* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-01-29 6:35 ` [PATCH 3/3] devfreq: add mediatek " Andrew-sh Cheng
@ 2019-02-01 3:43 ` Nicolas Boichat
2019-02-12 9:05 ` andrew-sh.cheng
2019-02-01 20:13 ` Matthias Kaehlcke
2019-02-12 4:06 ` Chanwoo Choi
2 siblings, 1 reply; 16+ messages in thread
From: Nicolas Boichat @ 2019-02-01 3:43 UTC (permalink / raw)
To: Andrew-sh Cheng
Cc: Mark Rutland, devicetree, srv_heupstream, linux-pm, Viresh Kumar,
Rafael J. Wysocki, lkml, Rob Herring, Chanwoo Choi,
Kyungmin Park, MyungJoo Ham,
moderated list:ARM/Mediatek SoC support, Matthias Brugger,
Fan Chen, linux-arm Mailing List
On Tue, Jan 29, 2019 at 1:36 PM Andrew-sh Cheng
<andrew-sh.cheng@mediatek.com> wrote:
>
> From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
>
> For big/little cpu cluster architecture,
> not only CPU frequency, but CCI frequency will also affect performance.
>
> Little cores and cci share the same buck in Mediatek mt8183 IC,
> so we add a CCI devfreq which will get notification when buck voltage
> is changed, then CCI devfreq can set cci frequency as high as possible.
>
> Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> ---
> drivers/devfreq/Kconfig | 9 ++
> drivers/devfreq/Makefile | 1 +
> drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> 3 files changed, 234 insertions(+)
> create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
>
> diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> index 6a172d3..0b14aab 100644
> --- a/drivers/devfreq/Kconfig
> +++ b/drivers/devfreq/Kconfig
> @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> It sets the frequency for the memory controller and reads the usage counts
> from hardware.
>
> +config ARM_MT8183_CCI_DEVFREQ
> + tristate "MT8183 CCI DEVFREQ Driver"
> + depends on ARM_MEDIATEK_CPUFREQ
> + help
> + This adds devfreq for cci clock
It might be nice to spell out that CCI stands for "Cache Coherent Interconnect".
> + which is shared the same regulator with cpu cluster.
> + It can track buck voltage and update a proper cci frequency.
> + Use notification to get regulator status.
> +
> source "drivers/devfreq/event/Kconfig"
>
> endif # PM_DEVFREQ
> diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> index 32b8d4d..25afe8c 100644
> --- a/drivers/devfreq/Makefile
> +++ b/drivers/devfreq/Makefile
> @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
>
> # DEVFREQ Event Drivers
> obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> new file mode 100644
> index 0000000..4837892
> --- /dev/null
> +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> @@ -0,0 +1,224 @@
> +// SPDX-License-Identifier: GPL-2.0
> +//
> +// Copyright (c) 2018 MediaTek Inc.
> +#include <linux/cpu.h>
> +#include <linux/cpufreq.h>
> +#include <linux/devfreq.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/regulator/consumer.h>
> +
> +#include "governor.h"
> +
> +struct cci_devfreq_data {
> + struct devfreq *devfreq;
> + struct regulator *proc_reg;
> + struct clk *cci_clk;
> + unsigned long buck_volt;
> + int volt_increasing;
> + struct notifier_block nb;
> +};
> +
> +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> + unsigned long *freq)
> +{
> + struct cci_devfreq_data *cci_devdata;
> + int i, opp_count;
No tab here: single space between int and i.
> + struct dev_pm_opp *opp;
> + unsigned long opp_rate, opp_voltage;
> +
> + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> +
> + /* find available frequency */
> + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> + &opp_rate);
This is quite inefficient as each call to dev_pm_opp_find_freq_floor
may go through the whole OPP table (so your loop is O(n**2), but this
can be done in O(n) by simply walking through the list).
I'd just copy what dev_pm_opp_find_freq_ceil does
(https://elixir.bootlin.com/linux/v4.19/source/drivers/opp/core.c#L425).
Or maybe we need a new function in the opp core.
> + opp_voltage = dev_pm_opp_get_voltage(opp);
> + dev_pm_opp_put(opp);
> +
> + if (opp_voltage <= cci_devdata->buck_volt)
> + break;
> + }
> + *freq = opp_rate;
> +
> + return 0;
> +}
> +
> +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> + unsigned int event, void *data)
> +{
> + return 0;
> +}
> +
> +static struct devfreq_governor mtk_cci_devfreq_governor = {
> + .name = "voltage_monitor",
> + .get_target_freq = mtk_cci_governor_get_target,
> + .event_handler = mtk_cci_governor_event_handler,
> +};
> +
> +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> + u32 flags)
> +{
> + struct cci_devfreq_data *cci_devdata;
> +
> + cci_devdata = dev_get_drvdata(dev);
> +
> + clk_set_rate(cci_devdata->cci_clk, *freq);
> +
> + return 0;
> +}
> +
> +static struct devfreq_dev_profile cci_devfreq_profile = {
> + .polling_ms = 0,
> + .target = mtk_cci_devfreq_target,
> +};
> +
> +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> + unsigned long val, void *data)
> +{
> + struct cci_devfreq_data *cci_devdata =
> + container_of(nb, struct cci_devfreq_data, nb);
> +
> + /* deal with reduce frequency */
> + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> + struct pre_voltage_change_data *pvc_data = data;
> +
> + if (pvc_data->old_uV > pvc_data->min_uV) {
> + cci_devdata->volt_increasing = 0;
> + cci_devdata->buck_volt =
> + (unsigned long)(pvc_data->min_uV);
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + } else {
> + cci_devdata->volt_increasing = 1;
> + }
> + }
> + /* deal with abort reduce frequency */
> + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
> + /* deal with increase frequency */
> + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> + cci_devdata->volt_increasing == 1)) {
> + cci_devdata->buck_volt = (unsigned long)data;
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + }
> +
> + return 0;
> +}
> +
> +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> +{
> + struct device *cci_dev = &pdev->dev;
> + struct cci_devfreq_data *cci_devdata;
> + struct notifier_block *nb;
> + int ret;
> +
> + dev_pm_opp_of_add_table(&pdev->dev);
> +
> + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> + if (!cci_devdata)
> + return -ENOMEM;
> + nb = &cci_devdata->nb;
> + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
This one is useless as you set cci_clk again just below.
> + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
For this one, I'd just create 2 goto targets below, instead of those
ifs in the error path.
> +
> + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci clock not ready, retry\n");
> + else
> + pr_err("no clock for cci: %d\n", ret);
> +
> + goto out_free_resources;
> + }
> +
> + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci regulator not ready, retry\n");
> + else
> + pr_err("no regulator for cci: %d\n", ret);
> +
> + goto out_free_resources;
> + }
> +
> + platform_set_drvdata(pdev, cci_devdata);
> +
> + /* create cci_devfreq and add to cci device.
> + * governor is voltage_monitor.
> + * governor will get platform_device data to make decision.
> + */
> + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> + &cci_devfreq_profile,
> + "voltage_monitor",
> + NULL);
> +
> + nb->notifier_call = cci_devfreq_regulator_notifier;
> + regulator_register_notifier(cci_devdata->proc_reg,
> + nb);
> +
> + return 0;
> +
> +out_free_resources:
> + if (!IS_ERR(cci_devdata->cci_clk))
> + clk_put(cci_devdata->cci_clk);
> + if (!IS_ERR(cci_devdata->proc_reg))
> + regulator_put(cci_devdata->proc_reg);
> +
> + return ret;
> +}
> +
> +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> + { .compatible = "mediatek,mt8183-cci" },
> + { },
> +};
> +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> +
> +static struct platform_driver cci_devfreq_driver = {
> + .probe = mtk_cci_devfreq_probe,
> + .driver = {
> + .name = "mediatek-cci-devfreq",
> + .of_match_table = mediatek_cci_devfreq_of_match,
> + },
> +};
> +
> +static int __init mtk_cci_devfreq_init(void)
> +{
> + int ret = 0;
> +
> + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> + if (ret) {
> + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> + return ret;
> + }
> +
> + ret = platform_driver_register(&cci_devfreq_driver);
> + if (ret)
> + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> +
> + return ret;
> +}
> +module_init(mtk_cci_devfreq_init)
> +
> +static void __exit mtk_cci_devfreq_exit(void)
> +{
> + int ret = 0;
> +
> + platform_driver_unregister(&cci_devfreq_driver);
> +
> + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> + if (ret)
> + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
> +}
> +module_exit(mtk_cci_devfreq_exit)
> +
> +MODULE_LICENSE("GPL v2");
> +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> +MODULE_AUTHOR("andrew-sh.cheng");
> +
> --
> 1.8.1.1.dirty
>
>
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-02-01 3:43 ` Nicolas Boichat
@ 2019-02-12 9:05 ` andrew-sh.cheng
0 siblings, 0 replies; 16+ messages in thread
From: andrew-sh.cheng @ 2019-02-12 9:05 UTC (permalink / raw)
To: Nicolas Boichat
Cc: Mark Rutland, devicetree, srv_heupstream, linux-pm, Viresh Kumar,
Rafael J. Wysocki, lkml, Chanwoo Choi, Kyungmin Park,
Rob Herring, moderated list:ARM/Mediatek SoC support,
MyungJoo Ham, Matthias Brugger, Fan Chen, linux-arm Mailing List
On Fri, 2019-02-01 at 10:43 +0700, Nicolas Boichat wrote:
> On Tue, Jan 29, 2019 at 1:36 PM Andrew-sh Cheng
> <andrew-sh.cheng@mediatek.com> wrote:
> >
> > From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
> >
> > For big/little cpu cluster architecture,
> > not only CPU frequency, but CCI frequency will also affect performance.
> >
> > Little cores and cci share the same buck in Mediatek mt8183 IC,
> > so we add a CCI devfreq which will get notification when buck voltage
> > is changed, then CCI devfreq can set cci frequency as high as possible.
> >
> > Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> > ---
> > drivers/devfreq/Kconfig | 9 ++
> > drivers/devfreq/Makefile | 1 +
> > drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> > 3 files changed, 234 insertions(+)
> > create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
> >
> > diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> > index 6a172d3..0b14aab 100644
> > --- a/drivers/devfreq/Kconfig
> > +++ b/drivers/devfreq/Kconfig
> > @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> > It sets the frequency for the memory controller and reads the usage counts
> > from hardware.
> >
> > +config ARM_MT8183_CCI_DEVFREQ
> > + tristate "MT8183 CCI DEVFREQ Driver"
> > + depends on ARM_MEDIATEK_CPUFREQ
> > + help
> > + This adds devfreq for cci clock
>
> It might be nice to spell out that CCI stands for "Cache Coherent Interconnect".
I will modify in next patch.
>
> > + which is shared the same regulator with cpu cluster.
> > + It can track buck voltage and update a proper cci frequency.
> > + Use notification to get regulator status.
> > +
> > source "drivers/devfreq/event/Kconfig"
> >
> > endif # PM_DEVFREQ
> > diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> > index 32b8d4d..25afe8c 100644
> > --- a/drivers/devfreq/Makefile
> > +++ b/drivers/devfreq/Makefile
> > @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> > obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> > obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> > obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> > +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
> >
> > # DEVFREQ Event Drivers
> > obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> > diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> > new file mode 100644
> > index 0000000..4837892
> > --- /dev/null
> > +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> > @@ -0,0 +1,224 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +//
> > +// Copyright (c) 2018 MediaTek Inc.
> > +#include <linux/cpu.h>
> > +#include <linux/cpufreq.h>
> > +#include <linux/devfreq.h>
> > +#include <linux/module.h>
> > +#include <linux/of.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/regulator/consumer.h>
> > +
> > +#include "governor.h"
> > +
> > +struct cci_devfreq_data {
> > + struct devfreq *devfreq;
> > + struct regulator *proc_reg;
> > + struct clk *cci_clk;
> > + unsigned long buck_volt;
> > + int volt_increasing;
> > + struct notifier_block nb;
> > +};
> > +
> > +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> > + unsigned long *freq)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
> > + int i, opp_count;
>
> No tab here: single space between int and i.
I will modify in next patch.
>
> > + struct dev_pm_opp *opp;
> > + unsigned long opp_rate, opp_voltage;
> > +
> > + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> > +
> > + /* find available frequency */
> > + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> > + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> > + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> > + &opp_rate);
>
> This is quite inefficient as each call to dev_pm_opp_find_freq_floor
> may go through the whole OPP table (so your loop is O(n**2), but this
> can be done in O(n) by simply walking through the list).
>
> I'd just copy what dev_pm_opp_find_freq_ceil does
> (https://elixir.bootlin.com/linux/v4.19/source/drivers/opp/core.c#L425).
> Or maybe we need a new function in the opp core.
I will add an API for opp framework for this in next patch
>
> > + opp_voltage = dev_pm_opp_get_voltage(opp);
> > + dev_pm_opp_put(opp);
> > +
> > + if (opp_voltage <= cci_devdata->buck_volt)
> > + break;
> > + }
> > + *freq = opp_rate;
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> > + unsigned int event, void *data)
> > +{
> > + return 0;
> > +}
> > +
> > +static struct devfreq_governor mtk_cci_devfreq_governor = {
> > + .name = "voltage_monitor",
> > + .get_target_freq = mtk_cci_governor_get_target,
> > + .event_handler = mtk_cci_governor_event_handler,
> > +};
> > +
> > +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> > + u32 flags)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
> > +
> > + cci_devdata = dev_get_drvdata(dev);
> > +
> > + clk_set_rate(cci_devdata->cci_clk, *freq);
> > +
> > + return 0;
> > +}
> > +
> > +static struct devfreq_dev_profile cci_devfreq_profile = {
> > + .polling_ms = 0,
> > + .target = mtk_cci_devfreq_target,
> > +};
> > +
> > +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> > + unsigned long val, void *data)
> > +{
> > + struct cci_devfreq_data *cci_devdata =
> > + container_of(nb, struct cci_devfreq_data, nb);
> > +
> > + /* deal with reduce frequency */
> > + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> > + struct pre_voltage_change_data *pvc_data = data;
> > +
> > + if (pvc_data->old_uV > pvc_data->min_uV) {
> > + cci_devdata->volt_increasing = 0;
> > + cci_devdata->buck_volt =
> > + (unsigned long)(pvc_data->min_uV);
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + } else {
> > + cci_devdata->volt_increasing = 1;
> > + }
> > + }
> > + /* deal with abort reduce frequency */
> > + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
> > + /* deal with increase frequency */
> > + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> > + cci_devdata->volt_increasing == 1)) {
> > + cci_devdata->buck_volt = (unsigned long)data;
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> > +{
> > + struct device *cci_dev = &pdev->dev;
> > + struct cci_devfreq_data *cci_devdata;
> > + struct notifier_block *nb;
> > + int ret;
> > +
> > + dev_pm_opp_of_add_table(&pdev->dev);
> > +
> > + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> > + if (!cci_devdata)
> > + return -ENOMEM;
> > + nb = &cci_devdata->nb;
> > + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
>
> This one is useless as you set cci_clk again just below.
>
> > + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
>
> For this one, I'd just create 2 goto targets below, instead of those
> ifs in the error path.
I will modify and change to 2 goto in next patch.
>
> > +
> > + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci clock not ready, retry\n");
> > + else
> > + pr_err("no clock for cci: %d\n", ret);
> > +
> > + goto out_free_resources;
> > + }
> > +
> > + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci regulator not ready, retry\n");
> > + else
> > + pr_err("no regulator for cci: %d\n", ret);
> > +
> > + goto out_free_resources;
> > + }
> > +
> > + platform_set_drvdata(pdev, cci_devdata);
> > +
> > + /* create cci_devfreq and add to cci device.
> > + * governor is voltage_monitor.
> > + * governor will get platform_device data to make decision.
> > + */
> > + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> > + &cci_devfreq_profile,
> > + "voltage_monitor",
> > + NULL);
> > +
> > + nb->notifier_call = cci_devfreq_regulator_notifier;
> > + regulator_register_notifier(cci_devdata->proc_reg,
> > + nb);
> > +
> > + return 0;
> > +
> > +out_free_resources:
> > + if (!IS_ERR(cci_devdata->cci_clk))
> > + clk_put(cci_devdata->cci_clk);
> > + if (!IS_ERR(cci_devdata->proc_reg))
> > + regulator_put(cci_devdata->proc_reg);
> > +
> > + return ret;
> > +}
> > +
> > +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> > + { .compatible = "mediatek,mt8183-cci" },
> > + { },
> > +};
> > +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> > +
> > +static struct platform_driver cci_devfreq_driver = {
> > + .probe = mtk_cci_devfreq_probe,
> > + .driver = {
> > + .name = "mediatek-cci-devfreq",
> > + .of_match_table = mediatek_cci_devfreq_of_match,
> > + },
> > +};
> > +
> > +static int __init mtk_cci_devfreq_init(void)
> > +{
> > + int ret = 0;
> > +
> > + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> > + if (ret) {
> > + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> > + return ret;
> > + }
> > +
> > + ret = platform_driver_register(&cci_devfreq_driver);
> > + if (ret)
> > + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > +
> > + return ret;
> > +}
> > +module_init(mtk_cci_devfreq_init)
> > +
> > +static void __exit mtk_cci_devfreq_exit(void)
> > +{
> > + int ret = 0;
> > +
> > + platform_driver_unregister(&cci_devfreq_driver);
> > +
> > + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > + if (ret)
> > + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
> > +}
> > +module_exit(mtk_cci_devfreq_exit)
> > +
> > +MODULE_LICENSE("GPL v2");
> > +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> > +MODULE_AUTHOR("andrew-sh.cheng");
> > +
> > --
> > 1.8.1.1.dirty
> >
> >
> > _______________________________________________
> > Linux-mediatek mailing list
> > Linux-mediatek@lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/linux-mediatek
>
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-01-29 6:35 ` [PATCH 3/3] devfreq: add mediatek " Andrew-sh Cheng
2019-02-01 3:43 ` Nicolas Boichat
@ 2019-02-01 20:13 ` Matthias Kaehlcke
2019-02-16 3:07 ` andrew-sh.cheng
2019-02-12 4:06 ` Chanwoo Choi
2 siblings, 1 reply; 16+ messages in thread
From: Matthias Kaehlcke @ 2019-02-01 20:13 UTC (permalink / raw)
To: Andrew-sh Cheng
Cc: Mark Rutland, devicetree, srv_heupstream, linux-pm, Viresh Kumar,
Rafael J. Wysocki, linux-kernel, Rob Herring, Chanwoo Choi,
Kyungmin Park, MyungJoo Ham, linux-mediatek, Matthias Brugger,
fan.chen, linux-arm-kernel
Hi Andrew,
On Tue, Jan 29, 2019 at 02:35:04PM +0800, Andrew-sh Cheng wrote:
> From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
the 'From' tag is only needed when you send patches not authored by
yourself. It doesn't do any harm either, but you can drop it in future
submissions.
> For big/little cpu cluster architecture,
> not only CPU frequency, but CCI frequency will also affect performance.
This doesn't provide much information, I suggest to drop it unless
others find it valuable.
> Little cores and cci share the same buck in Mediatek mt8183 IC,
> so we add a CCI devfreq which will get notification when buck voltage
> is changed, then CCI devfreq can set cci frequency as high as
> possible.
Possible alternative:
"This adds a devfreq driver for the Cache Coherent Interconnect (CCI)
of the Mediatek MT8183.
On the MT8183 the CCI is supplied by the same regulator as the LITTLE
cores. The driver is notified when the regulator voltage changes
(driven by cpufreq) and adjusts the CCI frequency to the maximum
possible value."
Feel free to ignore or to adjust.
> Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> ---
> drivers/devfreq/Kconfig | 9 ++
> drivers/devfreq/Makefile | 1 +
> drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> 3 files changed, 234 insertions(+)
> create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
>
> diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> index 6a172d3..0b14aab 100644
> --- a/drivers/devfreq/Kconfig
> +++ b/drivers/devfreq/Kconfig
> @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> It sets the frequency for the memory controller and reads the usage counts
> from hardware.
>
> +config ARM_MT8183_CCI_DEVFREQ
> + tristate "MT8183 CCI DEVFREQ Driver"
> + depends on ARM_MEDIATEK_CPUFREQ
> + help
> + This adds devfreq for cci clock
"This adds a devfreq driver for the Cache Coherent Interconnect (CCI)
of the Mediatek MT8183"?
> + which is shared the same regulator with cpu cluster.
> + It can track buck voltage and update a proper cci frequency.
> + Use notification to get regulator status.
not sure how important this info is here, sounds more like an
implementation detail to me.
> +
> source "drivers/devfreq/event/Kconfig"
>
> endif # PM_DEVFREQ
> diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> index 32b8d4d..25afe8c 100644
> --- a/drivers/devfreq/Makefile
> +++ b/drivers/devfreq/Makefile
> @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
>
> # DEVFREQ Event Drivers
> obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> new file mode 100644
> index 0000000..4837892
> --- /dev/null
> +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> @@ -0,0 +1,224 @@
> +// SPDX-License-Identifier: GPL-2.0
> +//
> +// Copyright (c) 2018 MediaTek Inc.
2019 :)
> +#include <linux/cpu.h>
> +#include <linux/cpufreq.h>
> +#include <linux/devfreq.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/regulator/consumer.h>
> +
> +#include "governor.h"
> +
> +struct cci_devfreq_data {
nit: consider dropping the '_data' suffix, it doesn't add much/any
value.
> + struct devfreq *devfreq;
> + struct regulator *proc_reg;
> + struct clk *cci_clk;
> + unsigned long buck_volt;
proc_reg_uV?
place after proc_reg
> + int volt_increasing;
should be bool
actually, is it really needed? See comment in cci_devfreq_regulator_notifier()
> + struct notifier_block nb;
> +};
> +
> +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> + unsigned long *freq)
> +{
> + struct cci_devfreq_data *cci_devdata;
nit: consider calling it 'cci_df' (here and elsewhere). In the context
the abbreviation is clear and there is less clutter when accessing
fields in the structure.
> + int i, opp_count;
> + struct dev_pm_opp *opp;
> + unsigned long opp_rate, opp_voltage;
> +
> + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> +
> + /* find available frequency */
> + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> + &opp_rate);
> + opp_voltage = dev_pm_opp_get_voltage(opp);
> + dev_pm_opp_put(opp);
> +
> + if (opp_voltage <= cci_devdata->buck_volt)
> + break;
> + }
> + *freq = opp_rate;
> +
> + return 0;
> +}
> +
> +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> + unsigned int event, void *data)
> +{
> + return 0;
> +}
> +
> +static struct devfreq_governor mtk_cci_devfreq_governor = {
> + .name = "voltage_monitor",
"mtk_cci_vmon" or similar?
> + .get_target_freq = mtk_cci_governor_get_target,
> + .event_handler = mtk_cci_governor_event_handler,
> +};
> +
> +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> + u32 flags)
> +{
> + struct cci_devfreq_data *cci_devdata;
> +
> + cci_devdata = dev_get_drvdata(dev);
> +
> + clk_set_rate(cci_devdata->cci_clk, *freq);
> +
> + return 0;
> +}
> +
> +static struct devfreq_dev_profile cci_devfreq_profile = {
> + .polling_ms = 0,
> + .target = mtk_cci_devfreq_target,
> +};
> +
> +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> + unsigned long val, void *data)
> +{
> + struct cci_devfreq_data *cci_devdata =
> + container_of(nb, struct cci_devfreq_data, nb);
> +
> + /* deal with reduce frequency */
> + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> + struct pre_voltage_change_data *pvc_data = data;
> +
> + if (pvc_data->old_uV > pvc_data->min_uV) {
I would suggest to invert the condition. For humans it is easier to
parse "is the new voltage smaller than the previous one", than
viceversa.
> + cci_devdata->volt_increasing = 0;
> + cci_devdata->buck_volt =
> + (unsigned long)(pvc_data->min_uV);
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + } else {
> + cci_devdata->volt_increasing = 1;
> + }
> + }
nit: add empty line
> + /* deal with abort reduce frequency */
Shouldn't this be "abort voltage change?". If so I suggest to drop the
comment, REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE in the condition makes
it clear.
> + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
> + /* deal with increase frequency */
> + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> + cci_devdata->volt_increasing == 1)) {
to check for a voltage increase you could use 'cci_devdata->buck_volt
< (unsigned long)data' instead of ->volt_increasing.
> + cci_devdata->buck_volt = (unsigned long)data;
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + }
> +
> + return 0;
> +}
> +
> +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> +{
> + struct device *cci_dev = &pdev->dev;
> + struct cci_devfreq_data *cci_devdata;
> + struct notifier_block *nb;
> + int ret;
> +
> + dev_pm_opp_of_add_table(&pdev->dev);
> +
> + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> + if (!cci_devdata)
> + return -ENOMEM;
> + nb = &cci_devdata->nb;
> + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
initialization not needed, the field is assigned a few lines below.
> + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
also not really needed with the right gotos in place
> +
> + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci clock not ready, retry\n");
> + else
> + pr_err("no clock for cci: %d\n", ret);
> +
> + goto out_free_resources;
nothing to be freed here, just return ret.
> + }
> +
> + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci regulator not ready, retry\n");
> + else
> + pr_err("no regulator for cci: %d\n", ret);
> +
> + goto out_free_resources;
> + }
> +
> + platform_set_drvdata(pdev, cci_devdata);
> +
> + /* create cci_devfreq and add to cci device.
> + * governor is voltage_monitor.
> + * governor will get platform_device data to make decision.
> + */
> + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> + &cci_devfreq_profile,
> + "voltage_monitor",
adjust name in case you follow my suggestion above.
check cci_devdata->devfreq for errors.
> + NULL);
> +
> + nb->notifier_call = cci_devfreq_regulator_notifier;
> + regulator_register_notifier(cci_devdata->proc_reg,
> + nb);
> +
> + return 0;
> +
> +out_free_resources:
> + if (!IS_ERR(cci_devdata->cci_clk))
> + clk_put(cci_devdata->cci_clk);
> + if (!IS_ERR(cci_devdata->proc_reg))
> + regulator_put(cci_devdata->proc_reg);
use individual labels and get rid of the IS_ERRs.
e.g.
err_put_reg:
regulator_put(cci_devdata->proc_reg);
err_put_clk:
clk_put(cci_devdata->cci_clk);
and jump to the corresponding lable above.
> +
> + return ret;
> +}
> +
> +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> + { .compatible = "mediatek,mt8183-cci" },
> + { },
> +};
> +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> +
> +static struct platform_driver cci_devfreq_driver = {
> + .probe = mtk_cci_devfreq_probe,
> + .driver = {
> + .name = "mediatek-cci-devfreq",
> + .of_match_table = mediatek_cci_devfreq_of_match,
> + },
> +};
> +
> +static int __init mtk_cci_devfreq_init(void)
> +{
> + int ret = 0;
initialization not needed
> +
> + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> + if (ret) {
> + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> + return ret;
> + }
> +
> + ret = platform_driver_register(&cci_devfreq_driver);
> + if (ret)
> + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> +
> + return ret;
> +}
> +module_init(mtk_cci_devfreq_init)
> +
> +static void __exit mtk_cci_devfreq_exit(void)
> +{
> + int ret = 0;
initialization not needed
> + platform_driver_unregister(&cci_devfreq_driver);
> +
> + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> + if (ret)
> + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
typically you reverse the order of initialization, i.e. remove the
governor, then unregister the driver.
> +}
> +module_exit(mtk_cci_devfreq_exit)
> +
> +MODULE_LICENSE("GPL v2");
> +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> +MODULE_AUTHOR("andrew-sh.cheng");
should be either "Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>" or
"Andrew-sh.Cheng"
Cheers
Matthias
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-02-01 20:13 ` Matthias Kaehlcke
@ 2019-02-16 3:07 ` andrew-sh.cheng
0 siblings, 0 replies; 16+ messages in thread
From: andrew-sh.cheng @ 2019-02-16 3:07 UTC (permalink / raw)
To: Matthias Kaehlcke
Cc: Mark Rutland, devicetree, srv_heupstream, linux-pm, Viresh Kumar,
Rafael J. Wysocki, linux-kernel, Chanwoo Choi, Kyungmin Park,
Rob Herring, linux-mediatek, MyungJoo Ham, Matthias Brugger,
fan.chen, linux-arm-kernel
On Fri, 2019-02-01 at 12:13 -0800, Matthias Kaehlcke wrote:
> Hi Andrew,
>
> On Tue, Jan 29, 2019 at 02:35:04PM +0800, Andrew-sh Cheng wrote:
> > From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
>
> the 'From' tag is only needed when you send patches not authored by
> yourself. It doesn't do any harm either, but you can drop it in future
> submissions.
Got it~
>
> > For big/little cpu cluster architecture,
> > not only CPU frequency, but CCI frequency will also affect performance.
>
> This doesn't provide much information, I suggest to drop it unless
> others find it valuable.
I will remove it in next patch.
>
> > Little cores and cci share the same buck in Mediatek mt8183 IC,
> > so we add a CCI devfreq which will get notification when buck voltage
> > is changed, then CCI devfreq can set cci frequency as high as
> > possible.
>
> Possible alternative:
>
> "This adds a devfreq driver for the Cache Coherent Interconnect (CCI)
> of the Mediatek MT8183.
>
> On the MT8183 the CCI is supplied by the same regulator as the LITTLE
> cores. The driver is notified when the regulator voltage changes
> (driven by cpufreq) and adjusts the CCI frequency to the maximum
> possible value."
>
> Feel free to ignore or to adjust.
Thank you. This description provide suitable information.
>
> > Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> > ---
> > drivers/devfreq/Kconfig | 9 ++
> > drivers/devfreq/Makefile | 1 +
> > drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> > 3 files changed, 234 insertions(+)
> > create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
> >
> > diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> > index 6a172d3..0b14aab 100644
> > --- a/drivers/devfreq/Kconfig
> > +++ b/drivers/devfreq/Kconfig
> > @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> > It sets the frequency for the memory controller and reads the usage counts
> > from hardware.
> >
> > +config ARM_MT8183_CCI_DEVFREQ
> > + tristate "MT8183 CCI DEVFREQ Driver"
> > + depends on ARM_MEDIATEK_CPUFREQ
> > + help
> > + This adds devfreq for cci clock
>
> "This adds a devfreq driver for the Cache Coherent Interconnect (CCI)
> of the Mediatek MT8183"?
Yes, I describe more detail in next patch.
>
> > + which is shared the same regulator with cpu cluster.
> > + It can track buck voltage and update a proper cci frequency.
> > + Use notification to get regulator status.
>
> not sure how important this info is here, sounds more like an
> implementation detail to me.
Yes, just describe the situation.
>
> > +
> > source "drivers/devfreq/event/Kconfig"
> >
> > endif # PM_DEVFREQ
> > diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> > index 32b8d4d..25afe8c 100644
> > --- a/drivers/devfreq/Makefile
> > +++ b/drivers/devfreq/Makefile
> > @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> > obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> > obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> > obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> > +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
> >
> > # DEVFREQ Event Drivers
> > obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> > diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> > new file mode 100644
> > index 0000000..4837892
> > --- /dev/null
> > +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> > @@ -0,0 +1,224 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +//
> > +// Copyright (c) 2018 MediaTek Inc.
>
> 2019 :)
Got it :)
>
> > +#include <linux/cpu.h>
> > +#include <linux/cpufreq.h>
> > +#include <linux/devfreq.h>
> > +#include <linux/module.h>
> > +#include <linux/of.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/regulator/consumer.h>
> > +
> > +#include "governor.h"
> > +
> > +struct cci_devfreq_data {
>
> nit: consider dropping the '_data' suffix, it doesn't add much/any
> value.
I will change to struct cci_devfreq in next patch.
>
> > + struct devfreq *devfreq;
> > + struct regulator *proc_reg;
> > + struct clk *cci_clk;
> > + unsigned long buck_volt;
>
> proc_reg_uV?
>
> place after proc_reg
OK~
>
> > + int volt_increasing;
>
> should be bool
>
> actually, is it really needed? See comment in cci_devfreq_regulator_notifier()
>
You are right. I will remove this and use notification data to judge the
condition in next patch.
> > + struct notifier_block nb;
> > +};
> > +
> > +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> > + unsigned long *freq)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
>
> nit: consider calling it 'cci_df' (here and elsewhere). In the context
> the abbreviation is clear and there is less clutter when accessing
> fields in the structure.
OK, I will change it in next patch.
>
> > + int i, opp_count;
> > + struct dev_pm_opp *opp;
> > + unsigned long opp_rate, opp_voltage;
> > +
> > + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> > +
> > + /* find available frequency */
> > + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> > + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> > + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> > + &opp_rate);
> > + opp_voltage = dev_pm_opp_get_voltage(opp);
> > + dev_pm_opp_put(opp);
> > +
> > + if (opp_voltage <= cci_devdata->buck_volt)
> > + break;
> > + }
> > + *freq = opp_rate;
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> > + unsigned int event, void *data)
> > +{
> > + return 0;
> > +}
> > +
> > +static struct devfreq_governor mtk_cci_devfreq_governor = {
> > + .name = "voltage_monitor",
>
> "mtk_cci_vmon" or similar?
OK. I will change it in next patch.
> > + .get_target_freq = mtk_cci_governor_get_target,
> > + .event_handler = mtk_cci_governor_event_handler,
> > +};
> > +
> > +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> > + u32 flags)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
> > +
> > + cci_devdata = dev_get_drvdata(dev);
> > +
> > + clk_set_rate(cci_devdata->cci_clk, *freq);
> > +
> > + return 0;
> > +}
> > +
> > +static struct devfreq_dev_profile cci_devfreq_profile = {
> > + .polling_ms = 0,
> > + .target = mtk_cci_devfreq_target,
> > +};
> > +
> > +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> > + unsigned long val, void *data)
> > +{
> > + struct cci_devfreq_data *cci_devdata =
> > + container_of(nb, struct cci_devfreq_data, nb);
> > +
> > + /* deal with reduce frequency */
> > + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> > + struct pre_voltage_change_data *pvc_data = data;
> > +
> > + if (pvc_data->old_uV > pvc_data->min_uV) {
>
> I would suggest to invert the condition. For humans it is easier to
> parse "is the new voltage smaller than the previous one", than
> viceversa.
Got it.
>
> > + cci_devdata->volt_increasing = 0;
> > + cci_devdata->buck_volt =
> > + (unsigned long)(pvc_data->min_uV);
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + } else {
> > + cci_devdata->volt_increasing = 1;
> > + }
> > + }
>
> nit: add empty line
I will modify in next patch.
>
> > + /* deal with abort reduce frequency */
>
> Shouldn't this be "abort voltage change?". If so I suggest to drop the
> comment, REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE in the condition makes
> it clear.
OK
>
> > + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
> > + /* deal with increase frequency */
> > + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> > + cci_devdata->volt_increasing == 1)) {
>
> to check for a voltage increase you could use 'cci_devdata->buck_volt
> < (unsigned long)data' instead of ->volt_increasing.
Thank you for suggestion.
I will modify in next patch.
>
> > + cci_devdata->buck_volt = (unsigned long)data;
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> > +{
> > + struct device *cci_dev = &pdev->dev;
> > + struct cci_devfreq_data *cci_devdata;
> > + struct notifier_block *nb;
> > + int ret;
> > +
> > + dev_pm_opp_of_add_table(&pdev->dev);
> > +
> > + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> > + if (!cci_devdata)
> > + return -ENOMEM;
> > + nb = &cci_devdata->nb;
> > + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
>
> initialization not needed, the field is assigned a few lines below.
>
> > + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
>
> also not really needed with the right gotos in place
OK. I will remove unnecessary initialization.
>
> > +
> > + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci clock not ready, retry\n");
> > + else
> > + pr_err("no clock for cci: %d\n", ret);
> > +
> > + goto out_free_resources;
>
> nothing to be freed here, just return ret.
OK.
>
> > + }
> > +
> > + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci regulator not ready, retry\n");
> > + else
> > + pr_err("no regulator for cci: %d\n", ret);
> > +
> > + goto out_free_resources;
> > + }
> > +
> > + platform_set_drvdata(pdev, cci_devdata);
> > +
> > + /* create cci_devfreq and add to cci device.
> > + * governor is voltage_monitor.
> > + * governor will get platform_device data to make decision.
> > + */
> > + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> > + &cci_devfreq_profile,
> > + "voltage_monitor",
>
> adjust name in case you follow my suggestion above.
>
> check cci_devdata->devfreq for errors.
Will add error check in next patch.
>
> > + NULL);
> > +
> > + nb->notifier_call = cci_devfreq_regulator_notifier;
> > + regulator_register_notifier(cci_devdata->proc_reg,
> > + nb);
> > +
> > + return 0;
> > +
> > +out_free_resources:
> > + if (!IS_ERR(cci_devdata->cci_clk))
> > + clk_put(cci_devdata->cci_clk);
> > + if (!IS_ERR(cci_devdata->proc_reg))
> > + regulator_put(cci_devdata->proc_reg);
>
> use individual labels and get rid of the IS_ERRs.
>
> e.g.
>
> err_put_reg:
> regulator_put(cci_devdata->proc_reg);
>
> err_put_clk:
> clk_put(cci_devdata->cci_clk);
>
>
> and jump to the corresponding lable above.
Got it. I will use individual label and remove the IS_ERR checking.
>
> > +
> > + return ret;
> > +}
> > +
> > +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> > + { .compatible = "mediatek,mt8183-cci" },
> > + { },
> > +};
> > +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> > +
> > +static struct platform_driver cci_devfreq_driver = {
> > + .probe = mtk_cci_devfreq_probe,
> > + .driver = {
> > + .name = "mediatek-cci-devfreq",
> > + .of_match_table = mediatek_cci_devfreq_of_match,
> > + },
> > +};
> > +
> > +static int __init mtk_cci_devfreq_init(void)
> > +{
> > + int ret = 0;
>
> initialization not needed
OK.
>
> > +
> > + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> > + if (ret) {
> > + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> > + return ret;
> > + }
> > +
> > + ret = platform_driver_register(&cci_devfreq_driver);
> > + if (ret)
> > + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > +
> > + return ret;
> > +}
> > +module_init(mtk_cci_devfreq_init)
> > +
> > +static void __exit mtk_cci_devfreq_exit(void)
> > +{
> > + int ret = 0;
>
> initialization not needed
OK.
>
> > + platform_driver_unregister(&cci_devfreq_driver);
> > +
> > + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > + if (ret)
> > + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
>
> typically you reverse the order of initialization, i.e. remove the
> governor, then unregister the driver.
OK. I will change it in next patch.
>
> > +}
> > +module_exit(mtk_cci_devfreq_exit)
> > +
> > +MODULE_LICENSE("GPL v2");
> > +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> > +MODULE_AUTHOR("andrew-sh.cheng");
>
> should be either "Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>" or
> "Andrew-sh.Cheng"
I will change to "Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>" :)
>
> Cheers
>
> Matthias
>
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-01-29 6:35 ` [PATCH 3/3] devfreq: add mediatek " Andrew-sh Cheng
2019-02-01 3:43 ` Nicolas Boichat
2019-02-01 20:13 ` Matthias Kaehlcke
@ 2019-02-12 4:06 ` Chanwoo Choi
2019-02-16 6:16 ` andrew-sh.cheng
2 siblings, 1 reply; 16+ messages in thread
From: Chanwoo Choi @ 2019-02-12 4:06 UTC (permalink / raw)
To: Andrew-sh Cheng, MyungJoo Ham, Kyungmin Park, Rob Herring,
Mark Rutland, Matthias Brugger, Rafael J. Wysocki, Viresh Kumar
Cc: devicetree, srv_heupstream, linux-pm, linux-kernel, fan.chen,
linux-mediatek, linux-arm-kernel
Hi Andrew-sh.Cheng,
On 19. 1. 29. 오후 3:35, Andrew-sh Cheng wrote:
> From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
>
> For big/little cpu cluster architecture,
> not only CPU frequency, but CCI frequency will also affect performance.
>
> Little cores and cci share the same buck in Mediatek mt8183 IC,
> so we add a CCI devfreq which will get notification when buck voltage
> is changed, then CCI devfreq can set cci frequency as high as possible.
>
> Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> ---
> drivers/devfreq/Kconfig | 9 ++
> drivers/devfreq/Makefile | 1 +
> drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> 3 files changed, 234 insertions(+)
> create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
>
> diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> index 6a172d3..0b14aab 100644
> --- a/drivers/devfreq/Kconfig
> +++ b/drivers/devfreq/Kconfig
> @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> It sets the frequency for the memory controller and reads the usage counts
> from hardware.
>
> +config ARM_MT8183_CCI_DEVFREQ
> + tristate "MT8183 CCI DEVFREQ Driver"
> + depends on ARM_MEDIATEK_CPUFREQ
> + help
> + This adds devfreq for cci clock
> + which is shared the same regulator with cpu cluster.
> + It can track buck voltage and update a proper cci frequency.
> + Use notification to get regulator status.
> +
Move ARM_MT8183_CCI_DEVFREQ under ARM_EXYNOS_BUS_DEVFREQ.
On later, I'll change the order between ARM_TEGRA_DEVFREQ and ARM_RK3399_DMC_DEVFREQ
alphabetically. So, you better to keep the order of name.
> source "drivers/devfreq/event/Kconfig"
>
> endif # PM_DEVFREQ
> diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> index 32b8d4d..25afe8c 100644
> --- a/drivers/devfreq/Makefile
> +++ b/drivers/devfreq/Makefile
> @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
Move it under obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ).
>
> # DEVFREQ Event Drivers
> obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> new file mode 100644
> index 0000000..4837892
> --- /dev/null
> +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> @@ -0,0 +1,224 @@
> +// SPDX-License-Identifier: GPL-2.0
> +//
> +// Copyright (c) 2018 MediaTek Inc.
Add blank line and change year 2018 -> 2019.
> +#include <linux/cpu.h>
> +#include <linux/cpufreq.h>
This driver doesn't use any helper function of cpufreq.h/cpu.h.
It depends on devfreq/regulator framework. Please remove them.
> +#include <linux/devfreq.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/regulator/consumer.h>
> +
> +#include "governor.h"
> +
> +struct cci_devfreq_data {
> + struct devfreq *devfreq;
> + struct regulator *proc_reg;
> + struct clk *cci_clk;
> + unsigned long buck_volt;
> + int volt_increasing;
> + struct notifier_block nb;
> +};
> +
> +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> + unsigned long *freq)
> +{
> + struct cci_devfreq_data *cci_devdata;
> + int i, opp_count;
> + struct dev_pm_opp *opp;
> + unsigned long opp_rate, opp_voltage;
Above three definitions use the tab for indentation. Please use space.
> +
> + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> +
> + /* find available frequency */
> + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> + &opp_rate);
> + opp_voltage = dev_pm_opp_get_voltage(opp);
> + dev_pm_opp_put(opp);
> +
> + if (opp_voltage <= cci_devdata->buck_volt)
> + break;
> + }
> + *freq = opp_rate;
> +
> + return 0;
> +}
> +
> +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> + unsigned int event, void *data)
> +{
> + return 0;
> +}
The devfreq core enables the governor with using DEVFREQ_GOV_START.
But, this driver just start after proving the device driver. It break
the previous sequence of governor.
So, if you register/unregister the regulator as following,
it can keep the existing sequence for governor starting.
case DEVFREQ_GOV_START:
case DEVFREQ_GOV_RESUME:
regulator_register_notifier(...);
case DEVFREQ_GOV_STOP:
case DEVFREQ_GOV_SUSPEND:
regulator_unregister_notifier(...);
break;
> +
> +static struct devfreq_governor mtk_cci_devfreq_governor = {
> + .name = "voltage_monitor",
> + .get_target_freq = mtk_cci_governor_get_target,
> + .event_handler = mtk_cci_governor_event_handler,
> +};
> +
> +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> + u32 flags)
> +{
> + struct cci_devfreq_data *cci_devdata;
> +
> + cci_devdata = dev_get_drvdata(dev);
You better to modify it as following:
struct cci_devfreq_data *cci_devdata = dev_get_drvdata(dev);
if (!cci_devdata)
return -EINVAL;
> +
> + clk_set_rate(cci_devdata->cci_clk, *freq);
> +
> + return 0;
> +}
> +
> +static struct devfreq_dev_profile cci_devfreq_profile = {
> + .polling_ms = 0,
Dont' need to initialize it as zero. just remove this line.
> + .target = mtk_cci_devfreq_target,
> +};
> +
> +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> + unsigned long val, void *data)
> +{
> + struct cci_devfreq_data *cci_devdata =
> + container_of(nb, struct cci_devfreq_data, nb);
> +
> + /* deal with reduce frequency */
> + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> + struct pre_voltage_change_data *pvc_data = data;
> +
> + if (pvc_data->old_uV > pvc_data->min_uV) {
> + cci_devdata->volt_increasing = 0;
IMO, 'volt_increasing' is not clear. Instead, cci_devdata
could have the 'previous_voltage' variable instead of volt_increasing.
And you could use 'previous_voltage' for comparision.
> + cci_devdata->buck_volt =
> + (unsigned long)(pvc_data->min_uV);
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + } else {
> + cci_devdata->volt_increasing = 1;
> + }
> + }
> + /* deal with abort reduce frequency */
> + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
if -> else if
At the same time, receives only one notification.
It is not necessary to check the two if statement always, .
> + /* deal with increase frequency */
> + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> + cci_devdata->volt_increasing == 1)) {
> + cci_devdata->buck_volt = (unsigned long)data;
> + mutex_lock(&cci_devdata->devfreq->lock);
> + update_devfreq(cci_devdata->devfreq);
> + mutex_unlock(&cci_devdata->devfreq->lock);
> + }
> +
> + return 0;
> +}
> +
> +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> +{
> + struct device *cci_dev = &pdev->dev;
> + struct cci_devfreq_data *cci_devdata;
> + struct notifier_block *nb;
> + int ret;
> +
> + dev_pm_opp_of_add_table(&pdev->dev);
Move it after getting clock/regulator. After checking that all resourcess are available,
and then you better to get the opp table. And check the return value of dev_pm_opp...()
> +
> + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> + if (!cci_devdata)
> + return -ENOMEM;
> + nb = &cci_devdata->nb;
> + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
It should be handled when error happen. It is not good for readability.
> + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
ditto.
> +
> + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci clock not ready, retry\n");
You should use 'dev_err' instead of pr_err on this driver.
I think that don't show the error when EPROBE_DEFER. you better to just return.
> + else
> + pr_err("no clock for cci: %d\n", ret);
How about changing the log like below?
"failed to get clock for CCI"
pr_err -> dev_err
> +
> + goto out_free_resources;
> + }
> +
> + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> + if (ret) {
> + if (ret == -EPROBE_DEFER)
> + pr_err("cci regulator not ready, retry\n");
ditto.
> + else
> + pr_err("no regulator for cci: %d\n", ret);
How about changing the log like below?
"failed to get regulator for CCI"
> +
> + goto out_free_resources;
> + }
> +
> + platform_set_drvdata(pdev, cci_devdata);
> +
> + /* create cci_devfreq and add to cci device.
> + * governor is voltage_monitor.
> + * governor will get platform_device data to make decision.
> + */
I think that you don't need to explain the kind of governor and how to
decide the next frequency by governor. It is enough to call the devm_devfreq_add_device.
> + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> + &cci_devfreq_profile,
> + "voltage_monitor",
> + NULL);
> +
> + nb->notifier_call = cci_devfreq_regulator_notifier;
> + regulator_register_notifier(cci_devdata->proc_reg,
> + nb);
As I commented, regulator_register_notifier() should be handled in event_handler function.
> +
> + return 0;
> +
> +out_free_resources:
> + if (!IS_ERR(cci_devdata->cci_clk))
> + clk_put(cci_devdata->cci_clk);
> + if (!IS_ERR(cci_devdata->proc_reg))
> + regulator_put(cci_devdata->proc_reg);
> +
> + return ret;
> +}
> +
> +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> + { .compatible = "mediatek,mt8183-cci" },
> + { },
> +};
> +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> +
> +static struct platform_driver cci_devfreq_driver = {
> + .probe = mtk_cci_devfreq_probe,
> + .driver = {
> + .name = "mediatek-cci-devfreq",
> + .of_match_table = mediatek_cci_devfreq_of_match,
> + },
> +};
> +
> +static int __init mtk_cci_devfreq_init(void)
> +{
> + int ret = 0;
> +
> + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> + if (ret) {
> + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> + return ret;
> + }
> +
> + ret = platform_driver_register(&cci_devfreq_driver);
> + if (ret)
> + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> +
> + return ret;
> +}
> +module_init(mtk_cci_devfreq_init)
> +
> +static void __exit mtk_cci_devfreq_exit(void)
> +{
> + int ret = 0;
> +
> + platform_driver_unregister(&cci_devfreq_driver);
> +
> + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> + if (ret)
> + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
> +}
> +module_exit(mtk_cci_devfreq_exit)
> +
> +MODULE_LICENSE("GPL v2");
nitpick. better to move it under MODULE_AUTHOR("")
> +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> +MODULE_AUTHOR("andrew-sh.cheng");
Add full name and email as following:
- MODULE_AUTHOR("Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>"
> +
>
Remove unneeded blank line.
--
Best Regards,
Chanwoo Choi
Samsung Electronics
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread* Re: [PATCH 3/3] devfreq: add mediatek cci devfreq
2019-02-12 4:06 ` Chanwoo Choi
@ 2019-02-16 6:16 ` andrew-sh.cheng
0 siblings, 0 replies; 16+ messages in thread
From: andrew-sh.cheng @ 2019-02-16 6:16 UTC (permalink / raw)
To: Chanwoo Choi
Cc: Mark Rutland, devicetree, srv_heupstream, linux-pm, Viresh Kumar,
Rafael J. Wysocki, linux-kernel, Rob Herring, fan.chen,
Kyungmin Park, MyungJoo Ham, linux-mediatek, Matthias Brugger,
linux-arm-kernel
On Tue, 2019-02-12 at 13:06 +0900, Chanwoo Choi wrote:
> Hi Andrew-sh.Cheng,
>
> On 19. 1. 29. 오후 3:35, Andrew-sh Cheng wrote:
> > From: "Andrew-sh.Cheng" <andrew-sh.cheng@mediatek.com>
> >
> > For big/little cpu cluster architecture,
> > not only CPU frequency, but CCI frequency will also affect performance.
> >
> > Little cores and cci share the same buck in Mediatek mt8183 IC,
> > so we add a CCI devfreq which will get notification when buck voltage
> > is changed, then CCI devfreq can set cci frequency as high as possible.
> >
> > Signed-off-by: Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>
> > ---
> > drivers/devfreq/Kconfig | 9 ++
> > drivers/devfreq/Makefile | 1 +
> > drivers/devfreq/mt8183-cci-devfreq.c | 224 +++++++++++++++++++++++++++++++++++
> > 3 files changed, 234 insertions(+)
> > create mode 100644 drivers/devfreq/mt8183-cci-devfreq.c
> >
> > diff --git a/drivers/devfreq/Kconfig b/drivers/devfreq/Kconfig
> > index 6a172d3..0b14aab 100644
> > --- a/drivers/devfreq/Kconfig
> > +++ b/drivers/devfreq/Kconfig
> > @@ -113,6 +113,15 @@ config ARM_RK3399_DMC_DEVFREQ
> > It sets the frequency for the memory controller and reads the usage counts
> > from hardware.
> >
> > +config ARM_MT8183_CCI_DEVFREQ
> > + tristate "MT8183 CCI DEVFREQ Driver"
> > + depends on ARM_MEDIATEK_CPUFREQ
> > + help
> > + This adds devfreq for cci clock
> > + which is shared the same regulator with cpu cluster.
> > + It can track buck voltage and update a proper cci frequency.
> > + Use notification to get regulator status.
> > +
>
> Move ARM_MT8183_CCI_DEVFREQ under ARM_EXYNOS_BUS_DEVFREQ.
> On later, I'll change the order between ARM_TEGRA_DEVFREQ and ARM_RK3399_DMC_DEVFREQ
> alphabetically. So, you better to keep the order of name.
OK.
>
>
> > source "drivers/devfreq/event/Kconfig"
> >
> > endif # PM_DEVFREQ
> > diff --git a/drivers/devfreq/Makefile b/drivers/devfreq/Makefile
> > index 32b8d4d..25afe8c 100644
> > --- a/drivers/devfreq/Makefile
> > +++ b/drivers/devfreq/Makefile
> > @@ -11,6 +11,7 @@ obj-$(CONFIG_DEVFREQ_GOV_PASSIVE) += governor_passive.o
> > obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ) += exynos-bus.o
> > obj-$(CONFIG_ARM_RK3399_DMC_DEVFREQ) += rk3399_dmc.o
> > obj-$(CONFIG_ARM_TEGRA_DEVFREQ) += tegra-devfreq.o
> > +obj-$(CONFIG_ARM_MT8183_CCI_DEVFREQ) += mt8183-cci-devfreq.o
>
> Move it under obj-$(CONFIG_ARM_EXYNOS_BUS_DEVFREQ).
OK.
>
> >
> > # DEVFREQ Event Drivers
> > obj-$(CONFIG_PM_DEVFREQ_EVENT) += event/
> > diff --git a/drivers/devfreq/mt8183-cci-devfreq.c b/drivers/devfreq/mt8183-cci-devfreq.c
> > new file mode 100644
> > index 0000000..4837892
> > --- /dev/null
> > +++ b/drivers/devfreq/mt8183-cci-devfreq.c
> > @@ -0,0 +1,224 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +//
> > +// Copyright (c) 2018 MediaTek Inc.
>
> Add blank line and change year 2018 -> 2019.
OK.
>
> > +#include <linux/cpu.h>
> > +#include <linux/cpufreq.h>
>
> This driver doesn't use any helper function of cpufreq.h/cpu.h.
> It depends on devfreq/regulator framework. Please remove them.
I will remove it in next patch.
>
> > +#include <linux/devfreq.h>
> > +#include <linux/module.h>
> > +#include <linux/of.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/regulator/consumer.h>
> > +
> > +#include "governor.h"
> > +
> > +struct cci_devfreq_data {
> > + struct devfreq *devfreq;
> > + struct regulator *proc_reg;
> > + struct clk *cci_clk;
> > + unsigned long buck_volt;
> > + int volt_increasing;
> > + struct notifier_block nb;
> > +};
> > +
> > +static int mtk_cci_governor_get_target(struct devfreq *devfreq,
> > + unsigned long *freq)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
> > + int i, opp_count;
> > + struct dev_pm_opp *opp;
> > + unsigned long opp_rate, opp_voltage;
>
> Above three definitions use the tab for indentation. Please use space.
OK.
>
> > +
> > + cci_devdata = dev_get_drvdata(devfreq->dev.parent);
> > +
> > + /* find available frequency */
> > + opp_count = dev_pm_opp_get_opp_count(devfreq->dev.parent);
> > + for (i = 0, opp_rate = ULONG_MAX; i < opp_count; i++, opp_rate--) {
> > + opp = dev_pm_opp_find_freq_floor(devfreq->dev.parent,
> > + &opp_rate);
> > + opp_voltage = dev_pm_opp_get_voltage(opp);
> > + dev_pm_opp_put(opp);
> > +
> > + if (opp_voltage <= cci_devdata->buck_volt)
> > + break;
> > + }
> > + *freq = opp_rate;
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_governor_event_handler(struct devfreq *devfreq,
> > + unsigned int event, void *data)
> > +{
> > + return 0;
> > +}
>
> The devfreq core enables the governor with using DEVFREQ_GOV_START.
> But, this driver just start after proving the device driver. It break
> the previous sequence of governor.
>
> So, if you register/unregister the regulator as following,
> it can keep the existing sequence for governor starting.
>
> case DEVFREQ_GOV_START:
> case DEVFREQ_GOV_RESUME:
> regulator_register_notifier(...);
>
> case DEVFREQ_GOV_STOP:
> case DEVFREQ_GOV_SUSPEND:
> regulator_unregister_notifier(...);
> break;
I will move regulator register/unregister notifier to
governor event handler as above in next patch.
>
> > +
> > +static struct devfreq_governor mtk_cci_devfreq_governor = {
> > + .name = "voltage_monitor",
> > + .get_target_freq = mtk_cci_governor_get_target,
> > + .event_handler = mtk_cci_governor_event_handler,
> > +};
> > +
> > +static int mtk_cci_devfreq_target(struct device *dev, unsigned long *freq,
> > + u32 flags)
> > +{
> > + struct cci_devfreq_data *cci_devdata;
> > +
> > + cci_devdata = dev_get_drvdata(dev);
>
> You better to modify it as following:
>
> struct cci_devfreq_data *cci_devdata = dev_get_drvdata(dev);
>
> if (!cci_devdata)
> return -EINVAL;
I will modify it in next patch
>
> > +
> > + clk_set_rate(cci_devdata->cci_clk, *freq);
> > +
> > + return 0;
> > +}
> > +
> > +static struct devfreq_dev_profile cci_devfreq_profile = {
> > + .polling_ms = 0,
>
> Dont' need to initialize it as zero. just remove this line.
OK.
>
> > + .target = mtk_cci_devfreq_target,
> > +};
> > +
> > +static int cci_devfreq_regulator_notifier(struct notifier_block *nb,
> > + unsigned long val, void *data)
> > +{
> > + struct cci_devfreq_data *cci_devdata =
> > + container_of(nb, struct cci_devfreq_data, nb);
> > +
> > + /* deal with reduce frequency */
> > + if (val & REGULATOR_EVENT_PRE_VOLTAGE_CHANGE) {
> > + struct pre_voltage_change_data *pvc_data = data;
> > +
> > + if (pvc_data->old_uV > pvc_data->min_uV) {
> > + cci_devdata->volt_increasing = 0;
>
> IMO, 'volt_increasing' is not clear. Instead, cci_devdata
> could have the 'previous_voltage' variable instead of volt_increasing.
> And you could use 'previous_voltage' for comparision.
OK, I will remove volt_increasing in next patch.
>
> > + cci_devdata->buck_volt =
> > + (unsigned long)(pvc_data->min_uV);
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + } else {
> > + cci_devdata->volt_increasing = 1;
> > + }
> > + }
> > + /* deal with abort reduce frequency */
> > + if ((val & REGULATOR_EVENT_ABORT_VOLTAGE_CHANGE) ||
>
> if -> else if
>
> At the same time, receives only one notification.
> It is not necessary to check the two if statement always, .
OK, I will separate it in next patch.
>
>
>
> > + /* deal with increase frequency */
> > + ((val & REGULATOR_EVENT_VOLTAGE_CHANGE) &&
> > + cci_devdata->volt_increasing == 1)) {
> > + cci_devdata->buck_volt = (unsigned long)data;
> > + mutex_lock(&cci_devdata->devfreq->lock);
> > + update_devfreq(cci_devdata->devfreq);
> > + mutex_unlock(&cci_devdata->devfreq->lock);
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static int mtk_cci_devfreq_probe(struct platform_device *pdev)
> > +{
> > + struct device *cci_dev = &pdev->dev;
> > + struct cci_devfreq_data *cci_devdata;
> > + struct notifier_block *nb;
> > + int ret;
> > +
> > + dev_pm_opp_of_add_table(&pdev->dev);
>
> Move it after getting clock/regulator. After checking that all resourcess are available,
> and then you better to get the opp table. And check the return value of dev_pm_opp...()
>
OK. I will move it after checking clock/regulator are available, and add
return value checking.
> > +
> > + cci_devdata = devm_kzalloc(cci_dev, sizeof(*cci_devdata), GFP_KERNEL);
> > + if (!cci_devdata)
> > + return -ENOMEM;
> > + nb = &cci_devdata->nb;
> > + cci_devdata->cci_clk = ERR_PTR(-ENODEV);
>
> It should be handled when error happen. It is not good for readability.
I will remove it in next patch.
>
> > + cci_devdata->proc_reg = ERR_PTR(-ENODEV);
>
> ditto.
I will remove it in next patch.
>
> > +
> > + cci_devdata->cci_clk = clk_get(cci_dev, "cci_clock");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->cci_clk);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci clock not ready, retry\n");
>
> You should use 'dev_err' instead of pr_err on this driver.
>
> I think that don't show the error when EPROBE_DEFER. you better to just return.
I will change pr_err to dev_err, and remove the log with EPROBE_DEFER in
next patch.
>
> > + else
> > + pr_err("no clock for cci: %d\n", ret);
>
> How about changing the log like below?
> "failed to get clock for CCI"
OK.
>
> pr_err -> dev_err
>
> > +
> > + goto out_free_resources;
> > + }
> > +
> > + cci_devdata->proc_reg = regulator_get_optional(cci_dev, "proc");
> > + ret = PTR_ERR_OR_ZERO(cci_devdata->proc_reg);
> > + if (ret) {
> > + if (ret == -EPROBE_DEFER)
> > + pr_err("cci regulator not ready, retry\n");
>
> ditto.
>
> > + else
> > + pr_err("no regulator for cci: %d\n", ret);
>
> How about changing the log like below?
> "failed to get regulator for CCI"
OK.
>
> > +
> > + goto out_free_resources;
> > + }
> > +
> > + platform_set_drvdata(pdev, cci_devdata);
> > +
> > + /* create cci_devfreq and add to cci device.
> > + * governor is voltage_monitor.
> > + * governor will get platform_device data to make decision.
> > + */
>
> I think that you don't need to explain the kind of governor and how to
> decide the next frequency by governor. It is enough to call the devm_devfreq_add_device.
I will remove these explanation in next patch.
>
> > + cci_devdata->devfreq = devm_devfreq_add_device(cci_dev,
> > + &cci_devfreq_profile,
> > + "voltage_monitor",
> > + NULL);
> > +
> > + nb->notifier_call = cci_devfreq_regulator_notifier;
> > + regulator_register_notifier(cci_devdata->proc_reg,
> > + nb);
>
> As I commented, regulator_register_notifier() should be handled in event_handler function.
OK. I will move it into governor event handler
>
> > +
> > + return 0;
> > +
> > +out_free_resources:
> > + if (!IS_ERR(cci_devdata->cci_clk))
> > + clk_put(cci_devdata->cci_clk);
> > + if (!IS_ERR(cci_devdata->proc_reg))
> > + regulator_put(cci_devdata->proc_reg);
> > +
> > + return ret;
> > +}
> > +
> > +static const struct of_device_id mediatek_cci_devfreq_of_match[] = {
> > + { .compatible = "mediatek,mt8183-cci" },
> > + { },
> > +};
> > +MODULE_DEVICE_TABLE(of, mediatek_cci_devfreq_of_match);
> > +
> > +static struct platform_driver cci_devfreq_driver = {
> > + .probe = mtk_cci_devfreq_probe,
> > + .driver = {
> > + .name = "mediatek-cci-devfreq",
> > + .of_match_table = mediatek_cci_devfreq_of_match,
> > + },
> > +};
> > +
> > +static int __init mtk_cci_devfreq_init(void)
> > +{
> > + int ret = 0;
> > +
> > + ret = devfreq_add_governor(&mtk_cci_devfreq_governor);
> > + if (ret) {
> > + pr_err("%s: failed to add governor: %d\n", __func__, ret);
> > + return ret;
> > + }
> > +
> > + ret = platform_driver_register(&cci_devfreq_driver);
> > + if (ret)
> > + devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > +
> > + return ret;
> > +}
> > +module_init(mtk_cci_devfreq_init)
> > +
> > +static void __exit mtk_cci_devfreq_exit(void)
> > +{
> > + int ret = 0;
> > +
> > + platform_driver_unregister(&cci_devfreq_driver);
> > +
> > + ret = devfreq_remove_governor(&mtk_cci_devfreq_governor);
> > + if (ret)
> > + pr_err("%s: failed to remove governor: %d\n", __func__, ret);
> > +}
> > +module_exit(mtk_cci_devfreq_exit)
> > +
> > +MODULE_LICENSE("GPL v2");
>
> nitpick. better to move it under MODULE_AUTHOR("")
I will move it after MODULE_AUTHOR :)
>
> > +MODULE_DESCRIPTION("Mediatek CCI devfreq driver");
> > +MODULE_AUTHOR("andrew-sh.cheng");
>
> Add full name and email as following:
> - MODULE_AUTHOR("Andrew-sh.Cheng <andrew-sh.cheng@mediatek.com>"
I will change to full name and email in next patch.
>
>
> > +
> >
>
> Remove unneeded blank line.
OK.
>
>
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 16+ messages in thread