu_dev = get_cpu_device(policy->cpu);
 	if (!cpu_dev) {
@@ -305,6 +383,8 @@  static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
 
 	index = args.args[0];
 
+	lmh_mitigation_enabled = of_property_read_bool(pdev->dev.of_node, "qcom,support-lmh");
+
 	res = platform_get_resource(pdev, IORESOURCE_MEM, index);
 	if (!res) {
 		dev_err(dev, "failed to get mem resource %d\n", index);
@@ -329,6 +409,11 @@  static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
 		goto unmap_base;
 	}
 
+	if (!alloc_cpumask_var(&data->cpus, GFP_KERNEL)) {
+		ret = -ENOMEM;
+		goto unmap_base;
+	}
+
 	data->soc_data = of_device_get_match_data(&pdev->dev);
 	data->base = base;
 	data->res = res;
@@ -347,6 +432,7 @@  static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
 		goto error;
 	}
 
+	cpumask_copy(data->cpus, policy->cpus);
 	policy->driver_data = data;
 
 	ret = qcom_cpufreq_hw_read_lut(cpu_dev, policy);
@@ -370,6 +456,20 @@  static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
 			dev_warn(cpu_dev, "failed to enable boost: %d\n", ret);
 	}
 
+	if (lmh_mitigation_enabled) {
+		data->lmh_dcvs_irq = platform_get_irq(pdev, index);
+		if (data->lmh_dcvs_irq < 0) {
+			ret = data->lmh_dcvs_irq;
+			goto error;
+		}
+		ret = devm_request_irq(dev, data->lmh_dcvs_irq, qcom_lmh_dcvs_handle_irq,
+				       0, "dcvsh-irq", data);
+		if (ret) {
+			dev_err(dev, "Error %d registering irq %x\n", ret, data->lmh_dcvs_irq);
+			goto error;
+		}
+		INIT_DEFERRABLE_WORK(&data->lmh_dcvs_poll_work, qcom_lmh_dcvs_poll);
+	}
 	return 0;
 error:
 	kfree(data);